text
stringlengths 254
1.16M
|
|---|
---
title: Circulating microRNAs as potential biomarkers of early vascular damage in vitamin
D deficiency, obese, and diabetic patients
authors:
- Adel B. Elmoselhi
- Mohamed Seif Allah
- Amal Bouzid
- Zeinab Ibrahim
- Thenmozhi Venkatachalam
- Ruqaiyyah Siddiqui
- Naveed Ahmed Khan
- Rifat A. Hamoudi
journal: PLOS ONE
year: 2023
pmcid: PMC10035929
doi: 10.1371/journal.pone.0283608
license: CC BY 4.0
---
# Circulating microRNAs as potential biomarkers of early vascular damage in vitamin D deficiency, obese, and diabetic patients
## Abstract
Vitamin D3 deficiency, obesity, and diabetes mellitus (DM) have been shown to increase the risk of cardiovascular diseases (CVDs). However, the early detection of vascular damage in those patients is still difficult to ascertain. MicroRNAs (miRNAs) are recognized to play a critical role in initiation and pathogenesis of vascular dysfunction. Herein, we aimed to identify circulating miRNA biomarkers of vascular dysfunction as early predictors of CVDs. We have recruited 23 middle-aged Emiratis patients with the following criteria: A healthy control group with vitamin D ≥ 20ng, and BMI < 30 (C1 group = 11 individuals); A vitamin D deficiency (Vit D level ≤ 20 ng) and obese (BMI ≥ 30) group (A1 group = 9 patients); A vitamin D deficiency, obese, plus DM (A2 group = 3 patients). Arterial stiffness via pulse wave velocity (PWV) was measured and the whole transcriptome analysis with qPCR validation for miRNA in plasma samples were tested. PWV relative to age was significantly higher in A1 group 19.4 ± 4.7 m/s and A2 group 18.3 ± 1.3 m/s compared to controls 14.7 ± 2.1 m/s ($p \leq 0.05$). Similar patterns were also observed in the Augmentation pressure (AP) and Alx%. Whole RNA-Sequencing revealed miR-182-5p; miR-199a-5p; miR-193a-5p; and miR-155-5p were differentially over-expressed (logFC > 1.5) in high-risk patients for CVDs vs healthy controls. Collectively, our result indicates that four specific circulating miRNA signature, may be utilized as non-invasive, diagnostic and prognostic biomarkers for early vascular damage in patients suffering from vitamin D deficiency, obesity and DM.
## Introduction
Cardiovascular diseases (CVDs) are the leading cause of mortality and morbidity, as well as an increasing financial burden almost all around the globe. Deaths attributed to CVDs were approximately estimated to be 19 million in 2020 globally, an increase of $18.7\%$ from 2010. In the United States, the estimated direct costs of CVDs have doubled to $226.2 billion in 2017–2018 [1].
Vitamin D3 deficiency has been shown in several studies to increase the risk of CVDs [2]. Moreover, obesity and overweight are considered as high-risk factors for CVDs [3]. Of note, in the United Arab Emirates (UAE) more than $30\%$ of the adult population are suffering from obesity compared to $13\%$ globally [4]. Furthermore, a cross-sectional study of obese and diabetic revealed a vast prevalence of $83\%$ of vitamin D deficiency, in the UAE population [5]. Another study showed that only $2.5\%$ of indoor employees have sufficient vitamin D levels [6]. Furthermore, vitamin D deficiency is well-known to be associated with various CVDs such as hypertension, myocardial infarction, stroke, congestive heart failure, peripheral vascular disease, and atherosclerosis [7].
Vitamin D3 is a steroid fat-soluble vitamin and is the active form of vitamin D mainly produced metabolically in the kidney as a result of skin exposure to sufficient ultraviolet B radiation, and a smaller amount is taken up via nutrition and absorbed from the gastrointestinal tract. Vitamin D3 has pleiotropic functions, including an anti-inflammatory effect that promotes the differentiation of monocytes to macrophages, lymphocytes and dendritic cells. Because these cells play an important role as the first line of defense in the immune system, vitamin D is important in infection control [8]. Vitamin D further enhances the immune system to combat inflammation via the differentiation of active CD4+ T cells and an increase in the inhibitory function of T cells [9]. Activation of local inflammatory cytokines has also been reported and is speculated to be due to a possible molecular mechanism that links vitamin D and endothelial dysfunction, atherogenesis development in coronary arteries as well as other CVDs [10]. Previously, vitamin D was shown to suppress the kappa-light-chain-enhancer nuclear factor of the activated B-cell (NF-κB) pathway. Activation of NF-κB occurs by targeting karyopherin subunit alpha 4 (KPNA4) which mitigates the progression of CVDs. Therefore, vitamin D deficiency enhances the regulation of KPNA4 and consequently increases the activation of NF-κB [11]. NF-κB binds to various κB elements in the nucleus and acts as a transcription factor that stimulates the transcription of inflammatory cytokines such as interleukin IL-6, IL-8 and tumor necrosis factor TNF-α [12].
Furthermore, vitamin D was shown to inhibit calcification in blood vessels via affecting interleukin activities [13]. Therefore, it is logical to hypothesize that vitamin D may improve endothelial dysfunction as its’ deficiency is associated with inflammation. Importantly, several studies in the last few years have been investigating this molecular aspect of endothelial dysfunction [14]. The protective role of Vitamin D in CVDs may also be attributed to the discovery of vitamin D receptors (VDR) in cardiomyocytes, endothelial cells and smooth muscle cells among other cells [15,16]. VDR is an intracellular receptor that binds to the active form of vitamin D, 1,25(OH)2D3. Once bound, it joins the retinoid X receptor and translocates to the nucleus where it binds to VDRE, the regulator site of the element promotor region of DNA, which may promote the synthesis of proteins associated with vitamin D [17]. The pleiotropic effects of vitamin D occur by activating VDR in vascular endothelial cells and cardiomyocytes and regulating the renin-angiotensin system and pancreatic cell activity [18]. Thus, patients with vitamin D deficiency were associated with several cardiovascular and metabolic disorders such as arterial stiffness, endothelial dysfunction, left ventricular hypertrophy, and diabetes mellites among others [19,20].
Nonetheless, the majority of randomized controlled clinical studies did not reveal the beneficial effects of vitamin D on endothelial dysfunction [21,22]. However, most of these studies utilised lower doses of vitamin D supplements, approximately 400–800 IU/day for 16–24 weeks duration. The Institute of Medicine (IOM) recommendation is 600IU/day as a dietary allowance for ages < 70 years and 800IU/day for ages > 70. The basis of these recommendations is mainly related to bone research and healthy conditions, but the optimum dose of vitamin D for other functions is not yet clear, although several researchers have investigated different doses and time frames [23,24]. An earlier study was conducted on nineteen obese adolescents aged 13–18, and vitamin D3 supplementation lasted for 12 weeks and showed increasing 25(OH)D levels with no effect on endothelial function [25]. Another study indicated that vitamin D3 supplementation improves arterial stiffness in a dose-response manner of up to ~4000 IU/day for 16 weeks in overweight African-Americans with vitamin D deficiency [26]. Further studies are still warranted to clarify the optimal dosage and duration of vitamin D supplementation to obtain benefits for vascular dysfunction and CVDs prevention.
Epigenetic mechanisms have been recently shown to regulate aging, genetic and lifestyle factors in determining the risk of vascular diseases [27]. Changes in the gene expression by epigenetic mechanisms include DNA methylation, post-translational histone modifications, and non-coding RNAs (ncRNAs) [28]. Increasing evidence in the last two decades points out the critical role of non-coding RNAs such as microRNA (miRNAs) and lncRNAs in various physiological and pathological conditions [29]. miRNAs are small RNA molecules with a powerful regulatory function for several biological processes since it regulates gene expression at the post-transcriptional level and affects protein translation [30]. It has been shown that miRNAs are present in the blood, plasma, erythrocytes and platelets [31]. These circulating endogenous miRNAs are very stable even under harsh conditions (e.g. extreme temperature) probably due to some protective mechanisms that prevent their degradation [32]. Therefore, circulatory miRNAs are now recognized to have a great potential to be used as biomarkers to predict and assess vascular diseases as well as possible novel therapeutic targets.
Since endothelial dysfunction is considered one of the early signs of CVDs, detection of endothelial dysfunction will have a significant positive impact on preventing CVDs, especially in high-risk individuals that suffer from vitamin D deficiency, obesity and diabetes mellitus (DM). Thus, there is a tremendous need to investigate and establish a prevention approach for early detection of cardiovascular complications, especially in high-risk individuals. Our study examined vascular dysfunction and determined its relevance in circulatory miRNAs and explored the whole transcriptome in adult Emiratis suffering from vitamin D deficiency and obesity, as well as patients with vitamin D deficiency, obesity and diabetes mellitus (DM) compared to a control group of participants with normal vitamin D levels and non-obese.
## Ethics statement
This study was reviewed and approved by the University of Sharjah Research Ethics Committee with Reference number REC 16–11–12, as well as by the ethics committee of University Hospital Sharjah (UHS). The procedures used in this study fully adhered to the tenets of the Declaration of Helsinki. The patients/participants provided their written informed consent to participate in this study.
## Study population
All participants were recruited at University Hospital Sharjah (UHS), between December, 2018 and March, 2020, from the Cardiology and Family medicine clinics. Participants were all Emirati nationals, and the inclusion and exclusion criteria were determined to achieve the main objectives as described in Fig 1. Briefly, we targeted middle-aged Emirati nationals with and without vitamin D (Vit D) deficiency with no previously diagnosed cardiovascular disorders or any other major debilitating diseases, to mainly detect the early changes in their vasculature. In particular, the inclusion criteria consisted of males > 45-year-old and females > 55-year-old (to avoid the changes in hormonal influence in pre-menopausal women); obese with body index mass (BMI) of more than 30 kg/m2; and blood Vit D, 25(OH)D3 status below 20 ng/ml. While the exclusion criteria comprised previously diagnosed with coronary vascular diseases; previously diagnosed with peripheral vascular diseases; any known signs and symptoms for CVDs; chronic debilitating diseases e.g. cancer. Out of the initial target of 90 participants, we managed to recruit 23, and the valid numbers included were: 11 controls (Vit D level > 20 ng, BMI < 30), 9 Vit D deficiency and obese (Vit D level ≤ 20 ng, BMI ≥ 30), and 3 Vit D deficiency, obese and diabetic (Vit D level ≤ 20 ng, BMI ≥ 30) as shown in Table 1. The low number of recruitments, which is considered as a limitation of the study, was due to the strict criteria of inclusion and exclusion, and the fact that the recruitment was conducted in only one university hospital that eventually was crippled by COVID-19 restrictions.
**Fig 1:** *Flow diagram of sample selection, inclusion and exclusion criteria, and baseline measurements.* TABLE_PLACEHOLDER:Table 1
## Clinical investigations
Medical history, physical examination and routine laboratory investigation were obtained for all participants plus pulse wave velocity (PWV) for assessing vascular stiffness. PWV measurement has been described in detail elsewhere [33]. Briefly, PWV measurement was performed using SphygmoCor (version 7.0, Atcor Medical, Sydney, Australia) for individuals in the supine position after a minimum of 10 min rest in a quiet temperature-controlled room. The PWV of the ‘aortic’ segment (aortic PWV) was recorded between the femoral and carotid arteries. The distance from the carotid recording site to the suprasternal notch was subtracted from the distance between the femoral recording site to the suprasternal notch. The quality demands of PWV were followed as suggested by the manufacturer, showing that a quality index ≥ $80\%$ is accepted. The PWV relative to age was calculated as follows: PWV (m/s) x 100 / Age (year old).
## Plasma samples collection
In the EDTA vacutainer tube, 10 ml of whole blood was collected from each participant. 2 ml of whole blood was transferred to a separate labeled tube and immediately placed in a -80°C freezer. Next, we centrifuged the rest of the whole blood at 4200 RPM for 10 minutes. 4 ml of the plasma was transferred to a tube and immediately placed in a -80°C freezer.
## Extraction of mRNA and miRNA from plasma samples
First, mRNA was isolated according to the TRIzol protocol (Invitrogen, USA), as described elsewhere [34]. In brief, in a 1.5ml microfuge tube, 1000 μl Trizol was added to 200 μl of whole blood. Then, the whole was incubated for 20 minutes on ice and 200 μl of chloroform was added. After 10 mins centrifugation, the upper aqueous solution was transferred to a new microfuge tube. To pellet the RNA, 500 μl of $100\%$ isopropanol was added, followed by 10 minutes of incubation at room temperature and 10 minutes of centrifugation. Following discarding the supernatant and washing steps, the pellet was air-dried and re-suspended in 30 μl RNase-free water. Second, miRNA was isolated using miRNeasy Serum/Plasma Kit (Qiagen, Valencia, CA) according to the manufacturer’s protocol.
## Whole RNA library preparation and sequencing
Whole transcriptome sequencing was performed for all 23 patients using RNA-Sequencing with Ion AmpliSeq Whole Transcriptome human gene expression kit (Thermo Fisher Scientific, Massachusetts, USA). Briefly, the purified RNAs were evaluated and quantified. Spectrophotometry was performed on the samples and the ratio of absorbance at 260 nm and 280 nm (A260/A280) was used to assess the purity of RNA. A ratio of around 2.0 indicated pure RNA without protein or organic contamination. cDNA library was constructed using Turbo DNase treated RNA using SuperScipt VILO cDNA synthesis kit (Invitrogen, Life Technologies, CA, USA). Samples were then tagged with unique Ion express barcodes and purified using AMPure XP Reagent (Beckman Coulter, USA). Barcoded libraries were assessed using the Ion Taqman library quantitation kit (Thermo Fisher Scientific) and then pooled equally. The pooled libraries were amplified using emulsion PCR on Ion One Touch2 instruments (OT2) and enriched using Ion One Touch ES as per the manufacturer’s instructions. The constructed cDNA libraries were sequenced using an Ion 540 Chip on an Ion S5 XL Semiconductor sequencer (Life Technologies).
## Transcriptomic data processing
*The* generated RNA-*Seq data* were processed using an in-house pipeline and analyzed using DESeq2 R/Bioconductor package to identify the differentially expressed genes in each comparison of patients with Vit D deficiency and obesity (Vit D level ≤ 20 ng, BMI ≥ 30) (A1 group) and Vit D deficiency, obese and diabetic (Vit D level ≤ 20 ng, BMI ≥ 30) (A2 group) against control healthy individuals with Vit D level > 20 ng and BMI < 30 (C1 group). Differentially overexpressed genes were selected with Fold change (FC) log2(FC) ≥ 1.5.
## Target miRNA identification
In order to identify circulating miRNAs that could serve as biomarkers of vascular dysfunction in high-risk patients with vitamin D deficiency, obesity and MD, a comprehensive bioinformatics analysis was performed using the transcriptomic data. The whole miRNA selection process was described in the flowchart in Fig 2. The up-expressed genes in patients compared to controls were reverse mapped to identify the target miRNAs. First, we were interested in the genes that are uniquely over-expressed in each of the subgroups and with (FC) log2(FC) ≥ 1.5. Genes with low read counts were filtered out; only genes with more than 30 normalized read counts were considered for further analysis. The predicted miRNA targets were identified using TargetScan (https://www.targetscan.org/vert_80/) and miRDB (http://mirdb.org/) databases. The miRNAs were then selected based on; the most prevalent transcript, a high conservation prediction and a context score percentile of > 70. Poorly conserved miRNAs were not considered. All predicted miRNAs were further cross-checked using another microRNA database: miRbase (https://www.mirbase.org/). Then, the Human microRNA Disease Database (HMDD) (http://www.cuilab.cn/hmdd) was searched for all predicted miRNAs. Only miRNAs that showed experiment-supported evidence and disease associations with heart disease, CVDs, obesity, vitamin D deficiency and MD were selected as potential biomarkers.
**Fig 2:** *Flow chart of gene targets pipeline and miRNA selection process with color coding of genes to miRNA (orange), miRNA identification (green), and miRNA validation (purple).*
## Validation of plasma miRNA using Quantitative real-time PCR
For plasma miRNA validation, cDNA was synthesized using microScript microRNA cDNA Synthesis Kit (Norgen, Canada) according to the manufacturer’s protocol. The cDNA concentration and quality were measured by NanoDrop (Thermo Fisher). qRT-PCR was performed in a Qiagen Rotor-Gene qPCR system using miScript SYBR Green PCR Kit (Qiagen). The Ct-value of the miRNA of interest was normalized against the expression of the housekeeping miRNA (miR-19b-3p) from each sample. The relative gene expression was calculated using the 2-ΔΔCt method [35]. The primer sequences are described in Table 2.
**Table 2**
| miRNA name | Forward primer | Reverse Primer |
| --- | --- | --- |
| hsa-miR-193a-5p | TCTTTGCGGGCGAGATG | GAACATGTCTGCGTATCTC |
| hsa-miR-182-5p | GGCAATGGTAGAACTCAC | GAACATGTCTGCGTATCTC |
| hsa-miR-200c-3p | GTCTTACCCAGCAGTGT | GAACATGTCTGCGTATCTC |
| hsa-miR-19b-3p | TGCAGGTTTGCATCCAG | GAACATGTCTGCGTATCTC |
| hsa-miR-199a-5p | CCAGTGTTCAGACTACC | GAACATGTCTGCGTATCTC |
| hsa-miR-155-5p | TGCTAATCGTGATAGGGG | GAACATGTCTGCGTATCTC |
## Statistical analysis
Statistical analysis was performed using GraphPad Prism v.5.0 (GraphPad Software, Inc. CA, USA). The clinical parameters were presented in mean ± standard deviation (SD). The unpaired t-test was used to compare statistical significance between two groups (Group A1 vs. group C1; group A2 vs C1, and group A1 vs. A2). Mann–Whitney U-tests were used to calculate the significance of differentiation of any two patient groups by various miRNAs. miRNA expression was presented as mean ± standard error. Pearson correlation testing was carried out to determine correlations between miRNA expression and various clinical characteristics in all patients of group A1 and group A2. A two-tailed p-value (p) < 0.05 was considered statistically significant for all statistical tests.
## Clinical characteristics of study participants
The clinical parameters of vitamin D deficiency and obese (A1 group) and DM (A2 group) compared to control individuals with vitamin D level > 20 ng and BMI < 30 (C1 group) are presented as means ± SD in Table 1. The age of the A1 group was 53 ± 11.7, while the C1 group was 58.4 ± 7.5. Blood Vit D levels showed a significant difference between A1 vs. C groups ($p \leq 0.0001$) in A1 group was 12.9 ± 4.3 ng, while in C1 group was 37.4 ± 12.6 ng. BMI was significantly different between the two groups ($p \leq 0.0001$); in A1 group the BMI was 33.6 ± 3.1, while in C1 group it was 26.4 ± 2.6. A significant increase in vascular stiffness measured in PWV relative to the participant’s age was observed in the A1 group compared to the control group with 19.4 ± 4.7 vs. 14.7 ± 2.1 m/s ($p \leq 0.05$). A similar trend was also noticed in ALX% (AP/pulse wave pressure) with 27.3 ± 9.5 vs. 24.5 ± 7 respectively. Augmentation pressure (AP), aortic/brachial systolic pressures and aortic/brachial diastolic pressures were also measured (Table 1). Furthermore, A2 group has shown a more significant increase in PWV relative to age compared to C1 group 18.3 ± 1.3 vs. 14.7± 2.1 m/s ($P \leq 0.05$), respectively. A similar trend was also noticed in ALX% between A2 group and C1 group 27 ± 9.5 vs. 24.1 ± 7 respectively, as well as in brachial systolic pressure and augmentation pressure.
## Identification of miRNAs in the plasma of high-risk patients for vascular damage
Based on our transcriptomic data, bioinformatics analysis was conducted comparing the three groups (C1, A1, and A2) to identify the potential miRNAs that are particularly associated with cardiovascular and metabolic disorders. The top miRNAs observed were miR-182-5p; miR-200c-3p; miR-199a-5p; miR-193a-5p; miR-155-5p. Between groups A1 and C1, the target genes of three miRNA -182-5p, miRNA-200-3p, and miRNA-199a-5p were CFL1, KIAA1432, and ZNF415, respectively as shown in Table 3. The target gene CFL1 was a 2.5 log2 fold increase in the A1 group compared to C1. *This* gene is described as Cofilin 1 in Gene Ontology (GO), in which its biological process is associated with actin filament fragmentation and depolymerization as well as regulation by the host viral process as shown in Table 4. *Target* gene KIAA1432 was a 1.66 log2 fold increase in the A1 group. It is described as RICI homolog RAB6A GEF complex partner and associated with the regulation of extracellular matrix constituent secretion and extracellular matrix organization. The target gene ZNF415 was a 2 log2 fold increase in the A1 group compared to the C1 group. *The* gene is described as Zinc finger protein 415 and is associated with regulation of transcription by RNA polymerase II and regulation of transcription, DNA template as shown in Table 4. For miRNA 193-5p, the target gene MTRNR2L8 was shown to increase 1.76 log2 fold in group A2 compared to the C1 group. *This* gene was described as MT-RNR2-like 8 and linked to the regulation of the execution phase of apoptosis and regulation of signaling receptor activity as shown in Table 4. In miRNA-155-5p, the target gene C9orf78 was shown to increase 1.53 log2 fold in group A1 compared to group A2. *This* gene was described as chromosome 9 open reading frame 78 and linked to the regulation of mRNA splicing via spliceosome, mRNA processing, and RNA splicing (Table 4).
## Validation of the down expression profiling of the top miRNAs
Validation of the top selected miRNAs was confirmed using qRT- PCR with the reference mir-19 as a normalization control since it has been reported to be stably expressed in different tissues and cell types, (Fig 3) [36,37]. We standardized the quantitative cycle (Cq) values based on the standardized Cq values of each miRNA. The lower the Cq value the higher the expression of the indicated miRNA in the plasma of the subject.
**Fig 3:** *qRT-PCR validation for individual miRNA: Expression levels of miR-182-5p, miR-200c-3p, miR-199a-5p (A-C) are between Vit D deficiency and obese patients (A1 group) compared to control individuals (C1 group); miR-193a-5p expression level (D) is between Vit D deficiency, obese, and diabetes mellitus (A2 group) and control individuals (C1 group); and miR155-5p expression level (E) is between A1 group and A2 group.*
## MiRNA 182-5p expression
Validation by qPCR has confirmed the down expression of miRNA 182-5p by showing a lower Cq in the A1 group compared to the C1 group. The expression of miRNA 182-5p was significantly ($p \leq 0.0001$) down expressed in vitamin D deficiency and obese patients (A1 group) by 1.524 folds compared to the control patients (C1 group) with normal vitamin D level and non-obese (Fig 4).
**Fig 4:** *miRNA down-expression validation by qPCR in affected patients compared to healthy control individuals.The gene expression comparison was between A1 and C1 groups for miR-182-5p, miR-200c-3p, miR-199a-5p; between A2 and C1 groups for miR-193p; and between A1 and A2 groups for miR155-5p expression.*
## MiRNA 200c-3p expression
The validation by qPCR showed lower expression in the A1 group compared to the C1 group as indicated in a higher quantitation cycle in the A1 group relative to the C1 group. The expression of miRNA 200c-3p was down expressed, but not statistically significant ($p \leq 0.5$), in vitamin D deficiency and obese patients (A1 group) by 15.2 folds compared to control patients (C1 group) with normal vitamin D level and non-obese (Fig 4).
## MiRNA 199a-5p expression
qPCR validation confirmed the down expression of miRNA 199a-5p in the A1 group compared to the C1 group by a significantly lower Cq in the A1 group compared to the C1 group as shown in Fig 4. The expression of miRNA 199a-5p was significantly lower ($p \leq 0.0001$) in vitamin D deficiency and obese patients (A1 group) by 1.992 folds compared to the control patients (C1 group) with normal vitamin D levels and non-obese.
## MiRNA 155-5p expression
The qPCR validation confirms the down expression of miRNA 155-5p in the A1 group compared to the A2 group by a lower quantitation cycle in the A2 group compared to the A1 group as shown in Fig 4. The miRNA 155-5p was very significantly down expressed ($p \leq 0.0003$) in vitamin D deficiency and obese patients (A1 group) by 87.674 folds compared to the patients suffering from vitamin D deficiency, obesity plus diabetes mellitus (A2 group).
## MiRNA 193-5p expression
The qPCR validation confirmed lower expression of miRNA 193-5p in the A2 group compared to the C1 group by a higher quantitation cycle in the A2 group compared to the C1 group (Fig 4). The expression of miRNA 193-5p was significantly lower ($p \leq 0.0002$) in vitamin D deficiency, obese patients plus diabetes mellitus (A2 group) by 70.763 folds compared to the control patients (C1 group).
## Circulating miRNA biomarkers correlate to vascular damage in high-risk patients
The matrix correlation between measured clinical parameters such as BMI, brachial and aortic systololic and diastolic blood pressure, Augmentation pressure (AP), Alx%, and PWV with circulating miRNAs, in particular miR-182-5p, miR-199a-5p, miR-155-5p, miR-193a-5p, and miR-155-5p as shown in Fig 5. Using Pearson’s correlation in data analysis; of particular interest the higher correction between PWV and Alx% were miR-182-5p, miR-199a-5p, and miR-155-5p. Also, the higher correlation of miR-182-5p, miR-193a-5p and miR-155-5-5p with brachial and aortic blood pressure.
**Fig 5:** *Correlation matrix among circulating miRNA expression and various clinical parameters in (A) patients with vitamin D deficiency and obesity and (B) patients with vitamin D deficiency, obesity, and diabetes mellitus.Analysis was performed by Pearson’s Correlation, p ≤ 0.05. BMI: Body Mass Index; VitD: Vitamin D level (ng/ml); BSP: Brachial systolic pressure (mmHg), BDP: Brachial diastolic pressure (mmHg), ASP: Aortic systolic pressure (mmHg), ADP: Aortic diastolic pressure (mmHg), AP: Augmentation pressure (mmHg), Alx%: AP/pulse wave pressure; PWV: Pulse wave pressure (m/s).*
## Discussion
Vascular dysfunction as an early sign of CVDs has been associated with vitamin D deficiency. However, its accurate and early diagnosis is difficult to achieve. Measuring circulatory miRNAs as a non-invasive diagnostic and prognostic marker for early vascular dysfunction may be a promising approach. Here, we report an increase in vascular stiffness in middle-aged obese Emirati patients suffering from vitamin D deficiency compared to control individuals with a normal level of vitamin D and non-obese, as well as to patients suffering from vitamin D deficiency, obesity, plus diabetes mellitus (DM). Furthermore, we have identified two miRNAs (miR-182-5p and miR-199a-5p) as promising biomarkers for vascular dysfunction in vitamin D deficiency and obese patients. Moreover, two other miRNAs (miR-193a-5p; miR-155-5p) have resulted in vascular dysfunction in patients suffering from vitamin D deficiency, obesity, and DM (Fig 6).
**Fig 6:** *Schematic summary for the novel circulatory miRNAs in patients suffering from vitamin D (Vit.D), obesity (high BMI), and diabetes mellitus (DM) which lead to early vascular dysfunction via expression of specific target genes with various biological processes via Gene Ontology (GO).*
Arterial stiffness has been reported to be an independent predictor of CVDs and mortality. Although arterial stiffness increases with aging, its progression increases further in patients suffering from vitamin D deficiency [38]. We have not observed statistically significant changes among our groups in both brachial and aortic pressures, most likely because of the small sample size. Nonetheless, the aortic systolic pressure was consistently lower than brachial systolic pressure as previously reported, which is more indicative of the actual afterload the heart has to encounter [38].
miR-182-5p has been reported in several studies to be associated with heart failure as prognostic biomarkers. In particular, miR-182-5p was found with miR-200a and miR-568 to be inversely correlated with Left ventricular mass index (LVMI) [39]. The target gene for miR-182-5p is CFL1 in Homo sapiens (Human) for Cofilin-1 protein (https://www.uniprot.org/uniprot/P23528). The function of the CFL1 gene includes the regulation of cell morphology and cytoskeletal organization in epithelial cells, as well as chemokine receptor ACKR2 up-regulation from the endosomal compartment to the cell membrane [40,41].
Downregulation of miR-199a-5p was shown in myocardial tissue of patients following coronary artery bypass graft surgery. It was also linked with an elevated level of cardioprotective protein Sirtuin 1 (SIRT 1) and in major adverse cardiac and cerebrovascular events at 3-year follow-ups [42]. Furthermore, alteration of miR-199a-5p has been associated with fibrosis and hypertrophic growth in diabetic cardiomyopathy, as well as with obesity and heart failure. The target gene of miR-199a-5p is ZNF415 which is the coding gene of the Zinc Finger Protein. It was reported to be differently expressed in vascular tumor-derived endothelium [43].
Comparing patients suffering from vitamin D deficiency, obesity and type 2 diabetes mellitus to control patients, significantly lower expression of miR-193a-5p was observed in our study. miR-193a-5p was also shown to be upregulated in heart hypertrophy for both in-vitro and in-vivo mice models induced by Ang II [44]. More importantly, it was associated in several studies with obesity, diabetes mellitus, as well as normoglycemic and hyperglycemic-affected adipogenesis [45]. In particular, mir-193a-5p was shown to fine-tune the adverse events of the MAPK signaling pathway to fight against obesity [46]. The miR-193a-5p target gene is MTRNR2L8. Recently, it has been reported that DNA methylation of MTRNR2L8 may play an important part in large-artery atherosclerotic stroke. Thus, it was suggested as a potential therapeutic target and diagnostic biomarker for stroke [47]. Further comparison between vitamin D and obese patients and vitamin D deficiency, obese and diabetic patients was a significantly higher expression of miR-155-5p in group A2 vs group A1. Changes in miR-155-5p were reported in several studies in both animal and human models related to diabetes mellitus and cardiovascular pathogeneses. A recent study has reported that miR-155-5p upregulation affects myocardial insulin resistance via mTOR signaling in chronic alcohol-drinking rats [48]. Additionally, miR-155 deletion in female mice increases adipogenic, insulin sensitivity, and energy uncoupling machinery, while limiting inflammation in white adipose tissue, which together could restrict high-fat diet-induced fat accumulation. These results identify miR-155 as a target candidate for improving obesity resistance [49]. Another study on humans has reported vitamin D-mediated attenuation of miR-155 in macrophages infected with dengue virus, which was implicated due to the cytokine response [50]. The target gene of miR-155-5p is C9orf78, which has been associated with several human cancer types and consider as a good prognostic maker for patient survival (https://www.proteinatlas.org/humanproteome/pathology).
On the whole, the new miRNAs we have identified here are consistent with previous studies in their association with cardiovascular and metabolic effects. The sample size was a limitation in our study because of the specific inclusion and exclusion criteria as detailed in the method section. However, we were able to overcome this limitation by using different bioinformatics tools to select and validate potential differentially expressed miRNAs. Further studies are warranted to confirm and refine the early detection of vascular dysfunction, especially in high-risk individuals with vitamin D deficiency, obesity and diabetes mellitus among the Emirati population. Such studies are essential in order to prevent and mitigate more serious consequences of cardiovascular and cerebrovascular diseases.
## Conclusion
Vascular stiffness is associated with vitamin D deficiency, obesity, and diabetes mellitus. Early detection of vascular dysfunction could prevent CVD, especially in high-risk individuals. The results of our study confirm the link between vitamin D deficiency, obesity, DM, and vascular dysfunction, as well as identifying novel mi-RNAs biomarkers for early diagnosis of vascular damage, especially for high-risk individuals. Herein we reported novel circulatory miRNAs biomarkers that are likely to provide specific non-invasive early diagnostic and prognostic tools for assessing vascular health. These comprise miR-182-5p and miR-199a-5p for vitamin D deficiency and obese patients, while miR-193a-5p and miR-155-5p for vitamin D, obese and diabetic patients. Considering the tremendous benefits of these novel biomarkers to prevent vascular complications in vitamin D deficiency, obese and diabetic patients, further larger studies are warranted. Finally, further research is needed to determine the precise role of vitamin D in CVD, in particular within the Emirati population.
## References
1. Tsao CW, Aday AW, Almarzooq ZI, Alonso A, Beaton AZ, Bittencourt MS. **Heart Disease and Stroke Statistics-2022 Update: A Report From the American Heart Association**. *Circulation* (2022) **145** e153-e639. DOI: 10.1161/CIR.0000000000001052
2. Autier P, Boniol M, Pizot C, Mullie P. **Vitamin D status and ill health: a systematic review**. *Lancet Diabetes Endocrinol* (2014) **2** 76-89. DOI: 10.1016/S2213-8587(13)70165-7
3. Maksimovic M, Vlajinac H, Radak D, Marinkovic J, Maksimovic J, Jorga J. **Association of overweight and obesity with cardiovascular risk factors in patients with atherosclerotic diseases**. *J Med Biochem* (2020) **39** 215-23. DOI: 10.2478/jomb-2019-0027
4. Sadiya A, Ahmed SM, Skaria S, Abusnana S. **Vitamin D status and its relationship with metabolic markers in persons with obesity and type 2 diabetes in the UAE: a cross-sectional study**. *J Diabetes Res* (2014) **2014** 869307. DOI: 10.1155/2014/869307
5. Al-Anouti F A-AS, Thomas J, Abdel-Wareth L, Devkaran S. **Sun avoidance among indoor employees leading to vitamin D deficiency and depression in the United Arab Emirates**. *International Journal of Medicine and Medical Sciences* (2013) **5** 503-509
6. McGreevy C WD. **New insights about vitamin D and cardiovascular disease: a narrative review**. *Ann Intern Med* (2011) **155** 20. DOI: 10.7326/0003-4819-155-12-201112200-00004
7. Lai YH, Fang TC. **The pleiotropic effect of vitamin d.**. *ISRN Nephrol* (2013) **2013** 898125. DOI: 10.5402/2013/898125
8. Liu M, Lee MH, Cohen M, Bommakanti M, Freedman LP. **Transcriptional activation of the Cdk inhibitor p21 by vitamin D3 leads to the induced differentiation of the myelomonocytic cell line U937**. *Genes Dev* (1996) **10** 142-53. DOI: 10.1101/gad.10.2.142
9. Chen S, Swier VJ, Boosani CS, Radwan MM, Agrawal DK. **Vitamin D Deficiency Accelerates Coronary Artery Disease Progression in Swine**. *Arterioscler Thromb Vasc Biol* (2016) **36** 1651-9. DOI: 10.1161/ATVBAHA.116.307586
10. Fatkhullina AR, Peshkova IO, Koltsova EK. **The Role of Cytokines in the Development of Atherosclerosis**. *Biochemistry (Mosc)* (2016) **81** 1358-70. DOI: 10.1134/S0006297916110134
11. Gholami F, Moradi G, Zareei B, Rasouli MA, Nikkhoo B, Roshani D. **The association between circulating 25-hydroxyvitamin D and cardiovascular diseases: a meta-analysis of prospective cohort studies**. *BMC Cardiovasc Disord* (2019) **19** 248. DOI: 10.1186/s12872-019-1236-7
12. Schurgers LJ, Dissel PE, Spronk HM, Soute BA, Dhore CR, Cleutjens JP. **Role of vitamin K and vitamin K-dependent proteins in vascular calcification**. *Z Kardiol* (2001) **3** 90. DOI: 10.1007/s003920170043
13. van den Oever IA, Raterman HG, Nurmohamed MT, Simsek S. **Endothelial dysfunction, inflammation, and apoptosis in diabetes mellitus.**. *Mediators Inflamm.* (2010) **2010** 792393. DOI: 10.1155/2010/792393
14. Lee JH O, ’Keefe JH, Bell D, Hensrud DD, Holick MF. **Vitamin D deficiency an important, common, and easily treatable cardiovascular risk factor?**. *J Am Coll Cardiol* (2008) **52** 1949-56. DOI: 10.1016/j.jacc.2008.08.050
15. Trehan N, Afonso L, Levine DL, Levy PD. **Vitamin D Deficiency, Supplementation, and Cardiovascular Health.**. *Crit Pathw Cardiol.* (2017) **16** 109-18. DOI: 10.1097/HPC.0000000000000122
16. Baeke F, Takiishi T, Korf H, Gysemans C, Mathieu C. **Vitamin D: modulator of the immune system**. *Curr Opin Pharmacol* (2010) **10** 482-96. DOI: 10.1016/j.coph.2010.04.001
17. Al Mheid I, Quyyumi AA. **Vitamin D and Cardiovascular Disease: Controversy Unresolved**. *J Am Coll Cardiol* (2017) **70** 89-100. DOI: 10.1016/j.jacc.2017.05.031
18. Weishaar RE, Simpson RU. **Involvement of vitamin D3 with cardiovascular function. II. Direct and indirect effects**. *Am J Physiol* (1987) **253** E675-83. DOI: 10.1152/ajpendo.1987.253.6.E675
19. Mancuso P, Rahman A, Hershey SD, Dandu L, Nibbelink KA, Simpson RU. **1,25-Dihydroxyvitamin-D3 treatment reduces cardiac hypertrophy and left ventricular diameter in spontaneously hypertensive heart failure-prone (cp/+) rats independent of changes in serum leptin.**. *J Cardiovasc Pharmacol* (2008) **51** 559-64. DOI: 10.1097/FJC.0b013e3181761906
20. Alyami A, Soares MJ, Sherriff JL, Mamo JC. **Vitamin D & endothelial function**. *Indian J Med Res* (2014) **140** 483-90. PMID: 25488441
21. Hussin AM, Ashor AW, Schoenmakers I, Hill T, Mathers JC, Siervo M. **Effects of vitamin D supplementation on endothelial function: a systematic review and meta-analysis of randomised clinical trials**. *Eur J Nutr* (2017) **56** 1095-104. DOI: 10.1007/s00394-016-1159-3
22. Kienreich K, Tomaschitz A, Verheyen N, Pieber T, Gaksch M, Grübler MR. **Vitamin D and cardiovascular disease.**. *Nutrients.* (2013) **5** 3005-21. DOI: 10.3390/nu5083005
23. Zittermann A, Gummert JF. **Nonclassical vitamin D action.**. *Nutrients.* (2010) **2** 408-25. DOI: 10.3390/nu2040408
24. Zittermann A.. **Vitamin D in preventive medicine: are we ignoring the evidence?**. *Br J Nutr* (2003) **89** 552-72. DOI: 10.1079/BJN2003837
25. Javed A, Kullo IJ, Balagopal PB, Kumar S. **Effect of vitamin D3 treatment on endothelial function in obese adolescents.**. *Pediatr Obes.* (2016) **11** 279-84. DOI: 10.1111/ijpo.12059
26. Raed A, Bhagatwala J, Zhu H, Pollock NK, Parikh SJ, Huang Y. **Dose responses of vitamin D3 supplementation on arterial stiffness in overweight African Americans with vitamin D deficiency: A placebo controlled randomized trial.**. *PLoS One* (2017) **12** e0188424. DOI: 10.1371/journal.pone.0188424
27. Zarzour A, Kim HW, Weintraub NL. **Epigenetic Regulation of Vascular Diseases**. *Arterioscler Thromb Vasc Biol* (2019) **39** 984-90. DOI: 10.1161/ATVBAHA.119.312193
28. Khyzha N, Alizada A, Wilson MD, Fish JE. **Epigenetics of Atherosclerosis: Emerging Mechanisms and Methods**. *Trends Mol Med* (2017) **23** 332-47. DOI: 10.1016/j.molmed.2017.02.004
29. Gangwar RS, Rajagopalan S, Natarajan R, Deiuliis JA. **Noncoding RNAs in Cardiovascular Disease: Pathological Relevance and Emerging Role as Biomarkers and Therapeutics**. *Am J Hypertens* (2018) **31** 150-65. DOI: 10.1093/ajh/hpx197
30. Ambros V.. **The functions of animal microRNAs**. *Nature* (2004) **431** 350-5. DOI: 10.1038/nature02871
31. Mitchell PS, Parkin RK, Kroh EM, Fritz BR, Wyman SK, Pogosova-Agadjanyan EL. **Circulating microRNAs as stable blood-based markers for cancer detection**. *Proc Natl Acad Sci U S A* (2008) **105** 10513-8. DOI: 10.1073/pnas.0804549105
32. van Rooij E, Marshall WS, Olson EN. **Toward microRNA-based therapeutics for heart disease: the sense in antisense**. *Circ Res* (2008) **103** 919-28. DOI: 10.1161/CIRCRESAHA.108.183426
33. Roth R, Madhani HD, Garcia JF. **Total RNA Isolation and Quantification of Specific RNAs in Fission Yeast.**. *Methods Mol Biol* (2018) **1721** 63-72. DOI: 10.1007/978-1-4939-7546-4_6
34. Schmittgen TD, Livak KJ. **Analyzing real-time PCR data by the comparative C(T) method.**. *Nat Protoc.* (2008) **3** 1101-8. DOI: 10.1038/nprot.2008.73
35. Laurent S, Cockcroft J, Van Bortel L, Boutouyrie P, Giannattasio C, Hayoz D. **Expert consensus document on arterial stiffness: methodological issues and clinical applications**. *Eur Heart J* (2006) **27** 2588-605. DOI: 10.1093/eurheartj/ehl254
36. Song X, Li W, Shen P, Bai B, Cao LL. **miR-19 Is a Potential Clinical Biomarker for Gastrointestinal Malignancy: A Systematic Review and Meta-analysis**. *Biomed Res Int* (2020) **2020** 2810150. DOI: 10.1155/2020/2810150
37. Duca RB, Massillo C, Dalton GN, Farré PL, Graña KD, Gardner K. **MiR-19b-3p and miR-101-3p as potential biomarkers for prostate cancer diagnosis and prognosis**. *Am J Cancer Res* (2021) **11** 2802-2820. PMID: 34249429
38. Gepner AD, Ramamurthy R, Krueger DC, Korcarz CE, Binkley N, Stein JH. **A prospective randomized controlled trial of the effects of vitamin D supplementation on cardiovascular disease risk.**. *PLoS One* (2012) **7** e36617. DOI: 10.1371/journal.pone.0036617
39. Ikitimur B, Cakmak HA, Coskunpinar E, Barman HA, Vural VA. **The relationship between circulating microRNAs and left ventricular mass in symptomatic heart failure patients with systolic dysfunction**. *Kardiol Pol* (2015) **73** 740-6. DOI: 10.5603/KP.a2015.0082
40. Bai SW, Herrera-Abreu MT, Rohn JL, Racine V, Tajadura V, Suryavanshi N. **Identification and characterization of a set of conserved and new regulators of cytoskeletal organization, cell morphology and migration**. *BMC Biol* (2011) **9** 54. DOI: 10.1186/1741-7007-9-54
41. Borroni EM, Cancellieri C, Vacchini A, Benureau Y, Lagane B, Bachelerie F. **β-arrestin-dependent activation of the cofilin pathway is required for the scavenging activity of the atypical chemokine receptor D6.**. *Sci Signal.* (2013) **6** ra30.1-11
42. Yamac AH, Huyut MA, Yilmaz E, Celikkale I, Bacaksiz A, Demir Y. **MicroRNA 199a Is Downregulated in Patients After Coronary Artery Bypass Graft Surgery and Is Associated with Increased Levels of Sirtuin 1 (SIRT 1) Protein and Major Adverse Cardiovascular Events at 3-Year Follow-Up.**. *Med Sci Monit* (2018) **24** 6245-54. DOI: 10.12659/MSM.912065
43. Ghilardi C, Chiorino G, Dossi R, Nagy Z, Giavazzi R, Bani M. **Identification of novel vascular markers through gene expression profiling of tumor-derived endothelium**. *BMC Genomics* (2008) **9** 201. DOI: 10.1186/1471-2164-9-201
44. Wang L, Qin D, Shi H, Zhang Y, Li H, Han Q. **MiR-195-5p Promotes Cardiomyocyte Hypertrophy by Targeting MFN2 and FBXW7**. *Biomed Res Int* (2019) **2019** 1580982. DOI: 10.1155/2019/1580982
45. Kim H, Bae YU, Lee H, Kim H, Jeon JS, Noh H. **Effect of diabetes on exosomal miRNA profile in patients with obesity**. *BMJ Open Diabetes Res Care* (2020) **8**. DOI: 10.1136/bmjdrc-2020-001403
46. Youssef EM, Elfiky AM, BanglySoliman N, Abu-Shahba MM. **Expression profiling and analysis of some miRNAs in subcutaneous white adipose tissue during development of obesity**. *Genes Nutr* (2020) **15** 8. DOI: 10.1186/s12263-020-00666-0
47. Shen Y, Peng C, Bai Q, Ding Y, Yi X, Du H. **Epigenome-Wide Association Study Indicates Hypomethylation of MTRNR2L8 in Large-Artery Atherosclerosis Stroke**. *Stroke* (2019) **50** 1330-8. DOI: 10.1161/STROKEAHA.118.023436
48. Li Z, Hu Z, Meng Y, Xu H, Wei Y, Shen D. **miR-155-5p upregulation ameliorates myocardial insulin resistance via mTOR signaling in chronic alcohol drinking rats**. *PeerJ* (2021) **9** e10920. DOI: 10.7717/peerj.10920
49. Gaudet AD, Fonken LK, Gushchina LV, Aubrecht TG, Maurya SK, Periasamy M. **miR-155 Deletion in Female Mice Prevents Diet-Induced Obesity.**. *Sci Rep.* (2016) **6** 22862. DOI: 10.1038/srep22862
50. Arboleda JF, Fernandez GJ, Urcuqui-Inchima S. **Vitamin D-mediated attenuation of miR-155 in human macrophages infected with dengue virus: Implications for the cytokine response**. *Infect Genet Evol* (2019) **69** 12-21. DOI: 10.1016/j.meegid.2018.12.033
|
---
title: Characterization of methylation profiles in spontaneous preterm birth placental
villous tissue
authors:
- Heather M. Brockway
- Samantha L. Wilson
- Suhas G. Kallapur
- Catalin S. Buhimschi
- Louis J. Muglia
- Helen N. Jones
journal: PLOS ONE
year: 2023
pmcid: PMC10035933
doi: 10.1371/journal.pone.0279991
license: CC BY 4.0
---
# Characterization of methylation profiles in spontaneous preterm birth placental villous tissue
## Abstract
Preterm birth is a global public health crisis which results in significant neonatal and maternal mortality. Yet little is known regarding the molecular mechanisms of idiopathic spontaneous preterm birth, and we have few diagnostic markers for adequate assessment of placental development and function. Previous studies of placental pathology and our transcriptomics studies suggest a role for placental maturity in idiopathic spontaneous preterm birth. It is known that placental DNA methylation changes over gestation. We hypothesized that if placental hypermaturity is present in our samples, we would observe a unique idiopathic spontaneous preterm birth DNA methylation profile potentially driving the gene expression differences we previously identified in our placental samples. Our results indicate the idiopathic spontaneous preterm birth DNA methylation pattern mimics the term birth methylation pattern suggesting hypermaturity. Only seven significant differentially methylated regions fitting the idiopathic spontaneous preterm birth specific (relative to the controls) profile were identified, indicating unusually high similarity in DNA methylation between idiopathic spontaneous preterm birth and term birth samples. We identified an additional 1,718 significantly methylated regions in our gestational age matched controls where the idiopathic spontaneous preterm birth DNA methylation pattern mimics the term birth methylation pattern, again indicating a striking level of similarity between the idiopathic spontaneous preterm birth and term birth samples. Pathway analysis of these regions revealed differences in genes within the WNT and Cadherin signaling pathways, both of which are essential in placental development and maturation. Taken together, these data demonstrate that the idiopathic spontaneous preterm birth samples display a hypermature methylation signature than expected given their respective gestational age which likely impacts birth timing.
## Introduction
Preterm birth (PTB), defined as delivery at less than 37 weeks of gestation is the leading cause of neonatal mortality worldwide. Prematurity affects an average of $10\%$ of infants born in the United States with rates increasing and costs approximately $26.2 billion dollars a year (annual societal cost including medical, educational, and lost productivity) [1, 2]. The majority ($50\%$) of preterm births are idiopathic and spontaneous (isPTB), rather than being medically indicated (e.g., pre-eclampsia). Risk factors include but are not limited to genetic ancestry, fetal sex, environmental exposures, and economic disparities [3]. Complications include developmental delays, growth restriction, chronic respiratory problems as well as adult sequelae [3]. Studies into the etiology of preterm birth have implicated a role for the placenta, a central component of the maternal-fetal interface, which has a vital role in pregnancy maintenance, communication, and birth timing as well as fetal growth and development [4]. As such, proper placental development, maturation, and function are essential for a successful pregnancy outcome and life-time offspring health. Each of these processes is an intricate balance of molecular interactions that are not fully understood even in healthy, normal, term pregnancies.
Placental maturation is accompanied by a marked increase in placental surface area due to placental remodeling initiated between 20–24 weeks gestation and continuing throughout the remainder of gestation which accommodates exponential fetal growth across the second half of gestation [4]. Under normal physiological conditions, placental maturation is recognized by specific histological hallmarks including increased quantities of terminal villi (<80 microns in diameter), syncytial nuclear aggregates (SNAs, 10+ syncytial nuclei being extruded from the syncytiotrophoblast), and formation of the vasculosyncytial membranes (VSM) which when observed in significant quantities prior to 37 weeks, signify placentas with advanced villous maturation (AVM) [5, 6]. Histological studies of pathological placentas indicate AVM occurs in 50–$60\%$ of isPTB and medically indicated preterm births [7, 8]. This indicates a potential developmental disconnect between placental maturation and the corresponding fetal maturation. In infection associated preterm births, AVM was observed in less than $20\%$ of pathologic placentas [7, 8]. These studies indicate multiple morphological endotypes exist, underlying the classical clinical PTB phenotypes, especially those of spontaneous PTB which are based on gestational age and simply defined as early, moderate, and late [9]. The identification of these morphological endotypes further highlights the heterogeneity confounding the identification of PTB etiology and potential diagnostic biomarkers.
Multiple levels of heterogeneity confound elucidation of molecular mechanisms involved in PTB, from inconsistent sampling of interface tissues to the numerous cell types within those tissues to individual differences within larger populations [10–13]. However, traditional epidemiological studies have not accounted for this morphological, molecular, and physiological heterogeneity. Instead, the use of extensive covariate data to attempt to overcome population-based heterogeneity has resulted in statistical overfit of models to specific datasets and loss of reproducibility and generalizability of biological inference across datasets [14, 15]. This has led to a dearth of robust biomarkers capable of assessing spontaneous PTB risk and managing real-time clinical care. Our approach differs from the population based epidemiological approaches in that we focus molecular profiling in smaller, prescreened datasets with combined with select harmonizable covariate data that can be obtained for any dataset.
We have previously identified transcriptomic profiles of AVM in a small cohort using clinically phenotyped placental villous samples from spontaneous PTB births, including isPTB and infection associated births, between 29 and 36 weeks and normal term births (TB) between 38 and 42 weeks [16]. In our datasets, we define infection associated preterm births as acute histologic chorioamnionitis (AHC) which have been identified via histological assessment of inflamed fetal membranes or molecular assessment [16]. Given the importance of DNA methylation (DNAm) to placental development and maturation [17–20], we hypothesized the gene expression differences we observed in our transcriptome data could be due to changes in DNAm at CpG islands between the birth types. Therefore, we sought to identify specific DNAm profiles of placental maturation associated with our transcriptional profiles of maturation.
## Study population
This study was approved by the Cincinnati Children’s Hospital Medical Center institutional review board (#IRB 2013–2243, 2015–8030, 2016–2033). De-identified TB ($$n = 8$$), isPTB ($$n = 11$$), and AHC ($$n = 8$$) placental villous samples along with appropriate covariate information were obtained from the following sources: The Global Alliance to Prevent Prematurity and Stillbirth (GAPPS) in Seattle Washington USA, the Research Centre for Women’s and Infant’s Health (RCWIH) at Mt Sinai Hospital Toronto Canada, the University of Cincinnati Medical Center (UCMC) and The Ohio State University College of Medicine, Department of Obstetrics & Gynecology. Samples contained only placental villous tissue originating from the fetus. Inclusion criteria included: maternal age 18 years or older, singleton pregnancies with either normal term delivery (38–42 weeks’ gestation) or preterm delivery (29–36 weeks’ gestation) without additional complications including maternal disease, fetal disease or genetic disorders.
## Statistical analyses
Cohort data were analyzed in Prism v8 (GraphPad). Data were evaluated for normality and non-parametric tests applied as appropriate. Parametric data are expressed as median and range and were analyzed by one-way ANOVA with Tukey’s Multiple Corrections testing across all group means. Categorical data were analyzed using Fisher’s Exact Test. These analyses were run independently of those included in [16].
## Intersection of transcriptomic candidate genes and CpG islands
Using the table function of the UCSC Genome Browser build hg38, we conducted a batch query using the 340 candidate genes from our previous transcriptome study [16]. Using these genes as identifiers, we created an intersection with the CpG Island Track [21]. This created an output table with gene names, genomic positions, and overlapping CpG islands including introns, exons, and regulatory regions 5’ and 3’. We then calculated the percentage of gene’s protein coding regions that overlapped with CpG islands for initial assessment of potential impact on transcription. We then utilized this table in subsequent analyses to determine the location of DMRs in relation to gene structure and regulation.
## DNA methylome generation
DNA was isolated from homogenized, snap frozen placental villous samples using the DNeasy Kit (Qiagen). DNA quantity and quality was assessed using Qubit 4 Fluorometer (Invitrogen) and Nanodrop Spectrophotometer (Thermo Fisher Scientific). A minimum of 500ng was submitted to the University of Minnesota Genomics Center and the University of Cincinnati Genomics, Epigenomics and Sequencing Core where DNA quantity and quality assessment were performed on a Bioanalyzer (Aligent), bisulfite conversion, and methylome generation conducted on the Illumina Methylation EPIC Bead Chip.
## DNA methylation array data processing
Methylation data processing and analyses were based on a previously developed workflow [22]. All packages are available within Bioconductor [23] and all package scripts were run in RStudio/R v4.0.2 [24, 25]. IDAT file preprocessing and probe quality control was conducted in R using scripts based on minfi [26] and methylumi [27]. IDAT files and a sample file containing covariate and BeadChip metadata were loaded into R where data quality was assessed using the mean detection p-values for the probes in each sample. We applied Functional Normalization(preprocessFunnorm) [28] for the algorithm’s ability to utilize the internal control probes for each individual sample in an unsupervised manner to control for unwanted variation.
After normalization, we excluded individual low-quality probes with a detection p-value > 0.1 in more than 2 samples or bead count <3 in at least $5\%$ of samples, sex chromosome probes, cross-hybridizing probes, and probes where SNPs (within the binding region or within 5-10bp of the binding region) could potentially affect hybridization [22]. To ensure appropriate filtering of problematic probes, we utilized several resources including the Illumina Methylation EPIC BeadChip hg38 manifest and Zhou et al. [ 29] to identify additional variation that would interfere with probe hybridization. We utilized McCartney et al. [ 30] to filter the cross-hybridizing probes that are not listed in the manifest. We removed all probes that reside in the ENCODE DAC black-list regions [31]. All filtering criteria and number of probes filtered can be found in S1 Table in S1 File.
Once probe filtering was complete, we assessed the data for batch effects using principal component analysis (PCA) and no significant batch effect was observed, therefore no correction was applied [32]. The resulting data matrix contained M-values which were utilized for the statistical analyses of the pairwise comparisons due to their statistical robustness. β-values, which are transformed M-values, represent the ratio of all methylated probe intensities over total signal intensities or a percentage of methylation [33]. All methylation values are delta M-values unless otherwise stipulated as they provide a better detection and true positive rates while reducing heteroscedasticity for methylation sites that are highly or non-methylated [33].
## Identification of differentially methylated positions
To assess differentially methylated positions (DMPs), we utilized generalized linear models within limma [34] to assess differential methylation for each individual probe within the M-value matrix as in [22] with adjustment for birth types and fetal sex as covariates within model. Due to the small sample numbers in our dataset, we did not assess any additional covariate data in this analysis as to not overfit the statistical models to this specific dataset and to increase generalizability of our findings in future studies. The following pairwise comparisons were used to identify significant positions of differential methylation: isPTB versus AHC, TB versus AHC and isPTB versus TB. The resulting output for these comparisons is a delta M-value representing the statistical difference in methylation at that position between the conditions being compared. Multiple corrections testing was conducted using the Benjamini Hochberg method [35] at multiple Q values: <0.05, <0.1, <0.2 and <0.3 (S2 Table in S1 File). We tested Q values to determine if our lack observations in one pairwise comparison at $Q = 0.05$ were due to a technical error or if these represented a true lack biological variability despite the statistical parameter selection. We opted to define significant DMPs with a Q <0.3 and a log2 fold-change of >±1.
## Methylome profile identification
To identify methylation profiles, we used Venny 2.0 [36] to generate Venn diagrams to intersect significant DMPs from each pairwise comparison to identify potential profiles for isPTB and AHC. An isPTB profile was defined as any DMP where the delta M-value of isPTB vs TB or AHC was differentially methylated compared to the delta M-values of AHC vs TB which were non-significant. An AHC profile was defined as any DMP where the AHC vs TB or AHC delta M-value was differentially methylated from the isPTB vs TB delta M-values which were non-significant. Heatmaps were generated in Prism v8 (GraphPad) using delta M-values. To assess if the differential methylation was influenced by outliers or by artifacts, we generated violin plots with β-values with median and quartiles in Prism v8 to check the distribution within selected individual samples.
## Differentially Methylation Region (DMR) identification
We used DMRcate v2.2.3 [22, 37] to identify differentially methylated regions comprised of significant DMPs within a specified distance using moderated t statistics. To identify significant DMPs within DMRcate, we used the M-value matrix (normalized and filtered) and set a threshold of Benjamini Hochberg adjusted p-value <0.3. Since DMRcate uses limma to determine the significant DMPs, we were able to utilize the same glm design from the initial DMPs analysis against adjusting for fetal sex. Once significant DMPs were identified, DMR identification thresholds were set at lamba = 1000, $C = 2$, and minimum cpgs = 5. As we are analyzing array data, we opted to use the default lambda and C (scaling factor) which allows for optimal differentiation with 1 standard deviation of support to account for Type 1 errors. Once significant DMRs were identified in each pairwise comparison, we intersected them using Venny 2.0 to identify isPTB and AHC specific DMRs. The isPTB profile was defined as any DMR that was differentially methylated when compared to the AHC and TB, with the AHC vs TB. The AHC profile was defined as any DMR that was differentially methylated compared to isPTB and TB and where the isPTB vs TB methylation was non-significant meaning no DMR was identified in DMRcate. We also set a mean difference in differentiation threshold of 0.01. Heatmaps were generated in Prism v8 (GraphPad) using delta M-values.
## Functional analyses of DMRs with associated genes
Genes with associated DMRs were entered into the Panther Pathway DB [38] for statistical overrepresentation analyses for Reactome Pathways and to assess gene ontology (GO) for biological and molecular processes. Fisher’s Exact tests were used to determine significance and Bonferroni correction for multiple comparisons. Pathways were considered significant if they had an adjusted p-value <0.05.
## Intersection of DMRs with transcriptome candidate genes
To determine if any of our significant DMR’s impacted candidate gene expression, we intersected the DMR’s genomic locations with our candidate gene locations. All genomics regions were mapped to hg38. Where there was overlap, indicating a potential regulatory event, we took those locations and intersected with using the UCSC Genome Table browser (hg38) and the CpG island tracks [21], using the feature-by-feature function. This allowed for identification of DMRs in CpG regions of our candidate genes.
## Methylation study characteristics
Maternal and fetal characteristics for the three different pregnancy outcomes included in the DNAm analyses are presented in Table 1. Transcriptomes from these samples were previously published [16]. Due to the amount of sample required for DNA extraction only a subset of the samples were used and the statistical analyses repeated but did not change. Significant differences were observed in gestational age and fetal weights between AHC and isPTB samples compared to the TB samples ($p \leq 0.05$). All AHC and TB for which there were fetal weights available were appropriate for gestational age. We included males and females in each sample set and adjusted the linear models for fetal sex in addition to birth outcome. It is important to note that in this study, we have mixed genetic ancestry within each of the sample sets.
**Table 1**
| Characteristics | Acute Histological Chorioamnionitis Births (AHC) | Idiopathic Spontaneous Preterm Births (isPTB) | Term Births | p-values |
| --- | --- | --- | --- | --- |
| Number of samples | 8 | 11 | 8 | |
| Maternal Age | 34.5(25–40) | 25(18–39) | 28(19–37) | NS1 |
| Gestational Age | 32(29–35)* | 33(30–36)* | 39(38–41) | <0.00011 |
| Fetal sex (% female) | 3(38%) | 6(55%) | 4(50%) | NS2 |
| Fetal weight (grams) | 1765(1360–2300)* | 2105(1450–2722)* | 3820(3650–4527) | <0.00011 |
| Birth weight percentile | 55(20–80)* | 60(3–80) | 90(60–99) | 0.041 |
| SGA % | 0 | 18.0% | 0 | |
| Delivery type | | | | |
| Cesarean (%) | 4(50%) | 4(37%) | 4(50%) | NS2 |
| Infection Status | | | | |
| (% Positive) | 8(100%)* | 0(0%) | 1(13%) | <0.00012 |
## Identification of transcriptomic profile candidate genes with overlapping CpG islands
The intersection of isPTB specific methylation profiles with the previously identified 170 upregulated genes in isPTB samples yielded 102 candidates ($60\%$) overlapping with CpG islands in their coding regions. In the AHC profile, $\frac{120}{170}$ ($81\%$) candidate genes intersected with CpG islands within coding regions.
## Identification of significant differentially methylated positions (DMP)
Preliminary quality control identified one sample with mean probe detection p-value >0.1 and it was subsequently removed from methylation analyses. Prior to normalization and subsequent probe filtering, there were 866,901 probes in the data matrix. After normalization and filtering, 108,691 probes were removed, leaving 758,210 probes in the matrix for analyses (S1 Table in S1 File).
Our initial statistical testing using the Benjamini Hochberg Q cutoff of 0.05 did not yield any significant DMPs in the isPTB vs TB pairwise comparison. With a Type 1 error rate of $5\%$, we expected to observe approximately 37,910 statistically significant DMPs in this comparison; however, we observed 0. By relaxing the rate of acceptable Type 1 errors to $30\%$, we would expect to observe 227, 463 statistically significant DMPs, yet we only observed a total of 662 significant DMPs (S2 Table in S1 File). We test modeled various statistical parameters to determine if our observations were due to technical errors or true biological differences. At every Q value tested and with different statistical models, we observed the number of DMPs between isPTB and TB to be significantly less than expected. Ultimately, we opted on a Q cutoff of 0.3 in limma [34].
We then set a threshold for differential methylation of log2 fold-change of >1. The DMP analysis identified a total of 24,202 significant DMPs across all pairwise comparisons in the model. In the isPTB vs AHC comparison we identified 8,309 DMPs, 4,334 with reduced methylation and 3,975 more methylated in isPTB compared to AHC. In the TB vs AHC comparison, we identified a total of 15,817 DMPs with 7,170 less methylated and 8,647 more methylated in TB. Lastly, in the isPTB vs TB comparison, 85 DMPs were identified as significant with 13 more methylated and 72 less methylated (Fig 1A).
**Fig 1:** *Identification of methylation profiles using a comparative approach.A. Differentially methylated positions were identified using pairwise comparisons in limma. Red points indicate significant DMPs with a threshold of log2 fold-change >1 and Benjamini Hochberg adjusted p-value <0.3. Blue lines represent log2 fold-change of 1. B. Genomic distribution of DMPs in the pairwise comparisons. The majority of DMPs in the isPTB and TB versus AHC comparisons are located inside or close to known CpG islands. However, in the isPTB versus TB comparison, the majority of DMPs are in open sea regions with no known islands within 4kb. C. The Venn diagram represents the intersection of pairwise comparisons to classify significant DMPs into isPTB and AHC specific methylation profiles.*
We observed differences in genomic location of the DMPs between the pairwise comparisons and thus, analyzed the genomic location distribution of the DMPs per comparison (Fig 1B). In the isPTB vs AHC and TB vs AHC comparisons the majority of DMPs were associated with CpG islands, shores, shelves (isPTB = $70\%$ and TB = $65\%$) while the remaining DMPs were in open sea locations which are typically 3-4kb away from CpG islands (isPTB = $30\%$ and TB = $35\%$ respectively). In contrast, in the isPTB vs TB comparison, $70\%$ of the DMPs were associated with open sea positions while only $30\%$ associated with CpG islands, shores, and shelves. The first step in identification of a DMP methylation profile was to intersect the significant DMPs from each pairwise comparison and determine which would potentially segregate into an isPTB or AHC profile (Fig 1C).
## Isolation of isPTB and AHC DNA methylation profiles using DMPs
As a result of the intersection of significant DMPs, we identified 47 potential isPTB specific DMPs. Upon examining the DNAm patterns for these DMPs across all pairwise comparisons, we wanted to know which DMPs has differential methylation in the isPTB versus the AHC and TB. We ultimately isolated 3 isPTB specific DMPs out of the 47 potential isPTB DMPs. Our examination of the individual sample beta values and their distribution for each DMP confirmed our findings were not due to artifacts or outliers (Fig 2A). Although we initially identified 8,306 potential AHC specific DMPs via the intersection, upon further examination of the DNAm pattern, we ultimately isolated 6,177 where the AHC samples were differentially methylated compared TB or isPTB (Fig 2B). Of these, 3,002 are more methylated and 3,175 are less methylated. We also examined the genomic location distribution of the AHC profile DMPs and found that $76\%$ were located within CpG islands, shores, and shelves with remaining $24\%$ located in open sea regions (S1 Fig).
**Fig 2:** *Identification of significant methylation profiles for isPTB and AHC DMPs.A. Three DMPs identified as having an isPTB specific methylation pattern where the isPTB samples were differentially methylated compared to the AHC or TB samples. The distribution of individual sample beta values was assessed to determine if there were outliers or artifacts influencing the methylation patterns. The dark bands represent the mean of the methylation values while the lighter grey bands represent the interquartile range. B. 6,177 DMPs demonstrating a methylation pattern where the AHC samples were differentially methylated compared to the isPTB or TB samples. The breakout heatmap shows the pattern or the top 25 more and less methylated samples and demonstrates the similarity of methylation between the isPTB and TB samples. The distribution of individual sample beta values was assessed to determine if there were outliers or artifacts influencing the methylation patterns.*
## Identification of differentially methylated regions (DMRs)
To identify differentially methylated regions, we used the M-value matrix of data values previously generated in our initial analyses. We utilized again a relaxed Q <0.3 to ensure we would be able to identify enough CpG sites to identify DMRs in the isPTB vs TB comparison (S3 Table in S1 File). Only then, we were able to identify significant DMRs within all pairwise comparisons (Table 2). 56 DMRs were observed within the isPTB vs TB comparison in contrast to the thousands significant DMRs identified in the isPTB and TB verses AHC pairwise comparisons. All isPTB vs TB DMRs were under 2000bp wide and had no more than 18 CpG sites in any given DMR. In contrast, the DMRs in the isPTB and TB vs AHC comparisons were wider and encompassed more probes (Table 2). We intersected the DMRs and identified potential candidate DMRs for isPTB and AHC methylation profiles (S2 Fig). Ultimately, we identified 51 potential isPTB specific and 12,843 AHC specific DMRs. These DMRs overlap with coding and non-coding loci across the genome as per the annotation from DMRcate package [37].
**Table 2**
| Pairwise comparison | Number of Significant DMRsIdentified* | Width of DMR(Range) | Number of Significant Probes in DMR (Range) |
| --- | --- | --- | --- |
| isPTB vs TB | 56 | 180-1750bp | 5–18 probes |
| isPTB vs AHC | 12883 | 83–9,386bp | 5–110 probes |
| TB vs AHC | 19006 | 37–14,383bp | 5–202 probes |
## Identification and function of DMRs specific to isPTB and AHC
Of the 51 candidate isPTB DMRs, only seven demonstrated an isPTB specific profile (Fig 3 and Table 3). Six isPTB specific DMRs overlap coding/non-coding loci with only one sitting in an upstream promoter region, LINC02028 (Table 4). This is the only isPTB-specific DMR that overlaps with a CpG island. Four of the DMRs sit within transcripts for FAM186A, NOD2, UBL7-AS1, and PDE9A, more specifically within introns or at intron/exon boundaries. The remaining two DMRs sit in the 3’UTR of genes, ZBTB4 and STXB6, with the ZBTB4 DMR crossing the last exon/UTR boundary (Table 4). No over-represented pathways were identified.
**Fig 3:** *isPTB specific DMR profile.Differentially methylated DMRs were identified by differences in the mean of the probe values across the DMR. Only 7 isPTB DMRs had an isPTB specific profile where the isPTB DMRs were less methylated than the TB or AHC DMRs. Two of the DMRs overlap non-coding regions. No DMRs were identified that were more methylated.* TABLE_PLACEHOLDER:Table 3 TABLE_PLACEHOLDER:Table 4 Of the 12,843 AHC specific DMRs, only 1,718 demonstrated an AHC specific methylation pattern. These DMRs include coding and non-coding loci (Fig 4A and S4 Table in S1 File). Of these, 801 DMRs are more methylated while 917 are less methylated than corresponding DMRs in the isPTB or TB pairwise comparison. In the top 25 more/less methylated loci, the lack of significant differences in methylation can clearly be observed in TB vs isPTB (Fig 4B). Of these, $19\%$ ($$n = 328$$) had direct overlap with CpG islands. The remaining $81\%$ had no overlap at all with CpG islands.
**Fig 4:** *AHC specific DMR profile.A. Differentially methylated DMRs were identified by differences in the mean of the probe values across the DMR. AHC specific DMRs are defined by when the AHC DMRs were differentially methylated compared to the TB or isPTB DMRs. B. The top 25 more and less methylated DMRs demonstrates the clarity of the molecular profile, as there is no significant differential methylation in the TB vs isPTB comparison.*
We assessed the potential implications of the AHC specific DMRs using statistical over-representation analyses for pathways and GO terms. In the more methylated DMRs, we identified two significantly over-represented pathways: WNT and Cadherin signaling (Table 5). Significant Biological Process GO terms included homophilic cell adhesion via plasma membrane adhesion molecules (GO:0007156) and cell-cell adhesion via plasma-membrane adhesion molecules (GO:0098742).
**Table 5**
| Unnamed: 0 | Homo sapiens (all genes in database) | Genes from input list | Expected | Fold Enrichment | Adjusted p-value* |
| --- | --- | --- | --- | --- | --- |
| PANTHER Pathways | | | | | |
| Cadherin signaling pathway (P00012) | 164.0 | 21.0 | 5.34 | 3.94 | 6.51e-05 |
| Wnt signaling pathway (P00057) | 317.0 | 30.0 | 10.31 | 2.91 | 0.000103 |
| GO biological process complete | | | | | |
| homophilic cell adhesion via plasma membrane adhesion molecules (GO:0007156) | 168.0 | 26.0 | 5.47 | 4.76 | 4.62e-06 |
| cell-cell adhesion via plasma-membrane adhesion molecules (GO:0098742) | 257.0 | 28.0 | 8.36 | 3.35 | 0.00105 |
| GO molecular function complete | | | | | |
| ion binding (GO:0043167) | 6354.0 | 277.0 | 206.71 | 1.34 | 5.61e-05 |
| binding (GO:0005488) | 16539.0 | 593.0 | 538.05 | 1.1 | 8.9e-05 |
| molecular_function (GO:0003674) | 18245.0 | 631.0 | 593.55 | 1.06 | 0.00423 |
| metal ion binding (GO:0046872) | 4268.0 | 192.0 | 138.85 | 1.38 | 0.00482 |
| cation binding (GO:0043169) | 4354.0 | 194.0 | 141.65 | 1.37 | 0.00908 |
| adenyl nucleotide binding (GO:0030554) | 1572.0 | 84.0 | 51.14 | 1.64 | 0.039 |
No significant over-represented pathways were identified in the less methylated DMRs. The significant Biological Process GO terms that were associated with the less methylated dataset include cell morphogenesis involved in differentiation (GO:0000904), cell morphogenesis (GO:0000902) and detection of chemical stimulus (GO:0009593). For Molecular Function, the following significant GO terms were identified: ion binding (GO:0043167), protein binding (GO:0005515), protein binding (GO:0005515), and olfactory receptor activity (GO:0004984) (Table 6).
**Table 6**
| Unnamed: 0 | Homo sapiens (all genes in database) | Genes from input list | Expected | Fold Enrichment | Adjusted p-value* |
| --- | --- | --- | --- | --- | --- |
| GO biological process complete | | | | | |
| cell morphogenesis involved in differentiation (GO:0000904) | 568.0 | 49.0 | 21.68 | 2.26 | 0.00515 |
| detection of chemical stimulus (GO:0009593) | 522.0 | 2.0 | 19.92 | 0.1 | 0.00802 |
| cell morphogenesis (GO:0000902) | 721.0 | 56.0 | 27.52 | 2.04 | 0.0196 |
| detection of chemical stimulus involved in sensory perception (GO:0050907) | 486.0 | 2.0 | 18.55 | 0.11 | 0.0364 |
| GO molecular function complete | | | | | |
| binding (GO:0005488) | 16539.0 | 689.0 | 631.2 | 1.09 | 0.000256 |
| protein binding (GO:0005515) | 14359.0 | 615.0 | 548.01 | 1.12 | 0.000439 |
| molecular_function (GO:0003674) | 18245.0 | 739.0 | 696.31 | 1.06 | 0.00133 |
| ion binding (GO:0043167) | 6354.0 | 310.0 | 242.5 | 1.28 | 0.00169 |
| olfactory receptor activity (GO:0004984) | 441.0 | 2.0 | 16.83 | 0.12 | 0.0487 |
## Identification of DMRs in regulatory elements of transcriptome candidate genes
Upon intersection of significant DMRs and the candidate genes, none of the isPTB DMRs intersected with any of the isPTB candidate genes. Out of the 1,718 significant AHC DMRs, only eight intersected with the AHC candidate genes (Table 7). Interestingly, six of these DMRs have methylation patterns, in all cases less methylated, that agree with upregulated transcription status. The remaining two have no correlation between profiles (S5 Table in S1 File).
**Table 7**
| DMR Genomic Location | DMR Associated Gene | DMR Size (bp) | Total CpGs in DMR | DMR location | Island Intersection | Methylation status at DMR | AHC Transcriptome profile | Methylation and Transcriptome Agreement |
| --- | --- | --- | --- | --- | --- | --- | --- | --- |
| chr9:21993972–21995735 | CDKN2A-CDKN2B-AS | 1764 | 13 | In promoter | chr9:21993972–21995735 | Less | Upregulated | Yes |
| chr12:6938111–6939048 | ATN1 | 938 | 6 | Mid-gene intron/exon boundary | No | Less | Upregulated | Yes |
| chr22:41939981–41941494 | CENPM | 1514 | 14 | Transcript Dependent | No | Less | Upregulated | Yes |
| chr7:108095719–108097606 | LAMB4 | 1888 | 11 | Mid Gene intronic | No | Less | Downregulated | No |
| chr16:23680392–23681287 | PLK1 | 896 | 5 | Mid Gene intronic | No | More | Upregulated | No |
| chr15:40731625–40735036 | RAD51 | 1605 | 15 | In promoter | No | Less | Upregulated | Yes |
| chr15:64752519–64753130 | RBPMS2 | 612 | 6 | Transcript Dependent | No | Less | Upregulated | Yes |
| chr22:24180492–24181665 | SUSD2 | 1174 | 11 | In promoter | No | Less | Upregulated | Yes |
For each of these eight genes, we examined the genomic location to determine if these DMRs were in promoters or CpG islands, potentially regulating gene expression. We observed only one DMR, CDKN2A, that overlapped with CpG islands 5’ upstream of their transcripts. The DMR upstream of CDKN2A also resides in the same genomic area as a non-coding transcript, CDKN2B. The remaining seven DMRs did not overlap any CpG islands although, two were in the promoter or first intronic region of their associated genes. CENPM and RBPMS2 have multiple transcripts and the location of the DMR varies depending on the specific transcript length and start site. Three DMRs reside in introns or across intron/exon boundaries (Table 7).
## Discussion
To gain insight into the role of DNA methylation in spontaneous preterm birth, we utilized pairwise comparisons of methylation between spontaneous preterm births and normal term births using a general linear model adjusting for fetal sex and gestational age at delivery. It is essential to note that normal gestationally age-matched placental samples are typically not available for studies such as this depending on ethical restrictions of the geographical locale of the study. Therefore, we opted to use with acute histologic chorioamnionitis samples (AHC) which been previously shown to have much lower occurrences of AVM than other clinically defined preterm birth types including PE and IUGR [7, 8] We were able to identify distinct methylation profiles at both the positional (DMP) and regional (DMR) levels in isPTB and AHC. Through bioinformatic functional assessment, we were able to identify pathways of interest pertaining to placental maturation.
Our preliminary analyses indicated that there were very few DMP and DMR between the isPTB and TB birth types regardless of the statistical parameters applied. We tested multiple parameters within the statistical models to ensure that lack of differences was likely due to biological factors, not technical errors. Given the sheer number of datapoints being examined, we felt that relaxing the Q value to 0.3 would not adversely affect our analyses and we were willing to accept the potential increase in false positives [39, 40]. This allowed us to better assess any potential differences between isPTB and TB despite the potential increase in false positives. The Benjamini Hochberg correction is dependent on the overall number of samples to be corrected and considered to be rather conservative. Regardless of the statistical parameters applied, the isPTB profile mimicked the TB profile to a high degree which, agrees with the transcriptomic profiles we previously identified [16] and provides additional evidence of a potential placental hypermaturity profile associated with isPTB. Although this a preliminary study investigating DNA methylation in spontaneous preterm birth, this pattern of DNA methylation was also observed in studies of iatrogenic preterm births in DMP and DMR analyses, for both PE and IUGR [20]. In the second study, focusing on imprinted regions found that IUGR samples also mimicked the PE and term controls [41]. Pyrosequencing from this second study confirmed no differences in the DMRs suggesting the detection of hypermaturity molecular profile. Given that hypermaturity is estimated to affect 50–$60\%$ of all preterm births including PE and IUGR [7, 8], these results provide additional evidence supporting the use of placental DNAm clinically to classify pathophysiologies such as hypermaturity [20, 42].
DMRs are associated with numerous disease pathologies in multiple tissues [43, 44]. While DNAm has been studied in the other adverse pregnancy outcomes such as PE and IUGR, this study is the first to look specifically at isPTB. Our analysis resulted in the identification of seven DMRs with isPTB specific methylation patterns; two are associated with non-coding transcripts (LINC02028 and UBL7-AS), five with genes (ZBTB4, STXBP6, PDE9A, NOD2, and FAM186A). Of these genes, four are of particular interest due to their potential function in or previous association with PTB.
ZBTB4 is a placentally expressed gene coding for a transcription factor that binds methylated CpGs in a repressive manner, controls TP53 responses in cells, and inhibits cell growth and proliferation [45–47]. TP53 was identified as a potential biological pathway of interest in our microarray meta-analysis of spontaneous PTB [48] and has been implicated in isPTB from a uterine perspective in mice [49]. STXBP6, also known as AMISYN, binds SNARE complex proteins together [50]. As SNARE complexes have been well described in synaptic vesicle formation and exocytosis [51] and regulation of membrane fusion dynamics [52, 53], the presence of this protein in the placenta suggests potential role in placental extracellular vesicle formation or the mediation of membrane fusion during cytotrophoblast differentiation [52, 54].
PDE9A functions in the hydrolysis of cAMP into monophosphates, modulating the bioavailability of cAMP and cGMP in cells [55]. cAMP signaling is essential to cytotrophoblast differentiation into syncytiotrophoblast [56]; therefore, alteration of PDE9A expression or function impacts cAMP bioavailability potentially altering this specific trophoblast differentiation pathway. In fact, PDE9A has been proposed as a potential first trimester maternal serum biomarker for Trisomy 21 [57]. Placentas from Trisomy 21 fetuses have multiple defects in cytotrophoblast differentiation, specifically cell fusion, resulting in what appears to be delayed villous maturation, indicating a key role for this gene in normal placental maturation [57–60].
NOD2 has a role in activation of the innate inflammatory response and has been implicated in NFKB activation [61–63]. NFKB activation is a central component of pro-inflammatory /labor pathways in both normal term and preterm pathophysiology [62, 64, 65]. As a member of the NOD-like receptor family, NOD2 has been previously associated with recognition of pathogen associated molecular patterns (PAMPs) and damage associated molecular patterns (DAMPs) both of which have been associated with preterm labor and birth [62]. The activation of pathways associated with PAMPs and DAMPs have previously been associated with sPTB and iatrogenic PTB [48, 66–68]. NOD2 has been studied primarily in the context of a proinflammatory factor in fetal membranes and myometrium; however, NOD2 is expressed in first trimester and term placental tissues, specifically in syncytiotrophoblast and stromal cells [61, 69]. Furthermore, NOD2 polymorphisms have been associated with preterm birth in several genetic studies examining innate immunity, preterm premature rupture of membranes (PPROM), and early onset PE and HELLP (Hemolysis, Elevated Liver enzymes and Low Platelets) syndromes [62, 67, 70, 71].
Taken together, these isPTB DMRs and their associated genes suggest that altered DNA methylation maybe highly influential in isPTB; however, from these data alone, it cannot be determined if this is causative or the result of isPTB as the samples were obtained at delivery. Although we cannot sample placental tissues throughout gestation to determine cause or effect, using DNAm profiling on delivered placental tissues will provide key insights in the pathophysiological underpinnings of adverse pregnancy outcomes.
In contrast to the isPTB DNAm profile, our examination of the AHC samples compared to the isPTB and TB samples identified 1,718 DMRs. We observed within the top 25 more/less methylated DMRs, multiple DMRs were associated with genes of interest that were previously associated with adverse pregnancy outcomes including IUGR and PE. Several have also been associated gestational diabetes mellitus (GDM) which can also result in preterm birth. *These* genes of interest include: MLLT1 [72], FGFR2 [72], CACNA1A [73], GCK [74, 75], FER1L6 [76], CTSH [77], and ACAP3 [78]. Additionally, GSE1 [79], VSTM1 [80], and ACSS1 [79] are expressed in the placenta but have not yet been associated with an adverse pregnancy outcome. Our pathway analyses of the more methylated DMRs, yielded two pathways with statistical over-representation, WNT and Cadherin signaling. Both pathways are necessary for placental development and maturation [81–84] and a prior methylation study in PE also identified differential methylation (increased methylation) in WNT and cadherin signaling [85], which agrees with our findings. Given that over $50\%$ of PE cases have hypermaturity along with the pathological hallmarks of PE, this may indicate a role for these pathways in placental maturation.
We initially hypothesized that changes in methylation at CpG islands could be driving the transcriptional differences we previously observed. However, when we intersected our significant DMRs with our candidate genes, we did not observe any overlap in the isPTB profiles and only eight examples of overlap in the AHC profiles. Of those eight DMR/gene combinations, only CDKN2A/CDKN2B-AS overlapped with a CpG island. CDKN2A, also known as p16, is a gene with multiple transcripts which have different first exons. Known as an important tumor suppressor, its primary role is regulating cell cycle progression through the regulation of TP53. Loss of function studies of Cdkn2a and Tp53 in mice have demonstrated histopathological changes in placenta and upregulated senescence markers as well as mitotic inhibition [86]. CDKN2B-AS is a functional RNA with regulatory roles via interaction with PRC1 and PRC1 which regulates the rest of the genes in this locus epigenetically [87]. Additionally, CDKN2B-AS, also known as ARNIL, has been implicated in preterm birth.
Interestingly, this DMR resides in locus consisting of CDKN2A/CDKN2A-DT/CDKN2B-AS/CDKN2B, a locus vital to cell cycle control and is dysregulated in many cancers. CDKNA-DT is a divergent transcript with no known function. However, CDKN2B, also known as p15, is another critical tumor suppressor, which inhibits cyclin kinases CDK4 and CDK6 [87]. These data along with our methylation data suggest the correct expression of the CDKN2A/CDKN2A-DT/CDKN2B-AS/CDKN2B locus is critical to the structure, function, and potentially the rate of maturity of the placenta and normal healthy pregnancy.
CENPM and SUSD2 have roles in cell cycling and proliferation with mutations associated with cancers. In many cancers the loss of methylation is associated with cell proliferation and migration via metastasis. However, in the developing and maturing placenta these processes are essential for growth, function, and maturation [42, 88, 89]. Less methylation at the DMRs associated with RAD51, RBPMS2, ATN1 and the corresponding upregulation could be indicative of senescence given their respective roles in DNA repair, regulation of cell differentiation, and transcriptional repression. While the intersection of our matched transcriptional and methylation data did not necessarily support our original hypothesis of gene regulation via CpG islands in promoter regions, we were able to identify a potentially critical biological function, cell proliferation and an essential locus, CDKN2A/CDKN2A-DT/CDKN2B-AS/CDKN2B, for further study.
One of the caveats to studying placental villous omics of any nature is the lack of normal gestational age matched tissue due to limited accessibility throughout gestation. We previously utilized infection associated samples in our transcriptome analyses as our gestational age controls as their villi did not appear to be inflamed via pathological assessment. While we cannot rule out that changes at AHC loci may be due to infection, we did not observe pathways or GO terms associated with immunity or infection. Our data suggests that the overall AHC DNAm profile is reflective of appropriate villous maturation rather than an infection profile as was observed in our transcriptome data [16].
This is the first study to examine DNAm in spontaneous preterm birth in the context of placental maturity. The identification of hypermaturity profiles by both positional and regional differences in methylation highlights importance of DNAm to placental maturation and thus warrants further study. These differences could be due to altered trophoblast biology. These data when taken in the context of a potential epigenetic clock, suggests that perhaps epigenetic aging may have a role as it has in other fetal tissue and stem cells [90, 91]. Future studies need to investigate the origin of the observed hypermaturity and its impact on the maternal-fetal interface and pregnancy outcomes.
## References
1. Blencowe H, Cousens S, Chou D, Oestergaard M, Say L, Moller A-B. **Born Too Soon: The global epidemiology of 15 million preterm births.**. *Reproductive Health.* (2013.0) **10** S2. DOI: 10.1186/1742-4755-10-S1-S2
2. Chawanpaiboon S, Vogel JP, Moller A-B, Lumbiganon P, Petzold M, Hogan D. **Global, regional, and national estimates of levels of preterm birth in 2014: a systematic review and modelling analysis**. *The Lancet Global Health* (2018.0) **7** e37-46. DOI: 10.1016/S2214-109X(18)30451-0
3. Monangi NK, Brockway HM, House M, Zhang G, Muglia LJ. **The genetics of preterm birth: Progress and promise.**. *Seminars in perinatology* (2015.0) **39** 574-83. DOI: 10.1053/j.semperi.2015.09.005
4. Burton GJ, Fowden AL. **The placenta: a multifaceted, transient organ.**. *Philosophical Transactions of the Royal Society of London B: Biological Sciences* (2015.0) **370** 20140066. DOI: 10.1098/rstb.2014.0066
5. Khong TY, Mooney EE, Ariel I, Balmus NCM, Boyd TK, Brundler M-A. **Sampling and Definitions of Placental Lesions: Amsterdam Placental Workshop Group Consensus Statement.**. *Archives of Pathology & Laboratory Medicine.* (2016.0) **140** 698-713. DOI: 10.5858/arpa.2015-0225-CC
6. Sankar DK, Bhanu SP, Kiran S, Ramakrishna B, Shanthi V. **Vasculosyncytial membrane in relation to syncytial knots complicates the placenta in preeclampsia: a histomorphometrical study.**. *Anatomy & Cell Biology.* (2012.0) **45** 86-91. DOI: 10.5115/acb.2012.45.2.86
7. Morgan TK. **Role of the Placenta in Preterm Birth: A Review.**. *American Journal of Perinatology* (2016.0) **33** 258-66. DOI: 10.1055/s-0035-1570379
8. Nijman TA, Vliet EO van, Benders MJ, Mol BW, Franx A, Nikkels PG. **Placental histology in spontaneous and indicated preterm birth: A case control study**. *Placenta* (2016.0) **48** 56-62. DOI: 10.1016/j.placenta.2016.10.006
9. Manuck TA, Esplin SM, Biggio J, Bukowski R, Parry S, Zhang H. **The phenotype of spontaneous preterm birth: application of a clinical phenotyping tool**. *American Journal of Obstetrics and Gynecology* (2015.0) **212** 487. DOI: 10.1016/j.ajog.2015.02.010
10. Benton SJ, Leavey K, Grynspan D, Cox BJ, Bainbridge SA. **The clinical heterogeneity of preeclampsia is related to both placental gene expression and placental histopathology**. *American journal of obstetrics and gynecology* (2018.0) **219** 604. DOI: 10.1016/j.ajog.2018.09.036
11. Moutquin J-M. **Classification and heterogeneity of preterm birth.**. *BJOG: An International Journal of Obstetrics and Gynaecology.* (2003.0) **110** 30-3. DOI: 10.1016/s1470-0328(03)00021-1
12. Yuen RKC, Robinson WP. **Review: A high capacity of the human placenta for genetic and epigenetic variation: Implications for assessing pregnancy outcome**. *Placenta* (2011.0) **32** S136-41. DOI: 10.1016/j.placenta.2011.01.003
13. Whigham C-AA, MacDonald TM, Walker SP, Hannan NJ, Tong S, Kaitu’u-Lino TJ. **The untapped potential of placenta-enriched molecules for diagnostic and therapeutic development**. *Placenta* (2019.0). DOI: 10.1016/j.placenta.2019.02.002
14. Ghaemi MS, Tarca AL, Romero R, Stanley N, Fallahzadeh R, Tanada A. **Proteomic signatures predict preeclampsia in individual cohorts but not across cohorts–implications for clinical biomarker studies.**. *J Maternal-fetal Neonatal Medicine.* (2021.0) 1-8
15. Steyerberg EW. **Clinical Prediction Models, A Practical Approach to Development, Validation, and Updating.**. *Statistics Biology Heal.* (2008.0) 83-100
16. Brockway HM, Kallapur SG, Buhimschi IA, Buhimschi CS, Ackerman WE, Muglia LJ. **Unique transcriptomic landscapes identified in idiopathic spontaneous and infection related preterm births compared to normal term births.**. *PloS one* (2019.0) **14** e0225062. DOI: 10.1371/journal.pone.0225062
17. Yuen RK, Peñaherrera MS, Dadelszen P von, McFadden DE, Robinson WP. **DNA methylation profiling of human placentas reveals promoter hypomethylation of multiple genes in early-onset preeclampsia.**. *European journal of human genetics: EJHG.* (2010.0) **18** 1006-12. DOI: 10.1038/ejhg.2010.63
18. Parets SE, Conneely KN, Kilaru V, Menon R, Smith AK. **DNA methylation provides insight into intergenerational risk for preterm birth in African Americans.**. *Epigenetics* (2015.0) **10** 784-92. DOI: 10.1080/15592294.2015.1062964
19. Avila L, Yuen RK, Diego-Alvarez D, Peñaherrera MS, Jiang R, Robinson WP. **Evaluating DNA methylation and gene expression variability in the human term placenta**. *Placenta* (2010.0) **31** 1070-7. DOI: 10.1016/j.placenta.2010.09.011
20. Wilson SL, Leavey K, Cox BJ, Robinson WP. **Mining DNA methylation alterations towards a classification of placental pathologies**. *Human molecular genetics* (2018.0) **27** 135-46. DOI: 10.1093/hmg/ddx391
21. Gardiner-Garden M, Frommer M. **CpG Islands in vertebrate genomes**. *J Mol Biol* (1987.0) **196** 261-82. DOI: 10.1016/0022-2836(87)90689-9
22. Maksimovic J, Phipson B, Oshlack A. **A cross-package Bioconductor workflow for analysing methylation array data.**. *F1000Research.* (2016.0) **5** 1281. DOI: 10.12688/f1000research.8839.3
23. Huber W, Carey VJ, Gentleman R, Anders S, Carlson M, Carvalho BS. **Orchestrating high-throughput genomic analysis with Bioconductor.**. *Nat Methods.* (2015.0) **12** 115-21. DOI: 10.1038/nmeth.3252
24. 24Team Rs. RStudio: Integrated Development for R. RStudio. 2020; Available from: http://www.rstudio.com/
25. 25Team RC. R: A language and environment for statistical ## computing. R Foundation for Statistical Computing [Internet]. 2020; Available from: https://www.R-project.org/
26. Aryee MJ, Jaffe AE, Corrada-Bravo H, Ladd-Acosta C, Feinberg AP, Hansen KD. **Minfi: a flexible and comprehensive Bioconductor package for the analysis of Infinium DNA methylation microarrays**. *Bioinformatics* (2014.0) **30** 1363-9. DOI: 10.1093/bioinformatics/btu049
27. 27S D, P D, S B, Triche, T Jr, M B. methylumi: Handle Illumina methylation data. 2020; Available from: https://www.bioconductor.org/packages/release/bioc/html/methylumi.html
28. Fortin J-P, Labbe A, Lemire M, Zanke BW, Hudson TJ, Fertig EJ. **Functional normalization of 450k methylation array data improves replication in large cancer studies**. *Genome Biology* (2014.0) **15** 503. DOI: 10.1186/s13059-014-0503-2
29. Zhou W, Laird PW, Shen H. **Comprehensive characterization, annotation and innovative use of Infinium DNA methylation BeadChip probes**. *Nucleic Acids Research* (2017.0) **45** e22. DOI: 10.1093/nar/gkw967
30. McCartney DL, Walker RM, Morris SW, McIntosh AM, Porteous DJ, Evans KL. **Identification of polymorphic and off-target probe binding sites on the Illumina Infinium MethylationEPIC BeadChip**. *Genomics Data* (2016.0) **9** 22-4. DOI: 10.1016/j.gdata.2016.05.012
31. Amemiya HM, Kundaje A, Boyle AP. **The ENCODE Blacklist: Identification of Problematic Regions of the Genome.**. *Scientific Reports* (2019.0) **9** 9354. DOI: 10.1038/s41598-019-45839-z
32. Price EM, Robinson WP. **Adjusting for Batch Effects in DNA Methylation Microarray Data, a Lesson Learned.**. *Frontiers in genetics.* (2018.0) **9** 83. DOI: 10.3389/fgene.2018.00083
33. Du P, Zhang X, Huang C-C, Jafari N, Kibbe WA, Hou L. **Comparison of Beta-value and M-value methods for quantifying methylation levels by microarray analysis**. *BMC Bioinformatics* (2010.0) **11** 587. DOI: 10.1186/1471-2105-11-587
34. Ritchie ME, Phipson B, Wu D, Hu Y, Law CW, Shi W. **limma powers differential expression analyses for RNA-sequencing and microarray studies**. *Nucleic Acids Research* (2015.0) **43** e47. DOI: 10.1093/nar/gkv007
35. Benjamini Y, Hochberg Y. **Controlling the False Discovery Rate: A Practical and Powerful Approach to Multiple Testing**. *Journal of the Royal Statistical Society: Series B (Methodological).* (1995.0) **57** 289-300
36. 36Oliveros, J.C. Venny. An interactive tool for comparing lists with Venn’s diagrams. 2015; Available from: https://bioinfogp.cnb.csic.es/tools/venny/index.html
37. Peters TJ, Buckley MJ, Statham AL, Pidsley R, Samaras K, Lord RV. **De novo identification of differentially methylated regions in the human genome.**. *Epigenetics & Chromatin.* (2015.0) **8** 6. DOI: 10.1186/1756-8935-8-6
38. Mi H, Huang X, Muruganujan A, Tang H, Mills C, Kang D. **PANTHER version 11: expanded annotation data from Gene Ontology and Reactome pathways, and data analysis tool enhancements**. *Nucleic Acids Research* (2017.0) **45** D183-9. DOI: 10.1093/nar/gkw1138
39. Li D, Xie Z, Pape M, Dye T. **An evaluation of statistical methods for DNA methylation microarray data analysis**. *BMC Bioinformatics* (2015.0) **16** 217. DOI: 10.1186/s12859-015-0641-x
40. Mansell G, Gorrie-Stone TJ, Bao Y, Kumari M, Schalkwyk LS, Mill J. **Guidance for DNA methylation studies: statistical insights from the Illumina EPIC array**. *Bmc Genomics* (2019.0) **20** 366. DOI: 10.1186/s12864-019-5761-7
41. Monteagudo-Sánchez A, Sánchez-Delgado M, Mora JRH, Santamaría NT, Gratacós E, Esteller M. **Differences in expression rather than methylation at placenta-specific imprinted loci is associated with intrauterine growth restriction.**. *Clin Epigenetics* (2019.0) **11** 35. DOI: 10.1186/s13148-019-0630-4
42. Wilson SL, Robinson WP. **Utility of DNA methylation to assess placental health**. *Placenta* (2018.0) **64** S23-8. DOI: 10.1016/j.placenta.2017.12.013
43. Wilson AS, Power BE, Molloy PL. **DNA hypomethylation and human diseases**. *Biochimica Et Biophysica Acta Bba—Rev Cancer* (2007.0) **1775** 138-62. DOI: 10.1016/j.bbcan.2006.08.007
44. Ehrlich M.. **DNA hypermethylation in disease: mechanisms and clinical relevance**. *Epigenetics* (2019.0) **14** 1-23. DOI: 10.1080/15592294.2019.1638701
45. Filion GJP, Zhenilo S, Salozhin S, Yamada D, Prokhortchouk E, Defossez P-A. **A Family of Human Zinc Finger Proteins That Bind Methylated DNA and Repress Transcription**. *Mol Cell Biol* (2006.0) **26** 169-81. DOI: 10.1128/MCB.26.1.169-181.2006
46. Yu Y, Shang R, Chen Y, Li J, Liang Z, Hu J. **Tumor suppressive ZBTB4 inhibits cell growth by regulating cell cycle progression and apoptosis in Ewing sarcoma**. *Biomed Pharmacother* (2018.0) **100** 108-15. DOI: 10.1016/j.biopha.2018.01.132
47. Weber A, Marquardt J, Elzi D, Forster N, Starke S, Glaum A. **Zbtb4 represses transcription of P21CIP1 and controls the cellular response to p53 activation**. *Embo J* (2008.0) **27** 1563-74. DOI: 10.1038/emboj.2008.85
48. Paquette AG, Brockway HM, Price ND, Muglia LJ. **Comparative transcriptomic analysis of human placentae at term and preterm delivery**. *Biology of reproduction* (2018.0) **98** 89-101. DOI: 10.1093/biolre/iox163
49. Hirota Y, Daikoku T, Tranguch S, Xie H, Bradshaw HB, Dey SK. **Uterine-specific p53 deficiency confers premature uterine senescence and promotes preterm birth in mice**. *Journal of Clinical Investigation* (2010.0) **120** 803-15. DOI: 10.1172/JCI40051
50. Scales SJ, Hesser BA, Masuda ES, Scheller RH. **Amisyn, a Novel Syntaxin-binding Protein That May Regulate SNARE Complex Assembly***. *J Biol Chem* (2002.0) **277** 28271-9. DOI: 10.1074/jbc.M204929200
51. Wang T, Li L, Hong W. **SNARE proteins in membrane trafficking**. *Traffic* (2017.0) **18** 767-75. DOI: 10.1111/tra.12524
52. Han J, Pluhackova K, Böckmann RA. **The Multifaceted Role of SNARE Proteins in Membrane Fusion.**. *Front Physiol* (2017.0) **8** 5. DOI: 10.3389/fphys.2017.00005
53. Bogaart G van den, Jahn R. **Counting the SNAREs needed for membrane fusion.**. *J Mol Cell Biol* (2011.0) **3** 204-5. DOI: 10.1093/jmcb/mjr004
54. Guček A, Gandasi NR, Omar-Hmeadi M, Bakke M, Døskeland SO, Tengholm A. **Fusion pore regulation by cAMP/Epac2 controls cargo release during insulin exocytosis.**. *Elife* (2019.0) **8** e41711. DOI: 10.7554/eLife.41711
55. Rentero C, Puigdomènech P. **Specific use of start codons and cellular localization of splice variants of human phosphodiesterase 9A gene**. *Bmc Mol Biol* (2006.0) **7** 39. DOI: 10.1186/1471-2199-7-39
56. Gerbaud P, Taskén K, Pidoux G. **Spatiotemporal regulation of cAMP signaling controls the human trophoblast fusion.**. *Front Pharmacol* (2015.0) **6** 202. DOI: 10.3389/fphar.2015.00202
57. Lim JH, Kim SY, Park SY, Lee SY, Kim MJ, Han YJ. **Non-Invasive Epigenetic Detection of Fetal Trisomy 21 in First Trimester Maternal Plasma.**. *Plos One.* (2011.0) **6** e27709. DOI: 10.1371/journal.pone.0027709
58. Frendo JL, Vidaud M, Guibourdenche J, Luton D, Muller F, Bellet D. **Defect of villous cytotrophoblast differentiation into syncytiotrophoblast in Down’s syndrome**. *The Journal of clinical endocrinology and metabolism* (2000.0) **85** 3700-7. DOI: 10.1210/jcem.85.10.6915
59. Pidoux G, Gerbaud P, Cocquebert M, Segond N, Badet J, Fournier T. **Review: Human trophoblast fusion and differentiation: Lessons from trisomy 21 placenta**. *Placenta* (2011.0) **33** S81-6. DOI: 10.1016/j.placenta.2011.11.007
60. Malassiné A, Frendo J-L, Evain-Brion D. **Trisomy 21- affected placentas highlight prerequisite factors for human trophoblast fusion and differentiation**. *International Journal of Developmental Biology* (2009.0) **54** 475-82
61. Bryant AH, Bevan RJ, Spencer-Harty S, Scott LM, Jones RH, Thornton CA. **Expression and function of NOD-like receptors by human term gestation-associated tissues**. *Placenta* (2017.0) **58** 25-32. DOI: 10.1016/j.placenta.2017.07.017
62. Lappas M.. **NOD1 and NOD2 Regulate Proinflammatory and Prolabor Mediators in Human Fetal Membranes and Myometrium via Nuclear Factor-Kappa B**. *Biol Reprod* (2013.0) **89**. DOI: 10.1095/biolreprod.113.110056
63. Bourhis LL, Benko S, Girardin SE. **Nod1 and Nod2 in innate immunity and human inflammatory disorders**. *Biochem Soc T* (2007.0) **35** 1479-84. DOI: 10.1042/BST0351479
64. Marchand M, Horcajadas JA, Esteban FJ, McElroy SL, Fisher SJ, Giudice LC. **Transcriptomic Signature of Trophoblast Differentiation in a Human Embryonic Stem Cell Model**. *Biology of Reproduction* (2011.0) **84** 1258-71. DOI: 10.1095/biolreprod.110.086413
65. Wang B, Palomares K, Parobchak N, Cece J, Rosen M, Nguyen A. **Glucocorticoid Receptor Signaling Contributes to Constitutive Activation of the Noncanonical NF-κB Pathway in Term Human Placenta**. *Molecular Endocrinology* (2013.0) **27** 203-11. PMID: 23239753
66. Tang D, Kang R, Coyne CB, Zeh HJ, Lotze MT. **PAMPs and DAMPs: signal 0s that spur autophagy and immunity**. *Immunological Reviews* (2012.0) **249** 158-75. DOI: 10.1111/j.1600-065X.2012.01146.x
67. Rijn BB van, Franx A, Steegers EAP, Groot CJM de, Bertina RM, Pasterkamp G. **Maternal TLR4 and NOD2 Gene Variants, Pro-Inflammatory Phenotype and Susceptibility to Early-Onset Preeclampsia and HELLP Syndrome.**. *Plos One.* (2008.0) **3** e1865. DOI: 10.1371/journal.pone.0001865
68. Robertson SA, Hutchinson MR, Rice KC, Chin P, Moldenhauer LM, Stark MJ. **Targeting Toll‐like receptor‐4 to tackle preterm birth and fetal inflammatory injury.**. *Clin Transl Immunol* (2020.0) **9** e1121. DOI: 10.1002/cti2.1121
69. Costello MJ, Joyce SK, Abrahams VM. **NOD Protein Expression and Function in First Trimester Trophoblast Cells**. *Am J Reprod Immunol* (2007.0) **57** 67-80. DOI: 10.1111/j.1600-0897.2006.00447.x
70. Strauss JF, Romero R, Gomez-Lopez N, Haymond-Thornburg H, Modi BP, Teves ME. **Spontaneous preterm birth: advances toward the discovery of genetic predisposition.**. *Am J Obstet Gynecol* (2018.0) **218** 294-314. DOI: 10.1016/j.ajog.2017.12.009
71. Ch Härtel, Finas D, Ahrens P, Kattner E, Schaible Th, Müller D. **Polymorphisms of genes involved in innate immunity: association with preterm delivery**. *Mhr Basic Sci Reproductive Medicine* (2004.0) **10** 911-5
72. Tekola-Ayele F, Zeng X, Ouidir M, Workalemahu T, Zhang C, Delahaye F. **DNA methylation loci in placenta associated with birthweight and expression of genes relevant for early development and adult diseases**. *Clin Epigenetics* (2020.0) **12** 78. DOI: 10.1186/s13148-020-00873-x
73. Vennou KE, Kontou PI, Braliou GG, Bagos PG. **Meta-analysis of gene expression profiles in preeclampsia**. *Pregnancy Hypertens* (2020.0) **19** 52-60. DOI: 10.1016/j.preghy.2019.12.007
74. Ramasammy R, Munisammy L, Sweta K, Selvakumar S, Velu K, Rani J. **Association between GCK gene polymorphism and gestational diabetes mellitus and its pregnancy outcomes**. *Meta Gene* (2021.0) **28** 100856
75. Beaumont RN, Warrington NM, Cavadino A, Tyrrell J, Nodzenski M, Horikoshi M. **Genome wide association study of offspring birth weight in 86577 women identifies five novel loci and highlights maternal genetic effects that are independent of fetal genetics**. *Human Molecular Genetics* (2018.0) **27** 742-56. PMID: 29309628
76. Lang CT, Markham KB, Behrendt NJ, Suarez AA, Samuels P, Vandre DD. **Placental Dysferlin Expression is Reduced in Severe Preeclampsia**. *Placenta* (2009.0) **30** 711-8. DOI: 10.1016/j.placenta.2009.05.008
77. Varanou A, Withington SL, Lakasing L, Williamson C, Burton GJ, Hemberger M. **The importance of cysteine cathepsin proteases for placental development**. *J Mol Med* (2006.0) **84** 305-17. DOI: 10.1007/s00109-005-0032-2
78. Jia R-Z, Zhang X, Hu P, Liu X-M, Hua X-D, Wang X. **Screening for differential methylation status in human placenta in preeclampsia using a CpG island plus promoter microarray**. *Int J Mol Med* (2012.0) **30** 133-41. DOI: 10.3892/ijmm.2012.983
79. Uhlén M, Fagerberg L, Hallström BM, Lindskog C, Oksvold P, Mardinoglu A. **Tissue-based map of the human proteome**. *Science* (2015.0) **347** 1260419. PMID: 25613900
80. Guo X, Zhang Y, Wang P, Li T, Fu W, Mo X. **VSTM1-v2, a novel soluble glycoprotein, promotes the differentiation and activation of Th17 cells**. *Cell Immunol* (2012.0) **278** 136-42. DOI: 10.1016/j.cellimm.2012.07.009
81. Sonderegger S, Pollheimer J, Knöfler M. **Wnt Signalling in Implantation, Decidualisation and Placental Differentiation–Review**. *Placenta* (2010.0) **31** 839-47. DOI: 10.1016/j.placenta.2010.07.011
82. Knöfler M, Pollheimer J. **Human placental trophoblast invasion and differentiation: a particular focus on Wnt signaling**. *Frontiers in Genetics* (2013.0) **4** 190. DOI: 10.3389/fgene.2013.00190
83. Kokkinos MI, Murthi P, Wafai R, Thompson EW, Newgreen DF. **Cadherins in the human placenta–epithelial–mesenchymal transition (EMT) and placental development.**. *Placenta* (2010.0) **31** 747-55. DOI: 10.1016/j.placenta.2010.06.017
84. Adu-Gyamfi EA, Czika A, Gorleku PN, Ullah A, Panhwar Z, Ruan L-L. **The Involvement of Cell Adhesion Molecules, Tight Junctions, and Gap Junctions in Human Placentation.**. *Reprod Sci.* (2021.0) **28** 305-20. DOI: 10.1007/s43032-020-00364-7
85. Yeung KR, Chiu CL, Pidsley R, Makris A, Hennessy A, Lind JM. **DNA methylation profiles in preeclampsia and healthy control placentas**. *Am J Physiol-heart C* (2016.0) **310** H1295-303. DOI: 10.1152/ajpheart.00958.2015
86. Gal H, Lysenko M, Stroganov S, Vadai E, Youssef SA, Tzadikevitch‐Geffen K. **Molecular pathways of senescence regulate placental structure and function**. *Embo J* (2019.0) **38** e100849. DOI: 10.15252/embj.2018100849
87. He S, Gu W, Li Y, Zhu H. **ANRIL/CDKN2B-AS shows two-stage clade-specific evolution and becomes conserved after transposon insertions in simians**. *Bmc Evol Biol* (2013.0) **13** 247. DOI: 10.1186/1471-2148-13-247
88. Gamage T, Schierding W, Hurley D, Tsai P, Ludgate JL, Bhoothpur C. **The role of DNA methylation in human trophoblast differentiation**. *Epigenetics* (2018.0) **13** 1154-73. DOI: 10.1080/15592294.2018.1549462
89. Mayne BT, Leemaqz SY, Smith AK, Breen J, Roberts CT, Bianco-Miotto T. **Accelerated placental aging in early onset preeclampsia pregnancies identified by DNA methylation**. *Epigenomics-uk* (2016.0) **9** 279-89. DOI: 10.2217/epi-2016-0103
90. Raj K, Horvath S. **Current perspectives on the cellular and molecular features of epigenetic ageing**. *Exp Biol Med* (2020.0) **245** 1532-42. DOI: 10.1177/1535370220918329
91. Lee Y, Choufani S, Weksberg R, Wilson SL, Yuan V, Burt A. **Placental epigenetic clocks: estimating gestational age using placental DNA methylation levels**. *Aging* (2019.0) **11** 4238-53. DOI: 10.18632/aging.102049
|
---
title: Plasma soluble tumor necrosis factor receptor I as a biomarker of lupus nephritis
and disease activity in systemic lupus erythematosus patients
authors:
- Xin-ran Liu
- Yuan-yuan Qi
- Ya-fei Zhao
- Yan Cui
- Zhan-zheng Zhao
journal: Renal Failure
year: 2023
pmcid: PMC10035946
doi: 10.1080/0886022X.2023.2174355
license: CC BY 4.0
---
# Plasma soluble tumor necrosis factor receptor I as a biomarker of lupus nephritis and disease activity in systemic lupus erythematosus patients
## Abstract
### Objectives
The goal of our study was to evaluate the potential role of sTNF-RI as a biomarker of renal involvement in SLE patients and active SLE.
### Methods
The study sample consisted of two cohorts. The discovery cohort included 16 SLE patients without renal involvement (non-LN), 60 lupus nephritis (LN) patients and 21 healthy controls (HCs) and the replication cohort included 18 SLE non-LN patients, 116 LN patients and 36 HCs.
### Results
The sTNF-RI levels differed significantly in the discovery cohort. The plasma sTNF-RI levels were higher in LN patients than in non-LN patients ($$p \leq .009$$) and HCs ($$p \leq 4$$ × 10−6). Plasma sTNF-RI levels were significantly higher in non-LN patients than in HCs ($$p \leq .03$$). The finding was confirmed in independent replication cohort (LNs vs. non-LN, $$p \leq 4.053$$ × 10−7; LNs vs. HCs, $$p \leq 2.395$$ × 10−18; non-LN vs. HCs, $$p \leq 2.51$$ × 10−4). The plasma sTNF-RI levels were associated with disease activity, renal function in SLE patients and urine protein in LN patients. The multivariate analysis revealed that high sTNF-RI was an independent risk factor for renal involvement. The multivariate logistic regression results suggested that high TNF-RI, high systolic blood pressure, high serum creatinine, low C4 and positive anti-dsDNA were independent risks of active SLE patients. A nomogram was constructed based on the results of multivariate logistic regression analysis and it was practical in predicting the risk of the active SLE patients. Immunohistochemistry suggested that the expression of TNF-RI in the kidney was increased.
### Conclusions
Plasma sTNF-RI might be a good biomarker of renal involvement and disease activity in SLE patients.
## Introduction
Systemic lupus erythematosus (SLE) is an autoimmune disorder that is characterized by the disruption of immune tolerance and autoantibody formation. It can involve many organs and systems, including the skin, kidney and hematopoietic system [1]. Lupus nephritis (LN) is a common complication and is associated with a poor prognosis in SLE. Asians reportedly have a higher incidence of LN [2], and the early diagnosis of LN is critical for timely management and patient survival. Kidney biopsy is the gold standard for diagnosing LN, but it is invasive [3,4]. Currently, renal involvement in SLE is usually evaluated by urinalysis or by measuring serum creatinine or eGFR. Therefore, finding earlier biomarkers is very important for the timely prediction of renal impairment in SLE patients.
Cytokines, particularly tumor necrosis factor-α (TNF-α), play an essential role in the pathogenesis of SLE. Previous studies had shown that SLE patients had high levels of serum TNF-α, and TNF-α was correlated with disease activity in SLE [5]. Increased renal expression of TNF-α was observed in lupus mice. Tumor necrosis factor receptor I (TNF-RI), the classic ligand for TNF-α, is correlated with chronic diseases, including autoimmune diseases. TNF-RI is ubiquitously expressed on almost all cell types in humans and plays an important role in proinflammatory effects. Moreover, it can activate signaling pathways that lead to cell death [6,7].
Accumulating evidence indicates the involvement of TNF-RI in the development of SLE. Sharapova et al. reported that autoantibodies from SLE patients interact with TNF-RI, thereby inducing murine fibroblast cell death [8]. In a retrospective study of Caucasians, soluble TNF-RI (sTNF-RI) was identified as a potential marker to distinguish SLE patients without renal impairment from LN patients [9].
The goal of our study was to evaluate the potential value of sTNF-RI as a biomarker for the prediction of renal involvement and renal function in Chinese SLE patients.
## Study population
SLE patients who were hospitalized at the Nephrology Hospital of the First Affiliated Hospital of Zhengzhou University between July 2019 and February 2021 were enrolled. Patients with tumors, pregnancy or other autoimmune diseases were excluded. The study sample consisted of two cohorts. The discovery cohort included 60 LN patients, 16 SLE patients without renal involvement (non-LN) and 21 healthy controls (HCs). The replication cohort included 116 LN patients, 18 non-LN patients and 36 HCs. All of the SLE patients met the American College of Rheumatology (ACR) revised criteria [10] as updated in 1997 [11]. The patients with LN were confirmed by renal biopsy and classified based on the current International Society of Nephrology/Renal Pathologic Society (ISN/RPS) 2003 criteria [12]. All patients were recruited from the First Affiliated Hospital of Zhengzhou University. All subjects were fully informed about the characteristics and purpose of our study and provided informed consent. Our study was approved by the Ethical Committee of the Medical Ethics Committee of Zhengzhou University First Hospital (2019-KY-134).
We collected clinical data at the time of collecting the samples and the data of pathological characteristics at the time of renal biopsy. The disease activity (SLEDAI) was assessed by the SLE Disease Activity Index 2000 [13]. In addition, we also calculated the estimated glomerular filtration rate (eGFR) by using the Modification of Diet in Renal Disease Study equations (eGFR-MDRD) [14] and Chronic Kidney Disease Epidemiology Collaboration equations (eGFR-EPI) [15].
## Definition of renal impairment
Renal impairment is defined by the following: 1. Urine protein/creatinine ratio ≥ 0.2; 2. Persistent proteinuria ≥ 0.15 g/d or greater than 1+ by dipstick; 3. Cellular casts including red blood cells (RBCs), hemoglobin, granular, tubular, or mixed, and cellular casts limited to RBC or WBC casts can be substituted for cellular casts in the absence of infection); 4. Active urinary sediment: 5 RBCs/HPF, 5 white blood cells (WBCs)/HPF; and 5. Renal biopsy supports the diagnosis of LN.
## Determination of plasma TNF-RI levels
Peripheral blood samples were collected with EDTA-K2 anticoagulant tubes after an overnight fast. Plasma was obtained by centrifugation at 3000 rpm for 10 min within 1 h. All samples were aliquoted and stored at −80 °C until this analysis. The levels of plasma sTNF-RI in the discovery cohort were detected and quantified by Quantibody® Human Inflammation Array 3 (Cat# QAH-INF-G3-4, Raybiotech), and the levels of plasma TNF-RI in the replication cohort were detected and quantified by the Human TNF-RI Quantikine ELISA Kit (R&D Systems, Minneapolis, MN) according to the manufacturer’s instructions.
## Immunohistochemistry of TNF-RI expression
Renal lesion tissues were harvested from a renal tissue biopsy, and normal kidney tissue samples adjacent to renal cancer served as normal controls. The deposition of TNF-RI in the kidneys of LN patients with different types was detected by immunohistochemistry (IHC). Two class II LN patients, 1 class III LN patient, 2 class IV LN patients, 3 class V patients and 5 normal controls were included. The renal tissues were prepared into 4 µm thick paraffin sections using polyclonal rabbit anti-human TNR-RI (21574-1-AP, Proteintech) at a 1:100 dilution. The TNF-RI-positive area was quantified using the ImageJ analysis system.
## Statistical analysis
χ2 tests or Fisher’s exact tests were used for categorical variables, and data are presented as n, %. Mann–Whitney tests or independent-samples T tests were used for continuous variables, and data are presented as the mean ± standard deviation (SD) or median and interquartile range (IQR). Pearson’s correlation coefficient was applied to detect the correlation between the level of plasma sTNF-RI and clinical indicators. Univariate and multivariate logistic regression analyses were used to explore the variables that were independently related to LN and active SLE. Then, the nomogram was generated based on multivariate logistic regression. In addition, calibration curves, decision curve analysis (DCA) and receiver operating characteristic (ROC) curves were plotted to determine the reliability of our nomogram. All data were analyzed by using R software version 4.1.2, and $p \leq .05$ was considered significant.
## Study population characteristics
The baseline characteristics of the patients and healthy controls are shown in Table 1. In the discovery cohort, non-LN patients had a median disease duration of 6.5 months (range 1–51.75 months), and the LN patients had a median disease duration of 12 months (range 1–63.25 months). In the replication cohort, the SLE without renal involvement patients had a median disease duration of 6.5 months (range 1–33.25 months), and the LN patients had a median disease duration of 12 months (range 1–48 months).
**Table 1.**
| Characteristics | The discovery cohort | The discovery cohort.1 | The discovery cohort.2 | The replication cohort | The replication cohort.1 | The replication cohort.2 |
| --- | --- | --- | --- | --- | --- | --- |
| Characteristics | SLE without renal impairment(N = 16) | LN(N = 60) | Healthy controls(N = 21) | SLE without renal impairment(N = 18) | LN(N = 116) | Healthy controls(N = 36) |
| Female, n, % | 14, 87.5% | 50, 83.33% | 14, 66.67% | 16, 88.89% | 92, 79.31% | 27, 75% |
| Age (year), mean ± S.D. | 31.81 ± 13.38 | 32.82 ± 12.42 | 25.25 ± 3.01 | 30.67 ± 13.01 | 34.84 ± 13.68 | 25.25 ± 3.01 |
| Disease duration (month), median (IQR) | 6.5 (1–51.75) | 12 (1–63.25) | | 6.5 (1–33.25) | 12 (1–48) | – |
| Onset age (year), median (IQR) | 29 (10–39.25) | 29.5 (18.75––37.5) | | 28 (19.75–37.75) | 29 (21–42) | – |
| SBP (mmHg), median (IQR) | 117.5 (106.75–120) | 122 (116–135) | 114 (106.5–118) | 119 (107.25–123) | 125 (120–135) | 111.5 (105–117) |
| DBP (mmHg), median (IQR) | 76 (69–80.25) | 80 (75–85) | 78 (71.5–80.5) | 75.5 (69–80) | 80 (75–86.75) | 74.5 (70.5–79.75) |
| Malar Rash, n, % | 6, 37.5% | 2, 3.33% | – | 8, 44.44% | 6, 5.17% | – |
| Arthritis, n, % | 1, 6.25% | 3, 5% | – | 1, 5.56% | 5, 4.31% | – |
| Serositis, n, % | 1, 6.25% | 13, 21.67% | – | 1, 5.56% | 28, 24.14% | – |
| Fever, n, % | 3, 18.75% | 1, 1.67% | – | 3, 16.67% | 3, 2.59% | – |
| Neurologic disorder, n, % | 0, 0% | 2, 3.33% | – | 0, 0% | 1, 0.86% | – |
| Leukopenia, n, % | 4, 25% | 3, 5% | – | 5, 27.78% | 10, 8.62% | – |
| Lyphopenia, n, % | 6, 37.5% | 23, 38.33% | – | 6, 33.33% | 47, 40.52% | – |
| Thrombocytopenia, n, % | 1, 6.25% | 5, 8.33% | – | 1, 5.56% | 14, 12.07% | – |
| Urea (mmol/l), median (IQR) | 3.8 (3.1–5.58) | 5.6 (4.03–8.48) | – | 4.45 (3.1–5.63) | 5.9 (4.43–8.5) | – |
| Serum creatinine (umol/l), median (IQR) | 55.5 (40.75–64.75) | 64 (56–90.25) | 51 (44–60.5) | 54.5 (42.25–64.25) | 66.5 (56–99) | 51.5(47.25–58.75) |
| Uric acid (umol/l), median (IQR) | 223 (188.25–263.5) | 300.5 (258.5–395.25) | – | 234 (196.75–258.5) | 309.5 (266.75–393.75) | – |
| eGFR-EPI (ml/min/1.73m2), median (IQR) | 124.06 (103.07–144.31) | 106.79 (76.77–121.55) | 132.11 (126.35–139.68) | 126.85 (103.78–140.88) | 104.32 (73.17–121.35) | 127.67 (122.09–134.13) |
| eGFR-MDRD (ml/min/1.73m2), median (IQR) | 142.92 (107.86–202.86) | 111.03 (75.78–131.64) | 160.76 (147.19–198.54) | 142.92 (111.1–202.68) | 102.85 (72.34–133.04) | 149.66 (125.35–168.34) |
| Serum albumin (g/l), median (IQR) | 41.6 (39.08–46.45) | 36.9 (30.23–40.45) | – | 40.5 (37.45–46.35) | 34.05 (25.9–38.28) | – |
| T-CHO (mmol/l), mean (S.D.) | 3.74 (3.04–4.91) | 3.9 (3.36–5.14) | – | 3.74 (3.06–4.7) | 4.34 (3.58–5.54) | – |
| TG (mmol/l), median (IQR) | 1.18 (1.03––1.89) | 1.31 (0.94–1.94) | – | 1.22 (1.05–2.03) | 1.47 (1.04–2.09) | – |
| Decreased C3, n, % | 7, 43.75% | 12, 20% | – | 9, 50% | 40, 34.48% | – |
| Decreased C4, n, % | 7, 43.75% | 12, 20% | – | 9, 50% | 30, 25.86% | – |
| Positive anti-ANAs, n, % | 15, 93.75% | 54, 90% | – | 17, 94.44% | 109, 93.97% | – |
| Anti-dsDNA Ab (U/ml), median (IQR) | 37.55 (10.85–135.2) | 20.3 (10–169.79) | – | 45.45 (12.55–152) | 36.89 (10–174.15) | – |
| Positive anti-dsDNA Ab, n, % | 4, 25% | 18, 30% | | 6, 33.3% | 36, 31.03% | |
| Positive anti-Sm Ab, n, % | 4, 25% | 7, 11.67% | – | 6, 33.33% | 22, 18.97% | – |
| SLEDAI score, median (IQR) | 3 (2–6.5) | 4 (0–11.75) | – | 4 (2–7.75) | 8 (2–12) | – |
| Renal biopsy, n, % a | | | | | | |
| Class II | | 6, 10% | | | 6, 5.17% | |
| Class III | | 13, 21.67% | | | 17, 14.66% | |
| Class IV | | 20, 33.33% | | | 42, 36.21% | |
| Class V | | 9, 15% | | | 21, 18.1% | |
| Class V + III/IV | | 12, 20% | | | 30, 25.86% | |
## Plasma sTNF-RI levels were increased in non-LN patients and LN patients
Of the 21 inflammatory factors tested, plasma sTNF-RI levels differed significantly in the discovery cohort. The plasma sTNF-RI levels were significantly higher in LN patients [8220.94 (6338.67–9645.84) pg/ml] than non-LN patients [6791.14 (5210.08–7308.11) pg/ml] ($$p \leq .009$$) and HCs [5706.72 (4779.93–6257.09) pg/ml] ($$p \leq 4$$ × 10−6). sTNF-RI levels were higher in the non-LN group than in the HCs ($$p \leq .03$$) (Figure 1(A)). The finding was confirmed in the independent replication cohort. ELISA determined that the plasma sTNF-RI levels of LN patients [1714.91 (1192.87–2704.57) pg/ml] were significantly higher than non-LN patients [870.58 (721.84–1208.55) pg/ml] ($$p \leq 4.053$$ × 10−7) and HCs [677.87(615.31–739.1) pg/ml]($$p \leq 2.395$$ × 10−18) and differed significantly between non-LN patients and HCs ($$p \leq 2.51$$ × 10−4) (Figure 1(B)).
**Figure 1.:** *Plasma sTNF-RI levels in SLE without renal impairment patients, LN patients and healthy controls (HCs) of the discovery cohort (A) and in the replication cohort (B).*
Because the disease activity was significantly different in the LN groups, we analyzed the difference between active LN patients (SLEDAI > 9) and inactive LN patients (SLEDAI ≤ 9). In the discovery cohort, we did not observe significant changes between active LN patients and inactive LN patients ($$p \leq .508$$). We noticed that the plasma sTNF-RI levels were significantly higher in active LN patients or in inactive LN patients than in HCs or non-LN patients ($p \leq .001$ and $$p \leq .021$$, respectively, in active LN; $p \leq .001$ and $$p \leq .017$$, respectively, in inactive LN). In the replication cohort, the plasma sTNF-RI levels gradually increased in HCs, non-LN patients, inactive LN patients and active LN patients. A significant difference was found between every two groups ($p \leq .001$) (Figure 2).
**Figure 2.:** *Plasma sTNF-RI levels in active LN patients (SLEDAI > 9), SLE without renal impairment patients and HCs of the discovery cohort (A) and in the replication cohort (C), plasma sTNF-RI levels in inactive LN patients (SLEDAI ≤ 9), SLE without renal impairment patients and HCs of the discovery cohort (B) and in the replication cohort (D). SLEDAI: SLE disease activity, *p < .05, ****p < .001.*
The level of plasma sTNF-RI was not different in different types of LN patients. Specifically, the expression of plasma sTNF-RI did not differ between proliferative LN patients (class III/IV, class V + III or IV) and membranous LN (class II or V) (Supplemental Material Table 1, Figures 1 and 2].
## Increased plasma sTNF-RI levels were associated with increased disease activity, decreased renal function in SLE patients and the level of urine protein in LN patients
We compared the plasma sTNF-RI levels between SLE patients with and without particular clinical features. In the discovery cohort, plasma sTNF-RI levels in SLE patients with serositis or thrombocytopenia were significantly higher than those in SLE patients without serositis or thrombocytopenia (serositis $$p \leq .015$$; thrombocytopenia $$p \leq .019$$). The SLE patients without anti-Sm antibodies showed significantly increased plasma sTNF-RI levels compared with those with anti-Sm antibodies ($$p \leq .039$$). However, this finding could not be verified in the replication cohort (Table 2).
**Table 2.**
| Characteristics | The discovery cohort | The discovery cohort.1 | The discovery cohort.2 | The replication cohort | The replication cohort.1 | The replication cohort.2 |
| --- | --- | --- | --- | --- | --- | --- |
| Characteristics | sTNF-RI (pg/mL) | sTNF-RI (pg/mL) | sTNF-RI (pg/mL) | sTNF-RI (pg/mL) | sTNF-RI (pg/mL) | sTNF-RI (pg/mL) |
| Characteristics | Presence (n) | Absence (n) | p | Presence (n) | Absence (n) | p |
| Malar rash | 6873.69 (5347.45–8082.09)(n = 8) | 7771.65 (6163.15–9574.56)(n = 68) | .204 | 1282.66 (777.17–1657.85)(n = 14) | 1658.28 (1106.7–2563.51)(n = 120) | .002 |
| Arthritis | 10,264.76 (6124.06–11,377.32)(n = 4) | 7707.11 (6163.15–9229.42)(n = 72) | .193 | 1914.9 (995.83–3012.73)(n = 6) | 1612.51 (1095.5–2338.95)(n = 128) | .683 |
| Serositis | 9491.71 (7505.03–11,014.25)(n = 14) | 7438.64 (5942.57–8925.46)(n = 62) | .015 | 1290.26 (989.35–3418.53)(n = 29) | 1653.2 (1111.34–2234.27)(n = 105) | .825 |
| Fever | 6957.01 (4972.53–8243.75)(n = 7) | 7768.46 (6216.58–9570.48)(n = 69) | .907 | 1400.38 (1062.34–2645.93)(n = 1) | 1637.97 (1085.95–2443.39)(n = 133) | .88 |
| Leukopenia | 6957.01 (4972.53–8243.75)(n = 7) | 7768.46 (6216.58–9570.48)(n = 69) | .232 | 1745.88 (1353.23–4637.57)(n = 15) | 1587.05 (1063.31–2272.38)(n = 119) | .091 |
| Lyphopenia | 8328.85 (6835.21–9628.29)(n = 29) | 7486.4 (5922.73–9159.02)(n = 47) | .229 | 1637.97 (1043.12–2576.3)(n = 49) | 1587.05 (1099.37–2415.67)(n = 85) | .563 |
| Thrombocytopenia | 9556.19 (8679.83–10,602.45)(n = 6) | 7507.24 (5942.57–9182.49)(n = 70) | .019 | 1673.53 (1219.19–4547.11)(n = 39) | 1637.97 (1082.37–2193.43)(n = 95) | .208 |
| Decreased C3 | 7733.96 (6569.79–9730.53)(n = 19) | 7768.46 (5935.96–9172.86)(n = 57) | .483 | 1637.97 (1064.52–2859.11)(n = 49) | 1637.97 (1099.37–2317.67)(n = 85) | .794 |
| Decreased C4 | 7528.08 (6569.79–9730.53)(n = 19) | 7774.83 (6029.45–9317.36)(n = 57) | .938 | 1516.26 (1006.43–4011.42)(n = 39) | 1637.97 (1149.45–2275.12)(n = 95) | .833 |
| Anti-ANA | 7745.04 (6216.58–9570.48)(n = 69) | 7390.89 (5949.19–9381.84)(n = 7) | .907 | 1637.97 (1100.32–2333.19)(n = 126) | 1616.68 (919.24–2443.39)(n = 8) | .693 |
| Anti-dsDNA Ab | 9035.21 (6715.75–10,283.27)(n = 22) | 7438.64 (5942.57–8847.72)(n = 54) | .062 | 1729.1 (1168–3298.87)(n = 42) | 1572.21 (1065.69–2254.69)(n = 92) | .218 |
| Anti-Sm Ab | 6109.72 (4789.18–7774.83)(n = 11) | 7968.16 (6454.51–9570.48)(n = 65) | .039 | 1446.36 (1165.71–3008.59)(n = 28) | 1637.97 (1062.12–2387.95)(n = 106) | .904 |
In the discovery cohort, we found that the level of plasma sTNF-RI was positively correlated with SLEDAI, serum creatinine, urea, uric acid, systolic blood pressure and total cholesterol, while it was inversely correlated with eGFR-EPI, eGFR-MDRD and serum albumin. In the replication cohort, we found that the level of plasma sTNF-RI was positively correlated with SLEDAI, anti-dsDNA Ab, serum creatinine, urea, uric acid, systolic blood pressure, diastolic blood pressure, total cholesterol, triglycerides, and low-density lipoprotein, while it was inversely correlated with C3, eGFR-EPI, eGFR-MDRD and serum albumin. We also recalculated the SLEDAI of LN patients excluding the renal domains, and our data showed that plasma sTNF-RI was positively correlated with SLEDAI (Supplemental Material Table 2, Figures 3 and 4].
**Figure 3.:** *Correlation of plasma sTNF-RI and urine protein-to-creatinine ratio in LN patients of the discovery cohort and in the replication cohort.*
To assess the association of plasma sTNF-RI concentrations with renal function decline, we divided LN patients with eGFR-EPI < 90 mL/min/1.73 m2 into two groups: chronic acute renal failure (CRF) groups and acute renal failure (ARF) groups. In the replication cohort, plasma sTNF-RI levels were significantly higher in the ARF group than in the CRF group (25 vs. 14, 3837.05 (2407.99–4858.57) vs. 2136.66 (1513.4–4003.82, $$p \leq .007$$).
One patient in the discovery cohort and 1 patient in the replication cohort lacked physical urine examination because of anuria from end-stage renal disease, and 5 patients in the discovery cohort and 10 patients in the replication cohort lacked urine spot urine protein-to-creatinine ratio. Our data showed that in the discovery cohort, LN patients who had urine protein greater than 1+ showed a higher incidence of plasma sTNF-RI levels [17 vs. 42, 9578.65 (7329.79–11,021.64) vs. 7771.65 (6069.59–8925.46), $$p \leq .016$$]. A more detailed analysis demonstrated that the level of plasma sTNF-RI was positively correlated with the spot urine protein-to-creatinine ratio (Spearman’s Rho correlation = 0.308, $$p \leq .022$$). This finding was confirmed in our replication cohort. In the replication cohort, LN patients who had urine protein greater than 1+ showed a higher incidence of plasma sTNF-RI levels [43 vs. 72, 2747.84 (1653.2–4547.11) vs. 1412.44 (1085.55–1861.94), $p \leq .001$]. The level of plasma sTNF-RI was positively correlated with the urine protein-to-creatinine ratio (Spearman’s Rho correlation = 0.577, $p \leq .001$) (Figure 3).
## Development of a disease diagnosis model for LN patients from non-LNs and active SLE from inactive SLE
To further evaluate the risk factors for LN in SLE patients, we performed univariate logistic regression in the validation cohort, and the model results showed that high TNF-RI, high SBP, low C4 and high UA were risk factors for LN in SLE patients. These factors were included in the multivariate logistic regression model, and the results showed that high TNF-RI was an independent risk factor for LN in SLE patients (Table 3). The ROC results suggested that the use of TNF-RI to distinguish LN from SLE patients had good efficacy [the value of the area under the ROC curve (AUC) 0.872, $95\%$ CI 0.794–0.950, cutoff value 949.47 pg/ml, sensitivity $93.1\%$, and specificity $66.7\%$] (Figure 4).
**Figure 4.:** *ROC of TNF-RI in differentiating LN patients from SLE without renal impairment patients. ROC: receiver operating characteristic; AUC: the value of the area under the ROC curve.* TABLE_PLACEHOLDER:Table 3.
To identify the risk factors for SLE patients with high disease activity (SLEDAI > 9), we performed univariate logistic regression in the validation cohort. The model results showed that high TNF-RI, high SBP, high DBP, high Scr, low C3, low C4, high UA and anti-dsDNA positivity were risk factors for SLE patients with high disease activity (SLEDAI > 9). In the multivariate logistic regression model, the results showed that high TNF-RI, high DBP, high serum creatinine, low C4 and anti-ds-DNA positivity were independent risk factors for active SLE patients (Table 4). We depicted a nomogram to visualize the model (Figure 5(A)). Next, we plotted ROC curves to assess relative accuracies (Figure 5(B)). The results showed that the ROC curve based on the obtained potential risk factors identified by multivariate logistic regression analysis had the best efficiency, with an AUC of 0.889 ($95\%$ CI 0.830–0.948), a specificity of $85.7\%$ and a sensitivity of $78\%$. The calibration curve indicated that the model had great calibration capability (Figure 5(C)). The results of decision curve analysis (DCA) showed that the established nomogram had the best net benefit and a wide range of threshold probabilities (Figure 5(D)).
**Figure 5.:** *(A) Nomogram based on the active SLE diagnosis model; (B) ROC curves comparing the nomogram (a) with anti-dsDNA (b), TNF-RI (c), serum creatinine (Scr) (d) and diastolic blood pressure (DBP) (e); (C) calibration curve of the nomogram; (D) DCA of the nomogram. ROC: receiver operating characteristic; AUC: the value of the area under the ROC curve; DCA: decision curve analysis.* TABLE_PLACEHOLDER:Table 4.
## Renal TNF-RI expression increased in LN patients
IHC results showed that TNF-RI was present predominantly in the nucleus. The expression levels of TMF-RI among different types of LN and the control group are listed in Figure 6. In the normal kidney and type II LN, TNF-RI was faintly stained in the tubulointerstitium and the glomeruli. TNF-RI in the glomeruli of patients with proliferative LN increased significantly. In contrast, the TNF-RI-positive area was significantly higher in the renal tubules of patients with type III, IV and V LN. ( Figure 6).
**Figure 6.:** *Renal TNF-RI expression in LN patients and controls. (A) Images show TNF-RI immunostaining in the glomeruli and tubulointerstitium. A representative image was shown (scale bar 50 µm); The positive area was quantified with different types and healthy control in glomerulus (B) and renal tubule (C).*p < .05、**p < .01.*
## Discussion
We performed a discovery-replication study to investigate the association between plasma sTNF-RI levels and SLE patients in the Chinese population. Our data showed that plasma sTNF-RI could distinguish between SLE patients with and without renal involvement. Moreover, plasma sTNF-RI was closely related to disease activity and renal function in SLE patients and the level of urine protein in LN patients.
Previous studies reported that TNF-RI could be a new biomarker for early renal decline in patients with type 2 diabetes and that serum sTNF-RI was correlated with plasma creatinine values in patients with sepsis syndrome [16–19], but relatively few investigations have evaluated the function of TNF-RI in patients with LN. As the predominant mediator of TNF signaling, increasing evidence supported the important role of TNF-RI signaling in the pathogenesis of SLE [20]. TNF-RI is expressed in a wide variety of cells, and it is present primarily in glomeruli and peritubular endothelial cells [21]. TNF-RI can promote both proinflammatory and proapoptotic functions, while chronic inflammation is postulated to be involved in deteriorating renal function [22,23]. A large body of evidence suggests that serum sTNF-RI is associated with inflammatory renal disease [24–26]. A previous study reported that serum sTNF-RI levels were associated with renal interstitial fibrosis in patients with glomerulopathies [27]. High levels of TNF-RI in the circulation could activate the production of proinflammatory cytokines and chemokines, which may induce direct renal injury [28]. Previous studies reported higher levels of urinary and serum TNF-RI in LN patients than in HCs and SLE patients. Moreover, TNF-RI was strongly correlated with disease activity in SLE patients and the level of urine protein in LN patients [9,29]. In our study, we set more stringent emission standards for renal impairment. We excluded patients who presented proteinuria during the course of the disease in the SLE without renal impairment group, and we still obtained the same result. Moreover, TNF-RI was associated with renal damage and disease activity after adjusting for potential confounding factors. Our data also showed that plasma sTNF-RI levels were significantly higher in LN patients with ARF than in LN patients with CRF. Furthermore, sTNF-RI could reportedly contribute directly to microvascular kidney injury, which in turn can lead to a sharp decline in renal function. The specific mechanism remains to be further studied. A particularly interesting novel finding was that plasma sTNF-RI levels and urine protein in LN patients were positively correlated. sTNF-RI is excreted in urine, so higher levels of TNF-RI may also indicate glomerular hypofiltration. However, an important consideration was that we also found such differences in the levels of plasma sTNF-RI between SLE patients without renal involvement and HCs, and the urine protein level of the latter was similar to that of HCs. We speculated that plasma TNF-RI might represent renal dysfunction more efficiently and sensitively than urine protein. This finding still merits further evaluation. Moreover, plasma sTNF-RI showed high specificity and sensitivity for the diagnosis of renal involvement and disease activity in SLE patients in our data.
As in previous reports [30], we found that glomerular TNF-RI was strongly increased in proliferative LN. We also observed that TNF-RI expression was increased, especially in renal tubular epithelial cells, in LN patients. Tubular damage plays an important role in the pathophysiology and progression of LN. Proximal renal tubular epithelial cells (PTECs) play an active role in the initiation of immune-mediated injury in LN [31]. However, the role of TNF-RI in the renal tubules has been neglected. Further studies are needed.
Our study also had some limitations. First, it is a monocentric trial with a small sample size. Second, because of the lower sensitivity of ACR criteria, patients with very recent onset of the disease or with less common manifestations may be missed.
As described previously, we suggest that plasma sTNF-RI could be used in clinics as a noninvasive and useful biomarker for SLE disease activity and prediction of renal involvement.
## Author contributions
XR.L collected the samples, performed the data and wrote this manuscript. YY.Q, YF.Z, Y.C helped in the samples collection and data analysis. ZZ.Z designed the entire study, the provided partial financial support and expert advice. All the authors mentioned had read and approved this final version of the manuscript.
## Disclosure statement
We declare that this study was conducted in the absence of any financial or commercial relationships that could be construed as a potential conflict of interest.
## Data availability statement
The data that support the findings of this study are available on request from the first author and corresponding author underlying reasonable request.
## References
1. Durcan L, O'Dwyer T, Petri M.. **Management strategies and future directions for systemic lupus erythematosus in adults**. *Lancet* (2019) **393** 2332-2343. PMID: 31180030
2. Jakes RW, Bae SC, Louthrenoo W. **Systematic review of the epidemiology of systemic lupus erythematosus in the Asia-Pacific region: prevalence, incidence, clinical features, and mortality**. *Arthritis Care Res* (2012) **64** 159-168
3. Almaani S, Meara A, Rovin BH.. **Update on lupus nephritis**. *Clin J Am Soc Nephrol* (2017) **12** 825-835. PMID: 27821390
4. Kiriakidou M, Ching CL.. **Systemic lupus erythematosus**. *Ann Intern Med* (2020) **172** Itc81-itc96. PMID: 32479157
5. Idborg H, Eketjäll S, Pettersson S. **TNF-α and plasma albumin as biomarkers of disease activity in systemic lupus erythematosus**. *Lupus Sci Med* (2018) **5** e000260. PMID: 29955370
6. Gane JM, Stockley RA, Sapey E.. **TNF-α autocrine feedback loops in human monocytes: the pro- and anti-Inflammatory roles of the TNF-α receptors support the concept of selective TNFR1 blockade in vivo**. *J Immunol Res* (2016) **2016** 1079851. PMID: 27747245
7. Borghi A, Verstrepen L, Beyaert R.. **TRAF2 multitasking in TNF receptor-induced signaling to NF-κB, MAP kinases and cell death**. *Biochem Pharmacol* (2016) **116** 1-10. PMID: 26993379
8. Sharapova TN, Romanova EA, Soshnikova NV. **Autoantibodies from SLE patients induce programmed cell death in murine fibroblast cells through interaction with TNFR1 receptor**. *Sci Rep* (2020) **10** 11144. PMID: 32636466
9. Adhya Z, El Anbari M, Anwar S. **Soluble TNF-R1, VEGF and other cytokines as markers of disease activity in systemic lupus erythematosus and lupus nephritis**. *Lupus* (2019) **28** 713-721. PMID: 31046570
10. Tan EM, Cohen AS, Fries JF. **The 1982 revised criteria for the classification of systemic lupus erythematosus**. *Arthritis Rheum* (1982) **25** 1271-1277. PMID: 7138600
11. Hochberg MC.. **Updating the American College of Rheumatology revised criteria for the classification of systemic lupus erythematosus**. *Arthritis Rheum* (1997) **40** 1725
12. Weening JJ, D'Agati VD, Schwartz MM. **The classification of glomerulonephritis in systemic lupus erythematosus revisited**. *Kidney Int* (2004) **65** 521-530. PMID: 14717922
13. Gladman DD, Ibañez D, Urowitz MB.. **Systemic lupus erythematosus disease activity index 2000**. *J Rheumatol* (2002) **29** 288-291. PMID: 11838846
14. Levey AS, Bosch JP, Lewis JB. **A more accurate method to estimate glomerular filtration rate from serum creatinine: a new prediction equation. Modification of diet in renal disease study group**. *Ann Intern Med* (1999) **130** 461-470. PMID: 10075613
15. Levey AS, Stevens LA, Schmid CH. **A new equation to estimate glomerular filtration rate**. *Ann Intern Med* (2009) **150** 604-612. PMID: 19414839
16. Schrauben SJ, Shou H, Zhang X. **Association of multiple plasma biomarker concentrations with progression of prevalent diabetic kidney disease: findings from the chronic renal insufficiency cohort (CRIC) study**. *J Am Soc Nephrol* (2021) **32** 115-126. PMID: 33122288
17. Nowak N, Skupien J, Smiles AM. **Markers of early progressive renal decline in type 2 diabetes suggest different implications for etiological studies and prognostic tests development**. *Kidney Int* (2018) **93** 1198-1206. PMID: 29398132
18. Froon AH, Bemelmans MH, Greve JW. **Increased plasma concentrations of soluble tumor necrosis factor receptors in sepsis syndrome: correlation with plasma creatinine values**. *Crit Care Med* (1994) **22** 803-809. PMID: 8181289
19. Bergquist M, Samuelsson L, Larsson A. **TNFR1, TNFR2, neutrophil gelatinase-associated lipocalin and heparin binding protein in identifying sepsis and predicting outcome in an intensive care cohort**. *Sci Rep* (2020) **10** 15350. PMID: 32948801
20. Chen G, Goeddel DV.. **TNF-R1 signaling: a beautiful pathway**. *Science* (2002) **296** 1634-1635. PMID: 12040173
21. Al-Lamki RS, Wang J, Vandenabeele P. **TNFR1- and TNFR2-mediated signaling pathways in human kidney are cell type-specific and differentially contribute to renal injury**. *FASEB J* (2005) **19** 1637-1645. PMID: 16195372
22. Feng YM, Thijs L, Zhang ZY. **Glomerular function in relation to circulating adhesion molecules and inflammation markers in a general population**. *Nephrol Dial Transplant* (2018) **33** 426-435. PMID: 28992257
23. Niewczas MA, Pavkov ME, Skupien J. **A signature of circulating inflammatory proteins and development of end-stage renal disease in diabetes**. *Nat Med* (2019) **25** 805-813. PMID: 31011203
24. Lee SM, Yang S, Cha RH. **Circulating TNF receptors are significant prognostic biomarkers for idiopathic membranous nephropathy**. *PLoS One* (2014) **9** e104354. PMID: 25098821
25. Bhatraju PK, Zelnick LR, Shlipak M. **Association of soluble TNFR-1 concentrations with long-term decline in kidney function: the multi-ethnic study of atherosclerosis**. *J Am Soc Nephrol* (2018) **29** 2713-2721. PMID: 30287518
26. Srivastava A, Schmidt IM, Palsson R. **The associations of plasma biomarkers of inflammation with histopathologic lesions, kidney disease progression, and mortality-the Boston kidney biopsy cohort study**. *Kidney Int Rep* (2021) **6** 685-694. PMID: 33732983
27. Sonoda Y, Gohda T, Suzuki Y. **Circulating TNF receptors 1 and 2 are associated with the severity of renal interstitial fibrosis in IgA nephropathy**. *PLoS One* (2015) **10** e0122212. PMID: 25860248
28. Navarro JF, Mora-Fernández C.. **The role of TNF-alpha in diabetic nephropathy: pathogenic and therapeutic implications**. *Cytokine Growth Factor Rev* (2006) **17** 441-450. PMID: 17113815
29. Zheng CZ, Yan WW, Luo YL. **Value of sTNF-R1 and linc0597 as indicators for disease activity and diagnosis of lupus nephritis**. *Eur Rev Med Pharmacol Sci* (2020) **24** 5582-5591. PMID: 32495893
30. Aten J, Roos A, Claessen N. **Strong and selective glomerular localization of CD134 ligand and TNF receptor-1 in proliferative lupus nephritis**. *J Am Soc Nephrol* (2000) **11** 1426-1438. PMID: 10906156
31. Hill GS, Delahousse M, Nochy D. **Proteinuria and tubulointerstitial lesions in lupus nephritis**. *Kidney Int* (2001) **60** 1893-1903. PMID: 11703608
|
---
title: Development of novel anilinoquinazoline-based carboxylic acids as non-classical
carbonic anhydrase IX and XII inhibitors
authors:
- Zainab M. Elsayed
- Hadia Almahli
- Alessio Nocentini
- Andrea Ammara
- Claudiu T. Supuran
- Wagdy M. Eldehna
- Sahar M. Abou-Seri
journal: Journal of Enzyme Inhibition and Medicinal Chemistry
year: 2023
pmcid: PMC10035947
doi: 10.1080/14756366.2023.2191163
license: CC BY 4.0
---
# Development of novel anilinoquinazoline-based carboxylic acids as non-classical carbonic anhydrase IX and XII inhibitors
## Abstract
As part of our ongoing endeavour to identify novel inhibitors of cancer-associated CA isoforms IX and XII as possible anticancer candidates, here we describe the design and synthesis of small library of 2-aryl-quinazolin-4-yl aminobenzoic acid derivatives (6a–c, 7a–c, and 8a–c) as new non-classical CA inhibitors. On account of its significance in the anticancer drug discovery and in the development of effective CAIs, the 4-anilinoquinazoline privileged scaffold was exploited in this study. Thereafter, the free carboxylic acid functionality was appended in the ortho (6a–c), meta (7a–c), or para-positon (8a–c) of the anilino motif to furnish the target inhibitors. All compounds were assessed for their inhibitory activities against the hCA I, II (cytosolic), IX, and XII (trans-membrane, tumour-associated) isoforms. Moreover, six quinazolines (6a–c, 7b, and 8a–b) were chosen by the NCI-USA for in vitro anti-proliferative activity evaluation against 59 human cancer cell lines representing nine tumour subpanels.
## Introduction
Carbonic anhydrases (CA, EC 4.2.1.1) are ubiquitous metalloenzymes that play a crucial role in catalysing the reversible hydration reaction of carbon dioxide to bicarbonate and protons.1 This reaction, catalysed by Zn+2 ion, has a critical role in many physiological and pathological processes such as gluconeogenesis and tumorigenicity.2,3 So far, fifteen human CA (hCA) isoforms have been identified, with varying distributions across tissues and cells.4 *As a* result of the dysfunction of different hCA isoforms activities, a number of pathological repercussions might occur, featuring these hCA isoforms as interesting pharmacological targets for a variety of therapeutic approaches using small molecule CA inhibitors (CAIs).4 Thus, the pharmacological applications of CAIs are identified for the management of diverse disorders such as ophthalmologic problems,5 epilepsy,6 obesity7 and human malignancies.8 Sulphonamides and their sulfamides and sulfamate bioisosteres are considered as classical hCA inhibitors with a high affinity to the zinc ion in the active site.3 *It is* worth to mention that although identification of several chemotypes of CAIs, like coumarins, phenols, thiocarbamates, and carboxylates,9–11 only primary sulfonamide-tethered CAIs have been clinically used for glaucoma (such as acetazolamide and dorzolamide), and investigated in the clinical trials for the treatment of human malignancies (SLC-0111), Figure 1.12,13 These sulfonamide-tethered CAIs produce strong CA inhibition, however, a number of them lack the necessary isoform selectivity. So, the design and synthesis of new nonclassical CAIs stands out as a promising strategy to discover effective and isoform-selective CAIs for the management of different diseases. The carboxylic acid-based derivatives represent an important non-classical CAIs chemotype that can exert the CA inhibitory effect through different modes of action, such as anchoring to the zinc-bound water-hydroxide ion through H-bonding, or direct binding to the catalytic zinc displacing bound water-hydroxide anion.14–16
**Figure 1.:** *Structure of acetazolamide, dorzolamide, SLC-0111, non-classical CAIs (I–III), and the target inhibitors (6, 7, and 8).*
In the few last years, we have reported several carboxylic acid-tethered small molecules as new CAIs.17–19 A novel series of benzofuran-based carboxylic acids was described as promising CA inhibitors in 2020.20 Among these benzofuran derivatives, compound I (Figure 1) with a meta-benzoic acid moiety inhibited hCA IX at a submicromolar concentration (KI = 0.79 μM), as well as exerted good hCA XII inhibitory activity (KI = 2.3 μM). Also in the same year, we have developed a small library of methylthiazolo[3,2-a]benzimidazole-based carboxylic acid derivatives as novel CA inhibitors.21 In particular, compound II (Figure 1) effectively suppressed CA isoforms IX and XII with inhibition constants equal 0.83 μM and 2.4 μM, respectively. Furthermore, we identified a new series of non-classical CA inhibitors that incorporates enaminone-based carboxylic acids.22 Compound III (Figure 1) endowed with a para-benzoic acid motif showed submicromolar hCA IX inhibitory activity (KI = 0.92 µM) and good hCA XII inhibitory activity (KI = 1.1 µM).
Based on the findings described above, and as part of our ongoing endeavour to identify novel inhibitors of cancer-associated CA isoforms IX and XII as possible anticancer candidates,23–29 here we describe the design and synthesis of a small library of 2-aryl-quinazolin-4-yl aminobenzoic acid derivatives (6a–c, 7a–c, and 8a–c) as new non-classical CA inhibitors (Figure 1). On account of its significance in the anticancer drug discovery and development,30–33 and in the development of effective CAIs,34,35 the 4-anilinoquinazoline privileged scaffold was exploited in this study. Thereafter, the free carboxylic acid functionality was appended in the ortho (6a–c), meta (7a–c), or para-positon (8a–c) of the anilino motif to furnish the target inhibitors.
All the newly synthesised quinazoline-based carboxylic acid derivatives (6a–c, 7a–c, and 8a–c) were assessed for their inhibitory activities against the hCA I, II (cytosolic), IX and XII (trans membrane, tumour associated) isoforms by the stopped-flow CO2 hydrase assay. Moreover, six quinazolines (6a–c, 7b, and 8a–b) were chosen by the NCI-USA for in vitro anti-proliferative activity evaluation against 59 human cancer cell lines representing nine tumour subpanels.
## Chemistry
Melting points (°C, uncorrected) were determined using a Stuart melting point apparatus. The IR spectra (KBr) were recorded on a SHIMADZU FT/IR spectrometer. The NMR spectra recorded by BRUKER 400 MHz NMR spectrometers using DMSO-d6 as the solvent. Chemical shifts were reported in parts per million (δ), and coupling constants (J) expressed in Hertz. 1H and 13C spectra were run at 400 and 101 MHz, respectively. Microanalytical data (C, H, and N) were obtained by FLASH 2000 CHNS/O analyser.
The synthetic strategy to develop the target 2-aryl-quinazolin-4-yl aminobenzoic acid derivatives (6a–c, 7a–c, and 8a–c) were represented in Schemes 1 and 2. Synthesis of intermediates (3a–c) was carried out by reacting different aldehydes (2a–c) with anthranilamide [1] in an aqueous solution of FeCl3. The key intermediates (4a–c) were then synthesised via a chlorination reaction of quinazolinone derivatives (3a–c) with phosphrous oxychloride in the presence of the catalytic amount of N,N-dimethylformamide (Scheme 1).
**Scheme 1.:** *Synthesis of chloroquinazolines (4a–c): Reaction conditions (i) FeCl3/H2O/heating 80 °C/3h, (ii) POCl3/N,N-dimethylformamide (cat.)/heating 90 °C/4h.*
The target 2-aryl-quinazolin-4-yl aminobenzoic acids (6a–c, 7a–c, and 8a–c) were obtained, with a yield of 65–$86\%$, by reacting 2-aryl-4-chloroquinazoline derivatives (4a–c) with aminobenzoic acid derivatives (5a–c) in refluxing isopropanol containing few drops of HCl (Scheme 2).
**Scheme 2.:** *Synthesis of 2-aryl-quinazolin-4-yl aminobenzoic acids (6a–c, 7a–c and 8a–c): Reaction conditions (i) Isopropanol/HCl (cat.)/reflux/2h.*
The target quinazoline derivatives (6a–c, 7a–c and 8a–c) were structurally confirmed by spectral and elemental analyses. The 1H NMR spectra of all compounds revealed a singlet signal around δ 11.70–12.34 ppm due to the proton of the NH group. Moreover, all compounds showed two doublet signals in the aromatic region around δ 7.49–8.40 and 8.61–9.03 ppm that are attributable to H5 and H8 of quinazoline moiety, respectively. In addition, 1H NMR spectra for derivatives (6a–b, 7a–b and 8a–b) showed another singlet signal for the CH3 group at the range of δ 2.41–2.43 ppm, whereas, the 1H NMR spectra for (6c, 7c and 8c) disclosed the singlet signal of the OCH3 group around δ 3.87–3.90 ppm. One the other hand, 13C NMR spectra for the target quinazoline derivatives confirmed the presence of the carboxylic C=O functionality at δ 162–170 ppm. Furthermore, 13C NMR spectra for compounds (6a–b, 7a–b, and 8a–b) showed a signal at δ 21.41–21.70 ppm for the CH3 carbon, whereas spectra of compounds (6c, 7c, and 8c) displayed a signal at δ 56.02–56.25 ppm for the OCH3 carbon.
## General procedures for the synthesis of 2-arylquinazolin-4(3H)-one derivatives (3a–c)
An aqueous solution of ferric chloride (5.4 g, 20 mmol) was added to a mixture of anthranilamide 1 (2.72 g, 20 mmol) and the appropriate aldehyde derivative 2a–c (20 mmol).36 The mixture was heated at 80 °C for 3 h. After completion of the reaction, as indicated by TLC (n-hexane: ethyl acetate 1:1), the formed solid was filtrated, washed with water (4 × 5 ml), dried, and finally recrystallized from dioxane to produce 2-arylquinazolin-4(3H)-ones 3a–c.
## General procedures for the synthesis of 2-aryl-4-chloroquinazolines (4a–c)
To a suspension of 2-arylquinazolinones 3a–c (1eq) in phosphorus oxychloride (10 eq), a catalytic amount of DMF was added.36 The reaction mixture was then heated at 90 °C for 4h. After cooling, the mixture was added drop-wise to ice-water with stirring, neutralised by ammonium hydroxide, and extracted by methylene chloride. The organic layer was washed with cold water, dried over anhydrous Na2SO4, and evaporated in vacuo. The obtained solid was crystallised from isopropanol to afford the key 2-aryl-4-chloroquinazolines intermediates 4a–c.
## Synthesis of 2/3/4-((2-arylquinazolin-4-yl)amino)benzoic acid derivatives (6a–c, 7a–c and 8a–c)
To a stirred solution of 4-chloroquinazoline derivatives 4a–c (1 mmol) in refluxing isopropanol (5 ml) containing a few drops of HCl, the appropriate aminobenzoic acid derivative 5a–c (1 mmol) was added. The reaction mixture was heated under reflux for 2 h. The solid formed upon cooling was collected by filtration, dried, and recrystallized from ethanol to afford the target quinazolines (6a–c, 7a–c, and 8a–c).
## 2-((2-(m-Tolyl)quinazolin-4-yl)amino)benzoic acid (6a)
White crystals, ($67\%$) yield; m.p. 198–200 °C; IR (KBr) νmax/cm−1; 1H NMR (DMSO-d6): 2.42 (s, 3H, CH3), 7.47–7.51 (m, 3H, Ar-H), 7.80 (t, 1H, H-6 quinazoline, $J = 7.2$ Hz), 7.86 (t, 1H, H-7 quinazoline, $J = 7.6$ Hz), 8.09–8.16 (m, 3H, Ar-H), 8.24–8.26 (m, 2H, Ar-H), 8.38 (d, 1H, H-5, quinazoline, $J = 8.4$ Hz), 8.61 (d, 1H, H-8 quinazoline, $J = 8.0$ Hz), 12.34 (s, 1H, NH); 13C NMR (DMSO-d6): 21.47 (CH3), 113.31, 122.0, 124.02, 125.95, 126.68, 126.85, 128.93, 129.36, 130.10, 131.54, 132.34, 133.78, 134.24, 136.35, 138.06, 138.74, 157.99, 159.29, 168.84 (C=O); Anal. Calcd. For: C22H17N3O2 (355.40): C, 74.35; H, 4.82; N, 11.82; Found: C, 74.15; H, 4.79; N, 11.85.
## 2-((2-(p-Tolyl)quinazolin-4-yl)amino)benzoic acid (6b)
White crystals, ($65\%$) yield; m.p. 200–202 °C; IR (KBr) νmax/cm−1; 1H NMR (DMSO-d6): 2.41 (s, 3H, CH3), 7.41 (d, 2H, Ar-H, $J = 8.0$ Hz), 7.47 (t, 1H, Ar-H, $J = 7.2$ Hz), 7.79 (t, 1H, H-6 quinazoline, $J = 7.6$ Hz), 7.85 (t, 1H, Ar-H, $J = 7.6$ Hz), 8.08–8.13 (m, 2H, Ar-H), 8.23–8.25 (m, 1H, Ar-H), 8.28 (d, 2H, Ar-H, $J = 8.4$ Hz), 8.41 (d, 1H, H-5 quinazoline, $J = 7.6$ Hz), 8.61 (d, 1H,H-8 quinazoline, $J = 8.4$ Hz), 12.34 (s, 1H, NH); 13C NMR (DMSO-d6): 21.67 (CH3), 113.21, 124.08, 126.03, 126.77, 128.85, 129.71, 130.04, 131.53, 133.79, 136.39, 144.36, 157.78, 159.32, 168.80 (C=O); Anal. Calcd. For: C22H17N3O2 (355.40): C, 74.35; H, 4.82; N, 11.82; Found: 74.25; H, 4.80; N, 11.84.
## 2-((2–(4-Methoxyphenyl)quinazolin-4-yl)amino)benzoic acid (6c)
White crystals, ($72\%$) yield; m.p. 204–206 °C; IR (KBr) νmax/cm−1; 1H NMR (DMSO-d6): 3.87 (s, 3H, OCH3), 7.15 (d, 2H, Ar-H, $J = 8.8$ Hz), 7.49 (t, 1H, Ar-H, $J = 7.2$ Hz), 7.79 (t, 1H, H-6 quinazoline, $J = 8.8$ Hz), 7.83 (t, 1H, Ar-H, $J = 8.8$ Hz), 8.08–8.20 (m, 3H, Ar-H), 8.39 (d, 2H, Ar-H, $J = 9.2$ Hz), 8.45 (d, 1H, H-5 quinazoline, $J = 8.4$ Hz), 8.62 (d, 1H, H-8 quinazoline, $J = 8.0$ Hz), 12.28 (s, 1H, NH); 13C NMR (DMSO-d6): 56.02 (OCH3), 56.25, 114.56, 115.03, 120.92, 126.45, 126.99, 130.34, 131.34, 131.84, 133.75, 135.30, 136.57, 159.33, 162.52, 162.70; Anal. Calcd. For: C22H17N3O3 (371.40): C, 71.15; H, 4.61; N, 11.31; Found: 71.37; H, 4.60; N, 11.25.
## 3-((2-(m-Tolyl)quinazolin-4-yl)amino)benzoic acid (7a)
Yellow crystals, ($72\%$) yield; m.p. 248–250 °C; IR (KBr) νmax/cm−1; 1H NMR (DMSO-d6):2.42 (s, 3H, CH3), 7.47–7.53 (m, 2H, Ar-H), 7.63 (t, 1H, H-6 quinazoline, $J = 8.0$ Hz), 7.80 (t, 1H, H7 quinazoline, $J = 7.6$ Hz), 7.89 (d, 1H, Ar-H, $J = 7.6$ Hz), 8.07–8.11 (m, 2H, Ar-H), 8.30–8.31 (m, 2H, Ar-H) 8.40 (d, 1H, H-5 quinazoline, $J = 8.4$ Hz), 8.75 (s, 1H, Ar-H), 9.03 (d, 1H, H-8 quinazoline, $J = 8.0$ Hz),11.82 (s, 1H, NH); 13C NMR (DMSO-d6): 21.41 (CH3), 25.96, 62.49, 113.26, 125.15, 125.52, 126.96, 127.27, 128.57, 129.31, 130.34, 131.64, 134.48, 136.38, 137.81, 138.96, 157.61, 159.31, 167.49 (C=O); Anal. Calcd. For: C22H17N3O2 (355.40): C, 74.35; H, 4.82; N, 11.82; Found: 74.38; H, 4.85; N, 11.78.
## 3-((2-(p-Tolyl)quinazolin-4-yl)amino)benzoic acid (7b)
Yellow crystals, ($67\%$) yield; m.p. 220–223 °C; IR (KBr) νmax/cm−1; 1H NMR (DMSO-d6):2.42 (s, 3H, CH3), 7.40 (d, 2H, Ar-H, $J = 8.0$ Hz), 7.64 (t, 1H, Ar-H, $J = 7.6$ Hz), 7.80 (t, 1H, H-6 quinazoline, $J = 7.6$ Hz), 7.89 (d, 1H, Ar-H, $J = 7.6$ Hz), 8.07 (t, 1H, H-7 quinazoline, $J = 8.0$ Hz), 8.12 (d, 1H, Ar-H, $J = 8.0$ Hz), 8.37 (d, 2H, Ar-H, $J = 8.0$ Hz), 8.40 (d, 1H, H-5 quinazoline, $J = 8.8$ Hz), 8.69 (s, 1H, H-2, Ar-H), 9.00 (d, 1H, H-8 quinazoline, $J = 8.4$ Hz), 11.79 (s, 1H, NH); 13C NMR (DMSO-d6): 21.69, 25.97, 62.48, 113.20, 120.96, 125.15, 125.51, 127.28, 128.49, 128.73, 129.44, 129.87, 130.05, 131.70, 136.38, 137.82, 144.55, 157.46, 159.35, 167.41; Anal. Calcd. For: C22H17N3O2 (355.40): C, 74.35; H, 4.82; N, 11.82; Found: 74.51; H, 4.79; N, 11.86.
## 3-((2–(4-Methoxyphenyl)quinazolin-4-yl)amino)benzoic acid (7c)
Off white crystals, ($75\%$) yield; m.p. 244–246 °C; IR (KBr) νmax/cm−1; 1H NMR (DMSO-d6): 3.90 (s, 3H, OCH3), 7.18 (d, 2H, Ar-H, $J = 8.8$ Hz), 7.66 (t, 1H, H-6 quinazoline, $J = 8.0$ Hz), 7.77 (d, 1H, Ar-H, $J = 7.2$ Hz), 7.81 (t, 1H, H-7 quinazoline, $J = 8.0$ Hz), 7.91 (d, 1H, Ar-H, $J = 7.6$ Hz), 8.10 (d, 1H, Ar-H, $J = 8.0$ Hz), 8.37 (d, 1H, H-5 quinazoline, $J = 8.4$ Hz), 8.48 (d, 2H, Ar-H, $J = 8.8$ Hz), 8.69 (s, 1H, H-2, Ar-H), 8.93(d, 1H, H-8 quinazoline, $J = 8.4$ Hz), 11.70 (s, 1H, NH); 13C NMR (DMSO-d6): 56.25 (OCH3), 113.02, 115.00, 123.62, 125.06, 125.53, 125.99, 127.35, 128.36, 128.79, 129.54, 130.40, 131.76, 132.06, 132.49, 136.47, 137.82, 157.04, 159.28, 164.16, 167.07, 167.42; Anal. Calcd. For: C22H17N3O3 (371.40): C, 71.15; H, 4.61; N, 11.31; Found: C, 71.27; H, 4.64; N, 11.26.
## 4-((2-(m-Tolyl)quinazolin-4-yl)amino)benzoic acid (8a)
White crystals, ($78\%$) yield; m.p. 201–204 °C; IR (KBr) νmax/cm−1; 1H NMR (DMSO-d6):2.43 (s, 3H, CH3), 7.52–7.53 (m, 2H, Ar-H), 7.81 (t, 1H, H-6 quinazoline, $J = 7.2$ Hz), 8.05–8.12 (m, 5H, Ar-H), 8.20–8.22 (m, 1H, Ar-H), 8.28 (s, 1H, Ar-H), 8.39 (d, 1H, H-5 quinazoline, $J = 8.4$ Hz), 9.01 (d, 1H, H-8 quinazoline, $J = 8.0$ Hz), 11.79 (s, 1H, NH); 13C NMR (DMSO-d6): 21.50 (CH3), 113.39, 124.27, 125.17, 127.03, 128.33, 128.59, 129.48, 130.11, 130.37, 132.13, 134.38, 136.43, 138.84, 141.66, 157.78, 159.42, 167.30 (C=O); Anal. Calcd. For: C22H17N3O2 (355.40): C, 74.35; H, 4.82; N, 11.82; Found: C, 74.32; H, 4.81; N, 11.86.
## 4-((2-(p-Tolyl)quinazolin-4-yl)amino)benzoic acid (8b)
Yellow crystals, ($70\%$) yield; m.p. 306–308 °C; IR (KBr) νmax/cm−1; 1H NMR (DMSO-d6):2.43 (s, 3H, CH3), 7.46 (d, 2H, Ar-H, $J = 8.0$ Hz), 7.82 (t, 1H, H-6 quinazoline, $J = 8.0$ Hz), 8.04 (d, 2H, Ar-H, $J = 8.8$ Hz), 8.09 (d, 2H, Ar-H, $J = 4$ Hz), 8.11 (t, 1H, H-7 quinazoline, $J = 2.8$ Hz), 8.32 (d, 2H, Ar-H, $J = 8.4$ Hz), 8.36 (d, 1H, H-5 quinazoline, $J = 8.4$ Hz), 8.96 (d, 1H, H-8 quinazoline, $J = 8.0$ Hz) 11.72 (s, 1H, NH); 13C NMR (DMSO-d6): 21.70, 113.35, 124.26, 125.10, 128.32, 128.51, 129.75, 130.18, 130.44, 136.45, 141.67, 144.41, 157.70, 159.44, 167.30; Anal. Calcd. For: C22H17N3O2 (355.40): C, 74.35; H, 4.82; N, 11.82; Found: C, 74.21; H, 4.80; N, 11.84.
## 4-((2–(4-Methoxyphenyl)quinazolin-4-yl)amino)benzoic acid (8c)
Off white crystals, ($78\%$) yield; m.p. >300 °C; IR (KBr) νmax/cm−1; 1H NMR (DMSO-d6): 3.90 (s, 3H, OCH3), 7.22 (d, 2H, Ar-H, $J = 8.8$ Hz,), 7.71 (d, 1H, Ar-H, $J = 8.8$ Hz), 7.81 (t, 1H, H-6 quinazoline, $J = 7.6$ Hz), 8.01 (d, 2H, Ar-H, $J = 8.4$ Hz), 8.09–8.13 (m, 3H, Ar-H and H-7 quinazoline), 8.42–8.47 (m, 3H, Ar-H and H-5 quinazoline), 8.97 (d, 1H, H-8 quinazoline, $J = 8.4$ Hz), 11.81 (s, 1H, NH); 13C NMR (DMSO-d6): 56.26, 113.06, 115.15, 120.53, 123.43, 124.56, 125.20, 128.44, 128.54, 130.45, 131.60, 132.08, 136.61, 141.48, 157.04, 159.39, 164.22, 167.29, 167.72; Anal. Calcd. For: C22H17N3O3 (371.40): C, 71.15; H, 4.61; N, 11.31; Found: C, 71.02; H, 4.59; N, 11.26.
## CA inhibitory assay
All the newly synthesised quinazoline-based carboxylic acid derivatives (6a–c, 7a–c and 8a–c) were assessed for their CA catalysed CO2 hydration activities against hCA isoforms I, II, IX and XII by the stopped flow CO2 hydrase assay as reported previously37–40 (Supporting Materials).
## In vitro NCI-59 cancer cell lines assays
The NCI-USA anticancer assays was performed utilising the NCI, Bethesda, Drug Evaluation Branch protocol,41–43 using the SRB cytotoxicity assay,44 as desribed earlier.45,46
## Carbonic anhydrase inhibition
All the newly synthesised quinazoline-based carboxylic acid derivatives (6a–c, 7a–c, and 8a–c) were assessed for their inhibitory activities against the hCA I, II (cytosolic), IX and XII (trans membrane, tumour associated) isoforms by the stopped-flow CO2 hydrase assay.37 Acetazolamide (AAZ) was used as a standard CA inhibitor. The data is summarised in Table 1.
**Table 1.**
| Unnamed: 0 | Unnamed: 1 | Unnamed: 2 | Unnamed: 3 | Unnamed: 4 | Unnamed: 5 | Unnamed: 6 |
| --- | --- | --- | --- | --- | --- | --- |
| | | | KI (µM)a | KI (µM)a | KI (µM)a | KI (µM)a |
| Cmpd | COOH | R | hCA I | hCA II | hCA IX | hCA XII |
| 6a | 2-COOH | 3-CH3 | >100 | 85.2 | 42.5 | 9.0 |
| 6b | 2-COOH | 4-CH3 | >100 | 48.2 | 34.4 | 7.1 |
| 6c | 2-COOH | 4-OCH3 | >100 | 83.1 | 46.4 | 8.9 |
| 7a | 3-COOH | 3-CH3 | >100 | 26.0 | 29.3 | 4.8 |
| 7b | 3-COOH | 4-CH3 | >100 | 41.7 | 24.2 | 0.91 |
| 7c | 3-COOH | 4-OCH3 | >100 | 85.8 | 31.6 | 0.48 |
| 8a | 4-COOH | 3-CH3 | 87.7 | 9.3 | 4.3 | 3.8 |
| 8b | 4-COOH | 4-CH3 | 73.2 | 3.9 | 1.6 | 0.42 |
| 8c | 4-COOH | 4-OCH3 | 66.3 | 4.6 | 4.5 | 0.25 |
| AAZ | – | – | 0.25 | 0.01 | 0.02 | 0.006 |
Only three of the tested quinazoline-based carboxylic acids (8a, 8b, and 8c) weakly inhibited the cytosolic hCA I isoform, with inhibition constants (KIs) equal 87.7, 73.2, and 66.3 µM, respectively, whereas quinazoline derivatives 6a–c and 7a–c could not inhibit hCA I up to 100 µM. These results revealed that grafting the carboxylic acids functionality at the para position (8a–c) could result in modest hCA I inhibitory activity, while shifting to ortho- (6a–c) or meta- (7a–c) positions resulted in the elimination of hCA I inhibitory activity (KIs > 100 M), Table 1.
The cytosolic hCA II was effectively inhibited by para-aminobenzoic acid-bearing quinazolines (8a–c) with KIs of 9.3, 3.9 and 4.6 µM, respectively, whereas, their ortho (6a–c) and meta (7a–c) regioisomers elicited modest inhibitory effects with inhibition constants spanning in the range of 26.0 − 85.8 µM. It is worth to mention that substitution of the 2-phenyl motif with a 4-methyl group, in series 8, led to compound 8b with the best hCA II inhibitory activity (KI = 3.9 µM).
Similar to the hCA I and hCA II inhibition profiles, the obtained KI values disclosed that the cancer-related hCA IX isoform was inhibited most effectively by para-aminobenzoic acid-bearing quinazolines (8a–c) with KIs equal 4.3, 1.6 and 4.5 µM, respectively. In addition, hCA IX was moderately affected by quinazolines decorated with ortho and meta aminobenzoic acid motifs with KIs ranging between 24.2 and 46.4 µM. The order of activities of target quinazoline-based carboxylic acids towards hCA IX was increased in the order of para isomers 8 (KIs: 1.6 − 4.5 µM) > meta isomers 7 (KIs: 24.2 − 31.6 µM) > ortho isomers 6 (KIs: 34.4 − 46.4 µM), Table 1. Regarding the impact of substitution on the 2-phenyl moiety, within series 6, 7, and 8, it was found that the order of hCA IX inhibitory activities was 4-methyl derivatives (6b, 7b, and 8b; KIs = 34.4, 24.2 and 1.6 µM) > 3-methyl derivatives (6a, 7a, and 8a; KIs = 42.5, 29.3 and 4.3 µM) > 4-methoxy derivatives (6c, 7c and 8c; KIs = 46.4, 31.6 and 4.5 µM), Table 1.
The second cancer-related isoform studied in this study is hCA XII, which is also the most vulnerable to the prepared molecules. All quinazoline-based carboxylic acids (6a–c, 7a–c, and 8a–c) exhibited good inhibition of hCA XII (KIs: 0.25 − 9.0 µM), as seen by the data in Table 1. In particular, the best hCA XII inhibitory effect was exerted by quinazoline 8c with a KI value equals 0.25 µM. Besides, quinazolines 7b, 7c, and 8b displayed also sub-micromolar inhibitory activity towards hCA XII with KI values 0.91, 0.48, and 0.42 µM, respectively. Similarly to the abovementioned deduced Structure-Activity Relationship (SAR) for hCA I, II, and IX isoforms, the order of hCA XII inhibitory activities was increased in the order of para isomers 8 (KIs: 0.25 − 3.8 µM) > meta isomers 7 (KIs: 0.48 − 4.8 µM) > ortho isomers 6 (KIs: 7.1–9 µM), Table 1. Also, it’s noteworthy that appending p-methoxyphenyl moiety at C-2 of quianzoline within series 7 and 8 (7c and 8c; KIs = 0.48 and 0.25 µM) resulted in a better hCA XII inhibitory activity than p-methylphenyl (7b and 8b; KIs = 0.91 and 0.42 µM) and m-methylphenyl (7a and 8a; KIs = 4.8 and 3.8 µM) moieties. The SAR for the inhibitory activity of the new quinazolines towards different hCA isoforms is summarised in Figure 2.
**Figure 2.:** *SARs summary for the hCA inhibitory activities of target quinazolines.*
As a result of the inhibitory profile for the reported quinazoline-based carboxylic acid derivatives (6a–c, 7a–c, and 8a–c) (Table 1), the selectivity index (SI) for each derivative was calculated and presented in Table 2. Regarding the selectivity towards CA IX and XII isoforms, most the examined quinazoline-based carboxylic acids (6a–c, 7a–c, and 8a–c) exhibited low to moderate selectivity, except compounds 7b and 7c that disclosed excellent selectivity towards hCA XII over hCA I (SI = 109.9 and 208.3, respectively), and hCA II (SI = 45.82 and 178.75, respectively), in addition, compounds 8b and 8c displayed outstanding selectivity towards hCA XII over hCA I (SI = 174.3 and 265.2, respectively), Table 2.
**Table 2.**
| Compound | I/IX | II/IX | I/XII | II/XII |
| --- | --- | --- | --- | --- |
| 6a | 2.4 | 2.0 | 11.1 | 9.47 |
| 6b | 2.9 | 1.4 | 14.1 | 6.79 |
| 6c | 2.2 | 1.79 | 11.2 | 9.34 |
| 7a | 3.4 | 0.89 | 20.8 | 5.42 |
| 7b | 4.1 | 1.72 | 109.9 | 45.82 |
| 7c | 3.2 | 2.72 | 208.3 | 178.75 |
| 8a | 20.4 | 2.16 | 23.1 | 2.45 |
| 8b | 45.8 | 2.44 | 174.3 | 9.29 |
| 8c | 14.7 | 1.02 | 265.2 | 18.4 |
| AAZ | 10.0 | 0.5 | 43.9 | 2.2 |
## In vitro anti-proliferative activity
The structures of all novel quinazoline based-carboxylic acids prepared in this study were submitted to the National Cancer Institute (NCI-USA), and six compounds (6a–c, 7b, and 8a–b) were chosen for in vitro anti-proliferative activity evaluation against fifty-nine human cancer cell lines representing nine tumour subpanels, according to Bethesda, Drug Evaluation Branch Protocol.41–43
## Preliminary single dose screening at 10 µM concentration
The anti-proliferative activities of the selected quinazoline derivatives (6a–c, 7b, and 8a–b) have been evaluated at single (10 μM) dose assay using the SRB protocol.44 The obtained data was presented as a mean-graph of the percentage growth of the various treated cancer cells and was displayed in Table 2 as the percentage growth inhibition (GI%) induced by the investigated compounds.
Examining the data in Table 3 revealed that the tested quinazoline-based-carboxylic acids (6a–c, 7b, and 8a–b) demonstrated diverse patterns of sensitivity and selectivity against the various NCI cancer cell panels. Quinazoline derivatives featuring ortho aminobenzoic acid (6a–c) showed excellent broad cell growth inhibitory activity (GI % mean = 63, 84, and 52, respectively) against most of all cancer cell lines, whereas compounds (7b and 8a–b) with meta and para-aminobenzoic acid moiety showed fair selective activity (GI % mean = 14, 16 and 20, respectively) towards certain cancer cell lines as shown in Table 3.
**Table 3.**
| Subpanel/cell line | Subpanel/cell line.1 | Compounda | Compounda.1 | Compounda.2 | Compounda.3 | Compounda.4 | Compounda.5 |
| --- | --- | --- | --- | --- | --- | --- | --- |
| Subpanel/cell line | Subpanel/cell line | 6a | 6b | 6c | 7b | 8a | 8b |
| Leukaemia | CCRF-CEM | 82 | 87 | 58 | 21 | 25 | 28 |
| Leukaemia | HL-60(TB) | 60 | 82 | 67 | – | – | 24 |
| Leukaemia | K-562 | 71 | 82 | 69 | 20 | – | 23 |
| Leukaemia | MOLT-4 | 65 | 89 | 64 | – | 22 | 24 |
| Leukaemia | RPMI-8226 | 71 | 83 | 80 | 29 | 30 | 45 |
| Leukaemia | SR | 77 | 81 | NT | 41 | 32 | 38 |
| Non-Small Cell Lung Cancer | A549/ATCC | 56 | 70 | 60 | – | 22 | – |
| Non-Small Cell Lung Cancer | EKVX | 79 | 87 | 66 | 41 | – | 28 |
| Non-Small Cell Lung Cancer | HOP-62 | 54 | 72 | 49 | – | – | – |
| Non-Small Cell Lung Cancer | HOP-92 | 105 | 120 | 78 | 29 | 29 | 25 |
| Non-Small Cell Lung Cancer | NCI-H226 | 64 | 85 | 47 | 25 | 52 | 62 |
| Non-Small Cell Lung Cancer | NCI-H23 | 79 | 104 | 38 | – | 52 | – |
| Non-Small Cell Lung Cancer | NCI-H322M | 52 | 66 | 24 | – | – | – |
| Non-Small Cell Lung Cancer | NCI-H460 | 81 | 109 | 89 | – | 22 | – |
| Non-Small Cell Lung Cancer | NCI-H522 | 40 | 67 | 63 | – | – | – |
| Colon cancer | COLO 205 | – | – | – | – | – | – |
| Colon cancer | HCC-2998 | 73 | 95 | 38 | – | 24 | 28 |
| Colon cancer | HCT-116 | 78 | 93 | 62 | 26 | – | 22 |
| Colon cancer | HCT-15 | 86 | 95 | 61 | 55 | 29 | 49 |
| Colon cancer | HT29 | 60 | 82 | 37 | – | – | – |
| Colon cancer | KM12 | 70 | 92 | 64 | 32 | 23 | 42 |
| Colon cancer | SW-620 | 65 | 86 | 61 | – | – | – |
| CNS cancer | SF-268 | 42 | 89 | 43 | 31 | 37 | 30 |
| CNS cancer | SF-295 | 73 | 94 | 60 | 34 | 36 | 42 |
| CNS cancer | SF-539 | 52 | 79 | 37 | 27 | – | 21 |
| CNS cancer | SNB-19 | 79 | 91 | 55 | 28 | 33 | 29 |
| CNS cancer | SNB-75 | 45 | 84 | 68 | – | – | – |
| CNS cancer | U251 | 73 | 82 | 66 | 21 | 27 | – |
| Melanoma | LOX IMVI | 64 | 91 | 50 | – | 29 | 43 |
| Melanoma | MALME-3M | 63 | 80 | 31 | – | – | – |
| Melanoma | M14 | 75 | 87 | 48 | – | – | – |
| Melanoma | MDA-MB-435 | 55 | 77 | 44 | – | – | – |
| Melanoma | SK-MEL-2 | 34 | 94 | 57 | – | – | – |
| Melanoma | SK-MEL-28 | 50 | 71 | 35 | – | – | – |
| Melanoma | SK-MEL-5 | 94 | 124 | 70 | – | 29 | 28 |
| Melanoma | UACC-257 | 51 | 71 | 28 | – | – | – |
| Melanoma | UACC-62 | 72 | 89 | 44 | – | – | – |
| Ovarian cancer | IGROV1 | 41 | 69 | 21 | – | – | – |
| Ovarian cancer | OVCAR-3 | 66 | 94 | 54 | – | – | 20 |
| Ovarian cancer | OVCAR-4 | 58 | 78 | 66 | 23 | – | 26 |
| Ovarian cancer | OVCAR-5 | 23 | 34 | 27 | – | – | – |
| Ovarian cancer | OVCAR-8 | 40 | 69 | 40 | – | – | – |
| Ovarian cancer | NCI/ADR-RES | 75 | 95 | 46 | – | – | 24 |
| Ovarian cancer | SK-OV-3 | 24 | 69 | 53 | – | – | – |
| Renal cancer | 786-0 | 75 | 81 | 48 | 24 | 29 | – |
| Renal cancer | A498 | 52 | 78 | 41 | – | 54 | – |
| Renal cancer | ACHN | 63 | 92 | 54 | 24 | – | 24 |
| Renal cancer | CAKI-1 | 48 | 79 | 53 | – | – | – |
| Renal cancer | SN12C | 59 | 78 | 32 | 23 | 42 | 51 |
| Renal cancer | TK-10 | – | 41 | 52 | – | – | – |
| Renal cancer | UO-31 | 69 | 87 | 25 | 51 | 54 | 58 |
| Prostate | PC-3 | 71 | 84 | 60 | – | – | 23 |
| Prostate | DU-145 | 50 | 81 | 43 | – | – | – |
| Breast cancer | MCF7 | 76 | 92 | 49 | – | – | 37 |
| Breast cancer | MDA-MB-231 | 34 | 65 | – | – | – | – |
| Breast cancer | HS 578 T | 84 | 109 | 72 | 34 | 48 | 32 |
| Breast cancer | BT-549 | 100 | 123 | 62 | – | 34 | 41 |
| Breast cancer | T-47D | 73 | 89 | 71 | – | 22 | 32 |
| Breast cancer | MDA-MB-468 | 115 | 126 | 57 | – | – | – |
| Mean GI % | Mean GI % | 63 | 84 | 52 | 14 | 16 | 20 |
| Number of sensitive cell lines | Number of sensitive cell lines | 56 | 58 | 56 | 21 | 25 | 30 |
In particular, quinazoline derivative 6b stood out as the most effective anti-proliferative compound (GI % mean = 84). Compound 6b exhibited excellent activity with GI% more than $75\%$ against the examined cancer cell lines from all subpanels, except non-small cell lung (A549, HOP-62, NCI-H322M and NCI-H522), colon (COLO 205), melanoma (SK-MEL-28 and UACC-257), ovarian (IGROV1, OVCAR-5, OVCAR-8 and SK-OV-3), renal (TK-10) and breast (MDA-MB-231) cancer cell lines. In addition, compound 6b showed good activity towards non-small cell lung (A549, HOP-62, NCI-H322M and NCI-H522), melanoma (SK-MEL-28 and UACC-257), ovarian (IGROV1, OVCAR-8, and SK-OV-3), breast (MDA-MB-231) cancer cell lines with GI% of 70, 72, 66, 67, 71, 71, 69, 69, 69 and $65\%$ respectively. It is noteworthy to mention that quinazoline derivative 6b had a lethal cytotoxic effect against non-small cell lung (HOP-92, NCI-H23, and NCI-H460), melanoma (SK-MEL-5), and breast (HS 578 T, BT-549, and MDA-MB-468) cancer cell lines with GI% equal 120, 104, 109,124, 109, 123 and $126\%$ respectively.
Moreover, quinazoline 6a disclosed a broad-spectrum anticancer effect against 56 cell lines representing all subpanels and emerged as the second most active compound in this assay (mean % GI = 63). Superiorly, quinazoline 6a exerted effective cell growth inhibitory activity with GI% more than $75\%$ against leukaemia (CCRF-CEM and SR), non-small cell lung (EKVX, NCI-H23 and NCI-H460), colon (HCT-116 and HCT-15), CNS (SNB-19), melanoma (MDA-MB-435 and SK-MEL-5), ovarian (NCI/ADR-RES), renal (786–0) and breast (MCF7, HS 578 T and BT-549) cancer cell lines. In addition, compound 6a possessed a lethal impact towards non-small cell lung (HOP-92) and breast (MDA-MB-468) cancer cell lines with GI% of 105 and $115\%$ respectively.
## In vitro NCI 5-dose assay
The preliminary screening data showed that quinazoline-carboxylic acid 6b (NSC: 835857) was the most active anticancer molecule in this study, with promising activity against numerous tumour cell lines. Thus, 6b was selected by NCI for further evaluations at a 5-doses (0.01–100 µM) level. Three dose-response parameters (GI50, TGI, and LC50) were calculated and displayed in Table 4.
**Table 4.**
| Cancer type/cells | Compound 6b (NSC: 835857) | Compound 6b (NSC: 835857).1 | Compound 6b (NSC: 835857).2 |
| --- | --- | --- | --- |
| Cancer type/cells | GI50 (µM) | TGI (µM) | LC50 (µM) |
| Leukaemia | Leukaemia | Leukaemia | Leukaemia |
| MOLT-4 | 11.7 | <100 | <100 |
| RPMI-8226 | 4.4 | <100 | <100 |
| K-562 | 7.05 | <100 | <100 |
| SR | 10.3 | <100 | <100 |
| CCRF-CEM | 5.7 | <100 | <100 |
| HL-60(TB) | 12.9 | 75.4 | <100 |
| Non-small cell lung cancer | Non-small cell lung cancer | Non-small cell lung cancer | Non-small cell lung cancer |
| HOP-92 | 2.9 | 29.3 | <100 |
| NCI-H226 | 10.2 | 54.2 | <100 |
| NCI-H522 | 4.4 | 27.1 | <100 |
| NCI-H322M | 13.6 | <100 | <100 |
| NCI-H460 | 8.5 | 27.7 | 84.0 |
| NCI-H23 | 15.7 | 44.4 | <100 |
| EKVX | 8.9 | 56.2 | <100 |
| HOP-62 | 10.8 | 33.7 | <100 |
| A549/ATCC | 8.31 | <100 | <100 |
| Colon cancer | Colon cancer | Colon cancer | Colon cancer |
| KM 12 | 13.7 | 81.2 | <100 |
| SW-620 | 18.2 | <100 | <100 |
| HT29 | 17.9 | <100 | <100 |
| HCT-15 | 5.8 | 28.7 | <100 |
| COLO 205 | 17.8 | 41.0 | 94.1 |
| HCC-2998 | 12.4 | 36.0 | <100 |
| HCT-116 | 10.1 | 36.1 | <100 |
| CNS cancer | CNS cancer | CNS cancer | CNS cancer |
| SNB-75 | 1.4 | 80.3 | <100 |
| U251 | 5.6 | 35.0 | <100 |
| SF-539 | 11.9 | 28.3 | 67.2 |
| SNB-19 | 8.1 | 81.7 | <100 |
| SF-295 | 7.04 | 27.0 | 86.0 |
| SF-268 | 14.3 | <100 | <100 |
| Melanoma | Melanoma | Melanoma | Melanoma |
| MDA-MB-435 | 13.9 | 87.2 | <100 |
| UACC-62 | 10.2 | 27.0 | 71.6 |
| M14 | 15.5 | 77.4 | <100 |
| UACC-257 | 15.3 | 90.3 | <100 |
| SK-MEL-5 | 5.3 | 20.3 | 57.5 |
| SK-MEL-28 | 15.9 | 97.4 | <100 |
| LOX IMVI | 10.2 | 31.6 | 97.7 |
| MALME-3M | 15.6 | 64.5 | <100 |
| SK-MEL-2 | 3.7 | 15.6 | 83.5 |
| Ovarian cancer | Ovarian cancer | Ovarian cancer | Ovarian cancer |
| IGROV1 | 19.1 | 87.1 | <100 |
| OVCAR-4 | 8.8 | <100 | <100 |
| OVCAR-5 | 21.8 | 68.9 | <100 |
| OVCAR-8 | 18.2 | <100 | <100 |
| NCI/ADR-RES | 16.8 | 98.5 | <100 |
| SK-OV-3 | 11.9 | 55.3 | <100 |
| Renal cancer | Renal cancer | Renal cancer | Renal cancer |
| 786-0 | 17.1 | <100 | <100 |
| A498 | 19.9 | 58.1 | <100 |
| ACHN | 6.4 | 91.1 | <100 |
| CAKI-1 | 13.8 | <100 | <100 |
| RXF 393 | 13.1 | 65.9 | <100 |
| SN 12 C | 11.0 | 60.4 | <100 |
| TK-10 | 29.3 | <100 | <100 |
| UO-31 | 7.3 | 77.9 | <100 |
| Prostate cancer | Prostate cancer | Prostate cancer | Prostate cancer |
| PC-3 | 12.7 | <100 | <100 |
| DU-145 | 17.7 | <100 | <100 |
| Breast cancer | Breast cancer | Breast cancer | Breast cancer |
| MCF7 | 9.4 | 89.9 | <100 |
| MDA-MB-231 | 17.2 | 59.7 | <100 |
| HS 578 T | 10.2 | 67.0 | <100 |
| BT-549 | 14.6 | 33.8 | 78.1 |
| T-47D | 6.3 | 70.0 | <100 |
| MDA-MB-468 | 8.3 | 42.0 | <100 |
Results displayed in Table 4, disclosed that compound 6b exhibited good anti-proliferative activities towards all the examined human cancer cell subpanels with GI50 values range 1.4 − 19.9 μM, except for renal TK-10 cell line (GI50 = 29.3 μM). In particular, the best anti-proliferative activity was noticed for non-small cell (HOP-92), CNS (SNB-75), and melanoma (SK-MEL-2) cancer cell lines with GI50 values equal 2.9, 1.4, and 3.7 μM, respectively (Table 4).
Concerning the cytostatic impact of quinazoline 6b, it showed moderate to good effect towards melanoma (SK-MEL-2), non-small cell (HOP-92, NCI-H522, NCI-H460, and HOP-62), colon (HCT-15, COLO 205, HCC-2998, and HCT-116), CNS (U251, SF-539, and SF-295), melanoma (UACC-62, SK-MEL-5, and LOX IMVI), and breast (BT-549, and MDA-MB-468) with TGI range = 15.6– 42 μM. It is worthy to mention that 6b exhibited LC50 values more than 100 μM and considered as non-lethal towards all the examined cell lines except for non-small cell (NCI-H460), colon (COLO 205), CNS (SF-539 and SF-295), melanoma (UACC-62, SK-MEL-5, LOX IMVI, and SK-MEL-2), and breast (BT-549) that possessed weak lethal effect with LC50 = 84.0, 94.1, 67.2, 86.0, 71.6, 57.5, 97.7, 83.5, and 78.1 μM, respectively (Table 3).
With regard to the sensitivity of the examined cell lines, quinazoline 6b elicited comparatively homogenous growth inhibitory activity throughout all NCI panels, with good growth inhibition full panel GI50 (MG-MID) equals 11.99 μM, as well as subpanel GI50 (MG-MID) values spanning from 8.04 to 15.66 μM. In particular, the most susceptible subpanels were CNS and Leukaemia with MG-MID of 8.04 and 8.68 μM, respectively (Table 5). In order to assess the selectivity of 6b, its full panel MG-MID is divided by its individual subpanel MG-MID (Table 5). The selectivity index for compound 6b ranged from 0.76 to 1.49 which points out that 6b has non-selective broad-spectrum anti-proliferative activity towards all NCI cancer subpanels. It is worth to mention that the best anti-proliferative counterpart 6b is not the most active inhibitor against CA IX or XII, thus the target of this compound could be other than CAs.
**Table 5.**
| Subpanel tumour cell line | 6b | 6b.1 |
| --- | --- | --- |
| Subpanel tumour cell line | MG-MID | Selectivity index |
| Leukaemia | 8.68 | 1.38 |
| Non-small cell lung cancer | 9.24 | 1.29 |
| Colon cancer | 13.70 | 0.87 |
| CNS cancer | 8.04 | 1.49 |
| Melanoma | 11.73 | 1.02 |
| Ovarian cancer | 15.66 | 0.76 |
| Renal cancer | 14.74 | 0.81 |
| Prostate cancer | 15.2 | 0.78 |
| Breast cancer | 10.99 | 1.09 |
| Full panel MG-MID | 11.99 | – |
## Conclusions
Three sets of 2-aryl-quinazolin-4-yl aminobenzoic acid regioisomers (6a–c, 7a–c, and 8a–c) were designed and synthesised as new non-classical CA inhibitors. Their CA inhibitory activities towards isoforms I, II, IX, and XII were evaluated. Only three of the tested quinazoline-based carboxylic acids (8a, 8b, and 8c) weakly inhibited the cytosolic hCA I isoform, with inhibition constants (KIs) equal 87.7, 73.2, and 66.3 µM. The cytosolic hCA II was effectively inhibited by para-aminobenzoic acid-bearing quinazolines (8a–c) with KIs of 9.3, 3.9, and 4.6 µM, respectively, whereas, their ortho (6a–c) and meta (7a–c) regioisomers elicited modest inhibitory effects. Moreover, the cancer-related hCA IX isoform was inhibited most effectively by quinazolines (8a–c) with KIs equal 4.3, 1.6, and 4.5 µM, respectively. Also, the results revealed that the cancer-related hCA XII isoform is the most vulnerable to the prepared molecules. In particular, the best hCA XII inhibitory effect was exerted by quinazoline 8c (KI = 0.25 µM), also, quinazolines 7b, 7c, and and 8b displayed sub-micromolar hCA XII inhibitory activity (KI = 0.91, 0.48, and 0.42 µM, respectively). The SAR analysis highlighted that the order of hCA inhibitory activities was increased in the order of para isomers 8 > meta isomers 7 > ortho isomers 6. On the other hand, the anti-proliferative activities of the quinazoline derivatives (6a–c, 7b, and 8a–b) have been evaluated at single (10 μM) dose assay against 59 cancer cell lines in the NCI-USA. Quinazoline derivatives featuring ortho aminobenzoic acid (6a–c) showed excellent broad cell growth inhibitory activity (GI % mean = 63, 84 and 52, respectively) against most of all cancer cell lines, whereas compounds (7b and 8a–b) with meta and para aminobenzoic acid moiety showed fair selective activity (GI % mean = 14, 16, and 20, respectively) towards certain cancer cell lines. Thereafter, 6b was selected by NCI for further evaluations at 5-doses (0.01–100 µM) level. Quinazoline 6b elicited comparatively homogenous growth inhibitory activity throughout all NCI panels, with good growth inhibition full panel GI50 (MG-MID) equals 11.99 μM, as well as subpanel GI50 (MG-MID) values spanning from 8.04 to 15.66 μM. In particular, the most susceptible subpanels were CNS and Leukaemia with MG-MID of 8.04 and 8.68 μM, respectively.
## Disclosure statement
CT *Supuran is* Editor-in-Chief of the Journal of Enzyme Inhibition and Medicinal Chemistry. He was not involved in the assessment, peer review, or decision-making process of this paper. The authors have no relevant affiliations of financial involvement with any organisation or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript. This includes employment, consultancies, honoraria, stock ownership or options, expert testimony, grants or patents received or pending, or royalties.
## References
1. Supuran CT.. **Structure and function of carbonic anhydrases**. *Biochem J* (2016) **473** 2023-2032. PMID: 27407171
2. Capasso C, Supuran CT.. **isoforms/isoforms**. *Targeting carbonic anhydrases* (2014) 7-16
3. Supuran CT.. **Advances in structure-based drug discovery of carbonic anhydrase inhibitors**. *Expert Opin Drug Discov* (2017) **12** 61-88. PMID: 27783541
4. Mishra CB, Tiwari M, Supuran CT.. **Progress in the development of human carbonic anhydrase inhibitors and their pharmacological applications: Where are we today?**. *Med Res Rev* (2020) **40** 2485-2565. PMID: 32691504
5. Mincione F, Nocentini A, Supuran CT.. **Advances in the discovery of novel agents for the treatment of glaucoma**. *Expert Opin Drug Discov* (2021) **16** 1209-1225. PMID: 33914670
6. Ciccone L, Cerri C, Nencetti S, Orlandini E.. **Carbonic anhydrase inhibitors and epilepsy: State of the art and future perspectives**. *Molecules* (2021) **26** 6380. PMID: 34770789
7. Costa G, Carta F, Ambrosio FA, Artese A, Ortuso F, Moraca F, Rocca R, Romeo I, Lupia A, Maruca A. **A computer-assisted discovery of novel potential anti-obesity compounds as selective carbonic anhydrase VA inhibitors**. *Eur J Med Chem* (2019) **181** 111565. PMID: 31387062
8. Angeli A, Carta F, Nocentini A, Winum J-Y, Zalubovskis R, Akdemir A, Onnis V, Eldehna WM, Capasso C, Simone GD. **Carbonic anhydrase inhibitors targeting metabolism and tumor microenvironment**. *Metabolites* (2020) **10** 412. PMID: 33066524
9. Al-Warhi T, Sabt A, Elkaeed EB, Eldehna WM.. **Recent advancements of coumarin-based anticancer agents: An up-to-date review**. *Bioorg Chem* (2020) **103** 104163. PMID: 32890989
10. Carta F, Aggarwal M, Maresca A, Scozzafava A, McKenna R, Supuran CT.. **Dithiocarbamates: a new class of carbonic anhydrase inhibitors. Crystallographic and kinetic investigations**. *Chem Commun* (2012) **48** 1868-1870
11. Karioti A, Carta F, Supuran CT.. **Phenols and polyphenols as carbonic anhydrase inhibitors**. *Molecules* (2016) **21** 1649. PMID: 27918439
12. Masini E, Carta F, Scozzafava A, Supuran CT.. **Antiglaucoma carbonic anhydrase inhibitors: a patent review**. *Expert Opin Ther Pat* (2013) **23** 705-716. PMID: 23627893
13. 13A study of SLC-0111 and gemcitabine for metastatic pancreatic ductal cancer in subjects positive for CAIX (SLC-0111-17-01). 2018. Available from: https://clinicaltrials.gov/ct2/show/NCT03450018. (2018)
14. D'Ambrosio K, Carradori S, Monti SM, Buonanno M, Secci D, Vullo D, Supuran CT, De Simone G.. **Out of the active site binding pocket for carbonic anhydrase inhibitors**. *Chem Commun* (2015) **51** 302-305
15. Supuran CT.. **Novel carbonic anhydrase inhibitors**. *Future Med Chem* (2021) **13** 1935-1937
16. Supuran CT.. **How many carbonic anhydrase inhibition mechanisms exist?**. *J Enzyme Inhib Med Chem* (2016) **31** 345-360. PMID: 26619898
17. Ibrahim HS, Abdelrahman MA, Nocentini A, Bua S, Abdel-Aziz HA, Supuran CT, Abou-Seri SM, Eldehna WM.. **Insights into the effect of elaborating coumarin-based aryl enaminones with sulfonamide or carboxylic acid functionality on carbonic anhydrase inhibitory potency and selectivity**. *Bioorg Chem* (2022) **126** 105888. PMID: 35661530
18. Elimam DM, Elgazar AA, Bonardi A, Abdelfadil M, Nocentini A, El-Domany RA, Abdel-Aziz HA, Badria FA, Supuran CT, Eldehna WM.. **Natural inspired piperine-based sulfonamides and carboxylic acids as carbonic anhydrase inhibitors: design, synthesis and biological evaluation**. *Eur J Med Chem* (2021) **225** 113800. PMID: 34482273
19. Cau Y, Vullo D, Mori M, Dreassi E, Supuran CT, Botta M.. **Potent and selective carboxylic acid inhibitors of tumor-associated carbonic anhydrases IX and XII**. *Molecules* (2017) **23** 17. PMID: 29271882
20. Eldehna WM, Nocentini A, Elsayed ZM, Al-Warhi T, Aljaeed N, Alotaibi OJ, Al-Sanea MM, Abdel-Aziz HA, Supuran CT.. **Benzofuran-based carboxylic acids as carbonic anhydrase inhibitors and antiproliferative agents against breast cancer**. *ACS Med Chem Lett* (2020) **11** 1022-1027. PMID: 32435420
21. Alkhaldi AAM, Al-Sanea MM, Nocentini A, Eldehna WM, Elsayed ZM, Bonardi A, Abo-Ashour MF, El-Damasy AK, Abdel-Maksoud MS, Al-Warhi T. **3-Methylthiazolo [3,2-**. *Eur J Med Chem* (2020) **207** 112745. PMID: 32877804
22. Abo-Ashour MF, Almahli H, Bonardi A, Khalil A, Al-Warhi T, Al-Rashood ST, Abdel-Aziz HA, Nocentini A, Supuran CT, Eldehna WM.. **Enaminone-based carboxylic acids as novel non-classical carbonic anhydrases inhibitors: design, synthesis and in vitro biological assessment**. *J Enzyme Inhib Med Chem* (2022) **37** 2256-2264. PMID: 36000171
23. Shaldam M, Eldehna WM, Nocentini A, Elsayed ZM, Ibrahim TM, Salem R, El-Domany RA, Capasso C, Abdel-Aziz HA, Supuran CT.. **Development of novel benzofuran-based SLC-0111 analogs as selective cancer-associated carbonic anhydrase isoform IX inhibitors**. *Eur J Med Chem* (2021) **216** 113283. PMID: 33667848
24. Bua S, Lomelino C, Murray AB, Osman SM, ALOthman ZA, Bozdag M, Abdel-Aziz HA, Eldehna WM, McKenna R, Nocentini A. **“A sweet combination”: developing saccharin and acesulfame K structures for selectively targeting the tumor-associated carbonic anhydrases IX and XII**. *J Med Chem* (2020) **63** 321-333. PMID: 31794211
25. Shaldam M, Nocentini A, Elsayed ZM, Ibrahim TM, Salem R, El-Domany RA, Capasso C, Supuran CT, Eldehna WM.. **Development of novel quinoline-based sulfonamides as selective cancer-associated carbonic anhydrase isoform IX inhibitors**. *IJMS* (2021) **22** 11119. PMID: 34681794
26. Al-Warhi T, Elbadawi MM, Bonardi A, Nocentini A, Al-Karmalawy AA, Aljaeed N, Alotaibi OJ, Abdel-Aziz HA, Supuran CT, Eldehna WM.. **Design and synthesis of benzothiazole-based SLC-0111 analogues as new inhibitors for the cancer-associated carbonic anhydrase isoforms IX and XII**. *J Enzyme Inhib Med Chem* (2022) **37** 2635-2643. PMID: 36146927
27. Eldehna WM, Abdelrahman MA, Nocentini A, Bua S, Al-Rashood ST, Hassan GS, Bonardi A, Almehizia AA, Alkahtani HM, Alharbi A. **Synthesis, biological evaluation and in silico studies with 4-benzylidene-2-phenyl-5 (4H)-imidazolone-based benzenesulfonamides as novel selective carbonic anhydrase IX inhibitors endowed with anticancer activity**. *Bioorg Chem* (2019) **90** 103102. PMID: 31299596
28. Eldehna WM, Taghour MS, Al-Warhi T, Nocentini A, Elbadawi MM, Mahdy HA, Abdelrahman MA, Alotaibi OJ, Aljaeed N, Elimam DM. **Discovery of 2, 4-thiazolidinedione-tethered coumarins as novel selective inhibitors for carbonic anhydrase IX and XII isoforms**. *J Enzyme Inhib Med Chem* (2022) **37** 531-541. PMID: 34991416
29. Tawfik HO, Shaldam MA, Nocentini A, Salem R, Almahli H, Al-Rashood ST, Supuran CT, Eldehna WM.. **Novel 3-(6-methylpyridin-2-yl) coumarin-based chalcones as selective inhibitors of cancer-related carbonic anhydrases IX and XII endowed with anti-proliferative activity**. *J Enzyme Inhib Med Chem* (2022) **37** 1043-1052. PMID: 35437108
30. Liu F, Tang B, Liu H, Li L, Liu G, Cheng Y, Xu Y, Chen W, Huang Y.. **4-Anilinoquinazoline derivatives with epidermal growth factor receptor inhibitor activity**. *Anti-Cancer Agents Med Chem* (2016) **16** 1652-1664
31. Ahmad I., Shagufta . **An insight into the therapeutic potential of quinazoline derivatives as anticancer agents**. *MedChemComm* (2017) **8** 871-885. PMID: 30108803
32. Ismail RS, Abou-Seri SM, Eldehna WM, Ismail NS, Elgazwi SM, Ghabbour HA, Ahmed MS, Halaweish FT, Abou El Ella DA.. **Novel series of 6-(2-substitutedacetamido)-4-anilinoquinazolines as EGFR-ERK signal transduction inhibitors in MCF-7 breast cancer cells**. *Eur J Med Chem* (2018) **155** 782-796. PMID: 30047410
33. Ismail RS, Ismail NS, Abuserii S, Abou El Ella DA.. **Recent advances in 4-aminoquinazoline based scaffold derivatives targeting EGFR kinases as anticancer agents**. *Future J Pharm Sci* (2016) **2** 9-19
34. Nada H, Elkamhawy A, Abdellattif MH, Angeli A, Lee CH, Supuran CT, Lee K.. **4-Anilinoquinazoline-based benzenesulfonamides as nanomolar inhibitors of carbonic anhydrase isoforms I, II, IX, and XII: design, synthesis, in-vitro, and in-silico biological studies**. *J Enzyme Inhib Med Chem* (2022) **37** 994-1004. PMID: 35350942
35. Alafeefy AM, Ahmad R, Abdulla M, Eldehna WM, Al-Tamimi A-MS, Abdel-Aziz HA, Al-Obaid O, Carta F, Al-Kahtani AA, Supuran CT.. **Development of certain new 2-substituted-quinazolin-4-yl-aminobenzenesulfonamide as potential antitumor agents**. *Eur J Med Chem* (2016) **109** 247-253. PMID: 26774930
36. Eldehna WM, Almahli H, Al-Ansary GH, Ghabbour HA, Aly MH, Ismael OE, Al-Dhfyan A, Abdel-Aziz HA.. **Synthesis and in vitro anti-proliferative activity of some novel isatins conjugated with quinazoline/phthalazine hydrazines against triple-negative breast cancer MDA-MB-231 cells as apoptosis-inducing agents**. *J Enzyme Inhib Med Chem* (2017) **32** 600-613. PMID: 28173708
37. Khalifah RG.. **The carbon dioxide hydration activity of carbonic anhydrase: I. stop-flow kinetic studies on the native human isoenzymes B and C**. *J Biol Chem* (1971) **246** 2561-2573. PMID: 4994926
38. Abdelrahman MA, Ibrahim HS, Nocentini A, Eldehna WM, Bonardi A, Abdel-Aziz HA, Gratteri P, Abou-Seri SM, Supuran CT.. **Novel 3-substituted coumarins as selective human carbonic anhydrase IX and XII inhibitors: synthesis, biological and molecular dynamics analysis**. *Eur J Med Chem* (2021) **209** 112897. PMID: 33038795
39. Fares M, Eldehna WM, Bua S, Lanzi C, Lucarini L, Masini E, Peat TS, Abdel-Aziz HA, Nocentini A, Keller PA. **Discovery of potent dual-tailed benzenesulfonamide inhibitors of human carbonic anhydrases implicated in glaucoma and in vivo profiling of their intraocular pressure-lowering action**. *J Med Chem* (2020) **63** 3317-3326. PMID: 32031797
40. Said MA, Eldehna WM, Nocentini A, Bonardi A, Fahim SH, Bua S, Soliman DH, Abdel-Aziz HA, Gratteri P, Abou-Seri SM. **Synthesis, biological and molecular dynamics investigations with a series of triazolopyrimidine/triazole-based benzenesulfonamides as novel carbonic anhydrase inhibitors**. *Eur J Med Chem* (2020) **185** 111843. PMID: 31718943
41. Monks A, Scudiero D, Skehan P, Shoemaker R, Paull K, Vistica D, Hose C, Langley J, Cronise P, Vaigro-Wolff A. **Feasibility of a highflux anticancer drug screen using a diverse panel of cultured human tumor cell lines**. *J Natl Cancer Inst* (1991) **83** 757-766. PMID: 2041050
42. Boyd MR., Teicher BA. *Drug discovery and development: anticancer drug development guide: preclinical screening, clinical trials and approval* (2014) 41-62
43. Boyd MR, Paull KD.. **Some practical considerations and applications of the National Cancer Institute in vitro anticancer drug discovery screen**. *Drug Dev Res* (1995) **34** 91-109
44. Skehan P, Storeng R, Scudiero D, Monks A, McMahon J, Vistica D, Warren JT, Bokesch H, Kenney S, Boyd MR.. **New colorimetric cytotoxicity assay for anticancer-drug screening**. *J Natl Cancer Inst* (1990) **82** 1107-1112. PMID: 2359136
45. Eldehna WM, Salem R, Elsayed ZM, Al-Warhi T, Knany HR, Ayyad RR, Traiki TB, Abdulla M-H, Ahmad R, Abdel-Aziz HA. **Development of novel benzofuran-isatin conjugates as potential antiproliferative agents with apoptosis inducing mechanism in colon cancer**. *J Enzyme Inhib Med Chem* (2021) **36** 1423-1434
46. Elbadawi MM, Eldehna WM, Wang W, Agama KK, Pommier Y, Abe M.. **Discovery of 4-alkoxy-2-aryl-6, 7-dimethoxyquinolines as a new class of topoisomerase I inhibitors endowed with potent in vitro anticancer activity**. *Eur J Med Chem* (2021) **215** 113261. PMID: 33631697
|
---
title: Association between participation in the Northern Finland Birth Cohorts and
cardiometabolic disorders
authors:
- Martta Kerkelä
- Mika Gissler
- Tanja Nordström
- Olavi Ukkola
- Juha Veijola
journal: Annals of Medicine
year: 2023
pmcid: PMC10035958
doi: 10.1080/07853890.2023.2186478
license: CC BY 4.0
---
# Association between participation in the Northern Finland Birth Cohorts and cardiometabolic disorders
## Abstract
### Background
We studied the association between participation in the longitudinal follow-up study and cardiometabolic disorders in two longitudinal studies which started prospectively in the antenatal period: the Northern Finland Cohort 1966 (NFBC1966) and the Northern Finland Birth Cohort 1986 (NFBC1986). Both birth cohorts have been followed up since birth with multiple follow-ups including questionnaires, and clinical examinations.
### Methods
The NFBC studies were compared to comparison cohorts of individuals who were born in the same area as the study cohorts, but in different years. The data for the comparison cohort were obtained from registers. The cumulative incidence rates of hospital-treated cardiometabolic disorders were calculated for study and comparison cohorts covering the age of 7–50 years in NFBC1966 and the age of 0–29 years in NFBC1986. Cardiometabolic-related causes of death were analysed in NFBC1966 and the comparison cohort from the age of 0–50 years. The analysed cardiometabolic disorders were diabetes mellitus, coronary artery disease, hyperlipidaemia, obesity, hypertension, and cerebrovascular disorders. The risk ratio (RR) with $95\%$ confidence intervals (CI) was calculated by sex.
### Results
In NFBC1966, no differences in cumulative incidences of cardiometabolic disorders or cardiometabolic-related deaths compared to the comparison cohort were found. Male members of NFBC1986 had decreased risk of obesity (RR: 0.45, $95\%$ CI: 0.27–0.75) and any cardiometabolic disorders (RR: 0.75, $95\%$ CI: 0.59–0.95) compared to the comparison cohort.
### Conclusions
The results suggest that participation in the NFBC1986 may have a weak positive health effect among men. Agreement to follow-up studies focusing on diet, substance use, and physical activity, may slightly decrease the incident risk of cardiometabolic disorders in the study population. KEY MESSAGESEven mild interventions, such as follow-up studies in the prospective follow-up studies, might affect participants’ behaviour and consequently the incidence of cardiometabolic disordersThe fact that follow-up itself might affect the study population in terms of risk factors, has to be taken into account when estimating the representativeness of the followed population.
## Introduction
In prospective population-based cohort studies, a defined population is selected for longitudinal assessment of exposure-outcome relations. Data collection procedures include clinical examinations, questionnaires, tests, interviews, or linkage to existing data [1]. The Hawthorne effect is a phenomenon that occurs when the mere act of being observed in a study changes the behaviour of the study’s participants. This can impact the validity of the study’s results, as the behaviour that is being observed may not be representative of the participants’ normal behaviour [2]. In the Derbyshire Smoking Study, the Hawthorne effect was observed in a group of approximately 6,000 adolescents whose smoking habits were surveyed with a yearly questionnaire from 1974 to 1978 in selected schools. The findings showed that the prevalence of smoking was lower in schools that had been surveyed for five years [3]. A meta-analysis of 15 studies published between 2012 and 2022 found that the Hawthorne effect is common in medical research, which can limit the study’s external validity and result in gaps in medical knowledge [4]. Nevertheless, whether the intensive follow-ups affect the study population has not been thoroughly studied. The follow-up studies conducted may have a wide range of effects on the target population, even though the purpose of the prospective birth cohort setting is not to intervene but to get a representative sample of the population.
Cardiovascular disorders are the leading cause of death globally, and a proportion are preventable through healthy lifestyle choices. Major risk factors for cardiovascular disorders are tobacco use, unhealthy diet and obesity, physical inactivity, and harmful use of alcohol [5]. One successful intervention in encouraging a healthy lifestyle was the Stanford Community study. The aim was to reduce fat content in the daily diet and the health information was broadcast via the mass media. After two years of intervention, a significant drop in the population’s negative health factors (cholesterol level, blood pressure, smoking rate) could be seen [6]. Another well-known population-level health intervention is the North Karelia project, in which the aim was to reduce cardiovascular risk factors through interventions including broad health promotion and policies in the 1970s onwards. The project was successful: the results indicated substantial and positive shifts in health-related behaviours, the community-based part of the project ‘hypertension control programme’ brought hypertension under control for a significant proportion of hypertensive individuals, and over the years, the coronary heart disease mortality rates decreased [7]. The North Karelia project and Stanford Community study powerfully demonstrate the effect of population-based intervention in cardiometabolic disorders.
We were able to study the association between participation in a longitudinal follow-up study and cardiometabolic disorders in two longitudinal studies which started prospectively in the antenatal period: namely the Northern Finland Cohort 1966 (NFBC1966) [8] and the Northern Finland Birth Cohort 1986 (NFBC1986) [9]. Since birth, both birth cohorts have undergone several follow-ups such as surveys and clinical examinations. Even though the follow-ups of the NFBC studies were not supposed to be interventions, many of them included questions and screenings on lifestyle, exercise, diet, and use of intoxicants. Several follow-ups in the NFBC studies also included clinical examinations, in which weight and height, blood pressure, physical activity, and numerous other clinical measurements were measured. The hypothesis is that the members of NFBC studies might live more healthily due to follow-up and therefore have fewer cardiometabolic disorders than comparison cohorts.
## Cohorts
The NFBC1966 covers people whose expected date of birth was in 1966 in the former two northernmost provinces of Finland, Oulu, and Lapland. The cohort included 12,055 mothers and 12,231 children. Of the NFBC1966 members, 189 ($1.5\%$) were born in 1965, 11,999 ($97.9\%$) in 1966 and 63 ($0.5\%$) in 1967. NFBC1966 is a longitudinal and prospective birth cohort with several follow-ups. The comparison cohort for the NFBC1966 comprises all liveborn children in the provinces of Lapland and Oulu in the years 1965 and 1967. The cohort included 24,471 participants, 12,465 ($50.9\%$) born in 1965 and 12,006 ($49.1\%$) born in 1967. All data for the comparison cohort was collected from Finnish registers: personal identification numbers were obtained from the Digital and Population Data Services Agency and medical histories were obtained from the Finnish Institute of Health and Welfare (THL).
The NFBC1986 covers people with an expected date of birth between 1 July 1985 and 30 June 1986 in the former Finnish provinces of Oulu and Lapland. The cohort consisted of 9,362 mothers and 9,479 children. The NFBC1986 is a prospective, longitudinal birth cohort with numerous follow-ups. The NFBC1986 comparison cohort consists of all people born in the provinces of Lapland and Oulu in 1987. A sub-sample from the register-based Finnish Birth Cohort 1987 (FBC 1987) was used as the comparison cohort. All children born in Finland in 1987 who survived the perinatal period and were alive at seven days after birth are included in the FBC 1987 study, with all data collected from several registers. Members of FBC 1987 have never been contacted [10]. Information on the place of birth was obtained from the Medical Birth Register, which is currently maintained by the Finnish Institute for Health and Welfare (THL).
## Follow-ups in NFBC1966
The first follow-up for the NFBC1966 cohort was conducted during the mother’s pregnancy. The follow-up included a questionnaire (including questions about background, life situation, and living habits) during pregnancy (from the 24th to the 28th gestational week) and delivery information. The data were collected by the midwives at antenatal clinics [11]. The second follow-up was conducted at age 1 year, including a questionnaire concerning children’s growth, health, and development, with a $91.2\%$ participation rate [12]. The next follow-up was at age 14 years, with a participation rate of $93.6\%$. The follow-up was conducted with a postal questionnaire, including questions for NFBC1966 members about their growth and health, physical exercise, substance use (smoking, alcohol, and other intoxicants), living habits, school performance, and family situation [13]. The questions regarding physical exercise, health and substance use are described in more detail in Supplement Table 1.
The next follow-up for the whole cohort was conducted at age 31 years. The follow-up included a postal questionnaire (participation rate $77.4\%$), including questions about a life situation, background information, exercise, physical performance (how often and how much), occupation, living environment, health (e.g. height, weight, history of cardiometabolic disorders, use of medications), diet, and living habits (questions on smoking, alcohol, and other intoxicant use) (Supplement Table 2). Clinical examination included a skin prick allergy test, physical performance tests (step test, grip strength test, back endurance test), blood pressure and pulse, blood tests, other laboratory tests (cholesterol, glucose, insulin, triglyceride), and spirometry (participation rate $71.3\%$) [14].
The latest follow-up has been conducted at age 46 years, with a $66.5\%$ participation rate in the questionnaire and $56.7\%$ in the clinical examination. The questionnaire included questions about background, lifestyle (questions about smoking, alcohol, and other intoxicant use), health (e.g. disorders and medications), economy, work, and mental resources (Supplement Table 3). In the clinical examination, data were collected on cardiovascular health (brachial pressure, blood pressure, 15-led rest ECG, echo and carotid ultrasound), allergies (skin allergy test, spirometry, whole body examination by dermatologist), eyes (eye examination), physical activity and fitness (two-week accelerometer measurements, back muscle strength test, step test, hand grip strength test), pain perception and tolerance (pressure pain and thermal perception), musculoskeletal health (spine, ankle, and lower back movement tests, knee, ankle foot radiography and lumbar MRI), height, weight, waist and hip circumferences, dental health, cognitive tests, and biological samples [15].
The follow-up in NFBC1966 also includes several subsamples. One of the subsamples focusing on the health and life satisfaction of male cohort members was conducted on a random subsample of males at age 24 years ($$n = 2$$,500) [16]. The follow-up consisted of a questionnaire including questions about physical activity, diet, smoking habits, overall health, and relationships. Another sub-sample focusing on health was conducted on a random subsample of 31-year-old follow-up participants ($$n = 196$$). The follow-up consisted of exercise and food diaries for 7 days [17]. Further details of all substudies conducted in NFBC1966 are provided in Supplement Table 4.
## Follow-ups in NFBC1986
The NFBC1986 data collection began throughout the mother’s pregnancy. At antenatal clinics, mothers were given three questionnaires that included questions on their background, smoking and alcohol use, general health, fatigue throughout pregnancy, and delivery information [18].
At age 7 years, the second follow-up was completed in the autumn of the first school year. The questionnaire, which included information about children’s growth, development, and health, as well as their socioeconomic factors, was sent to parents (participation rate $90\%$) [19]. The next follow-up was scheduled for the spring of the children’s first school year. The follow-up included two questionnaires for parents, one for themselves and one for the children’s teachers. With a participation rate of $90\%$, the parents’ questionnaire included a modified (one of the internalizing items was modified) Rutter A2 [20]. The teachers completed Rutter B2 [21], with a $92\%$ participation rate. Rutter scales assess children’s behavioural and emotional characteristics.
The next follow-up was undertaken at the age of 15–16 years, with a questionnaire for adolescents (participation rate $77.9\%$) and parents ($74.5\%$). The adolescent’s questionnaire included questions about family, school, health (history of heart diseases, weight and height, and blood pressure), physical activity (e.g. times and duration of physical exercise), sexual behaviour, substance use (smoking, alcohol, or other intoxicants), diet (eating habits, use of sugar and fat, amount and times of eating bread, dairy products, sweets, etc.), living habits and hobbies. The questions regarding physical exercise, diet, cardiovascular health, and substance use are described in Supplement Table 5. The adolescents were also invited to clinical examinations (participation rate of $73.5\%$). The clinical examination included weight, height, waist-hip measurements, sitting height, spirometry, blood pressure, pulse rate, blood samples, physical activity (bicycle ergometer) test, prick tests, and questions about puberty, nutrition, smoking, and use of alcohol.
NFBC1986 also includes multiple sub-studies. At age 18 years, a questionnaire including questions about low back pain history, medical history, quality of life, nutrition, socioeconomic status, leisure activities, history of injuries, smoking history, occupational exposure, sports activities, and psychological factors, was sent to cohort members living in Oulu and the surrounding municipalities [22]. The participants of the substudy were also invited to an MRI scan of the lumbar spine at age 19–22 years and 29–32 years [23]. At age 24 years, the ESTER- Preterm Birth, Pregnancy and Offspring Health in Adult Life study was conducted on a subpopulation of NFBC1986. The data collection consisted of blood sampling, blood pressure, BMI, waist, and hip measurements [24]. The gynaecological health of young females was studied by a questionnaire sent to female members of NFBC1986 at age 26 years. The questionnaire included questions on socio-demographic and health background, mainly about reproduction, menstruation, and infertility [25]. Further details of all substudies conducted in NFBC1986 are provided in Supplement Table 6.
## Final datasets of NFBC and comparison cohorts
From NFBC1966 we excluded those who were born in 1965 ($$n = 186$$) or 1967 ($$n = 63$$) and stillbirths ($$n = 153$$), while the comparison cohort comprises liveborn children in the years 1965 and 1967. The final dataset of NFBC1966 comprised 11,723 participants, 6,000 males and 5,723 females. The comparison dataset comprised 24,471 participants, 12,511 males and 11,960 females.
To make the study and comparison cohorts comparable, from NFBC1986, we excluded those who died during the perinatal period (stillbirths $$n = 47$$, died during the first seven days of life $$n = 36$$). The final NFBC1986 dataset included 9,396 participants, 4,839 males, and 4,557 females. The comparison dataset included 8,959 participants, 4,550 males, and 4,409 females.
The Finnish register data were given for this specific study, and the data cannot be shared without authorisation from the register keepers and the University of Oulu. In the use of data, we follow the EU General Data Protection Regulation ($\frac{679}{2016}$) and Finnish Data Protection Act.
## Cardiometabolic disorders
The following cardiometabolic disorders were considered as an outcome measure: *Diabetes mellitus* (differentiation into type 1, type 2 and unspecific types in ICD-9 and ICD-10); Coronary artery disease; hyperlipidaemia; overweight, obesity and other hyperalimentation; hypertension; cerebrovascular disorder, and any aforementioned cardiometabolic disorder. The Care Register for Health Care (CRHC), maintained by the Finnish Institute for Health and Welfare, was used to identify patients who had a diagnosis (primary or secondary diagnosis) of an above-described cardiometabolic disorder. We used the ICD-8 (1972–1986), ICD-9 (1987–1995) and ICD-10 (1996–2017) classifications (Table 1). The CRHC is one of the oldest individual-level hospital discharge registers and contains nationwide hospital discharge information on inpatient visits starting from 1967. From 1998 onwards, the register also includes information on specialised outpatient care in public hospitals. By law, all hospitals are obligated to report all inpatient care. The data covers public hospitals only. Several studies indicate that the quality of CRHC is high [26]. NFBC1966 and its comparison cohort were followed from age 7 to 50 years (1972–2017), while the information on the CRHC was incomplete before 1972. In NFBC1986 and its comparison cohort, the follow-up covered age 0 to 29 years (1985–2016). In addition, we examined cardiometabolic-related deaths in NFBC1966. Cardiometabolic-related deaths were defined with the corresponding ICD codes (Table 1) from statistical underlying cause-of-death diagnoses. The cause-of-death codes were obtained from the Cause of Death Register, maintained by Statistics Finland.
**Table 1.**
| Unnamed: 0 | ICD-81969–1986 | ICD-91987–1995 | ICD-101996–2017 |
| --- | --- | --- | --- |
| Diabetes mellitus | 250 | 250 | E10-E14 |
| Type 1 | | 2500B-2508B | E10 |
| Type 2 | | 2500A-2508A | E11 |
| Unspecific type | | 2500C-2508C, 2500X-2508X | E12-E14 |
| Coronary artery disease | 410–414 | 410–414 | I20–I25 |
| Hyperlipidaemia | 272 | 272 | E78 |
| Overweight, obesity, and other hyperalimentation | 277.99 | 278 | E65-E68 |
| Hypertension | 400–404 | 401–405 | I10–I15 |
| Cerebrovascular disorders | 430–438 | 430–438 | I60–I69 |
| Any cardiometabolic disorder | 250, 272, 277.99, 400–404, 410–414, 430–438 | 250, 272, 278, 401–405, 410–414, 430–438 | E10-E14, E65-E68, E78, I10-I15, I20-I25, I60-I69 |
## Statistical analysis
The cumulative incidence rates of cardiometabolic disorders in all hospital-treated cardiovascular disorders (including inpatient and specialized outpatient visits) were calculated for the study and comparison cohorts covering the full follow-up (age 7 to 50 years in NFBC1966; age 0 to 29 years in NFBC1986). Different types of diabetes mellitus were examined from 1987 onwards (age 2 to 29 years in NFBC1986 and age 22 to 50 years in NFBC1966). Due to the small number of cases of hyperlipidaemia and coronary artery disorders in the younger population (follow-up ends at age 29 years), the separate diagnosis classes are not included in the analysis. Risk ratios (RRs) with $95\%$ confidence intervals (CIs) were calculated by sex separately in each diagnosis group. The age of the first onset of cardiometabolic diagnosis (median with IQR) is reported in each diagnosis group. The difference between the medians is estimated using quantile estimation (QE) and Q with p-values are reported [27]. The age of the first onset of cardiometabolic disorders was plotted over the full follow-up period in both NFBCs, separated by sex. Cumulative incidences of cardiometabolic-related causes of death were calculated to NFBC1966 and comparison cohorts at age 0 to 50 years. The age of death caused by any cardiometabolic disorders (median with IQR) by sex is also reported. Analysis was performed using R version 1.4.1106.
## Results
Table 2 reports the cumulative incidences % (N), Pearson’s chi-square statistics, and risk ratios (RR) with $95\%$ confidence intervals (CI) of all hospital-treated cardiovascular disorders of NFBC1966 with the comparison cohort at age 7 to 50 years. No significant differences in cumulative incidences between NFBC1966 and the comparison cohort were found for cardiometabolic disorders. Female members of NFBC1966 had a lower age of onset of coronary artery disease (median age of onset 42.5 years vs 45.8 years; Q:3.85, $$p \leq 0.0.049$$), diabetes mellitus (median age of onset 38.1 years vs 41.7 years; Q:8.28, $$p \leq 0.004$$) and any cardiometabolic disorder (median age of onset 42.1 years vs 43.1 years; Q:4.16, $$p \leq 0.041$$) compared to the comparison cohort (Figure 1, Table 3). No difference in cumulative incidences or age of onset in different types of diabetes mellitus from age 22 to 50 years between NFBC1966 and the comparison cohort was found (Table 4).
**Figure 1.:** *Cumulative incidence of first cardiometabolic disorders in NFBC1966 and comparison cohort at age 7 to 50 years by sex.* TABLE_PLACEHOLDER:Table 2. TABLE_PLACEHOLDER:Table 3. TABLE_PLACEHOLDER:Table 4.
The cumulative incidences of cardiovascular disorders (age 0 to 29 years) in NFBC1986 and its comparison cohort are reported in Table 4. Male members of NFBC1986 had a lower risk of overweight, obesity, and other hyperalimentation (RR: 0.45, $95\%$ CI: 0.27–0.75), type 2 diabetes (RR 0.41, $95\%$ CI: 0.27–0.75) and any cardiometabolic disorder (RR: 0.75, $95\%$ CI: 0.59–0.95). In female members of NFBC1986, no difference in cumulative incidence rates between NFBC and the comparison cohort was found (Table 5). Figure 2 plots the age of the first onset of cardiometabolic disorders in the NFBC1986 and comparison cohort by sex. Male members of NFBC1986 had a higher age of onset in obesity (median age of onset 21.3 years vs. 15.1 years; Q: 10.76, $$p \leq 0.001$$) and any cardiometabolic disorder (median age of onset 17.6 years vs. 17.0 years; Q: 5.29, $$p \leq 0.022$$) and female members in diabetes mellitus (median age of onset 15.1 years vs.10.1 years; Q: 8.06, $$p \leq 0.005$$) than the comparison cohort. ( Table 6).
**Figure 2.:** *Cumulative incidence of first cardiometabolic disorders in NFBC1986 and comparison cohort at age of 0 to 29 years by sex.* TABLE_PLACEHOLDER:Table 5. TABLE_PLACEHOLDER:Table 6.
Cardiometabolic-related deaths in the NFBC1966 and comparison cohort by sex are reported in Table 7. No significant difference between NFBC1966 and the comparison cohort was found (males RR: 1.06, $95\%$ CI: 0.68–1.64, $$p \leq 0.794$$; females RR: 1.93, $95\%$ CI: 0.88–4.22, $$p \leq 0.095$$). Age of death did not differ between NFBC1966 and the comparison cohort in males (Q: 0.50, $$p \leq 0.481$$) or females (Q: 0.00, $$p \leq 0.973$$).
**Table 7.**
| Unnamed: 0 | Males | Males.1 | Females | Females.1 |
| --- | --- | --- | --- | --- |
| Cause of death | NFBC1966 N = 6000% (N) | Comparison N = 12,511% (N) | NFBC1966 N = 5723% (N) | Comparison N = 11,960% (N) |
| Cerebrovascular disorders | 0.1 (6) | 0.1 (10) | 0.1 (4) | 0.1 (6) |
| Coronary artery disease | 0.3 (18) | 0.3 (41) | 0.1 (4) | 0.0 (4) |
| Diabetes mellitus | 0.1 (4) | 0.0 (2) | 0.1 (4) | 0.0 (2) |
| Hypertension | 0.0 (2) | 0.0 (6) | 0.0 (0) | 0.0 (1) |
| Death caused by any cardiometabolic disorder | 0.5 (30) | 0.5 (59) | 0.2 (12) | 0.1 (13) |
| Age of death (Median (IQR)) | 42.6 (36.1–46.7) | 44.1 (38.2–47.9) | 43.0 (41.6–46.2) | 42.9 (40.0–44.3) |
## Discussion
The results partly supported our hypothesis of lower incidences of cardiometabolic disorders in the NFBC, where the male members of NFBC1986 had decreased risk of overweight, obesity, and other hyperalimentation, type II diabetes and any cardiometabolic disorder diagnosis at age of 0 to 29 years. In the NFBC1966 study, there was no significant difference in the overall incidence of coronary artery disease, diabetes mellitus, or any cardiometabolic disorder between the study and comparison cohorts. However, the onset of these disorders occurred at a younger age in females within the study cohort. This may indicate that female members of NFBC1966 tended to seek medical treatment more frequently, potentially due to the follow-up studies conducted in NFBC1966. The NFBC1966 participants had more frequent measurements of cardiometabolic risk factors during the follow-up (clinical examinations at ages 31 and 46 years), which could have contributed to the earlier age of onset of cardiometabolic disorders in NFBC1966. It is also possible that members of NFBC1966 are generally more self-aware of their health, leading to earlier detection and treatment of these conditions. Further research is needed to fully understand the factors that may have contributed to the observed differences in age of onset between the study and comparison cohorts.
The follow-ups of the NFBC studies differ considerably. In NFBC1986, the follow-up was more intense in childhood and adolescence (follow-ups for the whole cohort at age of 7, 8, and 15–16 years), while in NFBC1966 the follow-ups for the whole cohort included childhood and adolescent data collection only at age 14 with questionnaire years. The age 15–16 years follow-up study of NFBC1986 explored possible cardiovascular risk groups by surveying health, lifestyle choices, and eating habits widely. The follow-up study also included a clinical examination, in which blood pressure, height, and weight were measured and a physical fitness test was conducted. In NFBC1966, the follow-up at age of 14 years included only basic questions about activity, BMI, and substance use. The data collection regarding cardiovascular risk factors was limited. Moreover, the first clinical examination in the follow-up for the whole cohort in NFBC1966 was conducted during the age 31 years follow-up study. The younger cohort might have been influenced by information on healthier lifestyle choices in adolescence via follow-ups conducted at age 15–16 years. In the age 31 and 46 years follow-ups of NFBC1966, eating habits, physical exercise, and overall health were explored widely, but there was no difference in cardiometabolic disorders from age 7 to 50 years between NFBC1966 and the comparison cohort.
Eurostat reports that the prevalence of self-reported overweight and obesity (BMI ≥ 25) was $39.3\%$ among Finnish males aged 18 to 29 years and $66.0\%$ among Finnish males aged 25 to 64 years in 2014 [28]. Obesity is a relatively rare cause to use healthcare: $0.5\%$ in NFBC1986 and $1.0\%$ in the comparison cohort had a healthcare diagnosis of overweight, obesity, and other hyperalimentation until the age of 29; $1.3\%$ in males in NFBC1966, $1.5\%$ in the comparison cohort from age of 7 to 50 years. Diagnoses related to overweight and obesity cases are seldom given in specialised care. Nevertheless, obesity is preventable via lifestyle choices. The systematic review suggests that population reductions in weight are achievable through community-based interventions, including interventions that have incorporated educational, health promotion, social marketing, policy, or legislative reform strategies [29]. Multiple follow-up studies conducted on NFBC1986 during childhood and adolescence might have encouraged the participants of NFBC1986 to make healthier lifestyle choices, and therefore, the male members of NFBC1986 did have less overweight, obesity, or another hyperalimentation diagnosis, and were diagnosed at an older age than the comparison cohort.
Type 1 and type 2 diabetes develop due to interactions between genetic and environmental factors, but more behavioural factors, such as a sedentary lifestyle and poor diet, have been associated with type 2 diabetes [30,31]. Type 1 diabetes occurs predominantly in young people (diagnosis at 30 years of age or younger) [32], whereas type 2 diabetes generally manifests after age 40 years [33]. Male members of NFBC1986 had a lower risk for type 2 diabetes and female members had a higher age of onset of diabetes mellitus than the comparison cohort, but that was mostly due to the age at type 1 diabetes mellitus diagnosis. If the follow-up has encouraged the participants to make healthier lifestyle choices, it has likely affected the onset of type 2 diabetes, rather than type 1 diabetes.
In NFBC1966, no clear peaks were seen in the cumulative incidences of cardiometabolic disorders during the follow-up period. In NFBC1986, the difference in the cumulative incidence of cardiometabolic disorders between the study cohort and the comparison cohort appeared to steadily increase in males over the follow-up period, having a higher incidence in the comparison cohort. In females, in the comparison cohort cumulative incidence of cardiometabolic disorders peaked at an early age and then the difference between NFBC1986 and the comparison cohort remained stable over time. However, the members of NFBC1986 were a little too young for an accurate assessment of the differences in the incidence of cardiometabolic disorders. The preliminary results for NFBC1986 suggest possible health-promoting effects, but these need to be confirmed as the cohort members age.
## Strengths
The data used in this study were from Finnish registers, which have been found to have a good standard [26]. The NFBC studies and comparison cohorts were born in the same area one to two years apart, so it could be assumed that the sociodemographic background factors did not vary across the cohorts. The response rate in the NFBC studies can be considered high since poor response rates in follow-up cohort studies are causing increasing concern [34]. In addition, the study included two large birth cohorts, in which the follow-ups differ from each other.
## Limitations
The study also has some limitations. We could not identify outpatient and inpatient hospital visits in NFBC1966 and its comparison cohort from 1998 onwards. The comparison cohort of NFBC1966 includes a small percentage ($1.0\%$) of original members of NFBC1966. The prevalence rates were also relatively low in some diagnostics groups, especially for the younger birth cohort, which could have led to false-positive findings.
## Conclusion
The results partly supported our hypothesis of lower incidences of cardiometabolic disorders in NFBC, where the male members of NFBC1986 had decreased risk of overweight, obesity, and other hyperalimentation, type 2 diabetes and any cardiometabolic disorder diagnosis at age of 0 to 29 years. However, the members of NFBC1986 were a little too young for an accurate assessment of the differences in the incidence of cardiometabolic disorders. The preliminary results for NFBC1986 suggest possible health-promoting effects, but these need to be confirmed as the cohort members age. The participation bias in epidemiological follow-up studies has previously been under-examined, and the results need to be replicated. In a previous study, an association between the use of mental health care services and participation in the NFBC1986 study was found [35].
## Ethical approval
The Northern Finland Birth Cohorts are kept under review by the ethical committee of the Northern Ostrobothnia Hospital District and permission to gather data was obtained from the Finnish Ministry of Social Affairs and Health (NFBC1986: Northern Ostrobothnia Hospital District Ethical Committee §$\frac{108}{2017}$; NFBC1966: Northern Ostrobothnia Hospital District Ethical Committee §$\frac{94}{2011}$). Register-based 1987 FBC obtained ethics approval from the Finnish Institute for Health and Welfare (Ethical committee §$\frac{28}{2009}$).
## Author contributions
The authors confirm contribution to the paper as follows: study conception and design, or analysis and interpretation of the data: MK, MG, TN, OU, JV; the drafting of the paper, revising it critically for intellectual content: MK, MG, TN, OU, JV. All authors approved the final version of the manuscript and agree to be accountable for all aspects of the work.
## Disclosure statement
No potential competing interest was reported by the authors.
## Data availability statement
Data cannot be shared publicly because the utilised data have been given for this specific study, and the data cannot be shared without permission from the University of Oulu and Findata. The register data can be applied from Findata, the Finnish Health and Social Data Permit Authority by researchers who meet the criteria for access to confidential data. NFBC data is available from the University of Oulu, Infrastructure for Population Studies. Permission to use the data can be applied for research purposes via the electronic material request portal. In the use of data, we follow the EU general data protection regulation ($\frac{679}{2016}$) and Finnish Data Protection Act. The use of personal data is based on cohort participants written informed consent in his/her latest follow-up study, which may cause limitations to its use. Please, contact the NFBC project centre (NFBCprojectcenter(at)oulu.fi) and visit the cohort website for more information. Aggregated data underlying the results presented in the study can be asked from the corresponding author as long as the current permission is valid [2025].
## References
1. Szklo M.. **Population-based cohort studies**. *Epidemiol Rev* (1998) **20** 81-90. PMID: 9762511
2. Jones S.. **Was there a Hawthorne effect?**. *Am J Sociol* (1992) **98** 451-468
3. Murray M, Swan A V, Kiryluk S. **The hawthorne effect in the measurement of adolescent smoking**. *J Epidemiol Community Health* (1988) **42** 304-306. PMID: 3251014
4. Berkhout C, Berbra O, Favre J. **Defining and evaluating the hawthorne effect in primary care, a systematic review and meta-analysis**. *Front Med* (2022) **9** 1-15
5. 5WHO. Cardiovascular diseases (CVDs) 2021. https://www.who.int/en/news-room/fact-sheets/detail/cardiovascular-diseases-(cvds).. (2021)
6. Farquhar JW, Fortmann SP, Flora JA. **Effects of communitywide education on cardiovascular disease risk factors. The Stanford Five-City Project**. *JAMA* (1990) **264** 359-365. PMID: 2362332
7. Puska P, Vartiainen E, Laatikainen T. *The North Karelia project : from North Karelia to national action* (2009)
8. 8University of Oulu: Northern Finland Birth Cohort
1966. University of Oulu. Http://UrnFi/Urn:Nbn:Fi:Att:Bc1e5408-980e-4a62-B899-43bec3755243 n.d.. (1966)
9. 9University of Oulu: Northern Finland Birth Cohort
1986. University of Oulu. Http://UrnFi/Urn:Nbn:Fi:Att:F5c10eef-3d25-4bd0-Beb8-F2d59df95b8e n.d.. (1986)
10. Paananen R, Gissler M.. **Cohort profile: the 1987 Finnish birth cohort**. *Int J Epidemiol* (2012) **41** 941-945. PMID: 21378104
11. Järvelin MR, Hartikainen-Sorri AL, Rantakallio P.. **Labour induction policy in hospitals of different levels of specialisation**. *Br J Obstet Gynaecol* (1993) **100** 310-315. PMID: 8494831
12. von Wendt L, Mäkinen H, Rantakallio P.. **Psychomotor development in the first year and mental retardation-a prospective study**. *J Mental Deficiency Res* (1984) **28** 219-225
13. Rantakallio P.. **A follow-up study up to the age of 14 of children whose mothers smoked during pregnancy**. *Acta Paediatr Scand* (1983) **72** 747-753. PMID: 6637474
14. Marjo-Riitta J, Ulla S, Vanessa K. **Early life factors and blood pressure at age 31 years in the 1966 Northern Finland birth cohort**. *Hypertension* (2004) **44** 838-846. PMID: 15520301
15. Nordström T, Miettunen J, Auvinen J. **Cohort profile: 46 years of follow-up of the Northern Finland birth cohort 1966 (NFBC1966)**. *Int J Epidemiol* (2022) 1786-1787. PMID: 34999878
16. Pietilä A-M.. (1994)
17. Laitinen J, Kiukaanniemi K, Heikkinen J. **Body size from birth to adulthood and bone mineral content and density at 31 years of age: results from the Northern Finland 1966 birth cohort study**. *Osteoporos Int* (2005) **16** 1417-1424. PMID: 15782283
18. Järvelin M-R, Elliott P, Kleinschmidt I. **Ecologic and individual predictors of birthweight in a Northern Finland birth cohort 1986**. *Paediatr Perinat Epidemiol* (1997) **11** 298-312. PMID: 9246691
19. Taanila A, Ebeling H, Kotimaa A. **Is a large family a protective factor against behavioural and emotional problems at the age of 8 years?**. *Acta Paediatr* (2004) **93** 508-517. PMID: 15188979
20. Elander J, Rutter M.. **Use and development of the rutter parents’ and teachers’ scales**. *Int J Method Psychiat Res* (1996) **6** 63-78
21. Rutter M.. **A children’s behaviour questionnaire for completion by teachers: preliminary findings**. *J Child Psychol Psychiatry* (1967) **8** 1-11. PMID: 6033260
22. Auvinen JP, Paananen MVJ, Tammelin TH. **Musculoskeletal pain combinations in adolescents**. *Spine* (2009) **34** 1192-1197. PMID: 19444067
23. Mäki T, Oura P, Paananen M. **Longitudinal analysis of paraspinal muscle cross-sectional area during early adulthood - a 10-year follow-up MRI study**. *Sci Rep* (2019) **9** 19497. PMID: 31862980
24. Sipola-Leppänen M, Vääräsmäki M, Tikanmäki M. **Cardiometabolic risk factors in young adults who were born preterm**. *Am J Epidemiol* (2015) **181** 861-873. PMID: 25947956
25. West S, Lashen H, Bloigu A. **Irregular menstruation and hyperandrogenaemia in adolescence are associated with polycystic ovary syndrome and infertility in later life: northern Finland birth cohort 1986 study**. *Hum Reprod* (2014) **29** 2339-2351. PMID: 25085801
26. Sund R.. **Quality of the Finnish hospital discharge register: a systematic review**. *Scand J Public Health* (2012) **40** 505-515. PMID: 22899561
27. McGrath S, Sohn H, Steele R. **Meta-analysis of the difference of medians**. *Biom J* (2020) **62** 69-98. PMID: 31553488
28. 28Eurostat. Body mass index (BMI) by sex, age and income quintile. Https://AppssoEurostatEcEuropaEu/Nui/ShowDo 2021.. (2021)
29. Wolfenden L, Wyse R, Nichols M. **A systematic review and meta-analysis of whole of community interventions to prevent excessive population weight gain**. *Prev Med* (2014) **62** 193-200. PMID: 24518005
30. Gillespie KM. **Type 1 diabetes: pathogenesis and prevention**. *Can Med Assoc J* (2006) **175** 165-170. PMID: 16847277
31. Zimmet P, Alberti KGMM, Shaw J.. **Global and societal implications of the diabetes epidemic**. *Nature* (2001) 414(6865):782–787. PMID: 11473310
32. Bluestone JA, Herold K, Eisenbarth G.. **Genetics, pathogenesis and clinical interventions in type 1 diabetes**. *Nature* (2010) **464** 1293-1300. PMID: 20432533
33. Vaiserman A, Lushchak O.. **Developmental origins of type 2 diabetes: focus on epigenetics**. *Ageing Res Rev* (2019) **55** 100957. PMID: 31473332
34. Galea S, Tracy M.. **Participation rates in epidemiologic studies**. *Ann Epidemiol* (2007) **17** 643-653. PMID: 17553702
35. Kerkelä M, Gissler M, Veijola J.. **Association of participation in the Northern Finland birth cohort 1986 with mental disorders and suicidal behaviour**. *Epidemiol Health* (2022) **44** e2022005-0. PMID: 34990532
|
---
title: Triptolide protects against podocyte injury in diabetic nephropathy by activating
the Nrf2/HO-1 pathway and inhibiting the NLRP3 inflammasome pathway
authors:
- Chenlei Lv
- Tianyang Cheng
- Bingbing Zhang
- Ke Sun
- Keda Lu
journal: Renal Failure
year: 2023
pmcid: PMC10035962
doi: 10.1080/0886022X.2023.2165103
license: CC BY 4.0
---
# Triptolide protects against podocyte injury in diabetic nephropathy by activating the Nrf2/HO-1 pathway and inhibiting the NLRP3 inflammasome pathway
## Abstract
Objectives: Diabetic nephropathy (DN) is the most common microvascular complication of diabetes mellitus. This study investigated the mechanism of triptolide (TP) in podocyte injury in DN.
Methods: DN mouse models were established by feeding with a high-fat diet and injecting with streptozocin and MPC5 podocyte injury models were induced by high-glucose (HG), followed by TP treatment. Fasting blood glucose and renal function indicators, such as 24 h urine albumin (UAlb), serum creatinine (SCr), blood urea nitrogen (BUN), and kidney/body weight ratio of mice were examined. H&E and TUNEL staining were performed for evaluating pathological changes and apoptosis in renal tissue. The podocyte markers, reactive oxygen species (ROS), oxidative stress (OS), serum inflammatory cytokines, nuclear factor-erythroid 2-related factor 2 (Nrf2) pathway-related proteins, and pyroptosis were detected by Western blotting and corresponding kits. MPC5 cell viability and pyroptosis were evaluated by MTT and Hoechst 33342/PI double-fluorescence staining. Nrf2 inhibitor ML385 was used to verify the regulation of TP on Nrf2.
Results: TP improved renal function and histopathological injury of DN mice, alleviated podocytes injury, reduced OS and ROS by activating the Nrf2/heme oxygenase-1 (HO-1) pathway, and weakened pyroptosis by inhibiting the nod-like receptor (NLR) family pyrin domain containing 3 (NLRP3) inflammasome pathway. In vitro experiments further verified the inhibition of TP on OS and pyroptosis by mediating the Nrf2/HO-1 and NLRP3 inflammasome pathways. Inhibition of Nrf2 reversed the protective effect of TP on MPC5 cells.
Conclusions: Overall, TP alleviated podocyte injury in DN by inhibiting OS and pyroptosis via Nrf2/ROS/NLRP3 axis.
## Introduction
Diabetes mellitus (DM) is a chronic metabolic disorder chiefly characterized by hyperglycemia, and the incidence of DM is increasing worldwide [1]. Diabetic nephropathy (DN), the most common microvascular complication of DM, is primarily featured by albuminuria and progressive loss of kidney function [2,3]. Intrinsically, podocyte injury and loss are the main causes of proteinuria, and podocyte injury is mainly attributed to oxidative stress (OS) and inflammatory injury induced by high-glucose (HG) [4,5]. To date, the treatment and management strategies of DN mainly focus on reducing body weight, blood glucose, and blood pressure, and the commonly used first-line treatment is renin-angiotensin system inhibitors, including angiotensin-converting enzyme inhibitors or angiotensin receptor blockers [6,7]. Dapagliflozin (DAPA) is an inhibitor of sodium-glucose cotransporter-2 and is often used as the first choice for DN treatment in clinical practice [8,9]. However, these therapies have certain limitations and side effects. It is interesting to note that Chinese medicine has numerous targets and good potential in the treatment of DN [10]. Consequently, it is of great significance to find potent targets and therapeutic modalities for DN.
Recent evidence suggests that the production and accumulation of reactive oxygen species (ROS) caused by hyperglycemia and hyperlipidemia lead to OS, which is critical in DN [11,12]. In particular, excessive ROS can cause kidney damage by promoting lipid oxidation, thus resulting in podocyte injury and inflammation [13,14]. Nuclear factor-erythroid 2-related factor 2 (Nrf2), a key factor in cell regulation of OS, is the most sensitive signal for scavenging intracellular ROS to resist OS and is also one of the therapeutic targets of DN [15,16]. Interestingly, Nrf2 can inhibit the activation of Nod-like receptor (NLR) family pyrin domain containing 3 (NLRP3) inflammasome by scavenging ROS [17]. Inflammasomes are multi-protein complexes expressed in myeloid cells, present in the cytoplasm of many cell types, and can induce innate immune responses by sensing damage signals and microbial attacks [18]. NLRP3 inflammasome is the most widely studied complex, consisting of NLRP3, apoptosis-associated speck-like protein (ASC), and pro-caspase-1 [19]. Compelling evidence suggests the involvement of the NLRP3 inflammasome pathway in the pathological process of type 2 DM [20]. Meanwhile, the NLRP3 inflammasome pathway-mediated pyroptosis is essential in renal injury [21]. Recent research has been reported that Festine reduces DN-provoked podocyte injury by inhibiting NLRP3 inflammasome [22]. In particular, ROS is an important factor in NLRP3 inflammasome activation, and inhibition of ROS levels in cells can impede NLRP3 inflammasome activation [21,23]. The aforementioned studies imply that the Nrf2/ROS/NLRP3 pathway may be a therapeutic target for DN.
Triptolide (TP) is an alkaloid extracted from traditional Chinese medicine Tripterygium wilfordii, which has anti-inflammatory, antioxidant, hypoglycemic, lipid-lowering, and antitumor effects [24]. Cumulative evidence suggests the protective effect of TP on DN [25,26]. For instance, Han F et al. identified in their work that TP could inhibit the PDK1/Akt/mTOR pathway in human renal mesangial cells to protect against DN [27]. Hence, TP may be one of the candidate drugs for treating DN [28]. Considering all of this evidence, we hypothesized that TP can protect against DN podocyte injury by activating the Nrf2/HO-1 pathway to reduce ROS levels, inhibit the NLRP3 inflammatory pathway, and regulate OS and pyroptosis. The specific objective of this study was to investigate the therapeutic effect and mechanism of TP on DN, aiming to provide new insights into the pathogenesis and treatment of DN.
## Ethics statement
Animal experiments were approved by the Animal Ethics Committee of The Third Affiliated Hospital, Zhejiang Chinese Medical University (Approval number: IACUC-20210406-15), and adequate measures were taken to minimize the mouse number and pain or discomfort. The study was carried out under ARRIVE guidelines.
## Establishment of animal models
The male C57BL/6J mice aged 6–8 weeks were provided by Beijing Experimental Animal Research Center (Beijing, China) [Animal License No. SYXK (Beijing) 2021-0045]. The mice were fed with a standard diet and free drinking water and reared at 23 ± 1 °C under standard light/dark cycles and $60\%$ humidity for 1 week. As stated previously [29], the DN mouse model was established by feeding with a high-fat diet (HFD) for 2 months and then intraperitoneally injecting with streptozotocin (STZ, 50 mg/kg, Solarbio, Beijing, China) for continuous 7 days. Normal mice were administered with an equal amount of sodium citrate buffer. Seven days after injection, fasting blood glucose (FBG) levels were measured using a glucose meter (Roche, Germany) and urine samples within 24 h were collected in a metabolic cage, followed by the determination of 24 h urine albumin (24 h UAlb) using the Bradford method. Mice with FBG ≥ 16.9 mmol/L and increased 24 h UAlb were considered DN mice.
## Group treatment of animals
Normal mice or successfully-established DN mice were assigned as follows, with 8 mice per group: [1] normal control (NC) group; [2] DN group; [3] DN + dimethyl sulfoxide (DMSO) group: DN mice were administrated with 1 mL normal saline containing $0.4\%$ DMSO; [4] DN + TP group: DN mice were intragastrically administrated with 100 μg/kg/d TP (Yuanye Bio-Technology, Shanghai, China). The dosage of TP was determined based on previous research [30] and pre-experiments. The purity of TP determined by high-performance liquid chromatography purity was >$98\%$. TP was dissolved in DMSO and diluted with normal saline; [5] DN + DAPA group: DN mice were intragastrically administrated with 1 mg/kg/d DAPA (Yuanye Bio-Technology) [31]. After 12 weeks of continuous treatment, all urine samples were collected in the metabolic cage for 24 h, and FBG and 24 h UAlb were determined the next morning. The survival rate of mice was shown in Table 1, and eventually, six surviving mice in each group were selected for subsequent analyses.
**Table 1.**
| Group | NC | DN | DN + DMSO | DN + TP | DN + DAPA |
| --- | --- | --- | --- | --- | --- |
| Number of survivals | 8/8 | 6/8 | 7/8 | 8/8 | 8/8 |
| Survival rate | 100% | 75% | 87.5% | 100% | 100% |
## Determination of serum indexes
Following previous work [32], serum indexes were determined. After treatment, the mouse weight was weighed and recorded. Mice were deeply anesthetized with $1\%$ pentobarbital sodium (50 mg/kg body weight) and sacrificed by drawing cardiac blood. The blood samples were centrifuged at 1000 g for 5 min at 4 °C and the serum was collected. Part of the serum was used to detect serum creatinine (SCr) and blood urea nitrogen (BUN) in mice using an automated biochemical analyzer (ADVIA 1650 Chemistry System; Bayer, Leverkusen, Germany), and the rest of the serum was used for subsequent detection.
## Hematoxylin-eosin (H&E) staining
As previously described [33], after the mice were sacrificed, bilateral kidneys were removed, the left kidneys were weighed, and the ratio of kidney/body weight (K/B Weight) was calculated. Thereafter, the kidney was fixed with $10\%$ paraformaldehyde in a refrigerator at 4 °C for 24 h, embedded in paraffin, and sliced into sections of 5 μm thickness. Sections were stained using H&E staining kits (G1120-100, Solarbio). Paraffined sections were regularly dewaxed and hydrated, stained with hematoxylin for 5 min, and then rinsed with tap water. Following differentiation with $1\%$ hydrochloric acid ethanol for 5 s, the sections were rinsed with running water again and treated with $1\%$ ammonia for 3–5 s. After washing with tap water, the sections were restained with eosin for 3 min, rinsed with water, dehydrated with gradient ethanol, cleared with xylene, and then sealed with neutral gum. The sections were observed and photographed under a light microscope (Olympus, Tokyo, Japan) and the pathological changes and the area of inflammatory infiltration in mouse kidney tissues were evaluated by Image Pro-Plus (Media Cybernetics, MD, USA). Five slices from each of the six mice in each group were collected, and each slice was randomly selected for 5 fields, with the results averaged.
## Terminal deoxynucleotidyl transferase-mediated dUTP-biotin nick end labeling (TUNEL) staining
The apoptosis in mouse kidney tissues was measured using TUNEL detection kits (C1098, Beyotime, Shanghai, China) [33]. Paraffined sections of mouse kidney tissue were routinely dewaxed, hydrated, washed with phosphate-buffered solution (PBS), and incubated with 20 μg/mL protease K at room temperature for 30 min. After soaking with PBS, sections were incubated with a mixture of enzyme solution and TUNEL standard solution (diluted at 1:10) at room temperature for 60 min. Next, sections were restained with hematoxylin, dehydrated with gradient ethanol, cleared with xylene for 2 min, and sealed with neutral gum. The sections were observed and photographed under a light microscope, and five fields were randomly selected from each section. Image Pro-Plus software was utilized to count the percentage of apoptotic cells (%), and the average value was calculated.
## Transmission electron microscope (TEM) observation
Following previous work [33,34], the 1 mm3 of mouse kidney tissue was fixed with $3.75\%$ glutaraldehyde and $1\%$ osmium acid. Subsequently, the tissues were embedded in epoxy resin and stained with uranyl acetate and citric acid. Afterward, TEM (Olympus, Tokyo, Japan) was used to observe the ultrastructural changes in mouse podocytes. Image-Pro Plus software was used to analyze the changes in the average number of podocytes and podocyte foot processes in glomerulus. Five slices from each of the six mice in each group were collected, and each slice was randomly selected for five fields, with the results averaged.
## Immunohistochemistry
As mentioned in prior research [35], the sections of mouse kidney tissues were incubated with $3\%$ hydrogen peroxide to eliminate endogenous peroxidase activity and then heated using a microwave oven for antigen repair. Later, the sections were blocked with $5\%$ bovine serum albumin at 37 °C for 30 min and incubated with diluted primary anti-Nephrin (1:2000, ab216341, Abcam, UK) overnight at 4 °C. After washing with PBS, the sections were incubated for 30 min with biotin-labeled immunoglobulin G (IgG) H&L secondary antibody (1:1000, ab207995) at 37 °C. Following washing with PBS, the sections were incubated for 15 min with horseradish peroxidase (HRP)-labeled streptomycin at room temperature, dripped with 3,3′-diaminobenzidine chromogenic solution (P0202, Beyotime) in dark conditions, and rinsed with tap water. The sections were restained with hematoxylin, dehydrated with gradient ethanol, cleared with xylene, and sealed with neutral gum, followed by observation and photography under a light microscope. Five visual fields were randomly selected for each section. Image Pro-Plus software was adopted to analyze the percentage of positive cells (%). The brown-yellow area was regarded as the positive expression, and the results were averaged.
## Cell culture
Conditional immortalized mouse podocytes MPC5 (Cell Bank of Chinese Academy of Sciences, Shanghai, China) were cultured in RPMI 1640 medium supplemented with $10\%$ fetal bovine serum and $1\%$ penicillin-streptomycin (Sigma-Aldrich, MI, USA) [32] and were induced with recombinant interferon-γ (Sigma-Aldrich) at 33 °C for proliferation. After 10–12 days, cells were cultured in a condition free of interferon-γ at 37 °C until the cells were fully differentiated into mature podocytes. After differentiation, cells were cultured in a constant temperature incubator with $5\%$ CO2 at 37 °C.
## Cell grouping
As previously described [36], MPC5 cells at the logarithmic growth stage were harvested and grouped as follows: [1] blank group: treated with 5 mM glucose; [2] HG group: treated with 25 mM glucose; [3] HG + DMSO group: treated with 25 mM glucose and $0.1\%$ DMSO; [4] HG + TP group: treated with 25 mM glucose and 10 μM TP (the administration method and dosage of TP were referred to the literature [37]); [5] HG + DAPA group: treated with 25 mM glucose and 2 μM DAPA, with the DAPA dosage referred to the literature [31]; [6] HG + DMSO + ML385: treated with 25 mM glucose, $0.1\%$ DMSO, and 5 μM ML385 (Nrf2 inhibitor, ab287109, Abcam), with the concentration of ML385 referred to the instruction and reference [38]; [7] HG + TP + ML385: treated with 25 mM glucose, 10 μM TP, and 5 μM ML385; [8] HG + TP + si-NC: transfected with si-NC plasmid (GenePharma, Shanghai, China) for 24 h and then treated with 25 mM glucose and 10 μM TP; [9] HG + TP + si-Nrf2: transfected with si-Nrf2 plasmid (GenePharma) for 24 h and next treated with 25 mM glucose and 10 μM TP. After 48 h of treatment, MPC5 cells were collected for subsequent experimentation.
## 3-(4,5-Dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide (MTT) assay
The cell viability was detected by MTT assay [39]. MPC5 cells were cultured in 96-well plates (1 × 104 cells/well) and incubated with MTT solution (M405849-1Set, Aladdin, Shanghai, China) at a final concentration of 1 mg/mL for 4 h at 37 °C. Formazan crystal was dissolved with 150 μL/well of DMSO and the absorbance at 570 nm was measured with a microplate reader.
## Cell pyroptosis detection
The cell pyroptosis was examined by Hoechst 33342/propidium iodide (PI) double fluorescence staining kits (CA1120, Solarbio) [40]. After different treatments in groups, MPC5 cells at the logarithmic growth stage were cultured in 6-well plates and stained with 10 μL Hoechst 33342 solution at 37 °C for 10 min under dark conditions. Thereafter, the cells were stained with 5 μL PI in the dark at 25 °C for 15 min. Staining results were observed using a fluorescence microscope (Nikon 80i, Nikon, Japan) and images were collected.
## Immunofluorescent staining
Following previous work [41], cell slides were prepared and fixed with $4\%$ paraformaldehyde. Following rinsing with PBS, the slides were sealed with goat serum at room temperature for 30 min and then incubated with the anti-Nephrin (1:500, ab216341, Abcam) overnight at 4 °C. The slides were washed in PBS with $0.05\%$ Tween-20 three times and next incubated with goat anti-rabbit secondary antibody Alexa Fluor® 594 IgG H&L (2 μg/mL, ab150080) in dark conditions. The nuclei were stained with 4′,6-diamidino-2-phenylindole and photographed under a fluorescence microscope.
## Measurement of ROS
ROS levels in kidney tissue or MPC5 cells were detected by 2′,7′-dichlorodihydrofluorescein diacetate (DCFH-DA) fluorescence probe and ROS levels in fresh frozen sections of mouse kidney tissue were measured using the kits (HR7814, Biolab, Beijing, China) [41]. In brief, the unfixed frozen kidney tissue sections of 10 μm thickness were added with 200 μL washing solution at room temperature, with the solution spreading over the surface of sections. After the sections were allowed to stand for 5 min, the washing solution was carefully aspirated and then 100 μL staining working solution was added dropwise. Following incubation for 30 min at 37 °C under conditions devoid of light, the staining solution was removed. Thereafter, sections were rinsed 2–3 times with PBS and mounted with glycerol, followed by the detection of fluorescence intensity under a fluorescence microscope. Additionally, ROS levels in MPC5 cells were measured using the kits (S0033; Beyotime). The collected cells were diluted with PBS, seeded into 6-well plates, and then incubated with DCFH-DA solution at a final concentration of 5 mM for 30 min at 37 °C. After washing with PBS, the fluorescence intensity of the samples was detected by a fluorescence microscope, with the emission wavelength set at 530 nm and the excitation wavelength set at 485 nm. The operation steps were carried out according to the kit instructions.
## Detection of OS-related indicators and inflammatory factors
Following previous work [34], kidney tissue homogenate or MPC5 cells were collected. Total protein was extracted using radioimmunoprecipitation assay lysate (W063-1-1, Jiancheng Bioengineering Institute, Nanjing, China). Protein concentration in the supernatant was measured using the bicinchoninic acid method (P0012S, Beyotime). Subsequently, the supernatant was diluted 20 times and 50 μL of diluted supernatant was collected to detect the content of the target protein. The levels of OS-related indexes malondialdehyde (MDA) (A003-1-2), superoxide dismutase (SOD) (A001-3-2), and glutathione (GSH) (A006-1-1) were detected by kits (Jiancheng). The mouse serum or MPC5 cells were collected, followed by determining the levels of inflammatory cytokines interleukin (IL)-1β (H002) and IL-18 (H015) by ELISA kits (Jiancheng).
## Western blotting (WB)
The total protein was extracted from mouse kidney tissues or cells using radioimmunoprecipitation assay lysate (W063-1-1, Jiancheng) [34]. After the protein concentration was measured using the bicinchoninic acid kits (P0012S, Beyotime), the proteins (40 μg) were isolated by $10\%$ sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) and transferred onto polyvinylidene fluoride membranes. Next, the membranes were blocked with $5\%$ skim milk for 1 h and incubated with rabbit primary antibodies anti-Nephrin (1:1000, ab216341, Abcam), anti-Podocin (1:10000, ab181143), anti-Nrf2 (1:1000, ab92946), anti-heme oxygenase-1 (HO-1) (1:1000, ab68477), anti-NLRP3 (1:1000, ab270449), anti-ASC (1:5000, ab155970), anti-Pro Caspase-1 (1:200, ab238972), anti-Gasdermin D N-terminal domain (GSDMD-N) (1:1000, ab215203), and anti-glyceraldehyde-3-phosphate dehydrogenase (GAPDH) (1:1000, ab9485) overnight at 4 °C. After washing with PBS, the membranes were incubated with HRP-labeled goat anti-rabbit secondary antibody IgG H&L (1:20000, ab97051) at room temperature for 30 min. After that, the membranes were developed by enhanced chemiluminescence and then observed and photographed. Image-Pro Plus 6.0 (Media Cybernetics, Inc., MD, USA) was adopted to analyze the relative expression of different proteins, with GAPDH as an internal reference.
## Statistical analysis
SPSS 21.0 software (IBM Corp. Armonk, NY, USA) and GraphPad Prism 8.0 software (GraphPad Software Inc., CA, USA) were used for statistical analysis and plotting of data. Statistical data tested by Shapiro-Wilk were normally distributed and expressed as mean ± standard deviation (SD). The t-test was used for data comparison between two groups, one-way analysis of variance (ANOVA) was used for data comparison among multiple groups, and Tukey’s test was used for the post-hoc test. p-Value was obtained from the bilateral tests. $p \leq 0.05$ was considered statistically significant.
## TP improved renal function and histopathological injury in DN mice
The DN mouse model was established by HFD combined with STZ injection, followed by treatment with TP, with DAPA as the positive control. FBG, 24 h UAlb, SCr, and BUN of mice were measured, and the K/B weight ratios of mice were calculated. Compared with the NC group, FBG, 24 h UAlb, Scr, and BUN in the DN group were increased (Figures 1(A–D), $p \leq 0.001$), as well as the ratio of K/B weight (Figure 1(E), $p \leq 0.001$). Furthermore, H&E staining manifested that the glomerular morphology of mice in the NC group was regular and the renal cell morphology was normal; however, the renal tissue cells in the DN group were swollen and morphologically altered, with increased inflammatory infiltration areas (Figure 1(F), $p \leq 0.01$). TUNEL staining showed that apoptosis of renal tissue cells was enhanced in the DN group (Figure 1(G), $p \leq 0.001$). Compared with the DN group, FBG, 24 h UAlb, SCr, and BUN in the DN + TP group were decreased (Figures 1(A–D), all $p \leq 0.01$), along with a decrease in K/B weight ratio to the normal range (Figure 1(E), $p \leq 0.05$). After TP treatment, the edema of mouse kidney tissue was alleviated, the cell morphology tended to be normal, the inflammatory infiltration area was reduced (Figure 1(F)), and the apoptosis in renal tissues was reduced in DN mice (Figure 1(G), $p \leq 0.001$). In parallel, DMSO conferred no obvious effect on DN mice and the therapeutic effect of TP was comparable to that of DAPA (Figures 1(A–G), all $p \leq 0.05$). Altogether, TP can improve renal function and histopathological damage in DN mice.
**Figure 1.:** *TP improved renal function and histopathological injury in DN mice. The DN mouse model was induced by HFD feeding combined with STZ injection. After 12 weeks of treatment with TP, (A) FBG was detected by glucose meter; (B) 24 h UAlb in mice was detected by Bradford method; (C,D) The serum contents of SCr and BUN in mice were detected using an automated biochemical analyzer; (E) The ratio of K/B weight was examined; (F) H&E staining was used to analyze the pathological changes of renal tissues in mice and to quantify the percentage of inflammatory areas; (G) TUNEL staining was used to detect the percentage of apoptosis in renal tissues. Data were expressed as mean ± SD, N = 6. One-way ANOVA was used for data comparison among multiple groups and Tukey’s test was used for the post-hoc test. p-Value was obtained from the bilateral tests. **p < 0.01, ***p < 0.001.*
## TP alleviated podocyte injury in DN mice
The pathogenesis of DN and the production of proteinuria are extremely related to podocyte injury [42,43]. TEM observation revealed that the average number of podocytes and podocyte foot processes in glomerulus was decreased and the number of hiatus of foot processes was reduced in the DN group, with the disappearance of slit diaphragm (Figure 2(A)). Subsequently, immunohistochemistry discovered that DN mice had a reduced number of Nephrin-positive podocytes in kidney tissues (Figure 2(B), $p \leq 0.001$). WB manifested that the protein levels of Nephrin and Podocin in kidney tissues of DN mice were decreased (Figure 2(C), $p \leq 0.001$), indicating significant damage to renal podocytes in the DN group. After TP treatment, the number of podocytes, foot processes, and hiatus of foot processes was elevated, podocyte characteristics were partially restored (Figure 2(A)), Nephrin-positive cells were increased (Figure 2(B), $p \leq 0.01$), and the protein levels of Nephrin and Podocin were enhanced (Figure 2(C), both $p \leq 0.05$). Additionally, there was no significant impact of DMSO on podocytes and no significant difference between the DN + TP group and the DN + DAPA group (Figures 2(A–C), all $p \leq 0.05$). These results verified the protective effect of TP on glomerular podocytes.
**Figure 2.:** *TP alleviated podocyte injury in DN mice. (A) The morphological and structural changes of podocytes in mouse kidney tissue were observed by TEM and the average number of podocytes and podocyte foot processes in glomeruli were quantitatively analyzed; (B) The number of Nephrin-positive cells was detected by immunohistochemistry; (C) The protein levels of Nephrin and Podocin in mouse kidney tissues were detected by WB. Data were expressed as mean ± SD, N = 6. One-way ANOVA was used for data comparison among multiple groups and Tukey’s test was used for the post-hoc test. **p < 0.01, ***p < 0.001.*
## TP ameliorated OS injury and activated the Nrf2/HO-1 pathway in the renal tissue of DN mice
Evidence suggests the close association between DN occurrence and OS imbalance [11,12]. The ROS, MDA, SOD, and GSH levels in mouse renal tissue were determined. It was found that ROS and MDA levels were enhanced and SOD and GSH levels were decreased in the DN group (Figures 3(A–D), all $p \leq 0.05$). In comparison to the DN group, the DN + TP group had diminished ROS and MDA levels as well as increased SOD and GSH levels (Figures 3(A–D), all $p \leq 0.05$). These results uncovered that TP can ameliorate OS injury in the renal tissue of DN mice. It is noteworthy that the Nrf2/HO-1 pathway is the core pathway against OS [44,45]. WB assay unveiled that relative to the NC group, the protein levels of Nrf2 and HO-1 in the DN group were decreased, while TP activated the levels of Nrf2 and HO-1 (Figure 3(E), $p \leq 0.05$). In addition, DMSO exerted no significant effect on DN mice, and in comparison with the DN + DAPA group, DN + TP group showed no significant difference in the above indexes (Figures 3(A–E), all $p \leq 0.05$). In summary, TP may protect against DN by stimulating the Nrf2/HO-1 pathway to reduce the ROS level and alleviate OS injury.
**Figure 3.:** *TP improved OS injury and activated the Nrf2/HO-1 pathway in renal tissue of DN mice. (A) DCFH-DA kits were used to detect ROS levels in mouse renal tissues; (B–D) The levels of OS-related enzymes, such as MDA, SOD, and GSH were detected; E: The levels of the Nrf2/HO-1 pathway-related proteins in mouse kidney tissues were detected by WB. Data were expressed as mean ± SD, N = 6. One-way ANOVA was used for data comparison among multiple groups and Tukey’s test was used for the post-hoc test. **p < 0.01, ***p < 0.001.*
## TP reduced pyroptosis of renal tissue in DN mice by inhibiting the NLRP3 inflammasome pathway
DN is an inflammatory disease, and pyroptosis caused by excessive inflammatory response is closely related to the onset and progression of DN [21,46]. To delineate whether TP can reduce the inflammatory response in DN renal tissue, we determined the levels of inflammatory cytokines IL-1β and IL-18 in the serum of mice. As documented in Figure 4(A), the secretion levels of IL-1β and IL-18 in the DN group were visibly higher than those in the NC group, while decreased after TP treatment. The IL-1β and IL-18 are downstream inflammatory factors regulated by the NLRP3 inflammasome pathway. To ascertain the effect of TP on the NLRP3 inflammasome pathway, we further measured the levels of related proteins by WB. The protein levels of NLRP3, ASC, pro-Caspase-1, and GSDMD-N were up-regulated in DN mice but diminished upon TP treatment (Figures 4(B,C), $p \leq 0.05$). In addition, DMSO unleashed no evident impact on DN mice, and there was no significant difference in each index between the DN + TP group and the DN + DAPA group (Figures 4(A–C), all $p \leq 0.05$). In short, TP may inhibit the inflammatory response by governing the NLRP3 inflammasome pathway, thereby reducing pyroptosis of renal tissue in DN mice.
**Figure 4.:** *TP reduced pyroptosis of renal tissue in DN mice by inhibiting the NLRP3 inflammasome pathway. (A) ELISA kits were used to detect the secretion levels of inflammatory cytokines IL-1β and IL-18 in the serum of mice; (B,C) WB was used to detect the protein level of cell pyroptosis markers mediated by NLRP3 inflammasome in mouse kidney tissue. Data were expressed as mean ± SD, N = 6. One-way ANOVA was used for data comparison among multiple groups and Tukey’s test was used for the post-hoc test. **p < 0.01, ***p < 0.001.*
## TP attenuated MPC5 cell damage induced by HG
The in vitro cell model of DN was established by treating MPC5 cells with HG [47], followed by intervention with TP, with DAPA as a positive control drug. Subsequently, immunofluorescent staining revealed that HG treatment reduced the level of Nephrin in MPC5 cells, while TP treatment partially restored it (Figure 5(A), all $p \leq 0.01$). MTT assay was adopted to testify cell proliferation and Hoechst 33342/PI double fluorescence staining was utilized to examine pyroptosis. Compared to the blank group, the cell viability of the HG group was reduced (Figure 5(B), $p \leq 0.001$), and the number of PI-positive cells was up-regulated (Figure 5(C), $p \leq 0.001$). After TP treatment, the viability of MPC5 cells was increased and PI-positive cells were decreased (Figures 5(B,C), $p \leq 0.01$). Meanwhile, DMSO caused no obvious effect on HG-treated MPC5 cells, and the therapeutic effect of TP was not significantly different from that of DAPA (Figures 5(A–C), all $p \leq 0.05$). Overall, TP can alleviate HG-induced podocyte injury.
**Figure 5.:** *TP attenuated MPC5 cell damage induced by HG. (A) The positive expression of podocyte marker Nephrin in MPC5 cells was detected by immunofluorescent staining; (B) MPC5 cell viability was detected by MTT assay; (C) Hoechst 33342/PI double fluorescence staining was used to detect the pyroptosis level of MPC5 cells. Data were expressed as mean ± SD, and the cell experiments were independently repeated three times. One-way ANOVA was used for data comparison among multiple groups and Tukey’s test was used for the post-hoc test. **p < 0.01, ***p < 0.001.*
## TP activated the Nrf2 pathway to reduce HG-induced OS in MPC5 cells
Subsequently, the detection with DCFH-DA fluorescence probe revealed that HG treatment up-regulated ROS levels in MPC5 cells, while TP and DAPA both inhibited ROS levels (Figure 6(A), $p \leq 0.01$). WB assay revealed that Nrf2 and HO-1 levels were downregulated in the HG group but up-regulated by treatment of TP or DAPA (Figure 6(B), all $p \leq 0.05$). Relative to the blank group, MDA level in the HG group was enhanced, while SOD and GSH levels were reduced (Figures 6(C–E), all $p \leq 0.05$). However, after TP or DAPA treatment, MDA levels decreased while SOD and GSH levels increased (Figures 6(C–E), all $p \leq 0.05$). Likewise, DMSO resulted in no evident impact on HG-treated MPC5 cells, and there was no significant difference in the therapeutic effect between TP and DAPA. In brief, TP alleviated OS injury of MPC5 cells induced by HG by reducing ROS levels through the Nrf2/HO-1 pathway.
**Figure 6.:** *TP activated the Nrf2 pathway to reduce OS induced by HG in MPC5 cells. (A) ROS levels in MPC5 cells were detected by a DCFH-DA fluorescence probe; (B) WB was used to detect Nrf2 and HO-1 protein levels; (C–E) The MDA, SOD, and GSH levels were determined. Data were expressed as mean ± SD; cell experiments were independently repeated three times. One-way ANOVA was used for data comparison among multiple groups and Tukey’s test was used for the post-hoc test. *p < 0.05, **p < 0.01, ***p < 0.001.*
## TP attenuated pyroptosis of HG-induced MPC5 cells by blocking the NLRP3 inflammasome pathway
The NLRP3 inflammasome-related protein levels were determined by WB, which demonstrated that NLRP3, ASC, pro-Caspase-1, and GSDMD-N levels in the HG group were higher than those in the blank group, while these protein levels in the TP + HG group were lower than those in the HG group (Figure 7(A), all $p \leq 0.05$). ELISA discovered that the secretion levels of IL-1β and IL-18 were up-regulated in the HG group but reduced after TP intervention (Figure 7(B)). DMSO had no significant effect on HG-treated MPC5 cells, and there was no significant difference in the therapeutic effect between TP and DAPA (Figures 7(A,B), all $p \leq 0.05$). These results suggest that TP can reduce pyroptosis and inflammatory damage induced by HG in MPC5 cells by suppressing the NLRP3 inflammasome pathway.
**Figure 7.:** *TP alleviated pyroptosis in HG-induced MPC5 cells by inhibiting the NLRP3 inflammasome pathway. (A) WB was used to detect the protein levels of pyroptosis markers in NLRP3 inflammasome in MPC5 cells; (B) The secretion levels of IL-1β and IL-18 were detected by ELISA kits. Data were expressed as mean ± SD, and cell experiments were independently repeated three times. One-way ANOVA was used for data comparison among multiple groups and Tukey’s test was used for the post-hoc test. *p < 0.05, **p < 0.01, ***p < 0.001.*
## TP protected HG-treated MPC5 cells by activating Nrf2 to scavenge ROS and inhibit the NLRP3 inflammasome pathway
To further verify whether Nrf2 is a target of TP, MPC5 cells were treated with the Nrf2 inhibitor ML385. WB discovered that the HG + DMSO + ML385 group had lower protein levels of Nrf2 and HO-1 than the HG + DMSO group, and the HG + TP + ML385 group also had lower levels than the HG + TP group (Figure 8(A), all $p \leq 0.001$), indicating that Nrf2 pathway was inhibited by ML385. After ML385 treatment, cell viability was reduced (Figure 8(B), all $p \leq 0.01$), MDA levels were enhanced, SOD and GSH levels were decreased (Figures 8(C–E), $p \leq 0.01$), and ROS levels were elevated (Figure 8(F), $p \leq 0.001$). These results evinced that ML385 inhibited the activation of the Nrf2 pathway induced by TP and increased ROS levels. Additionally, after ML385 treatment, the protein levels of NLRP3, ASC, pro-Caspase-1, and GSDMD-N were up-regulated (Figure 8(G), $p \leq 0.001$) and the levels of IL-1β and IL-18 were elevated (Figure 8(H), $p \leq 0.01$). MPC5 cells were transfected with si-Nrf2 and treated with a combination of HG and TP. The results unraveled that si-Nrf2 reversed the protective effect of TP on MPC5 (Figures 8(A–H), all $p \leq 0.05$). The above results evinced that TP could impede the NLRP3 inflammasome pathway by activating Nrf2 and scavenging ROS, thereby protecting HG-treated MPC5 cells.
**Figure 8.:** *TP protected HG-treated MPC5 cells by activating Nrf2 to scavenge ROS and inhibit the NLRP3 inflammasome pathway. (A) The protein levels of Nrf2 and HO-1 in MPC5 cells were detected by WB; (B) MPC5 cell activity was detected by MTT; (C–E) The levels of MDA, SOD, and GSH in MPC5 cells were detected; (F) ROS levels in MPC5 cells were detected by a DCFH-DA fluorescence probe; (G) WB was used to detect the levels of NLRP3 inflammasome pathway-related proteins; (H) The secretion levels of inflammatory cytokines IL-1β and IL-18 were detected by ELISA; Data were expressed as mean ± SD, and the cell experiments were independently repeated three times. Comparison between two groups was performed by the t-test. *p < 0.05, **p < 0.01, ***p < 0.001.*
## Discussion
DN, as a major complication of diabetes, is the leading cause of end-stage renal disease worldwide [48]. Recently, the functional and structural abnormalities of glomerular podocytes have been identified as one of the earliest events in the development of diabetic glomerular injury [49–51]. It is worth noting that TP could considerably reduce proteinuria and podocyte injury in DN animal models [52]. Our finding illustrated that TP inhibited OS and cell pyroptosis through the Nrf2/ROS/NLRP3 regulatory axis to protect against podocyte injury in DN.
Indeed, the protective effect of TP on DN has been elucidated in many reports [53–55]. Yet, the behind mechanism has not been fully elucidated so far. By treating mice with HFD combined with STZ injection and inducing MPC5 cells with HG, the DN mouse model and cell model were established, followed by treatment with TP. DN is a serious kidney disease characterized by kidney injury and tissue fibrosis [56]. In our current study, we reproduced the typical pathological changes in DN mouse kidneys, and TP treatment alleviated renal dysfunction and histopathological damage in DN mice, consistent with the reported protection of TP against renal histopathological injury in diabetic rats [25,57,58]. Several reports have shown the essential role of podocytes in the pathological mechanism of DN [43,59]. Specifically, podocyte loss and injury frequently occur in early DN patients, which may lead to severe proteinuria and renal damage [60,61]. Expectedly, we observed evident damage of renal podocytes in DN mice, while all indexes were recovered after TP treatment, with enhanced Nephrin and Podocin levels, consistent with the reported protection of TP on glomerular podocytes in DN rats [26,52].
Accumulated ROS may interact with fatty acids (polyunsaturated), leading to the formation of lipid peroxidation in renal tissues, which may ultimately lead to damage and toxicity [62]. On the other hand, OS is widely acknowledged as a major factor in DM-related complications, including DN [63,64], while TP is widely known for its good antioxidant effects [65,66]. The current study found that ROS levels and MDA levels in DN mice and HG-induced MPC5 cells were diminished after TP treatment, while SOD and GSH levels were increased, consistent with the prior finding that TP can visibly reduce renal inflammation and OS level in DN mice [67]. Overall, TP alleviated the OS injury of kidney tissue in DN mice and HG-induced MPC5 cells. On a separate note, the Nrf2/HO-1 pathway is crucial in anti-oxidant stress [44,45]. Subsequent experimentation in our study revealed that the levels of Nrf2 and HO-1 in DN mice and HG-induced MPC5 cells were decreased, whereas they were up-regulated after TP treatment. TP is capable of reducing the production of ROS and M1-type polarization by activating the Nrf2/HO-1 pathway in inflammatory bowel diseases [68]. Meanwhile, TP alleviates myocardial ischemia/reperfusion injuries in rats by activating the Nrf2/HO-1 pathway [69]. For the first time, our results reveal that TP may protect against DN by activating the Nrf2/HO-1 pathway to reduce ROS levels and OS injury.
Pyroptosis caused by an excessive inflammatory response is closely associated with DN [21,46]. Increased levels of inflammatory cytokines IL-1β and IL-18 are found in DN podocytes [70]. As expected, TP intervention diminished the secretion levels of IL-1β and IL-18 in DN mice and HG-induced MPC5 cells. TP can improve DN by regulating Th1/Th2 cell balance and reducing macrophage infiltration and levels of related-inflammatory factors in the kidney [71]. NLRP3 inflammasome may be involved in DN through activation of pyroptosis, and IL-1β and IL-18 are downstream inflammatory cytokines regulated by the NLRP3 inflammasome pathway [21,46,72]. Besides, the accumulation of ROS activates the NLRP3 pathway [73]. It is interesting to note that TP treatment partially reduced the protein levels of NLRP3 inflammasome-mediated pyroptotic markers in DN mice and HG-induced MPC5 cells. Likewise, TP induces GSDME-mediated pyroptosis of head and neck tumor cells by inhibiting mitochondrial hexokinase-ΙΙ [74]. Moreover, TP could prevent IgAN progression [75] and improve myocardial fibrosis by down-regulating NLRP3 inflammasomes [76]. Overall, this study strengthens the idea that TP may inhibit the inflammatory response and reduce the pyroptosis of renal podocytes in DN through the NLRP3 inflammasome pathway.
Nrf2 is a key transcription factor for cell regulation of OS, which can activate the transcription and expression of downstream anti-OS-related enzymes, such as HO-1 and SOD to eliminate the abnormal accumulation of ROS in cells, thereby alleviating OS injury [77,78]. ROS is crucial in the stimulation of NLRP3 inflammasome, and suppression of ROS levels in cells can inhibit the stimulation of NLRP3 inflammasome [21,23]. It can therefore be assumed that TP protects renal podocytes by activating the Nrf2 pathway to reduce ROS levels and inhibit the NLRP3 inflammasome pathway. Nrf2 inhibitor ML385 can eliminate COQ10-mediated renal protection [38]. Not surprisingly, our study illustrated that ML385 inhibited Nrf2 and raised ROS levels and si-Nrf2 reversed the protective effect of TP on MPC5 cells. This observation may support the hypothesis that TP alleviated inflammatory damage and pyroptosis of podocytes in DN by regulating the Nrf2/ROS/NLRP3 axis.
Previous studies have confirmed that DAPA can attenuate STZ-induced DN [79,80], so we chose DAPA as a positive control drug for TP. In our studies, the therapeutic effect of TP was not significantly different from that of DAPA, which indicated the potential of TP as a candidate drug for the treatment of DN. DAPA is an SGLT2 inhibitor with a single site of action. In contrast to DAPA, TP may be a multi-target therapeutic agent for DN. For instance, TP can inhibit the PDK1/Akt/mTOR pathway to restrain glomerular mesangial cell proliferation in DN [30]. TP impedes extracellular matrix accumulation in experimental DN by targeting the microRNA-137/Notch1 pathway [57]. TP alleviates podocyte epithelial-mesenchymal transition in DN via the kindlin-2 and EMT-related TGF-β/Smad pathway [26]. The aforementioned evidence has further evinced the superiority of TP as a therapeutic candidate for DN.
## Conclusion
To conclude, through animal and cell experiments, this study highlighted for the first time that TP protected podocytes from OS and pyroptosis in DN by activating the Nrf2 pathway and inhibiting the NLRP3 inflammasome pathway, which further clarified the mechanism of TP in reducing DN and provided references and therapeutic targets for new therapeutic drugs for DN. In the future, we will conduct more animal experiments and explore the direct regulation of TP on Nrf2 and NLRP3 in cell experiments, and further explore whether TP can protect against DN through epigenetic regulation, hoping to further clarify the protective mechanism of TP in DN in clinical application
## Ethical approval
The animal experiments were approved by the Animal Ethics Committee of The Third Affiliated Hospital, Zhejiang Chinese Medical University(Approval number: IACUC-20210406-15), and adequate measures were taken to minimize the mouse number and pain or discomfort. The study was carried out in accordance with ARRIVE guidelines.
## Author contributions
CLV guarantor of integrity of the entire study. TYC is responsible for research concepts, study design, and clinical research. CLV is responsible for the definition of knowledge content, literature research, and experimental studies. BBZ is responsible for data collection, and manuscript preparation. KS is responsible for manuscript editing, and statistical analysis. KDL is responsible for manuscript review, and data analysis. All authors read and approved the final manuscript.
## Disclosure statement
No potential conflict of interest was reported by the author(s).
## Data availability statement
The data that support the findings of this study are available from the corresponding author upon reasonable request.
## References
1. Chung MY, Choi HK, Hwang JT.. **AMPK activity: a primary target for diabetes prevention with therapeutic phytochemicals**. *Nutrients* (2021) **13** 4050. PMID: 34836306
2. Li X, Lu L, Hou W. **Epigenetics in the pathogenesis of diabetic nephropathy**. *Acta Biochim Biophys Sin* (2022) **54** 163-172. PMID: 35130617
3. Chen G, Wang H, Zhang W. **Dapagliflozin reduces urinary albumin excretion by downregulating the expression of cAMP, MAPK, and cGMP-PKG signaling pathways associated genes**. *Genet Test Mol Biomarkers* (2021) **25** 627-637. PMID: 34672772
4. Liu S, Yuan Y, Xue Y. **Podocyte injury in diabetic kidney disease: a focus on mitochondrial dysfunction**. *Front Cell Dev Biol* (2022) **10** 832887. PMID: 35321238
5. Zheng W, Guo J, Lu X. **cAMP-response element binding protein mediates podocyte injury in diabetic nephropathy by targeting lncRNA DLX6-AS1**. *Metabolism* (2022) **129** 155155. PMID: 35093327
6. Doshi SM, Friedman AN.. **Diagnosis and management of type 2 diabetic kidney disease**. *Clin J Am Soc Nephrol* (2017) **12** 1366-1373. PMID: 28280116
7. Tripathi YB, Yadav D.. **Diabetic nephropathy: causes and managements**. *Recent Pat Endocr Metab Immune Drug Discov* (2013) **7** 57-64. PMID: 23227864
8. Wheeler DC, Stefansson BV, Jongs N. **Effects of dapagliflozin on major adverse kidney and cardiovascular events in patients with diabetic and non-diabetic chronic kidney disease: a prespecified analysis from the DAPA-CKD trial**. *Lancet Diabetes Endocrinol* (2021) **9** 22-31. PMID: 33338413
9. Heerspink HJL, Stefansson BV, Correa-Rotter R. **Dapagliflozin in patients with chronic kidney disease**. *N Engl J Med* (2020) **383** 1436-1446. PMID: 32970396
10. Tang G, Li S, Zhang C. **Clinical efficacies, underlying mechanisms and molecular targets of chinese medicines for diabetic nephropathy treatment and management**. *Acta Pharm Sin B* (2021) **11** 2749-2767. PMID: 34589395
11. Nicholas SB.. **Novel anti-inflammatory and anti-fibrotic agents for diabetic kidney Disease-From bench to bedside**. *Adv Chronic Kidney Dis* (2021) **28** 378-390. PMID: 34922694
12. Deng J, Liu Y, Liu Y. **The multiple roles of fibroblast growth factor in diabetic nephropathy**. *J Inflamm Res* (2021) **14** 5273-5290. PMID: 34703268
13. Jha JC, Ho F, Dan C. **A causal link between oxidative stress and inflammation in cardiovascular and renal complications of diabetes**. *Clin Sci* (2018) **132** 1811-1836
14. Ostergaard JA, Cooper ME, Jandeleit-Dahm KAM.. **Targeting oxidative stress and anti-oxidant defence in diabetic kidney disease**. *J Nephrol* (2020) **33** 917-929. PMID: 32447617
15. Li H, Wang F, Zhang L. **Modulation of Nrf2 expression alters high glucose-induced oxidative stress and antioxidant gene expression in mouse mesangial cells**. *Cell Signal* (2011) **23** 1625-1632. PMID: 21658446
16. Alshehri AS.. **Kaempferol attenuates diabetic nephropathy in streptozotocin-induced diabetic rats by a hypoglycaemic effect and concomitant activation of the Nrf-2/HO-1/antioxidants axis**. *Arch Physiol Biochem* (2021) 1-14
17. Du L, Wang J, Chen Y. **Novel biphenyl diester derivative AB-38b inhibits NLRP3 inflammasome through Nrf2 activation in diabetic nephropathy**. *Cell Biol Toxicol* (2020) **36** 243-260. PMID: 31768838
18. Rathinam VA, Fitzgerald KA.. **Inflammasome complexes: emerging mechanisms and effector functions**. *Cell* (2016) **165** 792-800. PMID: 27153493
19. Wang L, Hauenstein AV.. **The NLRP3 inflammasome: mechanism of action, role in disease and therapies**. *Mol Aspects Med* (2020) **76** 100889. PMID: 32859386
20. Jiao Y, Nan J, Mu B. **Discovery of a novel and potent inhibitor with differential species-specific effects against NLRP3 and AIM2 inflammasome-dependent pyroptosis**. *Eur J Med Chem* (2022) **232** 114194. PMID: 35183871
21. Gao Y, Ma Y, Xie D. **ManNAc protects against podocyte pyroptosis via inhibiting mitochondrial damage and ROS/NLRP3 signaling pathway in diabetic kidney injury model**. *Int Immunopharmacol* (2022) **107** 108711. PMID: 35338958
22. Dong W, Jia C, Li J. **Fisetin attenuates diabetic nephropathy-induced podocyte injury by inhibiting NLRP3 inflammasome**. *Front Pharmacol* (2022) **13** 783706. PMID: 35126159
23. Zhang K, Lin L, Zhu Y. **Saikosaponin d alleviates liver fibrosis by negatively regulating the ROS/NLRP3 inflammasome through activating the ERbeta pathway**. *Front Pharmacol* (2022) **13** 894981. PMID: 35694250
24. Tong L, Zhao Q, Datan E. **Triptolide: reflections on two decades of research and prospects for the future**. *Nat Prod Rep* (2021) **38** 843-860. PMID: 33146205
25. Liu P, Zhang J, Wang Y. **The active compounds and therapeutic target of**. *Front Med* (2021) **8** 747922
26. Ren L, Wan R, Chen Z. **Triptolide alleviates podocyte epithelial-mesenchymal transition via kindlin-2 and EMT-Related TGF-beta/smad signaling pathway in diabetic kidney disease**. *Appl Biochem Biotechnol* (2022) **194** 1000-1012. PMID: 34596829
27. Han F, Xue M, Chang Y. **Erratum: triptolide suppresses glomerular mesangial cell proliferation in diabetic nephropathy is associated with inhibition of PDK1/Akt/mTOR pathway: erratum**. *Int J Biol Sci* (2020) **16** 3037-3038. PMID: 33061815
28. Liang X, Chen B, Wang P. **Triptolide potentiates the cytoskeleton-stabilizing activity of cyclosporine a in glomerular podocytes via a GSK3beta dependent mechanism**. *Am J Transl Res* (2020) **12** 800-812. PMID: 32269713
29. Chen L, Mei G, Jiang C. **Carbon monoxide alleviates senescence in diabetic nephropathy by improving autophagy**. *Cell Prolif* (2021) **54** e13052. PMID: 33963627
30. Han F, Xue M, Chang Y. **Triptolide suppresses glomerular mesangial cell proliferation in diabetic nephropathy is associated with inhibition of PDK1/Akt/mTOR pathway**. *Int J Biol Sci* (2017) **13** 1266-1275. PMID: 29104493
31. Cai T, Ke Q, Fang Y. **Sodium-glucose cotransporter 2 inhibition suppresses HIF-1alpha-mediated metabolic switch from lipid oxidation to glycolysis in kidney tubule cells of diabetic mice**. *Cell Death Dis* (2020) **11** 390. PMID: 32444604
32. Wang Y, Feng F, Sun HW. **miR-188-3p abolishes germacrone-mediated podocyte protection in a mouse model of diabetic nephropathy in type I diabetes through triggering mitochondrial injury**. *Bioengineered* (2022) **13** 774-788. PMID: 34847832
33. Zheng T, Wang HY, Chen Y. **Src activation aggravates podocyte injury in diabetic nephropathy via suppression of FUNDC1-Mediated mitophagy**. *Front Pharmacol* (2022) **13** 897046. PMID: 35614934
34. Shao X, Zhang X, Hu J. **Dopamine 1 receptor activation protects mouse diabetic podocytes injury via regulating the PKA/NOX-5/p38 MAPK axis**. *Exp Cell Res* (2020) **388** 111849. PMID: 31954110
35. Wang S, Yang Y, He X. **Cdk5-mediated phosphorylation of Sirt1 contributes to podocyte mitochondrial dysfunction in diabetic nephropathy**. *Antioxid Redox Signal* (2021) **34** 171-190. PMID: 32660255
36. Fan X, Hao Z, Li Z. **Inhibition of miR-17 ∼ 92 cluster ameliorates high glucose-induced podocyte damage**. *Mediators Inflamm* (2020) **2020** 6126490. PMID: 32774146
37. Li XY, Wang SS, Han Z. **Triptolide restores autophagy to alleviate diabetic renal fibrosis through the miR-141-3p/PTEN/Akt/mTOR pathway**. *Mol Ther Nucleic Acids* (2017) **9** 48-56. PMID: 29246323
38. Sun J, Zhu H, Wang X. **CoQ10 ameliorates mitochondrial dysfunction in diabetic nephropathy through mitophagy**. *J Endocrinol* (2019)
39. Wu Q, Li W, Zhao J. **Apigenin ameliorates doxorubicin-induced renal injury via inhibition of oxidative stress and inflammation**. *Biomed Pharmacother* (2021) **137** 111308. PMID: 33556877
40. Gu J, Shi JZ, Wang YX. **LncRNA FAF attenuates hypoxia/ischaemia-induced pyroptosis via the miR-185-5p/PAK2 axis in cardiomyocytes**. *J Cell Mol Med* (2022) **26** 2895-2907. PMID: 35373434
41. Jiang W, Xiao T, Han W. **Klotho inhibits PKCalpha/p66SHC-mediated podocyte injury in diabetic nephropathy**. *Mol Cell Endocrinol* (2019) **494** 110490. PMID: 31207271
42. Zhu W, Li YY, Zeng HX. **Carnosine alleviates podocyte injury in diabetic nephropathy by targeting caspase-1-mediated pyroptosis**. *Int Immunopharmacol* (2021) **101** 108236. PMID: 34653727
43. Li MR, Lei CT, Tang H. **MAD2B promotes podocyte injury through regulating numb-dependent notch 1 pathway in diabetic nephropathy**. *Int J Biol Sci* (2022) **18** 1896-1911. PMID: 35342338
44. Ucar BI, Ucar G, Saha S. **Pharmacological protection against ischemia-reperfusion injury by regulating the Nrf2-Keap1-ARE signaling pathway**. *Antioxidants* (2021) **10** 823. PMID: 34063933
45. Loboda A, Damulewicz M, Pyza E. **Role of Nrf2/HO-1 system in development, oxidative stress response and diseases: an evolutionarily conserved mechanism**. *Cell Mol Life Sci* (2016) **73** 3221-3247. PMID: 27100828
46. Al Mamun A, Ara Mimi A, Wu Y. **Pyroptosis in diabetic nephropathy**. *Clin Chim Acta* (2021) **523** 131-143. PMID: 34529985
47. Tan HB, Zhao Q, Chen L.. **Penehyclidine hydrochloride suppresses inflammation response and reduces podocyte injury in diabetic nephropathy by targeting fibrinogen-like protein 2**. *Int Immunopharmacol* (2022) **107** 108680. PMID: 35303505
48. Kaewin S, Changsorn K, Sungkaworn T. **Fungus-derived 3-hydroxyterphenyllin and candidusin a ameliorate palmitic Acid-Induced human podocyte injury via anti-oxidative and anti-apoptotic mechanisms**. *Molecules* (2022) **27** 2109. PMID: 35408508
49. Sun Y, Cui S, Hou Y. **The updates of podocyte lipid metabolism in proteinuric kidney disease**. *Kidney Dis* (2021) **7** 438-451
50. Opazo-Rios L, Mas S, Marin-Royo G. **Lipotoxicity and diabetic nephropathy: novel mechanistic insights and therapeutic opportunities**. *Int J Mol Sci* (2020) **21**
51. Suk Kang J, Son SS, Lee JH. **Protective effects of klotho on palmitate-induced podocyte injury in diabetic nephropathy**. *PLOS One* (2021) **16** e0250666. PMID: 33891667
52. Ma R, Liu L, Liu X. **Triptolide markedly attenuates albuminuria and podocyte injury in an animal model of diabetic nephropathy**. *Exp Ther Med* (2013) **6** 649-656. PMID: 24137241
53. Li Y, Hu Q, Li C. **PTEN-induced partial epithelial-mesenchymal transition drives diabetic kidney disease**. *J Clin Invest* (2019) **129** 1129-1151. PMID: 30741721
54. Xue M, Cheng Y, Han F. **Triptolide attenuates renal tubular epithelial-mesenchymal transition via the MiR-188-5p-mediated PI3K/AKT pathway in diabetic kidney disease**. *Int J Biol Sci* (2018) **14** 1545-1557. PMID: 30263007
55. Pang R, Gu D.. **Triptolide improves renal injury in diabetic nephropathy rats through TGF-beta1/smads signal pathway**. *Endocr Metab Immune Disord Drug Targets* (2021) **21** 1905-1911. PMID: 34530721
56. Xu ZJ, Shu S, Li ZJ. **Liuwei dihuang pill treats diabetic nephropathy in rats by inhibiting of TGF-beta/SMADS, MAPK, and NF-kB and upregulating expression of cytoglobin in renal tissues**. *Medicine* (2017) **96** e5879. PMID: 28099346
57. Han F, Wang S, Chang Y. **Triptolide prevents extracellular matrix accumulation in experimental diabetic kidney disease by targeting microRNA-137/Notch1 pathway**. *J Cell Physiol* (2018) **233** 2225-2237. PMID: 28695984
58. Zhou Y, Hong Y, Huang H.. **Triptolide attenuates inflammatory response in membranous glomerulo-nephritis rat via downregulation of NF-kappaB signaling pathway**. *Kidney Blood Press Res* (2016) **41** 901-910. PMID: 27871079
59. Sago T, Nakayama R, Okumura T. **Metabolism of prostaglandins D2 and F2 alpha in primary cultures of rat hepatocytes**. *Biochim Biophys Acta* (1986) **879** 330-338. PMID: 3465372
60. Lal MA, Young KW, Andag U.. **Targeting the podocyte to treat glomerular kidney disease**. *Drug Discov Today* (2015) **20** 1228-1234. PMID: 26096184
61. Nagata M.. **Podocyte injury and its consequences**. *Kidney Int* (2016) **89** 1221-1230. PMID: 27165817
62. Perez Gutierrez RM, Garcia Campoy AH, Paredes Carrera SP. **3′-O-beta-d-glucopyranosyl-alpha,4,2′,4′,6′-pentahydroxy-dihydrochalcone, from bark of**. *Molecules* (2019) **24** 1214. PMID: 30925713
63. Dua K, Sheshala R, Al-Waeli HA. **Antimicrobial efficacy of extemporaneously prepared herbal mouthwashes**. *Recent Pat Drug Deliv Formul* (2015) **9** 257-261. PMID: 26051152
64. Gupta G, Chellappan DK, Kikuchi IS. **Nephrotoxicity in rats exposed to paracetamol: the protective role of moralbosteroid, a steroidal glycoside**. *J Environ Pathol Toxicol Oncol* (2017) **36** 113-119. PMID: 29199592
65. Yang J, Tang X, Ke X. **Triptolide suppresses NF-kappaB-Mediated inflammatory responses and activates expression of Nrf2-Mediated antioxidant genes to alleviate caerulein-induced acute pancreatitis**. *Int J Mol Sci* (2022) **23**
66. Chen Q, Lei YQ, Liu JF. **Triptolide improves neurobehavioral functions, inflammation, and oxidative stress in rats under deep hypothermic circulatory arrest**. *Aging* (2021) **13** 3031-3044. PMID: 33465048
67. Dong XG, An ZM, Guo Y. **Effect of triptolide on expression of oxidative carbonyl protein in renal cortex of rats with diabetic nephropathy**. *J Huazhong Univ Sci Technolog Med Sci* (2017) **37** 25-29. PMID: 28224432
68. Tang B, Zhu J, Zhang B. **Therapeutic potential of triptolide as an anti-Inflammatory agent in dextran sulfate Sodium-Induced murine experimental colitis**. *Front Immunol* (2020) **11** 592084. PMID: 33240279
69. Yu H, Shi L, Zhao S. **Triptolide attenuates myocardial ischemia/reperfusion injuries in rats by inducing the activation of Nrf2/HO-1 defense pathway**. *Cardiovasc Toxicol* (2016) **16** 325-335. PMID: 26391895
70. Cheng Q, Pan J, Zhou ZL. **Caspase-11/4 and gasdermin D-mediated pyroptosis contributes to podocyte injury in mouse diabetic nephropathy**. *Acta Pharmacol Sin* (2021) **42** 954-963. PMID: 32968210
71. Guo H, Pan C, Chang B. **Triptolide improves diabetic nephropathy by regulating Th cell balance and macrophage infiltration in rat models of diabetic nephropathy**. *Exp Clin Endocrinol Diabetes* (2016) **124** 389-398. PMID: 27328403
72. Wu M, Han W, Song S. **NLRP3 deficiency ameliorates renal inflammation and fibrosis in diabetic mice**. *Mol Cell Endocrinol* (2018) **478** 115-125. PMID: 30098377
73. Du L, Wang J, Chen Y. **Correction to: novel biphenyl diester derivative AB-38b inhibits NLRP3 inflammasome through Nrf2 activation in diabetic nephropathy**. *Cell Biol Toxicol* (2022) **38** 913-914. PMID: 35169962
74. Yang J, Guo W, Lu M.. **Confusion about the article: natural product triptolide induces GSDME-mediated pyroptosis in head and neck cancer through suppressing mitochondrial hexokinase-IotaIota**. *Cancer Lett* (2022) **534** 115568. PMID: 35131385
75. He L, Peng X, Liu G. **Anti-inflammatory effects of triptolide on IgA nephropathy in rats**. *Immunopharmacol Immunotoxicol* (2015) **37** 421-427. PMID: 26466641
76. Shen J, Ma H, Wang C.. **Triptolide improves myocardial fibrosis in rats through inhibition of nuclear factor kappa B and NLR family pyrin domain containing 3 inflammasome pathway**. *Korean J Physiol Pharmacol* (2021) **25** 533-543. PMID: 34697264
77. Chen Y, Li T, Tan P. **Kaempferol from penthorum Chinense pursh attenuates hepatic ischemia/reperfusion injury by suppressing oxidative stress and inflammation through activation of the Nrf2/HO-1 signaling pathway**. *Front Pharmacol* (2022) **13** 857015. PMID: 35431932
78. Chen Y, Zhu L, Meng H. **Ferulic acid protects human lens epithelial cells against ionizing radiation-induced oxidative damage by activating Nrf2/HO-1 signal pathway**. *Oxid Med Cell Longev* (2022) **2022** 6932188. PMID: 35592532
79. Sarafidis P, Pella E, Kanbay M. **SGLT-2 inhibitors and nephroprotection in patients with diabetic and non-diabetic chronic kidney disease**. *Curr Med Chem* (2022)
80. Guo R, Wang P, Zheng X. **SGLT2 inhibitors suppress epithelial-mesenchymal transition in podocytes under diabetic conditions via downregulating the IGF1R/PI3K pathway**. *Front Pharmacol* (2022) **13** 897167. PMID: 36225569
|
---
title: 'Exploring the relationship between intestinal flora and the pathological mechanism
of myopia in adolescents from the perspective of Chinese and Western medicine: A
review'
authors:
- Xiaoming Xi
- Liang Han
- Mengmeng Ding
- Jinglu Li
- Chenye Qiao
- Zongjian Liu
- Shuyan Qie
journal: Medicine
year: 2023
pmcid: PMC10035986
doi: 10.1097/MD.0000000000033393
license: CC BY 4.0
---
# Exploring the relationship between intestinal flora and the pathological mechanism of myopia in adolescents from the perspective of Chinese and Western medicine: A review
## Abstract
The etiology of adolescent myopia involves genetic and environmental factors. The pathological mechanism of modern medicine includes blood perfusion, changes in blood molecules, neurotransmitters, and sclera remodeling. Chinese medicine believes that myopia is mainly related to the deficiency of liver blood and spleen and stomach disorders. The prevention and treatment of myopia in adolescents are very important, but in terms of the current incidence of myopia in adolescents and the level of clinical diagnosis and treatment, its prevention and treatment are insufficient. Modern medicine and traditional Chinese medicine both pay attention to integrity, so adolescent myopia should not only pay attention to eye changes but also pay attention to other body systems and other aspects of change. Intestinal flora has become a research hotspot in recent years, and it has been found that it is closely associated with multi-system and multi-type diseases. No studies have directly investigated the link between *Intestinal flora* and myopia in adolescents. Therefore, by summarizing the pathological mechanism of adolescent myopia and the connection between intestinal flora and the pathological mechanism of adolescent myopia, this paper analyzes the possible pathological mechanism of the influence of intestinal flora on adolescent myopia, providing a theoretical basis for future studies on the correlation between changes of intestinal flora and its metabolites and the incidence of adolescent myopia, which is of great significance for the study on the risk prediction of adolescent myopia.
## 1. Introduction
Myopia, a refractive abnormality, is a global public health problem that produces blurred vision due to an increase in the axial length of the eye,[1] resulting in an elongated focal ratio that focuses light from distant objects in front of the retina.[2] The prevalence of myopia is increasing globally, with a $20\%$ prevalence in children under 6 years of age, with refractive abnormalities being the most common, followed by strabismus and amblyopia.[1] Severe pathological myopia is closely associated with other ocular diseases and is usually determined by a combination of genetic and environmental factors.[3] *Juvenile myopia* is the most common type of myopia, with the highest prevalence in grades 2, 5, and 6 of elementary school, and progresses rapidly during adolescence.[2] The most typical risk factor for adolescent myopia is close work (e.g., reading, studying, television viewing, computer use, etc), but longitudinal studies have not confirmed that close work is the main cause of adolescent myopia.[4] The treatment of myopia in adolescents is mainly optical correction and pharmacological intervention,[5] but the results are not satisfactory, so prevention of myopia occurrence is the key. At present, there is no effective method to substantially prevent the occurrence of myopia, except for maintaining the correct study posture, limiting the time of electronic product use, eye exercises, and other healthy eye care science propaganda, which is related to the insufficient means of monitoring myopia risk factors. The intestinal flora is a hot spot for research and has been shown to regulate the systemic organs and tissues in both directions through the brain-intestinal axis, lung-intestinal axis, and kidney-intestinal axis, involving a wide range of diseases. The intestine is an important organ for nutrient absorption, and all tissues in the body need nutrient support to maintain normal physiological activities, so it is easy to understand that intestinal flora is closely related to multi-system and multi-species diseases. However, there are no studies that directly investigate the association between intestinal flora and myopia in adolescents.
## 2. Pathogenesis of myopia in adolescents
Although myopia is a pathological change of the eye, the pathological mechanism is not limited to the eye and surrounding tissues but may be associated with multiple systemic diseases in the body. This article summarizes the pathogenesis of myopia from both modern medicine and Chinese medicine.
## 2.1. Modern medical pathogenesis
The modern medical pathogenesis of myopia is complex, mainly including scleral remodeling, scleral hypoxia, related factor changes, dopamine mechanism, choroidal blood perfusion deficiency, and inflammatory response,[6] etc, and each mechanism is interconnected and interacts with each other to influence the occurrence and development of myopia.
## 2.1.1. Scleral remodeling.
Inappropriate extension of the eye axis is associated with scleral extracellular matrix remodeling that can cause a decrease in scleral strength and thickness, and is associated with slower synthesis and accelerated degradation of scleral extracellular matrix components,[7] and a decrease in scleral extracellular matrix components weakens the scleral framework, resulting in a lengthening of the eye axis. For example, decreased expression of type I collagen, a major extra-scleral membrane component, weakens the structural framework of the sclera. Although a large number of experiments have supported this conclusion, the initiators and signaling pathway mechanisms that trigger these changes are not uniformly established but are thought to be mainly related to the hypoxia signaling pathway, eIF2 signaling pathway, and mTOR signaling pathway. Among them, the eIF2 signaling pathway and mTOR signaling pathway are in turn related to scleral hypoxia,[8] and the 3 interact with each other. In addition, the choroid can translate and transmit myopic visual signals to the sclera, which may affect scleral extracellular matrix remodeling and ocular growth.[7] However, the mechanisms regulating ocular development between the choroid and sclera are unclear.
## 2.1.2. Scleral hypoxia.
Since hypoxia is a risk factor for pathological myopia, 1 study[8] used the Fluidigm C1 System, and determined the transcriptomes of 93 single cells isolated from form-deprivation (FD) scleras; untreated fellow eyes served as control. Forty-nine of the cells were from 4 scleral samples (each sample contained 6 to 8 scleral tissues) of FD eyes; 44 cells originated from 3 samples (from similarly pooled scleras) of control eyes. Determined the dynamics of hypoxia-inducible factor 1α (HIF-1α) protein levels in mice, an indicator of tissue hypoxia, after 2 days and 2 weeks of FD, and compared to controls, form-deprivation myopia (FDM) had significantly higher scleral HIF-1α levels compared to controls, and increased axial elongation was observed after 2 weeks. This suggests that local scleral hypoxia occurs during the development of myopia. To further investigate the role of scleral hypoxia in myopia, this study used immunoblotting to discover that myofibroblasts in human scleral fibroblasts (HSFs) underwent trans differentiation and collagen production after reducing ambient oxygen levels to $5\%$; in HSFs exposed to hypoxic conditions, HIF-1α protein, adherent spot protein, and emyofibre increased expression of the fibroblast marker α-smooth muscle actin and decreased expression of type I collagen α1 protein in HSFs exposed to hypoxic conditions.[8] *The previous* section discussed that collagen production weakens the scleral structural framework, so scleral hypoxia may lead to myopia through scleral myofibroblast trans differentiation and cause scleral remodeling.
## 2.1.3. Choroidal blood perfusion.
The metabolic and nutritional processes of the body’s tissues are dependent on blood perfusion,[9] and if blood perfusion is insufficient, it can lead to tissue ischemia, hypoxia, and even necrosis, affecting the development of the body. Studies have shown that visual signals lead to a decrease in choroidal capillary patency and blood flow, which leads to a decrease in the level of oxygen and nutrient supply to the adjacent sclera with no vascular distribution and consequent scleral hypoxia, which promotes myofibroblast trans differentiation through the accumulation of HIF-1αwhile reducing collagen production.[10] Since the sclera has no vascular distribution, insufficient choroidal blood perfusion can indirectly lead to scleral hypoxia, which triggers a series of pathological changes leading to the development of myopia. Based on the above findings, it can be hypothesized that in addition to visual signals that can lead to inadequate choroidal perfusion, the body’s blood deficiency (e.g., anemia) may also lead to reduced choroidal blood flow and thus myopia. To verify this hypothesis, we can collect children or adolescents who meet the diagnostic criteria of anemia as the observation group and normal children or adolescents as the control group, and observe the incidence of myopia in the 2 groups or follow up after a few years to see if there is a difference in the incidence, which can provide a background for later studies.
## 2.1.4. Dopamine mechanisms.
Dopamine (DA) is an important neurotransmitter in the retina that mediates a variety of functions including development, visual signaling, and refractive development.[11] Since the first discovery of a link between DA and ocular growth control in 1989, more and more studies have adopted the DA hypothesis (i.e., that the retina releases DA to counteract myopic development) as the main hypothesis for myopia control. Light control methods such as outdoor activities may be used to suppress myopia through DA-mediated mechanisms,[12] such as controlling retinal DA signaling and promoting refractive development.[13] Enhanced neuronal activity maintains homeostatic storage of DA,[11] so neuronal activity with DA as a hub is associated with normal retinal development, and reduced neuronal activity may inhibit retinal refractive development, etc, and myopia occurs. Neurons and DA are the basic units and important molecules of the nervous system, respectively, so the relationship between myopia and the nervous system can be studied in depth. Studies have shown that increasing DA content by injecting DA directly into the eye or using levodopa to promote DA synthesis, or enhancing DA signaling by nonselective DA receptor agonists such as APO and 2-amino-6, 7-dihydroxy-1, 2, 3, 4-tetrahydronaphthalene hydrobromide can prevent the development of myopia.[11] Therefore, reduced DA content is an important cause of myopia, which may specifically include reduced DA synthesis, reduced release, and increased destruction, or DA receptors affected resulting in reduced DA content. DA may inhibit myopia by triggering the release of other transmitters from the retina or choroid and inducing choroidal thickening.[14]DA receptors are present in almost all neuron types in the retina, including D1, D2, D4, and D5 receptors, and are also associated with other intracellular signaling pathways. D1 and D5 receptors (D1-like receptors) promote intracellular cyclic adenosine monophosphate (cAMP) synthesis, while D2 and D4 receptors (D2-like receptors) inhibit cAMP synthesis.[15] cAMP promotes DA synthesis, so D1 and D5 receptors can increase DA levels to inhibit myopia, and their deficiency may be a risk factor for myopia; D2 and D4 receptors have opposite roles to D2 and D5 in maintaining DA levels and increased D2 and D4 can reduce DA levels to induce myopia, which may be a risk factor for myopia. It was demonstrated that D2-like receptor antagonists inhibited the transient protective effect on unrestricted vision in form-deprived myopia, but not in negative crystal-induced myopia.[16] This result suggests that DA may prevent the development of myopia by inhibiting form-deprivation and lens defocus.
## 2.1.5. Associated factors.
Related factors act as intermediate bridges involved in the development and progression of myopia. They include trans differentiation of HIF-1α and HSFs, increased expression of matrix metalloproteinase-2 and α-smooth muscle actin, and decreased expression of type I collagen α1 protein, which jointly affect myopia through a bidirectional effect with scleral hypoxia and scleral remodeling. Among them, HIF-1α has been studied more frequently and is also the main relevant factor affecting myopia. In 1 study, a total of 30 patients (13 males, 17 females, age: 24–50 years) without pathologic myopia and underwent physical examination at The University of Hong Kong-Shenzhen Hospital was selected for collection of fasting peripheral blood as a normal control. Besides, the fasting peripheral blood of 30 patients (15 males, 15 females, age: 22–50 years) diagnosed with pathologic myopia in the ophthalmology department of The University of Hong Kong-Shenzhen Hospital was collected as the experimental group. Used microarray-based pathological myopia gene expression profiles to identify differentially expressed genes and cultured HSFs under hypoxic conditions, and showed that HIF-1α reduced miR-150-5p expression and promoted LAMA4-mediated activation of the p38 MAPK signaling pathway, which prevented extracellular matrix degradation in HSFs, ultimately leading to myopia.[17] In another experiment, guinea pigs were randomly divided into 2 groups (30 guinea pigs in each group): FDM group and control group (without any treatment). Used a monocular form-deprivation method to induce high myopia in guinea pigs, and a 532 nm laser was used to induce choroidal neovascularization (CNV) generation; the results showed that CNV proliferation in the FDM group was proportional to time, the incidence of proliferation was higher than that in the control group, and the hypoxia-inducible factor HIF- 1α and vascular endothelial growth factor (VEGF) expression were increased in both groups, and in addition, miRNA-21 expression was positively correlated with VEGF and HIF-1α expression in both groups.[18] Therefore, it can be concluded that CNV causes severe damage to central vision by a mechanism in which miRNA-21 is associated with the HIF-1α-VEGF signaling pathway, possibly by promoting the formation of CNV in guinea pigs and thus leading to myopia development.
## 2.1.6. Inflammatory response.
Dysregulation of inflammatory pathways may contribute to the pathogenesis of myopic retinal degeneration, specifically including upregulation of the AGE-RAGE signaling pathway, complement cascade, NOD-like receptor signaling pathway, IL-17 signaling pathway, and TNF signaling pathway, while downregulation of antigen processing and cell adhesion pathways.[19] In addition, CCL2,[20] IL-6, matrix metalloproteinase-2, and angiopoietin-1 levels are higher in highly myopic eyes.[21,22] It can be seen that the inflammatory response may be involved in the development and progression of myopia through altered signaling pathways and the production of inflammatory factors, but it is unclear whether the inflammatory response is a consequence or a cause of myopic retinal degeneration.
## 2.1.7. Other.
Myopia is also associated with ocular degenerative changes, including posterior chylomicrons, lacunar fissures, optic disc abnormalities, and choroidal atrophy,[23] which may result from ocular neuronal dysfunction or apoptosis. In addition, women have risk factors for high myopia[24] and there may be gender differences in the development of myopia,[25] so it is speculated that there may be some degree of relationship between estrogen and myopia.
## 2.2. Pathogenesis in Chinese medicine
Myopia is referred to in Chinese medicine as “being able to be near and timid to be far,” and its symptoms were first recorded in the Treatise on the Origin of Diseases, and also described in the “Zhengzhi Zhunsheng - The 7 Focal Points” and “Dacheng - Volume 2 of the Eye Classic.” However, the relevant TCM pathogenesis has been recorded as early as in the Yellow Emperor’s Classic of Internal Medicine. According to TCM, myopia is mainly related to liver and blood deficiency, congenital endowment deficiency, acquired spleen, and stomach dysregulation, and Yang deficiency.[26] According to Su Wen - Generation of the 5 Organs, “All the veins belong to the eyes...... blood goes to the liver, and the liver receives blood and can see.” *It is* pointed out that the function of the eyes depends on liver blood to moisten them, and the liver opens the orifice of the eyes, and the eyes are closely related to the liver. Therefore, a deficiency of liver blood can cause the eyes to lose blood to moisten them, resulting in symptoms such as dry eyes and blurred vision. The Ling Shu points out that “the vein of the liver and the foot of the Fructus yin...... goes up into the forehead and connects to the eye system......,” which believes that the liver has a close connection with the eyes through the liver meridian. The Yellow Emperor’s Classic of Internal Medicine states that “the liver and kidney are of the same origin,” or “the essence and blood are of the same origin,” and that the kidney is the master of bone, bone produces marrow, and marrow produces blood. The spleen and the stomach are the origins of the latter, the source of biochemical energy and blood, and the blood of the whole body is based on the biochemical blood function of the spleen and stomach. If the spleen and stomach are dysfunctional, blood biochemistry will be blocked, which will not only lead to dysfunction of the internal organs and meridians but also dysfunction due to the loss of blood moistening. The Sibai point in the stomach meridian is located 1 inch directly below the pupil of the eye and has the function of treating blindness in the eyes. The Ling Shu states: “The vein of the small intestine of the hand sun...... follows the neck up the cheek from the absence of the basin to the eye sharp canthus...... not the cheek on the against the nose, to the eye inner canthus......,” it can be seen that the small intestine has a very close connection with the eyes through the small intestine meridian. The Yellow Emperor’s Needle Classic suggests that both the large intestine of the hand Yangming and the small intestine of the hand Sun belong to the stomach of the foot Yangming, and from the analysis of the meridian connection and the physiological functions of the large and small intestines and the stomach, it is also suggested that the large and small intestines are attached to the stomach.[27] Therefore, dysfunction of the spleen and stomach is one of the important pathogenic mechanisms of myopia. Studies have shown that outdoor activities can inhibit the occurrence of myopia, and people with Yang deficiency tend to be mentally uninspired, quiet, and introverted, and mostly dislike outdoor sports,[28,29] so the incidence of myopia may be high. In addition, the function of blood to moisten the organs and tissues is diminished in Yang deficiency patients, so Yang deficiency patients have the conditions for myopia to occur.
## 3.1. Modern medical connections
Through literature reading, it is found that intestinal flora is related to tissue ischemia and hypoxia, dopamine alteration, and inflammatory response, and these mechanism alterations are exactly the pathological mechanism of myopia, but there are no studies directly explore the connection between intestinal flora and myopia pathological mechanism, so this paper will analyze from this aspect.
## 3.1..1. Intestinal flora and tissue ischemia and hypoxia.
Recent studies have shown that oxygen dynamics plays an important role in regulating homeostasis in the intestine and has a bidirectional regulatory mechanism with the intestinal flora.[30] The intestine is highly dependent on adaptive pathways activated by hypoxia, and HIF-1α is an important hypoxic factor, so HIF-1α is inextricably linked to the intestinal flora. Then, intestinal flora may weaken the scleral structure by increasing HIF-1α levels, leading to myopia development. Studies have shown that disorders of intestinal flora can contribute to the development of ischemic stroke or aggravate cerebral ischemia by enhancing systemic inflammation.[31] A non-randomly control study by Fu Ke et al[32] intervened in the intestinal flora of patients with cerebral infarction by Astragalus Stomach Zhenzhu Pill and found that it reduced cerebral ischemia-reperfusion injury in patients with cerebral infarction. This further suggests that intestinal flora disorders can contribute to the development of cerebral ischemia. The choroidal artery (chA) is divided into the anterior choroidal artery (AchA), the medial posterior choroidal artery, and the lateral posterior choroidal artery. AchA originates from the internal carotid artery (The AchA originates from the internal carotid artery, the medial posterior choroidal artery from the P1 or P2a segment of the posterior cerebral artery, and the lateral posterior choroidal artery from the P1 or P2a or P2p segment of the posterior cerebral artery. It can be seen that the chA is closely connected to the blood supply arteries of the brain, and poor circulation in the brain can lead to insufficient blood in the chA. Insufficient blood in the chA can have a direct effect on vision. The intestinal flora is closely related to the blood supply of the brain, so there is an indirect connection between the intestinal flora and the blood supply of the chA, so it is theoretically very likely that there is a close connection between the intestinal flora and the visual acuity. This inference can guide the basic experimental research on the relationship between intestinal flora and ocular hemodynamic changes and visual acuity, which, if confirmed experimentally, will be of great significance for both theoretical research on the pathogenesis of myopia and clinical treatment research.
## 3.1.2. Intestinal flora and dopamine alterations.
The relationship between gut flora and dopamine has become a hot topic of research in recent years. It has not been fully demonstrated that intestinal flora regulates dopamine in vivo, but there is growing evidence that it may play a role in host biosynthesis or catabolism.[33] The existence of a phenylalanine-tyrosine-dopa-dopamine metabolic pathway in microorganisms[34] suggests that bacteria may contain homologs of the enzyme genes used by mammals to produce dopamine,[35,36] such as enterococci for example, which produce biogenic dopamine.[37] Tyrosine hydroxylase is a key enzyme in the dopamine synthesis pathway.[38] It was shown that oral administration of flavopiridol triggered the biosynthesis of BH4 in the intestinal flora, increased the concentration of dopa/dopamine in the blood and brain, enhanced tyrosine hydroxylase activity to produce L-dopa, and ultimately improved animal fitness; in contrast, oral antibiotics reduced the number of intestinal flora and thus the ability of the intestine to produce dopamine.[34] This further suggests that intestinal flora can promote dopamine synthesis and release into the bloodstream, where dopamine runs to the brain and other parts of the body to function. Dopamine is an important neurotransmitter in the retina, which is closely related to vision, and its deficiency can lead to abnormal visual development and vision loss.[11] Therefore, intestinal flora dysbiosis could theoretically lead to visual development and vision loss by inhibiting dopamine synthesis and decreasing dopamine concentration in the retina. There are no studies to investigate the relationship between intestinal flora and visual acuity, and this paper uses dopamine as a bridge to establish the link between intestinal flora and visual acuity, which can improve a new direction for basic research on the pathogenesis of visual loss. It is recommended that related basic research observe which intestinal flora and how they change can lead to myopia through bioassay technology. This will not only be an important tool to detect the risk of myopia in adolescents but also may provide a new method to treat myopia in adolescents, which is important for the detection, prevention, and treatment of myopia in adolescents.
## 3.1.3. Intestinal flora and inflammatory response.
Studies have shown that fermented foods increase intestinal flora diversity and decrease inflammatory markers.[39] *Gut flora* translocation promotes the inflammatory response of the immune system and can lead to inflammation in some tissues, resulting in loss of function.[40] This suggests that intestinal flora is closely related to the inflammatory response. Among them, CCL2, IL-6, IL-17, AGE-RAGE signaling pathway, and TNF signaling pathway are closely related to the intestinal flora,[41] and these inflammatory factors and signaling pathways are precisely the pathological factors and signaling pathways of myopia, so changes in the intestinal flora may produce inflammatory responses through these inflammatory factors and signaling pathways and affect the development and progression of myopia. In addition, based on the intestine-hepatic axis theory, after the intestinal barrier is damaged, bacterial translocation endotoxin enters the portal system and activates liver blast cells, etc, and then releases a series of inflammatory factors that further aggravate liver damage and disease progression.[42] The link between liver and vision loss was briefly summarized by Yu Zeng Fang et al[43] For example, hepatic-brain-eye syndrome, a form of cortical blindness secondary to severe liver disease, often manifests as transient visual impairment; in functional magnetic resonance imaging of patients with mild hepatic encephalopathy, a reduction in blood oxygen level-dependent signals associated with the right parietal cortex of the brain and visual judgment was found, with the potential for cortical blindness; in lenticular In the retinal pigment epithelium of lens-induced myopia guinea pigs, stable expression and elevated activity of hepatocyte growth factor, a major cytokine derived from hepatocytes and stimulating hepatocyte proliferation, were found. This suggests that the liver and eye are closely related, and liver disease can lead to abnormal visual function, while hepatitis is the main cause of many liver diseases. Therefore, intestinal flora may affect vision through the intestine-liver axis and liver-eye pathological signaling pathways. In TCM, the liver opens the eyes and the liver-eye connection is very close, which provides the theoretical basis and academic background for the above inference.
## 3.2. Chinese medicine connection
The function of intestinal flora is similar to that of the spleen and stomach in Chinese medicine, such as the main digestion, absorption, and nutrient delivery, and has a close connection with the functions of the organs, tissues, organs, and meridians, which can be understood as the qi of the spleen and stomach and the yin and yang of the spleen and stomach.[27] Therefore, the connection between intestinal flora and myopia can be converted into a connection between the spleen and stomach and myopia. The liver’s main blood collection function depends on the spleen and stomach’s biochemical function of qi and blood, and if the spleen and stomach function is impaired then the biochemical deficiency of qi and blood can directly lead to liver blood deficiency. The spleen and stomach are the basis of the postnatal essence, which can nourish the innate kidney essence. A deficiency of Yang in the spleen and stomach will not only lead to slow movement of Qi and blood, weakening the role of warming and nourishing local tissues but will also affect the normal physiological functions of other internal organs throughout the body. Therefore, abnormal function of the spleen and stomach can lead to pathological mechanisms closely related to myopia, such as liver and blood deficiency, innate endowment without nourishment, and yang deficiency, which can lead to myopia.
## 4. Summary and outlook
There is no study to investigate the relationship between intestinal flora and myopia. In this paper, by summarizing the pathogenesis of myopia, we found that there is an overlap with the pathological condition caused by intestinal flora, mainly in 3 aspects: tissue ischemia and hypoxia, dopamine alteration, and inflammatory response, so we think that the change of intestinal flora may lead to the occurrence of myopia. The relationship between the 2 was also analyzed from the perspective of Chinese medicine, providing an academic background and theoretical basis for this inference. It can provide a reference direction for research related to myopia in adolescents. If future research detects the relationship between intestinal flora changes and visual acuity, it will be of great significance for the risk detection, prevention, and treatment of myopia in adolescents.
## Author contributions
Resources: Xiaoming Xi, Mengmeng Ding, Jinglu Li, Chenye Qiao, Zongjian Liu, Shuyan Qie.
Supervision: Xiaoming Xi.
Writing – original draft: Xiaoming Xi.
Writing – review & editing: Xiaoming Xi, Liang Han.
## References
1. Bremond-Gignac D. **[Myopia in children].**. *Med Sci* (2020) **36** 763-8
2. Zadnik K, Sinnott LT, Cotter SA. **Prediction of Juvenile-onset myopia.**. *JAMA Ophthalmol* (2015) **133** 683-9. PMID: 25837970
3. Baird PN, Saw SM, Lanca C. **Myopia.**. *Nat Rev Dis Primers* (2020) **6** 99. PMID: 33328468
4. Kiefer AK, Tung JY, Do CB. **Genome-wide analysis points to roles for extracellular matrix remodeling, the visual cycle, and neuronal development in myopia.**. *PLoS Genet* (2013) **9** e1003299. PMID: 23468642
5. Gwiazda J, Hyman L, Hussein M. **A randomized clinical trial of progressive addition lenses versus single vision lenses on the progression of myopia in children.**. *Invest Ophthalmolo Vis Sci* (2003) **44** 1492-500
6. Mu JY, Wang Y, Yang YN. **Current status of research on the epidemiology, etiology and pathogenesis of myopia.**. *New Adv Ophthalmol* (2021) **41** 1089-96
7. Wallman J, Winawer J. **Homeostasis of eye growth and the question of myopia.**. *Neuron* (2004) **43** 447-68. PMID: 15312645
8. Wu H, Chen W, Zhao F. **Scleral hypoxia is a target for myopia control.**. *Proc Natl Acad Sci USA* (2018) **115** E7091-100. PMID: 29987045
9. Claassen J, Thijssen DHJ, Panerai RB. **Regulation of cerebral blood flow in humans: physiology and clinical implications of autoregulation.**. *Physiol Rev* (2021) **101** 1487-559. PMID: 33769101
10. Khanna K, Mishra KP, Ganju L. **Golden root: a wholesome treat of immunity.**. *Biomed Pharmacother* (2017) **87** 496-502. PMID: 28073099
11. Zhou X, Pardue MT, Iuvone PM. **Dopamine signaling and myopia development: what are the key challenges.**. *Prog Retin Eye Res* (2017) **61** 60-71. PMID: 28602573
12. Chen S, Zhi Z, Ruan Q. **Bright light suppresses form-deprivation myopia development with activation of dopamine D1 receptor signaling in the ON pathway in retina.**. *Invest Ophthalmol Vis Sci* (2017) **58** 2306-16. PMID: 28431434
13. Feldkaemper M, Schaeffel F. **An updated view on the role of dopamine in myopia.**. *Exp Eye Res* (2013) **114** 106-19. PMID: 23434455
14. Nickla DL, Damyanova P, Lytle G. **Inhibiting the neuronal isoform of nitric oxide synthase has similar effects on the compensatory choroidal and axial responses to myopic defocus in chicks as does the non-specific inhibitor L-NAME.**. *Exp Eye Res* (2009) **88** 1092-9. PMID: 19450449
15. Beaulieu J-M, Gainetdinov RR. **The physiology, signaling, and pharmacology of dopamine receptors.**. *Pharmacol Rev* (2011) **63** 182-217. PMID: 21303898
16. Dong F, Zhi Z, Pan M. **Inhibition of experimental myopia by a dopamine agonist: different effectiveness between form deprivation and hyperopic defocus in guinea pigs.**. *Mol Vis* (2011) **17** 2824-34. PMID: 22128230
17. Ren Y, Yang X, Luo Z. **HIF-1α aggravates pathologic myopia through the miR-150-5p/LAMA4/p38 MAPK signaling axis.**. *Mol Cell Biochem* (2022) **477** 1065-74. PMID: 35034256
18. Liu L, Zhu D, Ding W. **MiRNA-21- HIF-1α-VEGF axis is associated with myopic choroidal neovascularization in guinea pigs.**. *Ophthalmic Res* (2022) **65** 493-505. PMID: 35158368
19. Zeng L, Li X, Liu J. **RNA-Seq analysis reveals an essential role of the tyrosine metabolic pathway and inflammation in myopia-induced retinal degeneration in guinea pigs.**. *Int J Mol Sci* (2021) **22** 12598. PMID: 34830490
20. Zhu X, Zhang K, He W. **Proinflammatory status in the aqueous humor of high myopic cataract eyes.**. *Exp Eye Res* (2016) **142** 13-8. PMID: 25805322
21. Zhang JS, Da Wang J, Zhu GY. **The expression of cytokines in aqueous humor of high myopic patients with cataracts.**. *Mol Vis* (2020) **26** 150-7. PMID: 32180680
22. Yuan J, Wu S, Wang Y. **Inflammatory cytokines in highly myopic eyes.**. *Sci Rep* (2019) **9** 3517. PMID: 30837544
23. Ohno-Matsui K, Wu PC, Yamashiro K. **IMI pathologic myopia.**. *Invest Ophthalmol Vis Sci* (2021) **62** 5
24. Matamoros E, Ingrand P, Pelen F. **Prevalence of myopia in France: a cross-sectional analysis.**. *Medicine (Baltimore)* (2015) **94** e1976. PMID: 26559276
25. Shinojima A, Kurihara T, Mori K. **Association between ocular axial length and anthropometrics of Asian adults.**. *BMC Res Notes* (2021) **14** 328. PMID: 34446087
26. Zhang YM, Shen JR. **Progress of research on the prevention and treatment of myopia in adolescents with Chinese medicine.**. *Natl Med Forum* (2022) **37** 70-3
27. Xi XM, Bi HY, Cao HH. **A review of Chinese medicine rehabilitation model for Parkinson’s disease based on brain-gut axis theory.**. *Shandong J Tradit Chin Med* (2022) **41** 120-5
28. Wang YIY, Song JK, Yang ZN. **Research progress on myopia and TCM constitution.**. *Western Tradit Chin Med* (2022) **35** 153-7
29. Huang J, Liao J, Fang Y. **Six-week exercise training with dietary restriction improves central hemodynamics associated with altered gut microbiota in adolescents with obesity.**. *Front Endocrinol (Lausanne)* (2020) **11** 569085. PMID: 33365012
30. Singhal R, Shah YM. **Oxygen battle in the gut: hypoxia and hypoxia-inducible factors in metabolic and inflammatory responses in the intestine.**. *J Biol Chem* (2020) **295** 10493-505. PMID: 32503843
31. Xu K, Gao X, Xia G. **Rapid gut dysbiosis induced by stroke exacerbates brain infarction in turn.**. *Gut* (2021) **70** 1486-94
32. Fu K, Zhang D, Song Y. **Tibetan medicine Qishiwei Zhenzhu pills can reduce cerebral ischemia-reperfusion injury by regulating gut microbiota and inhibiting inflammation.**. *Evid Based Complement Alternat Med* (2021) **2021** 2251679. PMID: 34804175
33. Strandwitz P. **Neurotransmitter modulation by the gut microbiota.**. *Brain Res* (2018) **1693** 128-33. PMID: 29903615
34. Wang Y, Tong Q, Ma SR. **Oral berberine improves brain dopa/dopamine levels to ameliorate Parkinson’s disease by regulating gut microbiota.**. *Signal Transduct Target Ther* (2021) **6** 77. PMID: 33623004
35. Rekdal VM, Bess EN, Bisanz JE. **Discovery and inhibition of an interspecies gut bacterial pathway for Levodopa metabolism.**. *Science* (2019) **364** eaau6323. PMID: 31196984
36. O’neill C. **Gut microbes metabolize Parkinson’s disease drug.**. *Science* (2019) **364** 1030-1. PMID: 31196998
37. Divyashri G, Krishna G, Prapulla SG. **Probiotic attributes, antioxidant, anti-inflammatory and neuromodulatory effects of Enterococcus faecium CFR 3003: in vitro and in vivo evidence.**. *J Med Microbiol* (2015) **64** 1527-40. PMID: 26450608
38. Zhang T, Wang T, Chen X. **Gut microbiota relieves inflammation in the substantia nigra of chronic Parkinson’s disease by protecting the function of dopamine neurons.**. *Exp Ther Med* (2022) **23** 52. PMID: 34934429
39. Wastyk HC, Fragiadakis GK, Perelman D. **Gut-microbiota-targeted diets modulate human immune status.**. *Cell* (2021) **184** 4137-4153.e14. PMID: 34256014
40. Scheithauer TPM, Rampanelli E, Nieuwdorp M. **Gut microbiota as a trigger for metabolic inflammation in obesity and type 2 diabetes.**. *Front Immunol* (2020) **11** 571731. PMID: 33178196
41. Liu Y, Liu X, Zhou W. **Integrated bioinformatics analysis reveals potential mechanisms associated with intestinal flora intervention in nonalcoholic fatty liver disease.**. *Medicine (Baltim)* (2022) **101** e30184
42. Marshall JC. **The gut as a potential trigger of exercise-induced inflammatory responses.**. *Can J Physiol Pharmacol* (1998) **76** 479-84. PMID: 9839072
43. Yu ZF, Zhong AL, Zhao XS. **Theoretical discussion on the treatment of myopia in children from liver and kidney theory.**. *World Sci Technol Modernization Chin Med* (2022) **24** 795-801
|
---
title: 'Effect of sterile ice water versus menthol spray on thirst symptoms of fasted
children in the intensive care unit: A prospective cohort study'
authors:
- Fangyan Ma
- Haiting He
- Banghong Xu
- Jing Zhou
- Kai Pu
journal: Medicine
year: 2023
pmcid: PMC10036011
doi: 10.1097/MD.0000000000033315
license: CC BY 4.0
---
# Effect of sterile ice water versus menthol spray on thirst symptoms of fasted children in the intensive care unit: A prospective cohort study
## Background:
Thirst is a very common symptom in fasted children in intensive care unit (ICU). This study aimed to evaluate the effect of sterile ice water versus menthol spray in ICU fasted children, to provide insights to the clinical care of fasted children.
### Methods:
The children admitted to the ICU of our hospital from June 1, 2021 to August 31, 2022 and needed to fast were included. Children were randomly assigned to the ice water group or menthol group. We evaluated and compared the thirst distress scale (TDS), oral mucosa wetness scale (OMWS), children medical fear scale (CMFS), numerical rating scale (NRS), unstimulated whole saliva (UWS) flow rate between 2 groups.
### Results:
A total of 139 children were included, involving 69 children in ice water group and 70 children in menthol group. There were no significant differences in the baseline characteristics, TDS, OMWS, OMWS, CMFS, and NRS score, UWS flow rate before intervention between ice water group and menthol group (all $P \leq .05$). After intervention, the TDS, OMWS, NRS score of menthol group was statistically less than that of ice water group (all $P \leq .05$), the UWS flow rate of menthol group was statistically higher than that of ice water group ($$P \leq .034$$).
### Conclusions:
Compared with ice water spray, menthol spray may be more beneficial to relieve the thirst and increase the comfort in ICU fasted children. Future studies with larger sample size and rigorous design are needed to evaluate the effects and safety of ice water and menthol spray in the nursing care of children.
## 1. Introduction
The intensive care unit (ICU), as the main place for hospitals to treat critically ill patients, not only provides emergency care conditions for patients, but also brings great pressure to patients. The noise of instruments, the large number of medical staff, and the continuous artificial lighting are all sources of stress for ICU patients, which often cause various discomfort during hospitalization and even after leaving the ICU.[1–3] Some studies[4,5] have pointed out that thirst is the strongest and most common type of discomfort in ICU, and it is one of the biggest sources of stress for ICU patients. This is mainly due to the fact that critically ill patients often lose too much body fluid due to surgical intervention and renal insufficiency, and the effective circulating blood volume is insufficient, so thirst is very easy to occur in ICU patients.[6] Some studies[7,8] have shown that the risk factors of thirst include gastrointestinal diseases, inability to drink water by mouth, and the use of high-dose diuretics and opioid analgesics. However, patients in ICU who fast for surgery, disease or treatment cannot meet the basic needs of drinking water, they will also feel thirsty, and their thirst-associated pain is more obvious.[9] ICU patients often need to fast due to therapeutic factors. The incidence of thirst symptoms in ICU patients is high, which brings serious problems to ICU patients. Currently medical staff have gradually improved the recognition and intervention of thirst in ICU patients. However, at present, there are few studies on thirst in ICU fasted children. As a special population in ICU, fasted children are easy to be ignored in clinical work because their age, experience and stress state are different from adults.[10,11] Particularly, the separation anxiety, fear of medical treatment, pain and other symptoms of fasting children are more obvious, the treatment compliance and subjective comfort are reduced,[12,13] thus may affect the recovery of the disease and the hospital stay, which should attract the attention of medical care providers. Therefore, we aimed to evaluate the effect of sterile ice water versus menthol spray on thirst symptoms of fasted children in ICU, to provide reliable evidence for reducing the thirst and improving the comfort and prognosis of fasted children in ICU.
## 2. Methods
This study was a prospective cohort design. The study protocol had been analyzed and approved by the ethical committee of our hospital with approval number: 202207155-1. And written informed consents had been obtained from all the guardians of included children.
## 2.1. Sample size calculation
The sample size was calculated by the sample size estimation formula of 2 groups of independent samples.[14] *The formula* was as follows: n1=n2=2(uα+uβ[1])2σ2δ2. In this formula, n was the required content of each group of samples; σ2 was the population standard deviation (assuming that the standard deviation of 2 populations was equal), which could be expressed by the mean deviation of 2 samples; δ2 was the difference between the means of the 2 groups. α and β was the first type error rate and the second type error rate. We assumed that α = 0.05, β = 0.20, *The minimum* sample size for each group was 56 cases. Considering the rigor of the study, the convenience of clinical data collection, and the $20\%$ sample loss rate, the final sample size should be 136 cases with 68 cases in each group.
## 2.2. Study population
We selected the children who were hospitalized in the ICU of our hospital from June 1, 2021 to August 31, 2022 and needed to fast as the study population. The inclusion criteria of children in this study were as following: children who needed to fast for treatment according to the doctor advice; children aged 8 to 14 years; children who were conscious with Ramsay sedation scores score ranging from −1 to +1; the length of ICU stay was more than 24 hours; and the children and guardians were well informed and agreed to participate in this study. The exclusion criteria of children in this study were as following: children with obvious oral infection associated with salivary pancreatitis, chemotherapy, or Sjogren syndrome, which might significantly influence the thirst feelings of children; those who have consciousness barriers and were unwilling to communicate; children whom were unwilling to participate in this study.
## 2.3. Equipment
Spray bottle: The capacity was 20 mL, and the amount of spray liquid was 0.5 mL per pressing. Manufacturer: Miniso (Nanjing Co., Ltd.); Sterilized water: The capacity was 500 mL and placed in the refrigerator at 4 °C. And it was be directly loaded into the selected watering to make ice water spray. Manufacturer: Shiyao Yinhu Pharmaceutical Co., Ltd. Menthol: The menthol for food flavor was soluble in water to prepare the menthol spray liquid. The preparation method was as following: We took l mL of menthol, added it to 500 mL sterilized water at 4 °C in cold storage, shake well enough to prepare a saturated solution of menthol with a concentration of 2‰. Menthol manufacturer: Nanjing Hua Yang Essence and Fragrance Industry Co., Ltd.
## 2.4. Interventions
The included children were randomly assigned to the ice water group and menthol group accordingly. Ice water group: confirming the children were conscious, we began to intervene for 3 timepoints, namely 0 minute, 15 minutes, and 30 minutes after starting spray. Before operation, we sprayed the air for several times to remove the air from the spout, so as to make the spray even. Then we asked the child to open his mouth and lift his tongue, sprayed it to the sublingual area, the mucosa in the left and right cheeks, and the tongue surface, and asked the child to close his mouth for 5 minutes for every intervention. After 40 minutes, the pain degree of thirst, medical fear and comfort degree of the children were scored respectively.
Menthol group: The evaluation and intervention period was the same as that of Ice water group. The prepared ice menthol spray was used to spray the oral cavity. The spraying method was the same as that of Ice water group, and the children were instructed to close their mouths for 5 minutes. After 40 minutes, the pain degree of thirst, medical fear and comfort degree of the children were scored respectively.
## 2.5.1. Thirst Distress Scale (TDS).
TDS[15] consists of 3 dimensions and 6 items, including duration of thirst (2 items), frequency of thirst (2 items) and intensity of thirst (2 items). The Likert 4 subscale method is used in this scale. One point means irrelevant, and 4 points means very relevant. Children can choose the number that can represent their subjective feelings, and the total score is the evaluation result. The total score is 6~24 points, which respectively indicates that there is no discomfort to continuous thirst.
## 2.5.2. Oral mucosa wetness scale (OMWS).
This scale[16] evaluate the degree of wetness of oral mucosa according to the wetness of oral mucosa. The lips begin to crack and peel, the mouth is completely dry for 4 points, the lips and mouth are completely dry for 3 points, the mouth is wet and the lips is dry for 2 points, and the mouth and lips are wet for 1 point. The lower the score, the more moister the mouth.
## 2.5.3. Children medical fear scale (CMFS).
CMFS[17] consists of 4 dimensions and 14 items, including fear of medical environment (3 items), fear of medical operation (4 items), fear of interpersonal relationship (4 items) and fear of self (3 items). The Likert 3 subscale method is used in this scale. One point indicates no phobia, 2 points indicates some phobia, and 3 points indicates very phobia. Children can choose the number that can represent their subjective feelings, and the total score is the evaluation result. The total score is 14~42, which respectively indicates no fear to continuous fear.
## 2.5.4. Numerical rating scale (NRS).
NRS[18] was rated with a ruler printed with a total of 11 numbers from 0 to 10. 0 represents “no thirst,” 1 to 3 represents mild thirst, 4 to 7 represents moderate thirst, and 8 to 10 represents severe thirst. During the evaluation, the nurse explained the meaning of the ruler number to the patient in a unified language, so that the patient could tick the number representing the severity of his thirst. It has been reported that the internal consistency of thirst intensity of patients in ICU and anesthesia recovery room is good (Cronbach α = 0.82).
## 2.5.5. Unstimulated whole saliva (UWS).
In this study, the cotton swab method was used to measure UWS flow rate.[19] After the spray intervention, the saliva in the mouth was sucked dry with a cotton ball, and then 3 weighed dry cotton swabs were selected and placed under the tongue and parotid glands on both sides of the children respectively. The patient was instructed not to swallow, and the time was 2 minutes. When the cotton swab was taken out, the saliva on the tongue surface was sucked dry together, and then the gross weight was weighed, and then the value obtained was divided by 2, which was the value of static total saliva flow rate (mg/min). The UWS flow rate before and after intervention was compared between the 2 groups.
## 2.6. Statistical methods
In this study, SPSS 23.0 statistical software (IBM Company, Armonk, NY) was used for data statistical analysis. Descriptive statistics were used for general data of patients, frequency (%) was used for qualitative data, Chi square test or Fisher exact test was used for inter group comparison; mean ± standard deviation was used for quantitative data conforming to normal distribution, 2 independent samples t test was used for inter group comparison, and paired samples t test was used for intra group comparison. $P \leq .05$ meant that the difference was statistically significant between 2 groups.
## 3.1. The characteristics of included children
A total of 150 fasted children were initially identified, 8 children were excluded because of not meeting the inclusion criteria, 3 children did not receive the allocated interventions. Finally, 139 children were included, with 69 children in ice water group and 70 children in menthol group (Fig. 1).
**Figure 1.:** *The CONSORT flow diagram of children inclusion.*
The characteristics of included children are presented in Table 1. There were no significant differences in the age, gender, body mass index, acute physiology and chronic health evaluation score, disease types, diuretics use, opioid use, hyponatremia, hypernatremia, blood ionized calcium level, blood sugar, antihypertensive therapy, duration of fasting and length of ICU stay between ice water group and menthol group (all $P \leq .05$).
**Table 1**
| Characteristics | Ice water group (n = 69) | Menthol group (n = 70) | t/χ2 | P |
| --- | --- | --- | --- | --- |
| Age (yr) | 11.2 ± 3.2 | 11.1 ± 2.9 | 2.127 | 0.104 |
| Female/male | 33/36 | 31/39 | 1.503 | 0.087 |
| BMI (kg/m2) | 19.6 ± 3.1 | 19.6 ± 3.0 | 2.174 | 0.116 |
| APACHE-II score | 18.7 ± 2.7 | 18.9 ± 3.1 | 3.005 | 0.102 |
| Disease types | | | 2.482 | 0.058 |
| Cardiovascular disease | 41 (59.42%) | 44 (62.86%) | | |
| Respiratory diseases | 12 (17.39%) | 13 (18.57%) | | |
| Surgical disease | 16 (23.19%) | 13 (18.57%) | | |
| Undergoing continuous renal replacement therapy | 8 (11.59%) | 8 (11.43%) | 1.353 | 0.117 |
| Diuretic use | 21 (30.43%) | 20 (28.57%) | 1.154 | 0.081 |
| Opioid use | 14 (20.29%) | 17 (24.29%) | 1.612 | 0.079 |
| Hyponatraemia | 3 (4.35%) | 3 (4.29%) | 1.009 | 0.088 |
| Hypernatraemia | 5 (7.25%) | 4 (5.71%) | 1.281 | 0.059 |
| Blood ionized calcium level (mmol/L) | 2.5 ± 0.7 | 2.4 ± 0.6 | 1.446 | 0.274 |
| Blood sugar level (mmol/L) | 5.6 ± 2.4 | 5.8 ± 3.0 | 1.208 | 0.116 |
| Antihypertensive therapy | 10 (14.49%) | 9 (12.86%) | 1.952 | 0.083 |
| Duration of fasting (h) | 14.2 ± 4.5 | 14.2 ± 5.1 | 3.199 | 0.121 |
| Length of ICU stay (d) | 5.5 ± 2.2 | 5.6 ± 2.8 | 2.018 | 0.095 |
## 3.2. TDS score
As shown in Table 2, there was no significant difference in the TDS score before intervention between ice water group and menthol group ($$P \leq .095$$). After intervention, the TDS score of menthol group was statistically less than that of ice water group ($$P \leq .016$$), indicating that menthol spray may be more beneficial to reduce the thirst distress.
**Table 2**
| Unnamed: 0 | Ice water group (n = 69) | Menthol group (n = 70) | t | P |
| --- | --- | --- | --- | --- |
| Before intervention | 18.17 ± 3.53 | 18.12 ± 4.07 | 3.114 | 0.095 |
| After intervention | 16.03 ± 2.15 | 14.09 ± 3.12 | 2.506 | 0.016 |
| t | 2.105 | 2.531 | | |
| P | .043 | .008 | | |
## 3.3. OMWS score
As shown in Table 3, there was no significant difference in the OMWS score before intervention between ice water group and menthol group ($$P \leq .093$$). After intervention, the OMWS score of menthol group was statistically less than that of ice water group ($$P \leq .009$$), indicating that menthol spray may be more beneficial to increase the oral mucosa wetness.
**Table 3**
| Unnamed: 0 | Ice water group (n = 69) | Menthol group (n = 70) | t | P |
| --- | --- | --- | --- | --- |
| Before intervention | 3.44 ± 1.91 | 3.38 ± 1.02 | 1.452 | 0.093 |
| After intervention | 2.43 ± 1.04 | 1.27 ± 1.12 | 1.028 | 0.009 |
| t | 1.201 | 1.945 | | |
| P | .062 | .022 | | |
## 3.4. CMFS score
As shown in Table 4, there were no significant differences in the CMFS score before and after intervention between ice water group and menthol group (all $P \leq .05$).
**Table 4**
| Unnamed: 0 | Ice water group (n = 69) | Menthol group (n = 70) | t | P |
| --- | --- | --- | --- | --- |
| Before intervention | 31.05 ± 7.22 | 30.25 ± 6.79 | 6.244 | 0.101 |
| After intervention | 28.53 ± 6.95 | 27.94 ± 7.03 | 5.916 | 0.089 |
| t | 5.421 | 6.014 | | |
| P | .056 | .095 | | |
## 3.5. NRS score
As shown in Table 5, there was no significant difference in the NRS score before intervention between ice water group and menthol group ($$P \leq .112$$). After intervention, the NRS score of menthol group was statistically less than that of ice water group ($$P \leq .044$$), indicating that menthol spray may be more beneficial to reduce the thirsty feelings.
**Table 5**
| Unnamed: 0 | Ice water group (n = 69) | Menthol group (n = 70) | t | P |
| --- | --- | --- | --- | --- |
| Before intervention | 8.17 ± 3.53 | 8.23 ± 3.74 | 2.016 | 0.112 |
| After intervention | 5.46 ± 2.01 | 3.14 ± 1.95 | 1.933 | 0.044 |
| t | 1.255 | 1.402 | | |
| P | .049 | .017 | | |
## 3.6. UWS flow rate
As shown in Table 6, there was no significant difference in the UWS flow rate before intervention between ice water group and menthol group ($$P \leq .181$$). After intervention, the UWS flow rate of menthol group was statistically higher than that of ice water group ($$P \leq .034$$), indicating that menthol spray may be more beneficial to increase the UWS flow rate.
**Table 6**
| Unnamed: 0 | Ice water group (n = 69) | Menthol group (n = 70) | t | P |
| --- | --- | --- | --- | --- |
| Before intervention | 0.26 ± 0.11 | 0.26 ± 0.14 | 1.094 | 0.181 |
| After intervention | 0.49 ± 0.14 | 0.72 ± 0.22 | 1.217 | 0.034 |
| t | 1.103 | 1.295 | | |
| P | .077 | .041 | | |
## 4. Discussion
Thirst is a subjective feeling that can trigger the desire of the human body to drink water, a manifestation of body imbalance, and one of the important factors influencing the patient comfort.[20,21] The normal value of serum sodium concentration is 135 to 145 mmol/L. When the serum sodium concentration exceeds the normal range, hypernatremia will occur, and the pituitary osmoreceptor cells will dehydrate. The plasma alternating osmotic pressure is affected accordingly. As long as it is $1\%$ to $2\%$ higher than the normal range, it will stimulate the increase of excitatory output, thus triggering the release of antidiuretic hormone and the generation of thirst.[22,23] When the balance of volume inside and outside cells is broken, the body will trigger a negative feedback system to trigger thirst signals to restore this balance.[24] The pain of thirst in ICU fasting patients can cause anxiety, irritability, low treatment compliance, sleep disorders and other states.[25] Some studies[26,27] show that strong thirst lasting more than 24 hours will increase the risk of delirium. The experience of thirst discomfort will make ICU patients in a state of stress, significantly increase the patient oxygen consumption, increase the burden of organ metabolism, and affect the prognosis of patients.[28] The painful experience of thirst will continue until the patient leaves the ICU, which will lead to a decrease in the patient comfort, an increase in dissociative anxiety, and an increase in fear, affecting the recovery of the disease.[29,30] For children in ICU, prolonged fasting and water deprivation will significantly increase their psychological pressure, keep their bodies in an emergency state, delay their recovery, and cannot meet their comfort requirements.[31] The results of this study have found that compared with ice water spray, menthol spray may be more beneficial to reduce the thirst distress and feelings, increase the oral mucosa wetness and UWS flow rate, menthol spray may be promoted in the clinical nursing care of fasted children in ICU.
Some studies[32,33] have pointed out that, thirst can be relieved by “pre absorption” and “post absorption.” Before water is absorbed by the body, the liquid in the mouth and in the process of swallowing will stimulate the receptors in the mouth and pharynx, so that excitement will be transmitted to the brain to produce a sense of satisfaction for drinking water.[34] After water is absorbed by the body, the thirst is relieved mainly by maintaining the balance of plasma osmotic pressure and blood volume.[35] Previous studies[36,37] have showed that through “pre absorption” stimulating the receptors of oropharynx to relieve thirst of such patients is a very effective intervention strategy. Compared with normal temperature stimulation, stimulation of oropharyngeal receptors with low temperature has a significant effect on relieving the symptoms of thirst in patients.[38] It has been reported that cold stimulation can activate the TRPM8 cold receptor distributed in the oral mucosa, therefore leading to cold sensation.[39] There will be more comfort without taking a lot of water, and the discomfort of thirst will also be alleviated.[40] *Many previous* studies[7,8] have shown that the local intervention of ice water spray is more effective than the cotton swab lip wetting method. Ice water spray will change the liquid into water mist, and under the action of the nozzle, it will quickly and evenly reach each part of the oral cavity. At the same time, the water mist can form a protective film to protect the oral mucosa, maximize the moist effect of the oral mucosa, reduce the occurrence of dry lips and whitening, thus reducing the occurrence of dry mouth and thirst.[41] The spray range of ice water spray is up to about 40 cm, so as to ensure that the spray can reach the deep throat of the patient, produce a moist effect on the throat, and do not stimulate the deep throat of the patient, enhancing patient comfort.[42] The decrease of saliva secretion is also the main reason for patients to feel thirsty. Ice water spray can act on the throat of patients to stimulate them to produce more saliva.[43] Besides, it can also inhibit the release of antidiuretic hormone, reduce the frequency of patients’ thirst, thus reducing the number of additional interventions.[44] The results of this study show that menthol spray may be more beneficial because it can stimulate the taste buds to produce a sense of coolness, and reduce the patient thirst and discomfort caused by the taste. The nourishing effect of liquid on the mouth and pharynx will also make thirsty patients feel happy and rewarded. There are osmoreceptors in the oral and throat mucosa. When patients feel the liquid in the mouth and throat, they can quickly relieve thirst. Besides, both cold and menthol stimulation can activate the transient receptor potential channel of oropharynx, so that stimulation can be transmitted to the cerebral cortex of children, resulting in reflex inhibition of thirst and pleasure satisfaction.[45,46] This study has some limitations that are worth considering. First of all, this study is a single center study, involving a small number of children. Secondly, limited by the research conditions, the effect of this study on ice water and menthol is mainly judged by various scales, and there is still a lack of objective indicators to analyze the treatment effect. Thirst is primarily caused by an increase in fluid osmolality and decrease in blood volume. We did not regularly monitor the plasma osmolality and blood volume unless the children had obvious related symptoms, thus we lacked the data on the plasma osmolality and blood volume. Thirdly, we currently did not find any side effect associated with menthol spray, more long-term studies are needed to further evaluate the safety of menthol spray. Finally, this study did not apply blind methods to patients, nurses and outcome evaluators. It is necessary to conduct more high-quality researches with larger sample in the future to further analyze the effect and safety of ice water and menthol on fasting children.
## 5. Conclusions
In conclusion, we have found that compared with ice water spray, the menthol spray may be more advantageous to reduce the thirst distress and feelings, increase the oral saliva secretion and mucosa wetness. Menthol spray is a safe, convenient and effective method for preventing thirst, which may be worthy of promoted in clinical nursing practice for fasted children in ICU.
## Author contributions
Data curation: Kai Pu.
Formal analysis: Kai Pu.
Funding acquisition: Kai Pu.
Investigation: Fangyan Ma, Haiting He, Banghong Xu, Jing Zhou, Kai Pu.
Methodology: Kai Pu.
Visualization: Kai Pu.
Writing – original draft: Kai Pu.
## References
1. Arbios D, Srivastava J, Gray E. **Cumulative stress debriefings to combat compassion fatigue in a pediatric intensive care unit.**. *Am J Crit Care* (2022) **31** 111-8. PMID: 35229152
2. Joosten K, Verbruggen S. **PN administration in critically ill children in different phases of the stress response.**. *Nutrients* (2022) **14** 1819. PMID: 35565787
3. Pota V, Coppolino F, Barbarisi A. **Pain in intensive care: a narrative review.**. *Pain Ther* (2022) **11** 359-67. PMID: 35220551
4. Puntillo K, Nelson JE, Weissman D. **Palliative care in the ICU: relief of pain, dyspnea, and thirst--a report from the IPAL-ICU Advisory Board.**. *Intensive Care Med* (2014) **40** 235-48. PMID: 24275901
5. Kjeldsen CL, Hansen MS, Jensen K. **Patients’ experience of thirst while being conscious and mechanically ventilated in the intensive care unit.**. *Nurs Crit Care* (2018) **23** 75-81. PMID: 28124464
6. Lindner G, Funk GC. **Hypernatremia in critically ill patients.**. *J Crit Care* (2013) **28** 216.e11216 e211-216.e20
7. Puntillo K, Arai SR, Cooper BA. **A randomized clinical trial of an intervention to relieve thirst and dry mouth in intensive care unit patients.**. *Intensive Care Med* (2014) **40** 1295-302. PMID: 24894026
8. VonStein M, Buchko BL, Millen C. **Effect of a scheduled nurse intervention on thirst and dry mouth in intensive care patients.**. *Am J Crit Care* (2019) **28** 41-6. PMID: 30600225
9. Flim M, Hofhuis J, Spronk P. **Measuring thirst distress of patients in the intensive care unit.**. *Nurs Crit Care* (2022) **27** 576-82. PMID: 34612559
10. Nabialek T, Tume LN, Cercueil E. **Planned peri-extubation fasting in critically ill children: an international survey of practice.**. *Front Pediatr* (2022) **10** 905058. PMID: 35633966
11. Veldscholte K, Cramer ABG, Joosten KFM. **Intermittent fasting in paediatric critical illness: the properties and potential beneficial effects of an overnight fast in the PICU.**. *Clin Nutr* (2021) **40** 5122-32. PMID: 34461586
12. Park J, Choi EK. **Transfer anxiety in parents of children transferred from pediatric intensive care units to general wards in South Korea: a hybrid concept analysis.**. *Child Health Nurs Res* (2022) **28** 154-65. PMID: 35538727
13. Badanta B, Rivilla-Garcia E, Lucchetti G. **The influence of spirituality and religion on critical care nursing: an integrative review.**. *Nurs Crit Care* (2022) **27** 348-66. PMID: 33966310
14. Li L, Yuxin T, Wei H. **Sample size estimation of common clinical trial design.**. *Syst Med* (2018) **3** 3-6
15. Qixiang D, Yuhong C, Qiqi J. **Evaluation on the application of thirst pain scale in maintenance hemodialysis patients.**. *J Nurs Train* (2019) **34** 5-7
16. Kailing Z, Dan F. **Evaluation on the effect of local intervention of ice water spray on patients with severe water deprivation or water restriction combined with thirst.**. *Med Theory Pract* (2022) **28** 35-8
17. Huan L, Xuan W, Liping Z. **The influence of noise on children’s medical fear and anxiety: the regulatory role of will control.**. *Chin J Clin Psychol* (2021) **29** 5-8
18. Wanlu G, Xiaohai W. **Preoperative selection of patient pain scoring method and effect analysis of postoperative pain evaluation.**. *J Pract Med* (2013) **29** 3-6
19. Jingyun Y, Jieting Z, Peiyi C. **Observation on the effect of black plum spray on improving thirst in hemodialysis patients based on the theory of acid sweet yin.**. *J Nurs* (2017) **24** 4-7
20. Rui Q, Jihong Z, Xuemin Z. **Construction of nursing intervention process for patients with thirst after cardiac surgery in ICU.**. *Chin J Mod Nurs* (2021) **27** 8-10
21. Yaoning Z, Yingbin X, Meichun H. **Analysis of thirst level and influencing factors of patients with non-invasive mechanical ventilation in ICU.**. *J Nurs* (2019) **34** 3-6
22. Leemhuis A, Shichishima Y, Puntillo K. **Palliation of thirst in intensive care unit patients: translating research into practice.**. *Crit Care Nurse* (2019) **39** 21-8. PMID: 31575591
23. Lin R, Li H, Chen L. **Prevalence of and risk factors for thirst in the intensive care unit: an observational study.**. *J Clin Nurs* (2022) **10** 4-7
24. Xin L, Jing G, Hongjia Q. **Analysis on influencing factors of thirst in ICU critically ill patients.**. *Chin J Pract Nurs* (2018) **34** 5-8
25. Junshan C, Aiqin Z. **Research progress of thirst in ICU patients with mechanical ventilation.**. *J Nurs* (2018) **33** 102-6
26. Hong L, Rong L, Jing Y. **Evaluation of the application effect of humanistic care strategy based on comfort nursing mode in ICU thirsty patients.**. *J Nurs Train* (2019) **34** 4-7
27. Yanhua C, Yanjun M, Jihui G. **Application of spray method in relieving thirst symptoms of patients with noninvasive positive pressure ventilation in ICU.**. *Shanghai Nurs* (2022) **28** 22-7
28. Negro A, Villa G, Greco M. **Thirst in patients admitted to intensive care units: an observational study.**. *Ir J Med Sci* (2022) **191** 2283-9. PMID: 34668107
29. Stotts NA, Arai SR, Cooper BA. **Predictors of thirst in intensive care unit patients.**. *J Pain Symptom Manage* (2015) **49** 530-8. PMID: 25116914
30. Sato K, Okajima M, Taniguchi T. **Association of persistent intense thirst with delirium among critically ill patients: a cross-sectional study.**. *J Pain Symptom Manage* (2019) **57** 1114-20. PMID: 30831240
31. Luo SH, Guo Q, Liu GJ. **Fasting for haemostasis in children with gastrointestinal bleeding.**. *Cochrane Database Syst Rev* (2016) **23** CD010714
32. Lingjun Y, Boya W, Jingjing P. **Ice stimulation therapy to prevent thirst in patients undergoing oral tracheal intubation mechanical ventilation.**. *Evid Based Nurs* (2021) **17** 4-7
33. Shuli Z. **The construction and empirical study of oral nursing plan for patients with mechanical ventilation through oral tracheal intubation.**. *J Pract Clin Nurs* (2019) **4** 1-5
34. Rui Q, Lei Y, Aiqin Z. **Research progress on influencing factors and nursing intervention of thirst in critically ill patients.**. *Chin J Pract Nurs* (2021) **37** 6-8
35. Yating F, Changcui Q. **Effect of intermittent oxygen driven humidification on relieving thirst of patients after gastrointestinal surgery in ICU.**. *Nurs Res* (2021) **35** 4-7
36. Maddison S, Wood RJ, Rolls ET. **Drinking in the rhesus monkey: peripheral factors.**. *J Comp Physiol Psychol* (1980) **94** 365-74. PMID: 6767760
37. Cersosimo E, Solis-Herrera C, Triplitt C. **Inhibition of renal glucose reabsorption as a novel treatment for diabetes patients.**. *J Bras Nefrol* (2014) **36** 80-92. PMID: 24676619
38. Xiaoxue Z, Chaozhu H, Jiehua W. **Effect of drinking a small amount of ice water on thirst of patients undergoing oral endotracheal intubation based on the safety strategy of thirst management.**. *J Nurs* (2020) **27** 4-8
39. Eccles R, Du-Plessis L, Dommels Y. **Cold pleasure. why we like ice drinks, ice-lollies and ice cream.**. *Appetite* (2013) **71** 357-60. PMID: 24060271
40. Wenxuan Y, Qi Z. **Investigation and analysis of the current situation of thirst and thirst pain in critically ill patients after surgery.**. *Mod Clin Nurs* (2020) **19** 7-9
41. Ford C, McCormick D, Parkosewich J. **Safety and effectiveness of early oral hydration in patients after cardiothoracic surgery.**. *Am J Crit Care* (2020) **29** 292-300. PMID: 32607569
42. Xiulian Y, Rong L. **Effect of ice water spray on improving oropharyngeal discomfort in gastrointestinal tumor patients undergoing gastrointestinal decompression.**. *Med Theory Pract* (2021) **34** 3-6
43. Xiaoying L, Yuehong H, Mengjiao D. **Nursing effect of ice water wiping combined with cold mint mouthwash spray on thirsty and thirsty ICU patients.**. *Nurs Rehabil* (2021) **18** 10-3
44. Yucui X, Yuan Y, Yadi F. **Evaluation of the effect of comprehensive nursing measures on relieving xerostomia in severe neurosurgical patients.**. *Mod Clin Nurs* (2021) **20** 6-9
45. Xiantong S, Yishan S. **Application of menthol ice lolly in the treatment of postoperative thirst of patients in the post anesthesia recovery room.**. *China Rural Med* (2022) **11** 29-31
46. Xiaolan C, Silan L, Bing X. **The effect of menthol application on the prevention and treatment of thirst in patients with digestive tract tumors during perioperative period.**. *Hainan Med J* (2020) **31** 4-9
|
---
title: The serum creatinine to cystatin C ratio predicts the risk of acute exacerbation
of chronic obstructive pulmonary disease
authors:
- Liang He
- Yan Li
- Xijun Gou
- Ling Lei
journal: Medicine
year: 2023
pmcid: PMC10036013
doi: 10.1097/MD.0000000000033304
license: CC BY 4.0
---
# The serum creatinine to cystatin C ratio predicts the risk of acute exacerbation of chronic obstructive pulmonary disease
## Abstract
The purpose of acute exacerbation of chronic obstructive pulmonary disease (AECOPD) treatment is to minimize the negative impact of the current exacerbation and to prevent the development of subsequent events. Therefore, it is important to identify readily available serological indicators during hospital admission to assess the prognosis of patients with AECOPD. All patients hospitalized in a Department of Respiratory and Critical Care Medicine of tertiary care hospital between January 2021 and December 2021 for AECOPD were analyzed using univariate correlations and binary logistic regression analysis with 2 models for associations between demographic, clinical, and laboratory features and AECOPD risk. The ratio of creatinine to cystatin C (Cre/Cys C) ratio was significantly associated with age (r = −0.206, $$P \leq .000$$), weight ($R = 0.331$, $$P \leq .000$$), body mass index (BMI) ($R = 0.133$, $$P \leq .007$$), and forced vital capacity (FVC)% predicted ($R = 0.130$, $$P \leq .009$$). Multiple regression was performed to predict the Cre/Cys C ratio from age, weight, BMI, forced expiratory volume during 1 second/FVC ratio, and FVC% predicted FABP-4, with F [5, 405] = 24.571, $$P \leq .000$$, R2 = 0.233. The results showed that the most significant predictors of the Cre/Cys C ratio were age ($$P \leq .007$$), weight ($$P \leq .000$$), BMI ($$P \leq .000$$), and predicted forced expiratory volume during 1 second ($$P \leq .000$$). Multivariate analysis was performed to determine whether the Cre/Cys C ratio was a predictor of AECOPD risk. Model 1 showed that a low Cre/Cys C ratio was associated with an increased hospital length of stay (odds ratio: −0.114, $95\%$ confidence interval: −0.061 to −0.005) and admission to the intensive care unit (odds ratio: 0.951, $95\%$ confidence interval: 0.907–0.996). After adjustment for potential confounding factors, model 2 showed that a low Cre/Cys C ratio was not independently associated with AECOPD risk. The present study indicated that the Cre/Cys C ratio is an easy, cheap, repeatable, and promising tool that allows us to evaluate the risk of AECOPD using serum markers. A low Cre/Cys C ratio was associated with a prolonged hospital length of stay and admission to the intensive care unit in AECOPD patients. However, the associations were not independent.
## 1. Introduction
Chronic obstructive pulmonary disease (COPD) is currently the 4th leading cause of death in the world but is projected to be the 3rd leading cause of death by 2020. Many people suffer from this disease for years and die prematurely from it or its complications.[1] COPD has a chronic and progressive course that is often punctuated by “exacerbations,” which are driven by respiratory infections and multiple functional impairments and comorbidities.[2–4] The purposes of acute exacerbation of chronic obstructive pulmonary disease (AECOPD) treatment are to minimize the negative impact of the current exacerbation and to prevent the development of subsequent events.[5] Acute exacerbation of COPD (AECOPD) is a primary cause of hospitalization and is associated with high mortality.[6]Therefore, it is important to identify serological indicators that are readily available at the time of hospital admission to evaluate the prognosis of patients with AECOPD.
The ratio of creatinine to cystatin C (Cre/Cys C) was first reported as a surrogate marker of muscle mass in 2013.[7] *It is* noninvasive, fast, and less expensive than other sarcopenia examination methods, such as magnetic resonance imaging and computed tomography.[8] *On this* basis, Cre/Cys C has been proven to be an accurate and inexpensive predictor of sarcopenia in patients with COPD,[8] type 2 diabetes,[9] and cancer.[10] Moreover, it is correlated with glucose disposal ability and diabetic complications in patients with type 2 diabetes.[9] *Sarcopenia is* common in patients with COPD,[11] which may affect both respiratory muscles and limb muscles and could, therefore, have critical consequences related to the quality of life and prognosis of COPD patients.[11,12] However, to date, no studies have evaluated the relationship of Cre/Cys C levels upon admission for AECOPD with the outcome of hospitalization. As it is a routine blood test, a detailed analysis of Cre/Cys C levels relation to risk seems worthwhile. Therefore, this study aimed to investigate the association between Cre/Cys C levels and risk in patients with AECOPD.
## 2.1. Study design and patient recruitment
We retrospectively and consecutively enrolled AECOPD admissions to the Department of Respiratory and Critical Care Medicine of Xindu District People’s Hospital between January 2021 and December 2021. Patients who received treatment according to the global initiative for chronic obstructive lung disease (GOLD) were invited to participate. The participants were divided into 2 groups based on their Cre/Cys C ratio values: the low Cre/Cys C ratio group and the high Cre/Cys C ratio group. The median Cre/Cys C ratio was 74.49, which was used as the threshold value, below which the Cre/Cys C ratio value was considered low.[13] The median value and any values above it were considered high Cre/Cys C ratio values.
Criteria for inclusion: *The diagnosis* of AECOPD was established based on the criteria of the GOLD.[1] Exacerbation of COPD was defined as an acute worsening of respiratory symptoms that resulted in additional therapy. All patients included in the study have completed blood routine and biochemical indicators, and complete clinical characteristic data.
The following patients were excluded: Age <18 years; Other potential causes of sarcopenia (malignant diseases, heart failure, hyperthyroidism, or any other devastating chronic disease); Chronic kidney disease or acute kidney injury; Concomitant active malignancy; and Incomplete clinical records.
The study was carried out in accordance with the Declaration of Helsinki and was approved by the Ethical Review Committee of the Xindu District People’s Hospital of Chengdu. Due to the retrospective nature of the study and the anonymity of the data, informed consent was not required from the patients.
## 2.2. Pulmonary function tests
All the enrolled patients underwent spirometry, and the forced expiratory volume during 1 second (FEV1) and forced vital capacity (FVC) were measured according to the GOLD consensus guidelines.[1] The FEV1, FEV1/FVC, and the ratio of FEV1 to predicted FEV1 after inhaling bronchodilators were recorded. According to the category of airflow limitation in COPD (based on FEV1 after the administration of a bronchodilator) in patients with FEV1/FVC < 0.70, stable COPD was divided into 4 subgroups: mild (GOLD 1), ≥$80\%$ of predicted FEV1; moderate (GOLD 2), $50\%$ to $80\%$ of predicted FEV1; severe (GOLD 3), $30\%$ to $50\%$ of predicted FEV1; very severe (GOLD 4), <$30\%$ of predicted FEV1.
## 2.3. Data collection and outcome assessment
Patient demographics, including age, sex, and body mass index (BMI), were recorded. We collected initial laboratory findings obtained during hospitalization, including serum creatinine and cystatin C levels.
Serum creatinine and cystatin C levels were measured at the hospital laboratory. Serum creatinine levels were measured using the enzymatic method (Beckman Coulter, Japan), while serum cystatin C levels were measured using the latex agglutination turbidimetric immunoassay (Wan Tai Drd, Beijing, China). The Cre/Cys C ratio was calculated by dividing the serum creatinine value by the serum cystatin C value.
The primary outcomes included respiratory support, complications during hospitalization, hospital length of stay (LOS), admission to the intensive care unit (ICU), and mortality. Respiratory support refers to the use of noninvasive ventilators or invasive ventilators during hospitalization due to the condition. Complications are the combination of acute hypoxemic respiratory failure or acute hypercapnic respiratory failure during hospitalization. LOS was obtained by calculating the patient’s admission time and discharge time. Mortality refers to death during hospitalization at this visit.
## 2.4. Statistical analysis
SPSS 26.0 (IBM Corp, Armonk, NY) was used to analyze the results statistically. Continuous variables were first evaluated for normal distribution. The normally distributed variables are presented as the means ± standard deviations. Categorical variables are presented as frequency counts and percentages. The variables between 2 groups were compared using the independent samples t test or Mann–Whitney U test. Categorical variables were analyzed using the χ2 test. Univariable correlations between variables were assessed based on Pearson correlation coefficients. Binary logistic regression analysis was used to analyze the association between the Cre/Cys C ratio and AECOPD risk. Two models were constructed for the regression analysis: model 1 was unadjusted, while model 2 was adjusted for confounding variables, including age, sex, weight, BMI, and chronic diseases (hypertension, diabetes, and coronary heart disease). We adjusted these variables to determine whether the factors of interest were related to AECOPD risk. P value <.05 was defined as the significance threshold.
## 3.1. Study participant characteristics
A total of 411 patients were included in the present study. The mean age of the participants was 70.49 years, and the age range was 38 to 95 years. There were 246 males ($59.85\%$) and 165 females ($40.15\%$). The mean Cre/Cys C ratio was 75.95 ± 14.58. The noninvasive ventilation rate, acute hypercapnic respiratory failure rate, ICU admission rate, and mortality rate were $11.68\%$, $22.87\%$, $2.92\%$, and $0.24\%$, respectively.
In the low Cre/Cys C ratio group, there were no significant differences in the different airflow limitation severity groups, including weight ($$P \leq .113$$), BMI ($$P \leq .445$$), creatinine level ($$P \leq .419$$), cystatin C level ($$P \leq .875$$), and Cre/Cys C ratio ($$P \leq .255$$). However, the age of the mild group was significantly higher than that of the other group ($$P \leq .015$$). The male ratio was also higher than that in the other group ($$P \leq .000$$). ( Table 1).
**Table 1**
| Unnamed: 0 | Low Cre/Cys C ratio group (n = 206) | Low Cre/Cys C ratio group (n = 206).1 | Low Cre/Cys C ratio group (n = 206).2 | Low Cre/Cys C ratio group (n = 206).3 | Low Cre/Cys C ratio group (n = 206).4 | Low Cre/Cys C ratio group (n = 206).5 | High Cre/Cys C ratio group (n = 205) | High Cre/Cys C ratio group (n = 205).1 | High Cre/Cys C ratio group (n = 205).2 | High Cre/Cys C ratio group (n = 205).3 | High Cre/Cys C ratio group (n = 205).4 | High Cre/Cys C ratio group (n = 205).5 | F/X2 value | P value |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| | Mild (n = 18) | Moderate (n = 76) | Severe (n = 67) | Very Severe (n = 45) | F/X2 value | P value | Mild (n = 23) | Moderate (n = 92) | Severe (n = 56) | Very Severe (n = 34) | F/X2 value | P value | F/X2 value | P value |
| Age, yr, mean (SD) | 77.00 ± 8.71 | 73.14 ± 9.92 | 71.67 ± 8.45 | 68.91 ± 10.89 | 3.566 | .015 | 73.26 ± 10.86 | 67.30 ± 11.35 | 68.79 ± 8.96 | 70.44 ± 10.23 | 2.274 | .081 | 1.207 | .273 |
| Male, n (%) | 2 (11.1) | 27 (35.5) | 32 (47.8) | 29 (64.4) | 18.155 | .000 | 18 (78.3) | 70 (76.1) | 40 (71.4) | 28 (82.4) | 1.462 | .691 | 44.911 | .000 |
| Weight | 46.89 ± 8.70 | 53.08 ± 10.46 | 50.82 ± 10.71 | 53.67 ± 14.23 | 2.012 | .113 | 55.35 ± 9.48 | 58.65 ± 9.93 | 56.43 ± 9.36 | 55.24 ± 9.20 | 1.573 | .197 | 6.479 | .011 |
| BMI | 22.62 ± 3.03 | 22.85 ± 3.79 | 21.87 ± 4.30 | 21.77 ± 5.37 | 0.895 | .445 | 22.56 ± 2.86 | 23.31 ± 3.56 | 22.35 ± 3.09 | 21.47 ± 3.34 | 2.796 | .041 | 5.145 | .024 |
| FEV1/FVC | 62.99 ± 4.49 | 58.51 ± 6.82 | 48.40 ± 7.39 | 44.65 ± 8.90 | 53.185 | .000 | 64.38 ± 3.66 | 59.95 ± 6.86 | 48.01 ± 6.36 | 42.35 ± 7.22 | 97.111 | .000 | 1.047 | .307 |
| FEV1 (% of predicted value) | 100.53 ± 22.44 | 63.32 ± 8.94 | 38.14 ± 5.75 | 22.50 ± 4.10 | 396.019 | .000 | 91.13 ± 10.48 | 63.14 ± 8.66 | 39.47 ± 5.63 | 23.12 ± 4.18 | 483.836 | .000 | 1.087 | .298 |
| Creatinine (µmol/L) | 67.21 ± 13.80 | 68.83 ± 17.70 | 68.11 ± 18.62 | 63.49 ± 16.88 | 0.947 | .419 | 87.84 ± 37.10 | 87.73 ± 38.45 | 78.66 ± 18.96 | 87.93 ± 36.60 | 0.993 | .397 | 15.967 | .000 |
| Cystatin C (mg/L) | 1.05 ± 0.24 | 1.07 ± 0.29 | 1.05 ± 0.35 | 1.02 ± 0.26 | 0.230 | .875 | 0.97 ± 0.31 | 0.99 ± 0.40 | 0.92 ± 0.22 | 1.03 ± 0.38 | 0.856 | .465 | 0.773 | .380 |
| Cre/Cys C ratio | 64.39 ± 7.10 | 65.04 ± 7.52 | 65.51 ± 6.51 | 62.74 ± 8.64 | 1.363 | .255 | 89.17 ± 13.99 | 88.23 ± 10.64 | 86.20 ± 9.99 | 85.48 ± 8.61 | 1.012 | .388 | 11.509 | .001 |
In the high Cre/Cys C ratio group, the moderate group had a significantly elevated BMI compared with the mild, severe, and very severe groups ($$P \leq .041$$). There were no significant differences in other parameters. The specific results are presented in Table 1.
Comparisons of parameters between the low Cre/Cys C ratio group and the high Cre/Cys C ratio group showed that the high Cre/Cys C ratio group had a significantly higher male ratio (76.1 vs $43.7\%$, $$P \leq .000$$), weight (57.11 ± 9.65 vs 51.93 ± 11.42, $$P \leq .011$$), BMI (22.66 ± 3.37 vs 22.27 ± 4.29, $$P \leq .024$$). There were no significant differences in other parameters. The specific results are presented in Table 1.
## 3.2. Outcomes
In the low Cre/Cys C ratio group, the noninvasive ventilation ratio (0, $1.3\%$, $11.9\%$ vs $44.4\%$, $$P \leq .000$$) and acute hypercapnic respiratory failure ratio ($11.1\%$, $5.3\%$, $28.4\%$ vs $71.1\%$, $$P \leq .000$$) of the very severe group were significantly higher than those in the other group.
In the high Cre/Cys C ratio group, compared with the mild, moderate, and severe groups, the very severe group had a significantly elevated noninvasive ventilation ratio (0, $2.2\%$, $7.1\%$ vs $38.2\%$, $$P \leq .000$$) and acute hypercapnic respiratory failure ratio (0, $5.4\%$, $21.4\%$, vs $58.8\%$, $$P \leq .000$$). Furthermore, the hospital LOS of the very severe group was significantly higher than that of the other group (9.04 ± 1.97, 9.59 ± 2.41, 10.45 ± 3.51 vs 11.47 ± 4.13, $$P \leq .005$$). There were no significant differences in other parameters. The specific results are presented in Table 2.
**Table 2**
| Outcomes | Low Cre/Cys C ratio group | Low Cre/Cys C ratio group.1 | Low Cre/Cys C ratio group.2 | Low Cre/Cys C ratio group.3 | Low Cre/Cys C ratio group.4 | Low Cre/Cys C ratio group.5 | High Cre/Cys C ratio group | High Cre/Cys C ratio group.1 | High Cre/Cys C ratio group.2 | High Cre/Cys C ratio group.3 | High Cre/Cys C ratio group.4 | High Cre/Cys C ratio group.5 | F/X2 value | P value |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| Outcomes | Mild | Moderate | Severe | Very Severe | F/X2 value | P value | Mild | Moderate | Severe | Very Severe | F/X2 value | P value | F/X2 value | P value |
| Respiratory support | | | | | | | | | | | | | | |
| Non-invasive ventilation | 0 (0) | 1 (1.3) | 8 (11.9) | 20 (44.4) | 47.742 | .000 | 0 (0) | 2 (2.2) | 4 (7.1) | 13 (38.2) | 30.298 | .000 | 2.304 | .129 |
| Invasive ventilation | 0 | 1 (1.3) | 0 | 0 | 2.574 | 1.000 | 0 | 0 | 0 | 0 | | | - | 1.000 |
| Complications | | | | | | | | | | | | | | |
| Acute hypoxemic respiratory failure | 1 (5.6) | 5 (6.6) | 7 (10.4) | 1 (2.2) | 2.946 | .400 | 1 (4.3) | 9 (9.8) | 7 (12.5) | 3 (8.8) | 1.055 | .773 | 1.186 | .276 |
| Acute hypercapnic respiratory failure | 2 (11.1) | 4 (5.3) | 19 (28.4) | 32 (71.1) | 63.979 | .000 | 0 | 5 (5.4) | 12 (21.4) | 20 (58.8) | 53.612 | .000 | 5.392 | .02 |
| Clinical outcomes | | | | | | | | | | | | | | |
| Hospital LOS (d) | 9.33 ± 2.77 | 10.21 ± 6.42 | 11.39 ± 4.25 | 12.36 ± 4.03 | 2.498 | .061 | 9.04 ± 1.97 | 9.59 ± 2.41 | 10.45 ± 3.51 | 11.47 ± 4.13 | 4.354 | .005 | 5.243 | .023 |
| Admission to the ICU | 0 | 5 (6.6) | 2 (3.0) | 0 | 4.593 | .204 | 1 (4.3) | 1 (1.1) | 2 (3.6) | 1 (2.9) | 2.326 | .461 | 0.333 | .564 |
| Mortality | 0 | 1 (1.3) | 0 | 0 | 2.574 | 1.000 | 0 | 0 | 0 | 0 | | | 0.998 | .318 |
Comparisons of parameters between the low Cre/Cys C ratio group and the high Cre/Cys C ratio group showed that the low Cre/Cys C ratio group had a significantly higher acute hypercapnic respiratory failure ratio (27.7 vs $18\%$, $$P \leq .02$$), hospital LOS (10.99 ± 5.09 vs 10.07 ± 3.11, $$P \leq .023$$). There were no significant differences in other parameters. The specific results are presented in Table 2.
## 3.3. Analysis of the correlation between the Cre/Cys C ratio and clinical parameters
With regard to study parameters, the Cre/Cys C ratio was significantly associated with age (r = −0.206, $$P \leq .000$$; Fig. 1A, Table 3), weight ($R = 0.331$, $$P \leq .000$$; Fig. 1B, Table 3), BMI ($R = 0.133$, $$P \leq .007$$; Fig. 1C, Table 3), FEV1/FVC ratio($R = 0.088$,$$P \leq .074$$; Fig. 1D, Table 3) and predicted FVC% ($R = 0.130$, $$P \leq .009$$; Fig. 1E, Table 3). Finally, a multiple regression was run to predict the Cre/Cys C ratio from age, weight, BMI, FEV1/FVC ratio, and FVC% predicted. These variables significantly predicted FABP-4, with F [5, 405] = 24.571, $$P \leq .000$$, R2 = 0.233. The results showed that the most significant predictors of the Cre/Cys C ratio were age ($$P \leq .007$$), weight ($$P \leq .000$$), BMI ($$P \leq .000$$), and predicted FEV1 ($$P \leq .000$$) (Table 4).
Multivariate analysis was performed to determine whether the Cre/Cys C ratio was a predictor of AECOPD risk. Model 1 showed that a low Cre/Cys C ratio was associated with an increased hospital LOS (odds ratio: −0.114, $95\%$ confidence interval: −0.061 to −0.005) and admission to the ICU (odds ratio: 0.951, $95\%$ confidence interval: 0.907–0.996). After adjustment for potential confounding factors, model 2 showed that a low Cre/Cys C ratio was not independently associated with AECOPD risk (Table 5).
**Table 5**
| Risk | Model 1 | Model 1.1 | Model 1.2 | Model 2 | Model 2.1 | Model 2.2 |
| --- | --- | --- | --- | --- | --- | --- |
| Risk | 95% CI | OR | P value | 95% CI | OR | P value |
| Respiratory support non-invasive ventilation | | | | | | |
| Low Cre/Cys C | 0.921 to 1.035 | 0.977 | .424 | 0.930 to 1.065 | 0.995 | .889 |
| High Cre/Cys C | 0.956 to 1.034 | 0.994 | .772 | 0.950 to 1.037 | 0.992 | .730 |
| Total | 0.983 to 1.024 | 1.003 | .762 | 0.985 to 1.035 | 1.010 | .433 |
| Invasive ventilation | | | | | | |
| Low Cre/Cys C | | | | | | |
| High Cre/Cys C | | | | | | |
| Total | | | | | | |
| Complications acute hypoxemia respiratory failure | | | | | | |
| Low Cre/Cys C | 0.932 to 1.068 | 0.997 | .942 | 0.909 to 1.054 | 0.979 | .569 |
| High Cre/Cys C | 0.950 to 1.044 | 0.996 | .871 | 0.923 to 1.023 | 0.972 | .273 |
| Total | 0.978 to 1.026 | 1.002 | .858 | 0.959 to 1.014 | 0.986 | .327 |
| Acute hypercapnia respiratory failure | | | | | | |
| Low Cre/Cys C | 0.949 to 1.033 | 0.990 | .637 | 0.951 to 1.043 | 0.996 | .866 |
| High Cre/Cys C | 0.976 to 1.037 | 1.006 | .707 | 0.980 to 1.052 | 1.015 | .398 |
| Total | 0.982 to 1.013 | 0.997 | .743 | 0.984 to 1.022 | 1.003 | .757 |
| Clinical outcomes hospital LOS (d) | | | | | | |
| Low Cre/Cys C | −0.127 to 0.061 | −0.048 | .492 | −0.138 to 0.052 | −0.063 | .369 |
| High Cre/Cys C | | | | | | |
| Total | −0.061 to −0.005 | −0.114 | .021 | −0.061 to 0.002 | −0.101 | .068 |
| Admission to the ICU | | | | | | |
| Low Cre/Cys C | 0.877 to 1.034 | 0.952 | .245 | 0.829 to 1.019 | 0.919 | .108 |
| High Cre/Cys C | 0.820 to 1.103 | 0.951 | .508 | 0.819 to 1.160 | 0.975 | .771 |
| Total | 0.907 to 0.996 | 0.951 | .034 | 0.898 to 1.011 | 0.953 | .110 |
| Mortality | | | | | | |
| Low Cre/Cys C | 0.605 to 1.279 | 0.879 | .501 | | | |
| High Cre/Cys C | | | | | | |
| Total | 0.899 to 1.154 | 1.018 | .775 | | | |
## 4. Discussion
The Cre/Cys C has previously been used as a surrogate marker of muscle mass in amyotrophic lateral sclerosis,[7] in patients undergoing physical checkups,[14] in ICU patients,[15] in lung transplant candidates,[16] and in type 2 diabetic patients.[17] To the best of our knowledge, based on the available literature, this is the first study to explore the corporation of the Cre/Cys C ratio, a rapidly measured and widely available biomarker, with risk in patients with AECOPD.
Several previous reports showed that the Cre/Cys C ratio was negatively correlated with age,[8,18–20] sex,[19] BMI,[8,19,20] and waist circumference.[20] In addition, there were positive correlations between the Cr/Cys C ratio and FEV1,[18,21] FVC,[21] and FVC% predicted.[8] Our results also showed that the male ratio, weight, and BMI levels were significantly elevated with a high Cre/Cys C ratio in patients with acute exacerbation of COPD. There was a negative correlation between the Cre/Cys C ratio and age. We consider the reason that it has been reported that muscle mass decreases $1\%$ to $2\%$ annually in patients with COPD over 50 years of age,[22] but $5\%$ to $13\%$ of patients over 65 years of age without chronic diseases such as COPD develop sarcopenia.[23] On the other hand, there was a positive correlation between the Cre/Cys C ratio and weight, BMI, and FVC% predicted. Except for BMI, the findings were similar to those of previous studies. However, a prospective study consecutively recruited patients aged 60 years and older and found that the Cre/Cys C ratio was positively correlated with BMI.[24] Previous reports that the Cre/Cys C ratio is a marker for muscle-adjusted visceral fat mass[20] could explain this finding.
A number of studies have explored the risk factors associated with mortality in patients with AECOPD thus far, and it is known that albumin, respiratory rate, blood gas analysis (PCO2, hemoglobin, lactic acid, etc), inflammation-related indicators, etc, are important prognostic factors for mortality in these patients.[13,25–29] Our study used another easy and inexpensive tool to assess the risk of AECOPD.
The serum Cre/Cys C ratio has been reported as a predictive marker for the adverse effects of chemotherapy in lung cancer. In recent studies, the association of the Cre/Cys C ratio with clinical outcomes (e.g., malnutrition, frailty and hospital stay),[24] hospital admission,[21] 3-year all-cause mortality,[24] cardiovascular disease,[9] risk of pneumonia,[19] and predictors of severe exacerbations 1 has been reported. Our results indicated that the acute hypercapnic respiratory failure ratio and hospital LOS were significantly higher in patients with a low Cre/Cys C ratio than in those with a high Cre/Cys C ratio. The low Cre/Cys C ratio showed a significant association with risk in patients with AECOPD, including increased hospital LOS and admission to the ICU ratio. There was no independent association with AECOPD risk following multivariable adjustments. A low Cre/Cys C ratio in patients with COPD is considered to be associated with a high risk of hospitalization, as systemic inflammation due to acute exacerbations causes muscle weakness and poor health.[30] Our study had some limitations. Previous studies have shown that a higher Cre/Cys C ratio was independently associated with a lower risk of 3-year all-cause mortality after adjusting for potential confounders. The mortality in our study was low and we were unable to determine the correlation between mortality and the Cre/Cys C ratio. Furthermore, it must be replicated in multicentric studies, using a higher number of patients coming from different areas. Finally, this study did not consider the long-term prognosis of the patients, and the prognosis of the patients after discharge could be followed up with in future studies to better assess the long-term risk of patients with AECOPD.
## 5. Conclusion
In conclusion, the present study indicated that the Cre/Cys C ratio is an easy, cheap, repeatable, and promising tool that allows us to evaluate the risk of AECOPD using serum markers. Additionally, a low Cre/Cys C ratio may be a valuable predictor of the risk of AECOPD. A low Cre/Cys C ratio was associated with a prolonged LOS and admission to the ICU in AECOPD patients. However, the associations were not independent.
## Author contributions
Data curation: Liang He, Yan Li, Xijun Gou.
Formal analysis: Liang He.
Writing – original draft: ling lei.
Writing – review & editing: ling lei.
## References
1. López-Campos JL, Soler-Cataluña JJ, Miravitlles M. **Global strategy for the diagnosis, management, and prevention of chronic obstructive lung disease 2019 report: future challenges.**. *Arch Bronconeumol* (2020) **56** 65-7. PMID: 31320191
2. Kahnert K, Alter P, Young D. **The revised GOLD 2017 COPD categorization in relation to comorbidities.**. *Respir Med* (2018) **134** 79-85. PMID: 29413512
3. Vogelmeier CF, Criner GJ, Martinez FJ. **Global strategy for the diagnosis, management, and prevention of chronic obstructive lung disease 2017 report. GOLD executive summary.**. *Am J Respir Crit Care Med* (2017) **195** 557-82. PMID: 28128970
4. Singh D, Agusti A, Anzueto A. **Global strategy for the diagnosis, management, and prevention of chronic obstructive lung disease: the GOLD science committee report 2019.**. *Eur Respir J* (2019) **53** 1900164. PMID: 30846476
5. Martinez FJ, Han MK, Flaherty K. **Role of infection and antimicrobial therapy in acute exacerbations of chronic obstructive pulmonary disease.**. *Expert Rev Anti Infect Ther* (2006) **4** 101-24. PMID: 16441213
6. Lima FV, Yen TY, Patel JK. **Trends in in-hospital outcomes among adults hospitalized with exacerbation of chronic obstructive pulmonary disease.**. *COPD* (2015) **12** 636-42. PMID: 26263035
7. Tetsuka S, Morita M, Ikeguchi K. **Creatinine/cystatin C ratio as a surrogate marker of residual muscle mass in amyotrophic lateral sclerosis.**. *Neurol Clin Neurosci* (2013) **1** 32-7
8. Hirai K, Tanaka A, Homma T. **Serum creatinine/cystatin C ratio as a surrogate marker for sarcopenia in patients with chronic obstructive pulmonary disease.**. *Clin Nutr* (2021) **40** 1274-80. PMID: 32863062
9. Yang Q, Zhang M, Sun P. **Cre/CysC ratio may predict muscle composition and is associated with glucose disposal ability and macrovascular disease in patients with type 2 diabetes.**. *BMJ Open Diabetes Res Care* (2021) **9** e002430
10. Sun J, Yang H, Cai W. **Serum creatinine/cystatin C ratio as a surrogate marker for sarcopenia in patients with gastric cancer.**. *BMC Gastroenterol* (2022) **22** 26. PMID: 35045814
11. Maltais F, Decramer M, Casaburi R. **An official American thoracic society/European respiratory society statement: update on limb muscle dysfunction in chronic obstructive pulmonary disease.**. *Am J Respir Crit Care Med* (2014) **189** e15-62. PMID: 24787074
12. Barreiro E, Bustamante V, Cejudo P. **Guidelines for the evaluation and treatment of muscle dysfunction in patients with chronic obstructive pulmonary disease.**. *Arch Bronconeumol* (2015) **51** 384-95. PMID: 26072153
13. Martinez-Rivera C, Portillo K, Muñoz-Ferrer A. **Anemia is a mortality predictor in hospitalized patients for COPD exacerbation.**. *COPD* (2012) **9** 243-50. PMID: 22360381
14. Kim SW, Jung HW, Kim CH. **A new equation to estimate muscle mass from creatinine and cystatin C.**. *PLoS One* (2016) **11** e0148495. PMID: 26849842
15. Kashani KB, Frazee EN, Kukrálová L. **Evaluating muscle mass by using markers of kidney function: development of the sarcopenia index.**. *Crit Care Med* (2017) **45** e23-9. PMID: 27611976
16. Kashani K, Sarvottam K, Pereira NL. **The sarcopenia index: a novel measure of muscle mass in lung transplant candidates.**. *Clin Transplant* (2018) **32** e13182. PMID: 29274246
17. Osaka T, Hamaguchi M, Hashimoto Y. **Decreased the creatinine to cystatin C ratio is a surrogate marker of sarcopenia in patients with type 2 diabetes.**. *Diabetes Res Clin Pract* (2018) **139** 52-8. PMID: 29496508
18. Nishiki K, Nojiri M, Kato R. **Serum creatinine/cystatin C ratio associated with cross-sectional area of erector spinae muscles and pulmonary function in patients with chronic obstructive pulmonary disease.**. *Int J Chron Obstruct Pulmon Dis* (2021) **16** 3513-24. PMID: 34992359
19. Chen X, Shen Y, Hou L. **Sarcopenia index based on serum creatinine and cystatin C predicts the risk of postoperative complications following hip fracture surgery in older adults.**. *BMC Geriatr* (2021) **21** 541. PMID: 34641805
20. Li S, Lu J, Gu G. **Serum creatinine-to-cystatin C ratio in the progression monitoring of non-alcoholic fatty liver disease.**. *Front Physiol* (2021) **12** 664100. PMID: 33935810
21. Amado CA, García-Unzueta MT, Lavin BA. **The ratio serum creatinine/serum cystatin C (a surrogate marker of muscle mass) as a predictor of hospitalization in chronic obstructive pulmonary disease outpatients.**. *Respiration* (2019) **97** 302-9. PMID: 30481791
22. Santilli V, Bernetti A, Mangone M. **Clinical definition of sarcopenia.**. *Clin Cases Miner Bone Metab* (2014) **11** 177-80. PMID: 25568649
23. Schols AM, Broekhuizen R, Weling-Scheepers CA. **Body composition and mortality in chronic obstructive pulmonary disease.**. *Am J Clin Nutr* (2005) **82** 53-9. PMID: 16002800
24. Tang T, Zhuo Y, Xie L. **Sarcopenia index based on serum creatinine and cystatin C is associated with 3-year mortality in hospitalized older patients.**. *Sci Rep* (2020) **10** 1260. PMID: 31988356
25. Lim WS, van der Eerden MM, Laing R. **Defining community acquired pneumonia severity on presentation to hospital: an international derivation and validation study.**. *Thorax* (2003) **58** 377-82. PMID: 12728155
26. Farrar D, Fairley L, Santorelli G. **Association between hyperglycaemia and adverse perinatal outcomes in south Asian and white British women: analysis of data from the born in Bradford cohort.**. *Lancet Diabetes Endocrinol* (2015) **3** 795-804. PMID: 26355010
27. Matkovic Z, Huerta A, Soler N. **Predictors of adverse outcome in patients hospitalised for exacerbation of chronic obstructive pulmonary disease.**. *Respiration* (2012) **84** 17-26. PMID: 22327370
28. Yu S, Fang Q, Li Y. **Independent factors associated with pneumonia among hospitalized patients with acute exacerbations of chronic obstructive pulmonary disease.**. *Medicine (Baltim)* (2018) **97** e12844
29. Puskarich MA, Illich BM, Jones AE. **Prognosis of emergency department patients with suspected infection and intermediate lactate levels: a systematic review.**. *J Crit Care* (2014) **29** 334-9. PMID: 24559577
30. Vilaró J, Ramirez-Sarmiento A, Martínez-Llorens JM. **Global muscle dysfunction as a risk factor of readmission to hospital due to COPD exacerbations.**. *Respir Med* (2010) **104** 1896-902. PMID: 20541383
|
---
title: 'Financial toxicity of breast cancer over the last 30 years: A bibliometrics
study and visualization analysis via CiteSpace'
authors:
- Hui Cheng
- Lu Lin
- Tingting Liu
- Shaotong Wang
- Yueyue Zhang
- Li Tian
journal: Medicine
year: 2023
pmcid: PMC10036026
doi: 10.1097/MD.0000000000033239
license: CC BY 4.0
---
# Financial toxicity of breast cancer over the last 30 years: A bibliometrics study and visualization analysis via CiteSpace
## Abstract
This literature on financial toxicity (FT) of breast cancer aimed to identify the leading countries, institutions, key researchers, influential references, top journals, research hotspots, and frontiers in the field. Published articles on FT in breast cancer patients were systematically retrieved and screened from the Web of Science databases from inception to March 28, 2022. The CiteSpace software was used to generate knowledge maps to analyze bibliometric characteristics in FT research on breast cancer patients. A total of 615 publications were included, with a year-on-year increase in the number of publications. A total of 591 authors conducted research on the FT in breast cancer patients, with Yabroff KR being the most prolific author. The US was the absolute leader in this field, with almost all major research institutions and authors located in the US. Supportive Care in Cancer was the most productive journal, and the Journal of Clinical Oncology was the most co-cited journal. The keywords representing the research hotspots were “quality of life,” “care,” “cost,” etc. Keywords burst detection indicated that “financial toxicity,” “survivors,” “impact,” “burden,” “income,” and “experience” have become the new research frontiers in the last 5 years. There is an overall upward trend in the research on FT of breast cancer over the last 30 years, which has important and ongoing research value. There is still a paucity of relevant research and more collaboration between authors, institutions, and countries is needed in the future to identify future research directions.
## 1. Introduction
In 2020, the International Agency for Research on Cancer released the latest global cancer update with an estimated 19.3 million new cancer cases worldwide, including an estimated 2.3 million new cases of female breast cancer, accounting for $11.7\%$ of all cancer cases and surpassing lung cancer as the leading cause of global cancer incidence.[1] With the further development of anti-tumor therapy, new anti-cancer drugs, genetic testing, and other treatments have been continuously developed and improved. While improving the therapeutic effect and prolonging the survival of patients, the high cost of medical treatment imposes a heavy financial burden on patients.[2] According to statistics, the medical expenses for cancer in the US increased by approximately $53 billion from 2010 to 2020, an increase of nearly $27\%$.[3] The average expenditure of adult cancer patients is estimated to be $16,346, 4 times that of non-cancer patients.[4] Financial toxicity (FT) is common among women with breast cancer due to the long-term nature of the treatment, the accumulation of treatment-related costs, and the inability to continue working while undergoing treatment and recovery.[5] Although most breast cancer patients have insurance, there are still a majority of patients with FT.[6] A study has found that 12 to $62\%$ of cancer survivors report being in debt because of their treatment, 47 to $49\%$ of survivors report experiencing some form of financial hardship, and 4 to $45\%$ of survivors do not adhere to the recommended prescription drugs due to the cost of treatment.[7] FT, proposed by American scholar S. Yousuf Zafar in 2013, refers to the subjective and objective financial distress of patients caused by cancer treatment and ongoing costs.[8] Currently, FT has been listed as one of the potential adverse reactions in the treatment of malignant tumors.[9] FT is a wide-ranging term that encompasses the costs incurred by patients following a cancer diagnosis, which has an impact on personal and family budgets.[10] There are changes in activities of daily living due to the need for a special diet, escorting, work absenteeism and expenses associated with hiring caregivers, among others. The treatment, whether covered by social security or not, generates costs with tests and/or adjunctive medications, which can financially burden patients and families.[10] A meta-analysis showed that compared with cancer patients with a lower financial burden, those with a higher financial burden were approximately twice as likely to have non-adherence to cancer medication and worse overall physical, mental, emotional, and social functioning and well-being.[11] Because of this financial instability, negative outcomes can be seen on physical, mental, emotional, and economic levels.[12] A systematic review revealed a positive correlation between FT and psychological symptoms.[13] FT can increase the risk of depression,[14] reduce the quality of life of patients, pose a threat to their health and life, and hinder the recovery of their social functions.[15] With an ever-increasing prevalence of breast cancer, which is the most common malignant cancer among women worldwide in 140 of the 184 countries,[16] it is imperative to understand the research status of FT in relation to this disease.
Visualization analysis has become a widely used method for analyzing correlations in massive data in recent years. It uses software to conduct correlation analysis of data and convert the results into a visual atlas, allowing for a more intuitive understanding of relevant information and making it easier to uncover the patterns hidden in big data and to quickly integrate effective information.[17] Based on literature search, current studies and systematic reviews have summarized the evidence for the measures used to quantify FT,[18] the risk factors for FT,[11] and the prevalence and health-related consequences of FT in cancer patients.[19] However, there is a lack of bibliometric and visual studies on the current status of FT research and research hotspots in breast cancer patients. Therefore, the purpose of this study was to analyze the literature on the FT in breast cancer patients using bibliometrics and visualization analysis to understand the current status of research and publication trends, analyze the collaboration among academic organizations, identify current research hotspots, and predict future research trends, to provide a reference for future research.
## 2.1. Search strategies
We searched the Web of Science Core Collection (WoSCC) for articles on FT in breast cancer patients from the inception of the database to March 28, 2022, using the following conditions: breast cancer and FT (search terms), English (language), article and review (type of literature). The relevance of the results was checked. The detailed search strategies are shown in the Appendix.
## 2.2. Inclusion and exclusion criteria
Inclusion criteria: original peer-reviewed articles or reviews.
Exclusion criteria: conference abstracts or errata documents; unpublished articles; duplicate publications; irrelevant articles.
## 2.3. Analysis tool
CiteSpace is a visualization software for bibliometric analysis developed by Professor Chaomei Chen based on the Java platform.[20] *It is* an interactive analysis tool that enables visualization tasks in scientific mapping by combining bibliometrics, visualization analysis, and data mining algorithms.[20] CiteSpace provides a variety of functional options for bibliometric research, including collaborative network analysis, co-citation analysis, and co-occurrence analysis, and can generate visual maps. *By* generating a series of visual knowledge maps, CiteSpace explores the current state of research, hotspots, frontiers, and evolution processes in a certain scientific field, reveals the research directions, research stages, and frontier characteristics of institutions and authors, and finally determines the development trend of this field.
The software version used in this study is 5.8 R3 (64-bit). The parameters of CiteSpace were as follows: time slicing (1994.1–2022.4), years per slice [1], term source (all selection), node type (choose one at a time), selection criteria [50], pruning (pathfinder, pruning sliced Networks), and the others were in default setting. When the default number of CiteSpace network nodes is >350, the centrality calculation function will be disabled. The “Computing Node centrality” function in the node menu needs to be manually enabled, and the nodes and links are used to generate visual knowledge maps.
## 2.4. Bibliometrics and visualization analysis
The database used in this study was Web of Science, and the articles were retrieved from the WoSCC, including SCI-Expanded, SSCI, CCR-Expanded, and Index Chemicus. The Web of Science database contains a large number of multidisciplinary, high-impact, international, and comprehensive academic journals, which provide a better knowledge map when using CiteSpace for visualization analysis.[21,22] As the database is updated daily, the literature included in WoSCC was retrieved within 1 day on March 28, 2022, to avoid bias. The document data were saved in the form of complete records, the cited references were saved in plain text format. The data were then imported into CiteSpace for data format conversion, deduplication, and visualization before the results were presented bibliometrically and visually.[23]
## 3.1. Quantity and trend analysis of published papers
A total of 615 articles were included in this study, and the actual number of articles published each year was calculated using the bibliometrics online analysis platform (http://bibliometric.com/). In the early stage (1994–2010), the number of papers per year was <10, except in 2004, 2008, and 2009 (Fig. 1A). In 2021, the number of papers published on the topic peaked, showing that the FT in breast cancer has become a research hotspot and attracted the attention of researchers worldwide.
**Figure 1.:** *(A) Annual trend of publications. (B) Number of annual publications and growth trends of the top 10 countries/regions.*
In order to identify countries/regions leading research in this area, further analysis of the literature from different countries/regions was conducted. The bar chart (Fig. 1B) shows the top 10 countries/regions in the total number of published articles in the past 30 years. The results show that the number of publications is still growing steadily, and the US is identified as the founder of this field. It has also been found that annual publications in China are growing rapidly.
## 3.2. Analysis of international/interregional and interinstitutional cooperation
To identify interinstitutional collaborations to conduct such studies, international/interregional and interinstitutional analyses were performed using CiteSpace. The size of concentric circles indicates the number of articles published, and institutions that publish more articles tend to have larger concentric circles. The connection between the 2 circles means they published together, and the thickness of the lines indicates the strength of their collaboration. The results show that 86 countries have already established partnerships, with 286 links to each other. Centrality is also known as intermediate centrality, and nodes with high centrality (>0.1) are generally regarded as turning points or key points in this field.[24] The US had the highest centrality (0.64) and the best partnerships in this area, followed by the UK (0.22) and Australia (0.1). However, China had less centrality (0.01) and less international cooperation than the US (Fig. 2A).
**Figure 2.:** *(A) The network of co-country (with a threshold of 10 publications). (B) The network of co-institution (with a threshold of 10 publications).*
The interinstitutional cooperation network diagram (Fig. 2B) shows 447 nodes and 818 links. Institutions located in the US accounted for the vast majority of the total. The top 5 institutions, all in the US, were the National Cancer Center (25 articles), Harvard Medical School (16 articles), the University of Michigan (16 articles), the Dana-Farber Cancer Institute (15 articles), and Duke University (14 articles).
## 3.3. Author and author co-citation analysis
The visualization atlas of 591 nodes and 873 links formed by the combined coauthor network is shown in Figure 3A. The co-authorship network shows the prolific authors and their cooperation. The most productive author was Yabroff KR with 19 articles, followed by Sharp L (11 articles) and Guy GP (10 articles), with no Chinese authors in the top 10. Although many of the authors have published relevant articles, there is little collaboration between them. In addition, the relatively low centrality of the authors indicates that more high-quality and large-scale collaborations are needed in the future (Table 1).
A co-citation relationship between authors is established when 2 (or more) authors are simultaneously cited in 1 or more subsequent papers. An analysis of an author’s co-citation network provides a clear picture of the core authors and their contributions to a field, the strength of which indicates the level of author involvement. The most co-cited authors were Yabroff KR (115 citations), followed by Zafar SY (88 citations), Mariotto AB (63 citations), Bradley CJ (54 citations), and Jagsi R (53 citations). In addition, Brown ML (0.17), Baker MS (0.13), and Yabroff KR (0.13) were the authors of high centrality (Fig. 3B).
## 3.4. Journal and journal co-cited analysis
Journal analysis and co-cited journal analysis can provide important information to understand journals with a high volume of publications. It is not only beneficial for researchers to timely discover the latest research trends in the field and be selective when submitting manuscripts, thus improving the publication rate, but also conducive to finding suitable collaborating institutions and promoting domestic and international academic exchanges,[25] thus improving the overall research level of FT of breast cancer. The WoSCC search revealed that the 615 articles included in the current analysis were published in 298 different journals, and the 10 journals with the most publications were identified through bibliometrics online analysis methods (Table 2).
**Table 2**
| Journal | IF (JCR 2021) | Quartile in category (JCR) | Article counts | Percentage (%) |
| --- | --- | --- | --- | --- |
| SUPPORTIVE CARE IN CANCER | 3.603 | Q2 | 34 | 5.528 |
| BREAST CANCER RESEARCH AND TREATMENT | 4.872 | Q2 | 22 | 3.577 |
| CANCER | 6.86 | Q1 | 22 | 3.577 |
| JOURNAL OF CLINICAL ONCOLOGY | 44.544 | Q1 | 13 | 2.114 |
| PLOS ONE | 3.24 | Q2 | 12 | 1.951 |
| PSYCHO-ONCOLOGY | 3.894 | Q2 | 12 | 1.951 |
| JNCI JOURNAL OF THE NATIONAL CANCER INSTITUTE | 11.577 | Q1 | 11 | 1.789 |
| BMC CANCER | 4.43 | Q3 | 10 | 1.626 |
| BMC HEALTH SERVICES RESEARCH | 2.655 | Q3 | 9 | 1.463 |
| ONCOLOGIST | 5.462 | Q2 | 9 | 1.463 |
The analysis of the cited journals showed that the most cited journals were Journal of Clinical Oncology (393 citations), Cancer (306 citations), JNCI Journal of the National Cancer Institute (244 citations), New England Journal of Medicine (237 citations), and Breast Cancer Research and Treatment (224 citations) (Fig. 4A).
**Figure 4.:** *(A) The network of journal co-citation (with a threshold of 120 publications). (B) The network of co-cited references (with a threshold of 10 publications).*
## 3.5. References and co-cited references analysis
Figure 4B shows the top references with a high frequency of co-citations. The most co-cited reference published by Ramsey SD[7] in 2016 elaborated on the serious financial distress faced by cancer patients after diagnosis, which may be related to cancer mortality. The second was a systematic review of the financial difficulties of cancer patients published by Altice CK[26] in 2016. The top 10 cited references out of the 615 included articles were listed in Table 3.
**Table 3**
| Title | Author | Journal | Year | Citation |
| --- | --- | --- | --- | --- |
| Family caregiver burden: results of a longitudinal study of breast cancer patients and their principal caregivers | Grunfeld, E | Canadian Medical Association Journal | 2004 | 584 |
| The financial toxicity of cancer treatment: a pilot study assessing out-of-pocket expenses and the insured cancer patient’s experience | Zafar, S. | Oncologist | 2013 | 536 |
| Cancer-related fatigue: the scale of the problem | Hofman, Maarten | Oncologist | 2007 | 523 |
| Economic burden of cancer across the European Union: a population-based cost analysis | Luengo-Fernandez, Ramon | LANCET Oncology | 2013 | 513 |
| Cost of care for elderly cancer patients in the United States | Yabroff, KR | JNCI Journal of The National Cancer Institute | 2008 | 493 |
| Burden of illness in cancer survivors: findings from a population-based national sample | Yabroff, KR | Jnci-Journal of The National Cancer Institute | 2004 | 417 |
| Incidence, treatment costs, and complications of lymphedema after breast cancer among women of working age: a 2-year follow-up study | Shih, Ya-Chen Tina | Journal of Clinical Oncology | 2009 | 299 |
| Economic burden of cancer in the United States: estimates, projections, and future research | Yabroff, KR | Cancer Epidemiology Biomarkers & Prevention | 2011 | 289 |
| Financial hardships experienced by cancer survivors: a systematic review | Altice, Cheryl K | JNCI-Journal of The National Cancer Institute | 2017 | 284 |
| Financial hardship associated with cancer in the United States: findings from a population-based sample of adult cancer survivors | Yabroff, KR | Journal of Clinical Oncology | 2016 | 235 |
## 3.6. Analysis of keywords
Keywords are a summary of the research topic of an article, and the main research directions of a certain discipline can be understood by analyzing the frequency of keywords.[27] A total of 573 keyword nodes and 2330 keyword lines were obtained, with a density of 0.0142. Word frequency analysis is a measurement method to analyze the research hotspots by counting the frequency of keywords or subject terms in the literature.[28] Excluding the search terms, the top 5 keywords in frequency in this study were “care” [104] “quality of life” [90] “cost” [80] “women” [78], and “impact” [62], as shown in Table 4. Centrality analysis is a measure of the importance of nodes in the network. The greater its value, the higher the representation and attention of nodes in the network.[24] Among the high-frequency keywords, “quality of life” (0.14) had the highest centrality, indicating the large number and influence of related studies conducted with this keyword. Keywords cluster analysis was performed on the retrieved literature on FT in breast cancer patients to explore research hotspots in this field. The calculated Q value (Q) and silhouette value (S) were indicators representing the modularity and homogeneity of the cluster network, respectively. A Q > 0.3 identified the cluster structure as significant, and an S > 0.5 or > 0.7 indicated that the clustering result was reasonable or highly credible, respectively.[29] In the cluster map of this study, $Q = 0.4598$ and $S = 0.7483$, indicating significant cluster structure and reasonable results. The cluster labels and main keywords are shown in Figure 5A.
The timeline distribution of keywords not only reveals the specific keywords included in each cluster but also shows the beginning and ending time nodes of each cluster theme, thus helping to sort out the development path of FT research in breast cancer patients (Fig. 5B).
Keywords burst detection are used to determine the development trend of the research field based on the trend of word frequency changes of the subject terms. Burst detection reveals the sudden changes in terms or references within a specific period of time. Words with high burst intensity represent frontier issues of research in the corresponding era, thus identifying emerging research trends.[30] “Financial toxicity,” “survivor,” “impact,” “burden,” “outcome,” and “experience” are the newly emerging terms in the last 5 years (Fig. 5C).
## 4. Discussions
The number of articles published is an important indicator of scientific research activities, which can reflect, to a certain extent, the development of the discipline and the attention of society.[31] This study used CiteSpace as a quantitative assessment tool to examine the literature on FT in breast cancer patients from inception to 2022. Analysis of the annual volume of publications in this field revealed an overall upward trend in the global literature on FT of breast cancer in recent years. As many as 191 articles ($31\%$) were published from 2020 to 2021, indicating the rapid development of research on FT of breast cancer and the increasing attention from clinical researchers.
Many countries/regions in the world have conducted research on FT of breast cancer, with the US in the leading position. Part of the supporting evidence shows that the top 5 publishers were all in the US, and 9 of the top 10 most productive authors were from the US. Moreover, articles published in the US had the highest average citation rates, which further demonstrates their strong academic influence. The reason may be that the concept of “financial toxicity” was first proposed by the American scholar Yousuf Zafar,[32] and the American Society of Clinical Oncology established the Cancer Treatment Cost Panel in 2009, which proposed the intervention guidelines for FT of cancer,[33] resulting in more mature research and higher quality of literature in this area in the US. Secondly, the FT prevalence survey and the construction of screening tools in the US are gradually taking shape. In 2015, the American Society of Clinical Oncology proposed a scoring system for comprehensive evaluation of anti-tumor treatment options according to the net health benefit and drug price of tumor treatment options, to make the limited medical resources more reasonably allocated by accurately measuring the value of new treatment, providing a reference for FT research in other countries.[34] In addition, in the US, the cost of cancer treatment drugs and corresponding adjuvant therapies has been rising year by year, and many drugs are self-paying. A study of breast cancer patients showed that $17\%$ of patients had out-of-pocket costs of >$5000.[35] At the same time, the consumption concept of the American people, the low personal asset reserves, and reduced income due to limited ability to work as a result of the disease further increase the FT of cancer, which makes the US pay more and more attention to research on FT.[10] Impact factor and journal partition are important indicators to evaluate the quality of journals, and the journal partition can weaken the structural imbalance factors between disciplines and facilitate the comparison and evaluation of journals in different disciplines.[36] *In this* study, 7 of the top 10 journals in terms of volume of articles were in Q1 or Q2, indicating the overall high quality of the current literature. Published and frequently cited papers also have great academic influence. Four of the top 10 highly cited articles were authored by Yabroff KR, the most productive author and a key figure in this field. A randomized controlled trial article published in 2004 referred to the higher disease burden in cancer survivors than in individuals without cancer and the need for increased attention to this population.[37] A questionnaire survey in 2016 determined that working-age cancer survivors usually experienced material, psychological, and financial difficulties, reminding healthcare professionals of the psychological state of these patients.[38] An article published in JNCI Journal of the National Cancer Institute in 2008 estimated the cost of care for cancer patients and informed the development of national cancer plans and policies.[39] Another article cited 299 times proposed a measure of the economic burden of cancer, estimated and predicted the cancer burden in the US at that time, identified key areas of future work, and informed healthcare policymakers, healthcare systems, and employers to improve cancer survivor experience in the US.[40] The keyword analysis of the literature revealed that the high-frequency keywords for FT of breast cancer mainly focused on the impact of financial burden, patient care, quality of life, and treatment, and the research hotspots were traceable. Cluster analysis of the keywords found that the research focus was mainly on the cost of breast cancer treatment, the effect of drug treatment, the side effects of treatment, and theoretical research. One study showed that $25\%$ of women with breast cancer experienced FT, and $12\%$ of patients with early-stage breast cancer still had medical debt 4 years after diagnosis.[41] FT not only affects the mental health and quality of life of patients but even increases their mortality in severe cases.[42] In addition, since breast cancer patients usually undergo a series of adjuvant treatments, it can cause a series of treatment side effects.[13–15] For some specific symptoms, enhancing care and preventing and reducing complications can also reduce the harm of FT. Analysis of the timeline view and burst keywords indicated that the demographic research on FT in breast cancer patients and its influencing factors were highly emergent, making it a frontier term in this field. Studies have shown that female, younger age, lower baseline income, adjuvant therapy, and recent disease diagnosis were the most frequently reported factors associated with FT.[43] According to bibliometric analysis, the experience, burden, and work management of breast cancer survivors have also received increasing attention in the last 5 years. The research has shown that the FT in breast cancer patients is a multidisciplinary and multi-faceted social hotpot that requires the cooperation of the whole society and the collaboration of multidisciplinary researchers to apply the research results to clinical practice, so as to benefit more patients.[44]
## 5. Limitations
This study used CiteSpace to analyze and present the research trends in the field of FT of breast cancer over the past 30 years by taking the literature retrieved from the WoSCC as the object of study, and the results have a reference value. However, some limitations still exist in this study. First, although the WoSCC is the most commonly used database in scientometrics research, our data were only obtained from WoSCC without searching other databases such as Embase or PubMed. Secondly, our data analysis used computer tools instead of manual selection, and the analysis of the content was presented in keyword fields with possible omissions and biases in the detection of details. In the future, it will be necessary to thoroughly outline the study designs of the included literature and conduct a comprehensive analysis of the full text. In addition, considering the large time span of the study and the lack of standardization process for keywords in the WoSCC, keyword omissions may have occurred.
## 6. Conclusions
This study conducted a bibliometric analysis of the literatures on FT in breast cancer patients over the past 30 years and revealed an overall upward trend in research in this field. The US is the absolute leader in this field, and the UK and Australia have also proven to be major research forces in this field with high publication rates and centrality. Strong collaborations found among many developed countries and prestigious institutions suggest that FT of breast cancer is attracting increasing attention. Current research hotspots in this field are the influencing factors of FT, post-treatment side effects, and quality of life. *In* general, there is a paucity of studies on FT of breast cancer, with a narrow scope and limited collaboration among authors. In the future, more collaboration among authors, institutions, and countries is needed to promote the further development of research in this field.
## Author contributions
Conceptualization: Hui Cheng, Lu Lin, Li Tian.
Data curation: Hui Cheng.
Formal analysis: Hui Cheng, Tingting Liu.
Funding acquisition: Li Tian.
Methodology: Lu Lin, Tingting Liu, Shaotong Wang, Yueyue Zhang, Li Tian.
Resources: Hui Cheng, Yueyue Zhang.
Software: Hui Cheng, Shaotong Wang.
Supervision: Li Tian.
Writing – original draft: Hui Cheng, Lu Lin, Li Tian.
Writing – review & editing: Lu Lin, Li Tian.
## References
1. Sung H, Ferlay J, Siegel RL. **Global cancer statistics 2020: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries.**. *CA Cancer J Clin* (2021.0) **71** 209-49. PMID: 33538338
2. Jiayu YAO, Xia WU. **Financial toxicity: an overlooked problem in the era of molecular and immune therapy of cancer treatment.**. *Chinese Clin Oncol* (2019.0) **24** 183-7
3. Cheng KKF, Wong WH, Koh C. **Unmet needs mediate the relationship between symptoms and quality of life in breast cancer survivors.**. *Support Care Cancer* (2016.0) **24** 2025-33. PMID: 26530229
4. So WK, Chow KM, Chan HY. **Quality of life and most prevalent unmet needs of Chinese breast cancer survivors at one year after cancer treatment.**. *Eur J Oncol Nurs* (2014.0) **18** 323-8. PMID: 24703094
5. Jing J, Gao J, Dong L. **Progress of research on financial toxicity of breast cancer.**. *China Med Herald* (2019.0) **16** 30-3
6. Rosenzweig M, West M, Matthews J. **Financial toxicity among women with metastatic breast cancer.**. *Oncol Nurs Forum* (2019.0) **46** 83-91. PMID: 30547962
7. Ramsey SD, Bansal A, Fedorenko CR. **Financial insolvency as a risk factor for early mortality among patients with cancer.**. *J Clin Oncol* (2016.0) **34** 980-6. PMID: 26811521
8. **The national cancer survivorship initiative vision.**
9. McDermott C. **Financial toxicity: a common but rarely discussed treatment side effect.**. *Ann Am Thorac Soc* (2017.0) **14** 1750-2. PMID: 28957637
10. Nogueira LA, Lenhani BE, Tomim DH. **Financial toxicity.**. *Asian Pac J Cancer Prev* (2020.0) **21** 289-93. PMID: 32102501
11. Smith GL, Lopez-Olivo MA, Advani PG. **Financial burdens of cancer treatment: a systematic review of risk factors and outcomes.**. *J Natl Compr Canc Netw* (2019.0) **17** 1184-92. PMID: 31590147
12. **The costs of cancer care 2020.**. (2020.0)
13. Chan RJ, Gordon LG, Tan CJ. **Relationships between financial toxicity and symptom burden in cancer survivors: a systematic review.**. *J Pain Symptom Manage* (2019.0) **57** 646-660.e1. PMID: 30550833
14. Sharp L, Carsin AE, Timmons A. **Associations between cancer-related financial stress and strain and psychological well-being among individuals living with cancer.**. *Psychooncology* (2013.0) **22** 745-55. PMID: 22411485
15. Collado L, Brownell I. **The crippling financial toxicity of cancer in the United States.**. *Cancer Biol Ther* (2019.0) **20** 1301-3. PMID: 31291813
16. Verhoeven D, Allemani C, Kaufman C. **Breast Cancer: global quality care optimizing care delivery with existing financial and personnel resources.**. *ESMO Open* (2020.0) **4** e000861. PMID: 32895233
17. Gu Y, Rong L, Yu X. **A visualized analysis of research articles on senile dementia nursing in China and foreign countries in 2012-2017.**. *J Chongqing Med Univ* (2019.0) **44** 456-60
18. Xu B, Hu L, Cheng Q. **A systematic review of financial toxicity among cancer patients in China.**. *Asia Pac J Oncol Nurs* (2022.0) **9** 100071. PMID: 35692729
19. Zhu Z, Xing W, Zhang X. **Cancer survivors’ experiences with financial toxicity: a systematic review and meta-synthesis of qualitative studies.**. *Psycho Oncol* (2020.0) **29** 945-59
20. Chen C. **Science mapping: a systematic review of the literature.**. *J Data Inf Sci* (2017.0) **2** 1-40
21. Falagas ME, Pitsouni EI, Malietzis GA. **Comparison of PubMed, Scopus, Web of Science, and Google Scholar: strengths and weaknesses.**. *FASEB J* (2008.0) **22** 338-42. PMID: 17884971
22. Martín-Martín A, Thelwall M, Orduna-Malea E. **Google Scholar, Microsoft Academic, Scopus, Dimensions, Web of Science, and OpenCitations’ COCI: a multidisciplinary comparison of coverage via citations.**. *Scientometrics* (2021.0) **126** 871-906. PMID: 32981987
23. Donthu N, Kumar S, Mukherjee D. **How to conduct a bibliometric analysis: an overview and guidelines.**. *J Bus Res* (2021.0) **133** 285-96
24. Chen C. **Searching for intellectual turning points: progressive knowledge domain visualization.**. *Proc Natl Acad Sci USA* (2004.0) **101** 5303-10. PMID: 14724295
25. Yang D, Liu ZY, Gao LL. **Bibliometric analysis of study on diabetic foot based on Web of Science database.**. *Chin Nurs Res* (2019.0) **33** 643-8
26. Altice CK, Banegas MP, Tucker-Seeley RD. **Financial hardships experienced by cancer survivors: a systematic review.**. *J Natl Cancer Inst* (2016.0) **109** djw205. PMID: 27754926
27. Bu ZH, Huang AL, Xue MT. **Visual analysis of current status and hot spots of nursing cost based on CiteSpace software in China.**. *Chin J Mod Nurs* (2020.0) **26** 4908-13
28. Xiao R-rong, Liu Y-li. **Features analysis of digital library high-impact papers.**. *Inf Sci* (2014.0) **32** 44-9
29. Chen Y, Chen CM, Liu ZY. **The methodology function of CiteSpace mapping knowledge domains.**. *Stud Sci Sci* (2015.0) **33** 242-53
30. Sun CY, Lin Z, Gu FC. **Research status and hotspots of positive psychology based on Web of Science database.**. *Chin J Mod Nurs* (2020.0) **26** 3765-71
31. Ye R, Xu GH. **Application status of bibliometrics in nursing in China.**. *J Qilu Nurs* (2011.0) **17** 38-40
32. Zafar SY, Abernethy AP. **Financial toxicity, Part I: a new name for a growing problem.**. *Oncology (Williston Park)* (2013.0) **27** 80-1, 149. PMID: 23530397
33. Meropol NJ, Schrag D, Smith TJ. **American Society of Clinical Oncology guidance statement: the cost of cancer care.**. *J Clin Oncol* (2009.0) **27** 3868-74. PMID: 19581533
34. Schnipper LE, Davidson NE, Wollins DS. **American Society of Clinical Oncology Statement: a conceptual framework to assess the value of cancer treatment options.**. *J Clin Oncol* (2015.0) **33** 2563-77. PMID: 26101248
35. Finkelstein EA, Tangka FK, Trogdon JG. **The personal financial burden of cancer for the working-aged population.**. *Am J Manag Care* (2009.0) **15** 801-6. PMID: 19895184
36. Lu YM, Ju JW. **Studies on the academic quality indexes and evaluation model of SCI papers.**. *Sci Technol Manag Res* (2012.0) **32** 71-4
37. Yabroff KR, Lawrence WF, Clauser S. **Burden of illness in cancer survivors: findings from a population-based national sample.**. *J Natl Cancer Inst* (2004.0) **96** 1322-30. PMID: 15339970
38. Yabroff KR, Dowling EC, Guy GP. **Financial hardship associated with cancer in the United States: findings from a population-based sample of adult cancer survivors.**. *J Clin Oncol* (2016.0) **34** 259-67. PMID: 26644532
39. Yabroff KR, Lamont EB, Mariotto A. **Cost of care for elderly cancer patients in the United States.**. *J Natl Cancer Inst* (2008.0) **100** 630-41. PMID: 18445825
40. Yabroff KR, Lund J, Kepka D. **Economic burden of cancer in the United States: estimates, projections, and future research.**. *Cancer Epidemiol Biomarkers Prev* (2011.0) **20** 2006-14. PMID: 21980008
41. Jagsi R, Pottow JA, Griffith KA. **Long-term financial burden of breast cancer: experiences of a diverse cohort of survivors identified through population-based registries.**. *J Clin Oncol* (2014.0) **32** 1269-76. PMID: 24663041
42. Kale HP, Carroll NV. **Self-reported financial burden of cancer care and its effect on physical and mental health-related quality of life among US cancer survivors.**. *Cancer* (2016.0) **122** 283-9. PMID: 26991528
43. Gordon LG, Merollini KMD, Lowe A. **A systematic review of financial toxicity among cancer survivors: we can’t pay the co-pay.**. *Patient* (2017.0) **10** 295-309. PMID: 27798816
44. Boulanger M, Mitchell C, Zhong J. **Financial toxicity in lung cancer.**. *Front Oncol* (2022.0) **12** 1004102. PMID: 36338686
|
---
title: 'Efficacy of lutein supplements on macular pigment optical density in highly
myopic individuals: A randomized controlled trial'
authors:
- Takeshi Yoshida
- Yasutaka Takagi
- Tae Igarashi-Yokoi
- Kyoko Ohno-Matsui
journal: Medicine
year: 2023
pmcid: PMC10036027
doi: 10.1097/MD.0000000000033280
license: CC BY 4.0
---
# Efficacy of lutein supplements on macular pigment optical density in highly myopic individuals: A randomized controlled trial
## Introduction:
Lutein supplementation is beneficial in preventing maculae from developing serious ocular diseases. This study aimed to evaluate the efficacy and safety of lutein administration in patients with high myopia (HM).
### Methods:
In a single-center randomized double-blinded placebo-controlled trial conducted over 24 months, 22 eyes were enrolled in lutein and control groups. Among them, 15 eyes in the lutein group and 13 eyes in the control group completed the study. All patients with HM (axial length > 26.00) were administered lutein (20 mg) or placebo once daily for 6 months. The macular pigment optical density (MPOD), rate of change in MPOD, visual acuity, contrast sensitivity, and electroretinogram after administration were examined at baseline, 3 months, and 6 months.
### Results:
The baseline MPOD in the control and lutein groups was 0.71 ± 0.21 and 0.70 ± 0.22, respectively. The MPOD in the control and lutein groups at 3 months was 0.70 ± 0.21 and 0.70 ± 0.25, respectively, and at 6 months was 0.66 ± 0.20 and 0.72 ± 0.27, respectively, which was not significantly different from those at baseline or between the groups. The MPOD significantly increased from baseline in the lutein group with less than 28.25 mm of axial length at 6 months (from 0.71 ± 0.20 to 0.78 ± 0.22, $$P \leq .02$$, t test). visual acuity, contrast sensitivity, and electroretinogram values were similar between the groups.
### Conclusion:
Lutein supplementation showed significant benefits in MPOD augmentation in patients with HM.
## 1. Introduction
The macula in the retina is a yellow-pigmented area at the posterior pole of the eye that allows central vision and provides the most acute visual acuity (VA) and best color identification.[1] Macular yellow pigments primarily consist of lutein and its structural isomer zeaxanthin.[2] *Lutein is* one of the few xanthophyll carotenoids found in high concentrations in the macula. Since de novo synthesis of lutein within the human body is impossible, it can only be obtained from the diet.[3] Macular pigments (MP), including lutein, are concentrated in the photoreceptor axons of the Henle nerve fiber layer and rod outer segments[4] where they easily undergo oxidative attack. Lutein is a potent antioxidant. Several basic and clinical studies have reported the antioxidative and anti-inflammatory properties of lutein in the eye, suggesting that lutein plays a major role in protecting the retina and retinal pigment epithelium from light-initiated oxidative damage by scavenging reactive oxygen species and filtering blue light. Thus, lutein is involved in the putative pathogenesis of many age-related eye diseases such as age-related macular degeneration (AMD),[5] retinitis pigmentosa,[6] and diabetic retinopathy.[7] Furthermore, lutein plays a key role in maintaining macular morphology and function.[8,9] Several studies have suggested that dietary supplementation with lutein prevents eye diseases, such as AMD, and improves visual function.[10–13] Feng et al[1] demonstrated that supplementation with 20-mg lutein increases macular pigment optical density (MPOD). Furthermore, lutein and zeaxanthin supplementation can improve contrast sensitivity (CS),[10] and low MP levels are associated with poor visual function in both healthy eyes and eyes with early AMD.[11] Therefore, it is used as a key ingredient in ocular supplements.
Myopia is a serious global health concern, with an increasing prevalence among teenagers and young adults of approximately $90\%$ in East Asia and $50\%$ in the United States and Europe.[12] *High myopia* (HM), characterized by progressive elongation of the eyeball (defined as an axial length [AL] > 26.0 mm), is associated with serious ocular complications such as retinal detachment, macular schisis, macular holes,[13] chorioretinal atrophy, and choroidal neovascularization.[14,15] Excessive elongation of the eyeball in HM results in consequential thinning of the retina, including the macula, which induces HM-related complications, for which there are no effective treatment options. In the AL > 26 mm group, a significant inverse correlation was reported between MPOD and AL.[16] in vivo measurements of retinal thickness using optical coherence tomography revealed an inverse relationship between AL and macular thickness.[17–20] *This is* the reason for the low MPOD in HM because MP, including lutein, is concentrated in the photoreceptor axons of the Henle nerve fiber layer and rod outer segments.[4] However, the advantages of high MP and the efficacy of lutein supplementation in patients with HM remain uncertain, although MP decreases in eyes with HM.
Most ocular complications of HM are irreversible and it is difficult to recover the VA and visual field; hence, therapeutic options are required to prevent ocular complications. Lutein supplementation improves visual function in patients with AMD,[10] and studies have explored whether dietary supplementation with lutein might prevent AMD or improve the condition of patients with AMD.[14,21,22] The efficacy of lutein supplementation in patients with HM remains unclear, and to our knowledge, no studies have been conducted in these patients with HM. Therefore, we aimed to evaluate the efficacy and safety of lutein supplementation in HM patients.
## 2.1. Design overview
This single-center, randomized, double-blind, placebo-controlled study was conducted between September 2017 and March 2021. We used placebo as a control to evaluate the efficacy and safety of lutein in patients with HM. The main components of the placebo were safflower oil and a stirring agent that had no drug effect. In total, 44 patients were enrolled in the trial in either the intervention or control group (ratio 1:1). All patients were administered the investigational drug or placebo orally, once daily for 6 months, and were interviewed during follow-up visits 3 and 6 months after the initiation of drug administration to complete the efficacy and safety assessments.
The study design was approved by the ethics committee of Tokyo Medical and Dental University. All patients were informed of the possible benefits and risks of the study, and provided signed informed consent. This study was performed in accordance with the ethical standards of the 1964 Declaration of Helsinki and its later amendments, and the Clinical Trials Act in Japan. This study was registered as a clinical trial in the Japan Registry of Clinical Trials (jRCT; ID: jRCTs031180168).
## 2.2. Setting and participants
Patients were recruited from the Tokyo Medical and Dental University Hospital. The inclusion criteria were as follows: Japanese origin, AL of eyeball 26.5 mm or more, and less than 30.0 mm; age 20 to 50 years, and patient consent. The exclusion criteria were as follows: comorbid severe primary diseases, such as cardiovascular disease, cerebrovascular disease, and diseases of the liver, kidneys, or hematopoietic system; pupil diameter less than 6.5 mm in mydriasis; vision-affected ocular diseases, such as glaucoma, AMD, and diseases of the retina; administration of lutein within the last 6 months; allergy to multiple drugs or known allergy to lutein capsule components; and other conditions considered unsuitable for participation after discussion.
## 2.3. Randomization and masking
Stratified random numbers were generated by statisticians using the SAS version 9.3 software (SAS Institute Inc., Cary, NC), which denoted the drug number as a running serial number for the sample size. The investigator administered the corresponding drugs to participants according to the sequence in which they were enrolled.
Unified manufacturing of the study drugs (lutein and placebo) was conducted by Santen Pharmaceutical Co. Ltd. Patients in the lutein group were orally administered lutein, 1 capsule, once daily (20 mg/capsule), whereas those in the placebo group were orally administered placebo identical to that in the lutein group daily. The treatment duration was 6 months. There were 2 levels of blinding: first, blinding of each drug to its corresponding group; and second, blinding of the drug was conducted by the blinding staff after the blinding status examination and completion of data locking. The second round of unblinding was performed after the completion of the statistical analysis and unmasking of the placebo and experimental groups. All participants, investigators, and staff members who performed the analyses were blinded to the study protocols.
## 2.4. Outcomes and measurements
The primary efficacy marker was the change in MPOD after 6 months of administration relative to that at baseline. To improve data accuracy, the MPOD was evaluated at least 3 times. The secondary efficacy markers were change in VA, change in CS, and change in electroretinogram (ERG) measurements (all relative to baseline values). The safety marker was the presence or absence of adverse events.
## 2.5. Measurement of macular pigment optical density
The MPOD measurements were performed using a standardized protocol based on the psychophysical method of heterochromatic flicker photometry. This protocol had high test–retest reliability ($R = 0.9$), and the participant responses at the 2 wavelengths were consistent with the absorption spectrum of lutein.[15,17,18,23] Briefly, participants were fitted with trial frames and appropriate lenses for testing after refraction. The optimal flicker frequency for heterochromatic flicker photometry was determined for each participant. Measurements were made with the eye at the foveal center, and the participant made 4 separate determinations. The participants viewed a small test field superimposed on a blue background with their right eye. The test field alternated between a wavelength (blue or blue-green) that was absorbed by the MP and a reference wavelength (green to yellow-green) outside the absorption band of the MP. The test field appeared to be flicker when the frequency of alternation was chosen correctly. While performing the measurements, the participants were instructed to adjust the energy of the bluish test light to stop flickering. The amount of bluish light required to nullify the flicker provides a measure of the absorption of MP (i.e., MPOD) at the retinal location of the test light. The participants were instructed to fixate on the center of the target.
## 2.6. Visual acuity
The VA of the patients was measured with the best refractive correction using the Nidek system Chart SC-1600 (Nidek Co., Aichi, Japan). The best-corrected visual acuity (BCVA) was compared before and after therapy. VA was converted to the logarithm of the minimal angle resolution and analyzed.
## 2.7. Contrast sensitivity
The CS was measured using a CSV-1000E contrast testing instrument (Vector Vision, Greenville, OH) at a distance of 2.4 m under standard brightness (85 cd/m2). This test consisted of the following 4 spatial frequencies: 3, 6, 12, and 18 cycles per degree (cpd). If the eye showed a refractive error, refractive correction was performed during the CS test.
## 2.8. Electroretinogram
Full-field ERGs were obtained. The ERG was performed according to the standards of the International Society for Clinical Electrophysiology of Vision.[19] The ERG was elicited using a light stimulator and recorded using a contact lens electrode (LE4000; Tomey, Japan). Possible correlations between the amplitudes of the a- and b-waves of the dark-adapted ERGs elicited by 200 cds/m2 and the MPOD were analyzed.
## 2.9. Datasets
The per-protocol set (PPS) was defined a priori in the protocol; therefore, analyses were performed on the PPS for lutein efficacy. For PPS, we excluded patients with poor compliance to the study protocol and those who refused to administer lutein tablets after baseline. In addition, the MPOD value was investigated using axial length-stratified analysis.
## 2.10. General analytical principles
Statistical analyses were performed using the SAS 9.3 software. The statistical description of quantitative data included the number of patients, mean, standard deviation, median, maximum, minimum, test statistic, and P value. The statistical description of the qualitative data included the frequency distribution, composition ratio, test statistics, and P value. Suitable statistical analysis methods were used for the intergroup and intragroup comparisons. All statistical tests were performed to obtain the test statistics and corresponding P values. A paired t test was used to directly obtain P values. All statistical tests were 2 tailed. Statistical significance was set at $P \leq .05.$
## 3.1. Patient characteristics
Of the initially enrolled 44 patients with HM ($$n = 44$$ eyes), with 22 patients in the lutein and control groups, 5 refused to continue the study, and 11 were excluded. Thus, 28 patients ($$n = 28$$ eyes) completed the study as PPS, with 15 ($$n = 15$$ eyes) in the lutein group and 13 ($$n = 13$$ eyes) in the control group. The participant distribution flowchart is shown in Figure 1. The mean ages in the lutein and control groups were 46.54 ± 3.48 years and 42.80 ± 6.55 years, respectively. There were 3 men and 12 women in the lutein group and 1 man and 12 women in the control group. The average of AL, BCVA, and MPOD in the lutein and control groups were 28.21 ± 1.01 mm and 28.83 ± 0.98 mm, −0.01 ± 0.13 and −0.09 ± 0.09, and 0.70 ± 0.22 and 0.71 ± 0.21, respectively. There were no significant differences in age, sex, AXL, BCVA, or MPOD between the lutein and control groups at the baseline. Furthermore, there were no significant differences between the lutein and control groups at baseline in the CS and ERG (a-wave, b-wave, and b/a ratio). The results are presented in Table 1 presents the results.
## 3.2. Changes after treatment
The MPOD values are listed in Table 2. After 3 and 6 months of treatment from baseline, the changes in MPOD in the lutein and control groups were 0.00 ± 0.20 and 0.00 ± 0.12 at 3 months and 0.03 ± 0.12 and −0.07 ± 0.13 at 6 months, respectively, with no significant differences between the groups. There were no significant differences in the baseline values. The rate of changes in the lutein and control groups were 0.00 ± $0.20\%$ and 0.02 ± $0.20\%$, respectively with no significant differences between the groups.
**Table 2**
| Unnamed: 0 | Control group (N = 13) | Lutein group (N = 15) | P value |
| --- | --- | --- | --- |
| MPOD scores comparison of the change | MPOD scores comparison of the change | MPOD scores comparison of the change | MPOD scores comparison of the change |
| Months 3 after treatment minus baseline | 0.00 (0.12) | 0.00 (0.20) | .95 |
| Months 6 after treatment minus baseline | −0.04 (0.09) | 0.03 (0.12) | .10 |
| MPOD scores comparison of the rate of change | MPOD scores comparison of the rate of change | MPOD scores comparison of the rate of change | MPOD scores comparison of the rate of change |
| Months 3 after treatment minus baseline, % | 0.02 (0.20) | 0.00 (0.28) | .87 |
| Months 6 after treatment minus baseline, % | −0.07 (0.13) | 0.02 (0.25) | .26 |
The BCVA (logarithm of minimal angle resolution) values are presented in Table 3. After 3 and 6 months of treatment from baseline, the changes in BCVA in the lutein and control groups were 0.01 ± 0.10 and 0.04 ± 0.07 at 3 months and 0.01 ± 0.10 and 0.04 ± 0.08 at 6 months, respectively with no significant differences between the groups.
**Table 3**
| LogMAR comparison of the change | Control group (N = 13) | Lutein group (N = 15) | P value |
| --- | --- | --- | --- |
| Months 3 after treatment minus baseline | 0.04 (0.07) | 0.01 (0.10) | 0.35 |
| Months 6 after treatment minus baseline | 0.04 (0.08) | 0.01 (0.10) | 0.35 |
The CS values are listed in Table 4. After 3 months of treatment, 3, 6, 12, and 18 cpd of the CS change from baseline in the lutein and control groups were 0.80 ± 1.82 and 0.23 ± 2.95, −0.27 ± 1.71 and −0.38 ± 2.81, −0.27 ± 2.66 and −0.15 ± 2.76, and −0.73 ± 1.67 and 0.62 ± 1.71, respectively. The differences between the 2 groups were not significant at 3, 6, and 12 cpd, but at 18 cpd ($$P \leq .05$$). After 6 months of treatment, 3, 6, 12, and 18 cpd of the CS change from baseline in the lutein and control groups were 0.47 ± 2.72 and −0.08 ± 3.23, −0.40 ± 2.85 and −0.69 ± 3.54, 0.00 ± 3.23 and −0.15 ± 2.64, and −0.53 ± 2.26 and 0.23 ± 2.01, respectively. The differences between the 2 groups were not significant at any point.
**Table 4**
| Contrast sensitivity comparison of the change | Contrast sensitivity comparison of the change.1 | Control group (N = 13) | Lutein group (N = 15) | P value |
| --- | --- | --- | --- | --- |
| Months 3 after treatment minus baseline | 3 | 0.23 (2.95) | 0.80 (1.82) | 0.54 |
| Months 3 after treatment minus baseline | 6 | −0.38 (2.81) | −0.27 (1.71) | 0.89 |
| Months 3 after treatment minus baseline | 12 | −0.15 (2.76) | −0.27 (2.66) | 0.91 |
| Months 3 after treatment minus baseline | 18 | 0.62 (1.71) | −0.73 (1.67) | 0.05 |
| Months 6 after treatment minus baseline | 3 | −0.08 (3.23) | 0.47 (2.72) | 0.63 |
| Months 6 after treatment minus baseline | 6 | −0.69 (3.54) | 0.40 (2.85) | 0.37 |
| Months 6 after treatment minus baseline | 12 | −0.15 (2.64) | 0.00 (3.23) | 0.89 |
| Months 6 after treatment minus baseline | 18 | 0.23 (2.01) | −0.53 (2.26) | 0.36 |
The rate of change in the amplitude of the a-wave (μV), b-wave (μV), and ratio of a/b-wave in the ERG are listed in Table 5. After 3 months of treatment, the rate of change in the a-wave, b-wave, and ratio of a/b-wave from baseline in the lutein and control groups was −0.03 ± 0.25 and 0.06 ± 0.22, −0.02 ± 0.19 and 0.08 ± 0.17, and 0.03 ± 0.15 and 0.04 ± 0.26, respectively, and the differences between the 2 groups were not significant for each factor. After 6 months of treatment, the rate of changes in the a-wave, b-wave, and ratio of a/b-wave from baseline in the lutein and control groups was 0.00 ± 0.26 and 0.04 ± 0.20, −0.02 ± 0.23 and 0.04 ± 0.23, and −0.02 ± 0.14 and −0.01 ± 0.10, respectively, and the differences between the 2 groups were not significant for each factor.
**Table 5**
| ERG scores comparison of the rate of change | ERG scores comparison of the rate of change.1 | Control group (N = 13) | Lutein group (N = 15) | P value |
| --- | --- | --- | --- | --- |
| Months 3 after treatment minus baseline, % | a wave | 0.06 (0.22) | −0.03 (0.25) | 0.32 |
| Months 3 after treatment minus baseline, % | b wave | 0.08 (0.17) | −0.02 (0.19) | 0.15 |
| Months 3 after treatment minus baseline, % | b/a ratio | 0.04 (0.26) | 0.03 (0.15) | 0.93 |
| Months 6 after treatment minus baseline, % | a wave | 0.04 (0.20) | 0.00 (0.26) | 0.69 |
| Months 6 after treatment minus baseline, % | b wave | 0.04 (0.23) | −0.02 (0.23) | 0.52 |
| Months 6 after treatment minus baseline, % | b/a ratio | −0.01 (0.10) | −0.02 (0.14) | 0.81 |
## 3.3. Stratified macular pigment optical density analysis based on axial length
MPOD outcomes were further stratified using different AL strategies (Table 6). The participants were classified into 2 groups according to AL ≤ 28.25 and >28.25 mm. In the lutein group, 9 eyes had AL ≤ 28.25 mm and 6 had AL > 28.25 mm AL. In the control group, 4 eyes had an AL ≤ 28.25 mm and 9 had an AL > 28.25 mm.
**Table 6**
| MPOD scores | AL < 28.25 mm | AL < 28.25 mm.1 | AL < 28.25 mm.2 | AL ≥ 28.25 mm | AL ≥ 28.25 mm.1 | AL ≥ 28.25 mm.2 |
| --- | --- | --- | --- | --- | --- | --- |
| MPOD scores | Control group (N = 4) | Lutein group (N = 9) | P value | Control group (N = 9) | Lutein group (N = 6) | P value |
| Baseline | 0.67 (0.25) | 0.71 (0.17) | .69 | 0.72 (0.21) | 0.67 (0.29) | .70 |
| Comparison of the change | Comparison of the change | Comparison of the change | Comparison of the change | Comparison of the change | Comparison of the change | Comparison of the change |
| Months 3 after treatment minus baseline | −0.04 (0.20) | 0.06 (0.17) | .35 | 0.02 (0.08) | −0.09 (0.22) | .22 |
| Months 6 after treatment minus baseline | −0.09 (0.05) | 0.09 (0.10) | .01 | −0.02 (0.10) | −0.07 (0.07) | .35 |
| Comparison of the rate of change | Comparison of the rate of change | Comparison of the rate of change | Comparison of the rate of change | Comparison of the rate of change | Comparison of the rate of change | Comparison of the rate of change |
| Months 3 after treatment minus baseline, % | −0.02 (0.30) | 0.09 (0.23) | .46 | 0.03 (0.14) | −0.14 (0.27) | .14 |
| Months 6 after treatment minus baseline, % | −0.12 (0.04) | 0.13 (0.15) | .01 | −0.03 (0.13) | −0.17 (0.23) | .16 |
In the AL ≤ 28.25 mm subgroups, the average of AL at baseline was 27.50 ± 0.46 mm in the lutein group and 27.66 ± 0.59 mm in the control with no significant difference between the groups. After 3 months from baseline, the change in MPOD in the lutein and the control group was 0.06 ± 0.17 and −0.04 ± 0.20, respectively, with no significant difference between the groups. The rate of MPOD changes in the lutein and the control group was 0.09 ± $0.23\%$ and −0.02 ± $0.30\%$, respectively with no significant difference between the groups. In contrast, after 6 months, the changes in MPOD from baseline in the lutein and control groups were 0.09 ± 0.10 and −0.09 ± 0.05, respectively, and the rate of MPOD changes from baseline in the lutein and control groups was 0.13 ± $0.15\%$ and −0.12 ± $0.04\%$, respectively, with a significant difference in both the values and rate ($$P \leq .01$$, $$P \leq .01$$, respectively). In the AL > 28.25 mm subgroups, the average of AL at baseline was 29.27 ± 0.53 mm in the lutein group and 29.35 ± 0.54 mm in the control with no significant difference between the groups. There were no significant differences between the groups either in the change in MPOD or in the change in rates after 3 and 6 months from baseline. In the stratified analysis, we also evaluated BCVA, CS, and ERG; however, we found no significant changes in the baseline values for each parameter.
## 3.4. Adverse events
None of the patients reported any adverse events or complications.
## 4. Discussion
In this RCT, we evaluated the effects of lutein supplementation on MPOD. We observed that lutein supplementation significantly increased the MPOD level over 6 months and might be beneficial in preventing the loss of macular pigments in patients with HM having <28.25 mm of AL. To our knowledge, this is the first prospective, randomized, double-blind, placebo-controlled clinical trial to confirm the clinical efficacy and safety of lutein administration in HM patients.
HM is a clinical risk factor of serious ocular complications. The complications increase proportionally with an increase in AL.[14,15,22] HM-related complications are often observed in the macular areas. Clinically, there appears to be a myopic maculopathy pattern. It ranges from the early appearance of a tesselated fundus to progressive development of diffuse atrophy and lacquer cracks, followed by progression to patchy atrophy. Choroidal neovascularization (CNV) generally develops adjacent to areas of patchy atrophy or lacquer cracks, resulting in irreversible and severe VA loss. VA in HM may be subnormal, even before advanced myopic maculopathy sets in. One of the reasons for this may be the alteration in the arrangement of photoreceptors, which is affected by excessive stretching of the posterior pole. This may lead to a subnormal visual function in the macula. In eyes with HM, the photoreceptors in the nasal hemiretina are aligned toward the optic nerve by eyeball stretching and the retina becomes thin. MPs are mainly distributed in the Henle fiber layer of the macular fovea and the outer ganglion layer of photoreceptors in the rod cells around the fovea.[20] Therefore, the concentration of MP in the fovea of HM is thought to be low due to retinal thinning. A previous study showed that MPOD was negatively correlated with the degree of myopia.[24] *In this* study, we showed that the average MPOD in patients was approximately 0.70 at baseline. Zhang et al[24] showed that the MPOD in HM with an of 27.87 mm of AL was 0.55which is similar to our results. MPOD positively correlated with central foveal thickness in patients with low-to-moderate myopia.[25] In the stratified analysis by AL of HM eyes in our study, we showed that MPOD values after supplementation significantly increased in individuals with HM with AL ≦ 28.25 mm, which is in line with previous reports because the retinal thickness in HM becomes thinner with AL elongation and the degree of myopia. Retinal thickness may also be an important factor when considering the efficacy of lutein supplementation in HM. However, the efficacy of lutein supplementation in patients with HM remains controversial, although the benefits may be dose dependent. Tanito et al[26] found that MPOD levels after 3 months of lutein supplementation did not increase in individuals with myopia exceeding −4 diopters. They were orally administered 10 mg lutein for 3 months. In this study, we orally administered 20 mg lutein for 6 months, and the MPOD level did not increase at 3 months; however, it significantly increased at 6 months in individuals with HM and AL ≤ 28.25 mm. Another study demonstrated that low-dose lutein supplementation (6 mg) did not significantly improve the MPOD in patients with early AMD.[27] Furthermore, Sasamoto et al[27] observed that daily supplementation with 6 mg lutein did not affect the MPOD level in healthy individuals over 1 year, and speculated that 6 mg lutein may be insufficient to increase the MPOD level. Thus, high-dose and long-term administration may be required to increase the MPOD in eyes with or without HM.
A decrease in MP is related to functional abnormalities of the macula, which eventually lead to age-related degenerative eye diseases.[28,29] *It is* hypothesized that carotenoids, including lutein, could protect the photoreceptors and the retinal pigment epithelium by screening these susceptible retinal structures for actinic blue light and quenching reactive oxygen species.[30] Barker et al[31] demonstrated that carotenoid supplementation resulted in the accumulation of MP and significant foveal protection against short-wavelength photochemical damage. In humans, dietary lutein and zeaxanthin intake is inversely associated with AMD risk.[32–34] In addition, Ma et al[35] found that supplementation with these macular carotenoids partially reversed the loss of visual function in patients with early AMD by elevating MPOD, suggesting a causative role of MPOD in the maintenance of normal visual function. Thus, some studies have suggested the importance of MP in protecting visual function from age-related damage.[36–38] However, whether lutein supplementation improves visual function remains controversial. We showed that the MPOD increased in HM individuals with AL less than 28.25 mm, and their visual functions, including VA, CS, and ERG, were maintained, but did not improve compared with that at baseline. Moreover, MPOD was not increased in the HM eyes with AL > 28.25 mm by lutein supplementation. One possibility is retinal thinning in the eyes with HM. Lutein is concentrated in the photoreceptor axons of the Henle nerve fiber layer and rod outer segments of the retina.[4][34][4][34][4,39] Furthermore, we observed that the MPOD of the HM eye is low because the retinal thickness of HM eyes is extremely thin owing to extreme eyeball elongation, indicative a lack of space for MP accumulation in retina. Another possible explanation could be the duration of lutein administration. Greater MP accumulation may be required to improve the visual function in eyes with HM. In the present study, patients with HM were administered lutein supplements for 6 months, and the MPOD was increased in the eyes with less than AL < 28.25 mm with the supplementation; however, visual functions were not improved. Longer administration of lutein may induce further MP accumulation, and it may be possible to be improved visual function in HM eyes. Further investigation with longer observation is necessary for a complete understanding of the association between the MPOD increase by lutein and the visual function in HM eyes.
Although myopia is not often considered a serious eye disorder because vision can be easily corrected with glasses or contact lenses, it increases the risk of other ocular pathologies, such as glaucoma, retinal detachment, and lacquer cracks.[40] A low level of MPOD may predispose myopic patients to lacquer cracks,[16] which is one of the complications of high myopia characterized by rupture of Bruch’s membrane and retinal pigment epithelium.[16,41] A previous report noted that eyes with lacquer cracks around the macula have a higher risk of developing myopic CNV, which is one of the most serious complications in HM eyes.[41,42] Furthermore, a cross-sectional demonstrated approximately $40\%$ reduced odds of myopia (OR = 0.57) among individuals with the highest lutein concentration of $20\%$ in plasma.[43] Thus, lutein supplementation may have the potential to prevent ocular complications, such as CNV development, and restore visual function in HM eyes. Lutein is generally regarded as safe, with minimal side effects associated with its long-term consumption. Currently, lutein is commercially available and consumed by many people worldwide. No side effects of lutein supplementation were observed in our study. Lutein supplementation has been extensively researched and is the current clinical standard for treating individuals at a risk of AMD. In addition, when lutein supplementation was coupled with omega-3 supplementation, carotenoid bioavailability was enhanced.[44,45] In the present study, we observed no development of myopic CNV in the both group during the administration; however, it still remains unclear about the protective effect in the development of myopic CNV in HM eyes. Taken together with the previous studies, lutein supplementation may have a potential to reduce the risk of HM-specific ocular complications, and further investigation with longer observation is necessary for complete understanding.
This study had several limitations. First, the intervention time was relatively short (6 months) and was performed using a single dosing strategy. Long-term observation is needed to determine the functional effect of lutein because HM usually occurs in childhood and gradually develops over the long term, sometimes even >50 years. However, it is uncertain whether a higher dosing strategy would provide greater benefits. A previous trial illustrated the role of nutritional supplementation in maintaining lutein levels in the blood and MPOD, and clarified its safety in normal subjects.[46] Second, our cohort had a relatively small sample size, which reduced the statistical power to assess the association with MPOD supplementation. Third, other variables such as dietary supplementation with carotenoid-rich foods were not regulated in this study. Further research is needed to study the association between different responses to dietary supplementation with carotenoid-rich foods.
## 5. Conclusions
To the best of our knowledge, this is the first randomized clinical study to assess the benefits of lutein supplementation in highly myopic individuals. Our findings showed significant benefits of lutein supplementation in MPOD augmentation in patients with HM. As this study included patients who underwent 6 months of follow-up, this might limit the evaluation of the extended effects. Further larger-scale and longer-term studies are required to strengthen these associations and evaluate the effects of lutein on visual function in patients with HM.
## Acknowledgments
The authors would like to thank Aiko Ibuki, Natsumi Miyagawa, and Chihiro Ono for MPOD examination and data collection.
## Author contributions
Conceptualization: Takeshi Yoshida, Yasutaka Takagi, Kyoko Ohno-Matsui.
Data curation: Takeshi Yoshida, Tae Igarashi.
Formal analysis: Takeshi Yoshida.
Funding acquisition: Takeshi Yoshida.
Investigation: Takeshi Yoshida.
Methodology: Takeshi Yoshida, Yasutaka Takagi.
Project administration: Takeshi Yoshida.
Resources: Takeshi Yoshida.
Software: Takeshi Yoshida, Tae Igarashi.
Supervision: Takeshi Yoshida.
Validation: Takeshi Yoshida, Tae Igarashi.
Visualization: Takeshi Yoshida.
Writing – original draft: Takeshi Yoshida.
Writing – review & editing: Takeshi Yoshida, Yasutaka Takagi, Kyoko Ohno-Matsui.
## References
1. Feng L, Nie K, Jiang H. **Effects of lutein supplementation in age-related macular degeneration.**. *PLoS One* (2019) **14** e0227048. PMID: 31887124
2. Weigert G, Kaya S, Pemp B. **Effects of lutein supplementation on macular pigment optical density and visual acuity in patients with age-related macular degeneration.**. *Invest Ophthalmol Vis Sci* (2011) **52** 8174-8. PMID: 21873668
3. Chung RWS, Leanderson P, Gustafsson N. **Liberation of lutein from spinach: effects of heating time, microwave-reheating and liquefaction.**. *Food Chem* (2019) **277** 573-8. PMID: 30502187
4. Snodderly DM, Auran JD, Delori FC. **The macular pigment. II. Spatial distribution in primate retinas.**. *Invest Ophthalmol Vis Sci* (1984) **25** 674-85. PMID: 6724837
5. Kijlstra A, Tian Y, Kelly ER. **Lutein: more than just a filter for blue light.**. *Prog Retin Eye Res* (2012) **31** 303-15. PMID: 22465791
6. Kilbride PE, Fishman M, Fishman GA. **Foveal cone pigment density difference and reflectance in retinitis pigmentosa.**. *Arch Ophthalmol* (1986) **104** 220-4. PMID: 3947297
7. Neelam K, Goenadi CJ, Lun K. **Putative protective role of lutein and zeaxanthin in diabetic retinopathy.**. *Br J Ophthalmol* (2017) **101** 551-8. PMID: 28232380
8. Bernstein PS, Li B, Vachali PP. **Lutein, zeaxanthin, and meso-zeaxanthin: the basic and clinical science underlying carotenoid-based nutritional interventions against ocular disease.**. *Prog Retin Eye Res* (2016) **50** 34-66. PMID: 26541886
9. Eggersdorfer M, Wyss A. **Carotenoids in human nutrition and health.**. *Arch Biochem Biophys* (2018) **652** 18-26. PMID: 29885291
10. Fujimura S, Ueda K, Nomura Y. **Preliminary analysis of the relationship between serum lutein and zeaxanthin levels and macular pigment optical density.**. *Clin Ophthalmol* (2016) **10** 2149-55. PMID: 27826180
11. Puell MC, Palomo-Alvarez C, Barrio AR. **Relationship between macular pigment and visual acuity in eyes with early age-related macular degeneration.**. *Acta Ophthalmol* (2013) **91** e298-303. PMID: 23575039
12. Dolgin E. **The myopia boom.**. *Nature* (2015) **519** 276-8. PMID: 25788077
13. Yan YN, Wang YX, Yang Y. **Ten-year progression of myopic maculopathy: the Beijing eye study 2001–2011.**. *Ophthalmology* (2018) **125** 1253-63. PMID: 29602569
14. Tan JS, Wang JJ, Flood V. **Dietary antioxidants and the long-term incidence of age-related macular degeneration: the blue mountains eye study.**. *Ophthalmology* (2008) **115** 334-41. PMID: 17664009
15. Stringham JM, Hammond BR, Nolan JM. **The utility of using customized heterochromatic flicker photometry (cHFP) to measure macular pigment in patients with age-related macular degeneration.**. *Exp Eye Res* (2008) **87** 445-53. PMID: 18778703
16. Benoudis L, Ingrand P, Jeau J. **Relationships between macular pigment optical density and lacquer cracks in high myopia.**. *J Fr Ophtalmol* (2016) **39** 615-21. PMID: 27544327
17. Mares JA, LaRowe TL, Snodderly DM. **Predictors of optical density of lutein and zeaxanthin in retinas of older women in the carotenoids in age-related eye disease study, an ancillary study of the women’s health initiative.**. *Am J Clin Nutr* (2006) **84** 1107-22. PMID: 17093164
18. Hammond BR, Wooten BR, Smollon B. **Assessment of the validity of in vivo methods of measuring human macular pigment optical density.**. *Optom Vis Sci* (2005) **82** 387-404. PMID: 15894915
19. McCulloch DL, Marmor MF, Brigell MG. **ISCEV standard for full-field clinical electroretinography (2015 update).**. *Doc Ophthalmol* (2015) **130** 1-12
20. Wiegand RD, Giusto NM, Rapp LM. **Evidence for rod outer segment lipid peroxidation following constant illumination of the rat retina.**. *Invest Ophthalmol Vis Sci* (1983) **24** 1433-5. PMID: 6618806
21. Arslan S, Kadayifcilar S, Samur G. **The potential role of dietary antioxidant capacity in preventing age-related macular degeneration.**. *J Am Coll Nutr* (2019) **38** 424-32. PMID: 30570376
22. Chew EY, Clemons TE, Agron E. **Effect of omega-3 fatty acids, lutein/zeaxanthin, or other nutrient supplementation on cognitive function: the AREDS2 randomized clinical trial.**. *JAMA* (2015) **314** 791-801. PMID: 26305649
23. Snodderly DM, Mares JA, Wooten BR. **Macular pigment measurement by heterochromatic flicker photometry in older subjects: the carotenoids and age-related eye disease study.**. *Invest Ophthalmol Vis Sci* (2004) **45** 531-8. PMID: 14744895
24. Zhang Y, Hao J, Cao K. **Macular pigment optical density responses to different levels of zeaxanthin in patients with high myopia.**. *Graefes Arch Clin Exp Ophthalmol* (2022) **260** 2329-37. PMID: 35044504
25. Zheng W, Zhang Z, Jiang K. **Macular pigment optical density and its relationship with refractive status and foveal thickness in Chinese school-aged children.**. *Curr Eye Res* (2013) **38** 168-73. PMID: 22906053
26. Tanito M, Obana A, Gohto Y. **Macular pigment density changes in Japanese individuals supplemented with lutein or zeaxanthin: quantification via resonance Raman spectrophotometry and autofluorescence imaging.**. *Jpn J Ophthalmol* (2012) **56** 488-96. PMID: 22699751
27. Sasamoto Y, Gomi F, Sawa M. **Effect of 1-year lutein supplementation on macular pigment optical density and visual function.**. *Graefes Arch Clin Exp Ophthalmol* (2011) **249** 1847-54. PMID: 21850440
28. Beatty S, Murray IJ, Henson DB. **Macular pigment and risk for age-related macular degeneration in subjects from a Northern European population.**. *Invest Ophthalmol Vis Sci* (2001) **42** 439-46. PMID: 11157880
29. Neuringer M, Sandstrom MM, Johnson EJ. **Nutritional manipulation of primate retinas, I: effects of lutein or zeaxanthin supplements on serum and macular pigment in xanthophyll-free rhesus monkeys.**. *Invest Ophthalmol Vis Sci* (2004) **45** 3234-43. PMID: 15326146
30. Kumar N, Mrejen S, Fung AT. **Retinal pigment epithelial cell loss assessed by fundus autofluorescence imaging in neovascular age-related macular degeneration.**. *Ophthalmology* (2013) **120** 334-41. PMID: 23137630
31. Barker FM, Snodderly DM, Johnson EJ. **Nutritional manipulation of primate retinas, V: effects of lutein, zeaxanthin, and n-3 fatty acids on retinal sensitivity to blue-light-induced damage.**. *Invest Ophthalmol Vis Sci* (2011) **52** 3934-42. PMID: 21245404
32. Ho L, van Leeuwen R, Witteman JC. **Reducing the genetic risk of age-related macular degeneration with dietary antioxidants, zinc, and omega-3 fatty acids: the Rotterdam study.**. *Arch Ophthalmol* (2011) **129** 758-66. PMID: 21670343
33. Connolly EE, Beatty S, Loughman J. **Supplementation with all three macular carotenoids: response, stability, and safety.**. *Invest Ophthalmol Vis Sci* (2011) **52** 9207-17. PMID: 21979997
34. Joachim N, Mitchell P, Rochtchina E. **Incidence and progression of reticular drusen in age-related macular degeneration: findings from an older Australian cohort.**. *Ophthalmology* (2014) **121** 917-25. PMID: 24332537
35. Ma L, Liu R, Du JH. **Lutein, zeaxanthin and meso-zeaxanthin supplementation associated with macular pigment optical density.**. *Nutrients* (2016) **8** 426. PMID: 27420092
36. Li B, Ahmed F, Bernstein PS. **Studies on the singlet oxygen scavenging mechanism of human macular pigment.**. *Arch Biochem Biophys* (2010) **504** 56-60. PMID: 20678467
37. Trieschmann M, van Kuijk FJ, Alexander R. **Macular pigment in the human retina: histological evaluation of localization and distribution.**. *Eye (Lond)* (2008) **22** 132-7. PMID: 17401321
38. Thurnham DI, Howard AN. **Studies on meso-zeaxanthin for potential toxicity and mutagenicity.**. *Food Chem Toxicol* (2013) **59** 455-63. PMID: 23819916
39. Bone RA, Landrum JT. **Macular pigment in Henle fiber membranes: a model for Haidinger’s brushes.**. *Vision Res* (1984) **24** 103-8. PMID: 6546825
40. Saw SM, Gazzard G, Shih-Yen EC. **Myopia and associated pathological complications.**. *Ophthalmic Physiol Opt* (2005) **25** 381-91. PMID: 16101943
41. Ohno-Matsui K, Kawasaki R, Jonas JB. **International photographic classification and grading system for myopic maculopathy.**. *Am J Ophthalmol* (2015) **159** 877-83.e7. PMID: 25634530
42. Ohno-Matsui K, Yoshida T, Futagami S. **Patchy atrophy and lacquer cracks predispose to the development of choroidal neovascularisation in pathological myopia.**. *Br J Ophthalmol* (2003) **87** 570-3. PMID: 12714395
43. Williams KM, Bentham GC, Young IS. **Association between myopia, ultraviolet B radiation exposure, serum vitamin D concentrations, and genetic polymorphisms in vitamin D metabolic pathways in a ulticountry European study.**. *JAMA Ophthalmol* (2017) **135** 47-53. PMID: 27918775
44. **Lutein + zeaxanthin and omega-3 fatty acids for age-related macular degeneration: the Age-Related Eye Disease Study 2 (AREDS2) randomized clinical trial.**. *JAMA* (2013) **309** 2005-15. PMID: 23644932
45. Korobelnik JF, Rougier MB, Delyfer MN. **Effect of dietary supplementation with lutein, zeaxanthin, and omega-3 on macular pigment: a randomized clinical trial.**. *JAMA Ophthalmol* (2017) **135** 1259-66. PMID: 28973076
46. Akuffo KO, Beatty S, Stack J. **Central Retinal Enrichment Supplementation Trials (CREST): design and methodology of the CREST randomized controlled trials.**. *Ophthalmic Epidemiol* (2014) **21** 111-23. PMID: 24621122
|
---
title: 'Exploring the molecular mechanism of Ling-Gui-Zhu-Gan decoction for the treatment
of type 2 diabetes mellitus based on network pharmacology and molecular docking:
A review'
authors:
- Feng Long
- Zhe Zhang
- Chunxiu Luo
- Xiao Lei
- Jinlian Guo
- Lin An
journal: Medicine
year: 2023
pmcid: PMC10036033
doi: 10.1097/MD.0000000000033210
license: CC BY 4.0
---
# Exploring the molecular mechanism of Ling-Gui-Zhu-Gan decoction for the treatment of type 2 diabetes mellitus based on network pharmacology and molecular docking: A review
## Abstract
To investigate the mechanism of action of the classical formula Ling-Gui-Zhu-Gan (LGZG) decoction in treating type 2 diabetes mellitus based on network pharmacology and molecular docking. The active ingredients and targets of LGZG decoction were collected by the Traditional Chinese Medicine Systems Pharmacology Database and Analysis Platform database and mapped using Cytoscape software to show their interrelationships. GeneCards, Pharmacogenomics Knowledge Base, OMIM, Therapeutic Target Database, and Drugbank databases were used to obtain targets related to type 2 diabetes; protein-protein interaction networks were established with the help of the STRING platform. Gene Ontology and Kyoto Encyclopedia of Genes and Genomes analyses were performed on selected core targets with the help of the Metascape platform. Finally, the AutoDock platform was used to perform molecular docking and display the results by Pymol software. One hundred twenty-one active ingredients, 216 effective target genes, 11,277 type 2 diabetes mellitus-related genes, 210 crossover genes, and 18 core genes were obtained for LGZG decoction. The results obtained by Kyoto Encyclopedia of Genes and Genomes indicated that the advanced glycosylation end products-receptor of advanced glycosylation end products signaling pathway, the phosphatidylinositol 3 kinase-Akt signaling pathway, and HIF-1 signaling pathway might be the key signaling pathways. Molecular docking showed that the binding energy of quercetin, kaempferol, naringenin, and licorice chalcone A to the core target genes were all <5.0 kJ-mol−1, with good affinity. In this study, the potential active ingredients and mechanisms of action of LGZG decoction in the treatment of type 2 diabetes were initially investigated, which provided a basis for the in-depth study of its drug basis and mechanisms of action.
## 1. Introduction
Diabetes mellitus is a group of chronic hyperglycemic metabolic diseases caused by defective insulin secretion or utilization, with polydipsia, polyphagia, polyuria, and weight loss as typical clinical manifestations.[1] According to the International Diabetes Federation, the number of people with diabetes is expected to increase to about 700 million by 2045.[2] It has been documented that type 2 diabetes mellitus (T2DM) accounts for about $95\%$ of diabetic patients,[3] making it a global health problem. T2DM is a non-insulin-dependent type of diabetes, which can induce various complications, including diabetic nephropathy and cardiovascular. The clinical treatment of T2DM mainly includes lifestyle and pharmacological interventions. However, the long-term application of simple hypoglycemic drugs is often accompanied by different toxic side effects, such as heart failure and increased risk of fracture in women.[4,5] In contrast, the treatment of T2DM in traditional Chinese medicine (TCM) can make up for the shortcomings of Western medicine through its holistic concept and evidence-based treatment. It can regulate blood glucose by improving pancreatic β-cell function, promoting insulin secretion, and improving insulin sensitivity,[6,7] which has the advantages of less adverse reactions and stable efficacy and is receiving more and more attention from medical practitioners. The etiology of diabetes mellitus belongs to the category of “emaciation-thirst disease” in Chinese medicine. It is generally believed that the etiology of thirst is based on yin deficiency, dryness, and heat. The disease is usually located in the lung, stomach, and kidneys.[8] However, the clinical etiology is complex and varies, and it is not uncommon to have atypical symptoms and spleen deficiency with dampness.[9–12] *In this* paper, we use network pharmacology and molecular docking methods to predict the core target genes of Ling-Gui-Zhu-Gan (LGZG) decoction for the treatment of T2DM and investigate the potential pathway mechanisms to pave the way for further research. The detailed workflow of this study is presented in Figure 1.
**Figure 1.:** *Flow chart of the study.*
## 2.1. Acquisition of active ingredients and targets of TCM
The chemical composition of “Fu Ling,” “Gui Zhi,” “Bai Zhu,” and “Gan Cao” was retrieved from the database of Traditional Chinese Medicine Systems Pharmacology Database and Analysis Platform (TCMSP) (http://tcmspw.com/tcmsp.php). The chemical compositions of “Fu Ling,” “Gui Zhi,” “Bai Zhu,” and “Gan Cao” were searched. Based on the absorption distribution and metabolic excretion screening conditions, with oral bioavailability ≥ $30\%$ and drug similarity ≥ 0.18, the active ingredients and potential targets of Fu Ling, Gui Zhi, Bai Zhu, and Gan Cao in LGZG decoction were obtained. The obtained compound target genes were entered into the Uniprot (https://www.uniprot.org/) database, normalized, and transformed to find the corresponding gene symbols (standard gene names).
## 2.2. Collection of potential pathogenic targets of diseases
The GeneCards (http://www.genecards.org/), OMIM (http://www.omim.org/), Pharmacogenomics Knowledge Base (https://www.pharmgkb.org), Therapeutic Target Database (http://db.idrblab.net/ttd), and DrugBank (https://www.drugbank.ca) databases for the keyword “type 2 diabetes mellitus.” The search results of the 5 databases were combined, all disease-related genes were merged, and duplicate target genes were removed to obtain the potential pathogenic targets of T2DM.
## 2.3. Venny analysis of intersecting genes
The data of the potentially active compounds and the pathogenic targets of T2DM were mapped into Venny 2.1.0. The intersection was taken, and the “intersection gene Venn map” was created.
## 2.4. “Active ingredient-target” network construction
Based on the obtained active ingredients and intersecting genes, we construct the active ingredient-target network relationship, import the constructed network relationship into Cytoscape 3.8.0 software, and build a network map, with nodes representing active ingredients and action targets respectively; edges are used to connect the nodes and show the connection between active ingredients and action targets. Furthermore, according to the software set graphics, color, character size, and other parameters to build a visual production map.
## 2.5. Construction of protein interaction network and key target screening
Protein-protein interaction (PPI) studies the correlation between compounds and disease-related protein molecules from the perspective of biochemical, signal transduction, and genetic networks. The common target genes of LGZG decoction and T2DM were imported into the STRING database (https://string-db.org/), the study species was limited to “human,” the confidence level was set to 0.9, the free nodes were hidden, and the rest of the parameters were set to default to obtain the PPIs, and exported to Save as TSV format file and save the original image. The resulting files were imported into Cytoscape 3.8.0 software for visualization. The CytoNCA plug-in was used to calculate the Betweenness, Closeness, Degree, Eigenvector, Network, and local average connectivity-based scores for each node. Each node is scored, and the node with a higher score is retained by filtering the value > median value to obtain the core gene of the network.
## 2.6. Gene Ontology (GO) functional analysis and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis
To further determine the functions of the intersection target genes and their signaling pathways, set $P \leq .05$, with the help of the “ggplot2” “stringi” “colorspace” package in R language and the “cluster Profiler” “enrichplot” “DOSE” library in Bioconductor “ggplot2” “stringi” “colorspace” in R language and “clusterProfiler” “enrichplot” and “DOSE” in Bioconductor library; “language package to perform GO enrichment analysis on the intersection genes of LGZG decoction and T2DM, and clarify the mode of action; meanwhile, the KEGG pathway analysis was performed in the above way to clarify the pathway of action, and visualization analysis was performed to obtain cluster plots and bubble plots.
## 2.7. Pharmacophore-target molecular docking validations
The above analysis selected top-ranked active ingredients and core genes, and molecular docking was verified according to whether there was an interaction between the core compounds and the core target genes. 2D structure files of small molecule ligands were downloaded from the PubChem database, and the 2D structure was converted to a 3D structure in ChemOffice software and saved in mol2 format. The pdb format of protein 3D structure was downloaded from the Protein Data Bank database (http://www.rcsb.org), and its water molecules and small molecule ligands were removed by PyMOL software. The resulting small molecule ligand and protein receptor files were converted into PDBQT format, and the active pockets for molecular docking were determined simultaneously. Finally, molecular docking was performed using Vina software.
## 3.1. Active ingredients and potential targets
As shown in Table 1, after screening in the TCMSP database according to the conditions, a total of 121 active compounds were collected, including 15 active compounds for Fuling, 7 active compounds for Gui Zhi, 7 active compounds for Bai Zhu, and 92 active compounds for Gan Cao. Sitosterol was common to both Gui Zhi and Gan Cao among these active compounds. After searching the corresponding targets in the TCMSP database based on the screened active ingredients, 216 annotated genes were obtained by normalizing the target genes to the UniProt database (as shown in Supplemental File S1, Supplemental Digital Content, http://links.lww.com/MD/I644, which illustrates the related targets and the basic information of all active compounds).
**Table 1**
| Mol ID | OB (%) | OB (%).1 | DL | Source |
| --- | --- | --- | --- | --- |
| MOL000273 | (2R)-2-[(3S,5R,10S,13R,14R,16R,17R)-3,16-dihydroxy-4,4,10,13,14-pentamethyl-2,3,5,6,12,15,16,17-octahydro-1H-cyclopenta[a]phenanthren-17-yl]-6-methylhept-5-enoic acid | 30.93 | 0.81 | Fu Ling |
| MOL000275 | Trametenolic acid | 38.71 | 0.8 | Fu Ling |
| MOL000276 | 7,9 (11)-Dehydropachymic acid | 35.11 | 0.81 | Fu Ling |
| MOL000279 | Cerevisterol | 37.96 | 0.77 | Fu Ling |
| MOL000280 | (2R)-2-[(3S,5R,10S,13R,14R,16R,17R)-3,16-dihydroxy-4,4,10,13,14-pentamethyl-2,3,5,6,12,15,16,17-octahydro-1H-cyclopenta[a]phenanthren-17-yl]-5-isopropyl-hex-5-enoic acid | 31.07 | 0.82 | Fu Ling |
| MOL000282 | Ergosta-7,22E-dien-3beta-ol | 43.51 | 0.72 | Fu Ling |
| MOL000283 | Ergosterol peroxide | 40.36 | 0.81 | Fu Ling |
| MOL000285 | (2R)-2-[(5R,10S,13R,14R,16R,17R)-16-hydroxy-3-keto-4,4,10,13,14-pentamethyl-1,2,5,6,12,15,16,17-octahydrocyclopenta[a]phenanthren-17-yl]-5-isopropyl-hex-5-enoic acid | 38.26 | 0.82 | Fu Ling |
| MOL000287 | 3Beta-Hydroxy-24-methylene-8-lanostene-21-oic acid | 38.7 | 0.81 | Fu Ling |
| MOL000289 | Pachymic acid | 33.63 | 0.81 | Fu Ling |
| MOL000290 | Poricoic acid A | 30.61 | 0.76 | Fu Ling |
| MOL000291 | Poricoic acid B | 30.52 | 0.75 | Fu Ling |
| MOL000292 | Poricoic acid C | 38.15 | 0.75 | Fu Ling |
| MOL000296 | Hederagenin | 36.91 | 0.75 | Fu Ling |
| MOL000300 | Dehydroeburicoic acid | 44.17 | 0.83 | Fu Ling |
| MOL001736 | (−)-Taxifolin | 60.51 | 0.27 | Gui Zhi |
| MOL000358 | Beta-sitosterol | 36.91 | 0.75 | Gui Zhi |
| MOL000359 | Sitosterol | 36.91 | 0.75 | Gui Zhi |
| MOL000492 | (+)-Catechin | 54.83 | 0.24 | Gui Zhi |
| MOL000073 | Ent-epicatechin | 48.96 | 0.24 | Gui Zhi |
| MOL004576 | Taxifolin | 57.84 | 0.27 | Gui Zhi |
| MOL011169 | Peroxyergosterol | 44.39 | 0.82 | Gui Zhi |
| MOL000020 | 12-Senecioyl-2E,8E,10E-atractylentriol | 62.4 | 0.22 | Bai Zhu |
| MOL000021 | 14-Acetyl-12-senecioyl-2E,8E,10E-atractylentriol | 60.31 | 0.31 | Bai Zhu |
| MOL000022 | 14-Acetyl-12-senecioyl-2E,8Z,10E-atractylentriol | 63.37 | 0.3 | Bai Zhu |
| MOL000028 | α-Amyrin | 39.51 | 0.76 | Bai Zhu |
| MOL000033 | (3S,8S,9S,10R,13R,14S,17R)-10,13-dimethyl-17-[(2R,5S)-5-propan-2-yloctan-2-yl]-2,3,4,7,8,9,11,12,14,15,16,17-dodecahydro-1H-cyclopenta[a]phenanthren-3-ol | 36.23 | 0.78 | Bai Zhu |
| MOL000049 | 3β-Acetoxyatractylone | 54.07 | 0.22 | Bai Zhu |
| MOL000072 | 8β-Ethoxy atractylenolide III | 35.95 | 0.21 | Bai Zhu |
| MOL004941 | (2R)-7-hydroxy-2-(4-hydroxyphenyl)chroman-4-one | 71.12 | 0.18 | Gan Cao |
| MOL001792 | DFV | 32.76 | 0.18 | Gan Cao |
| MOL004835 | Glypallichalcone | 61.6 | 0.19 | Gan Cao |
| MOL004841 | Licochalcone B | 76.76 | 0.19 | Gan Cao |
| MOL004985 | Icos-5-enoic acid | 30.7 | 0.2 | Gan Cao |
| MOL004996 | Gadelaidic acid | 30.7 | 0.2 | Gan Cao |
| MOL003896 | 7-Methoxy-2-methyl isoflavone | 42.56 | 0.2 | Gan Cao |
| MOL000500 | Vestitol | 74.66 | 0.21 | Gan Cao |
| MOL004957 | HMO | 38.37 | 0.21 | Gan Cao |
| MOL004328 | Naringenin | 59.29 | 0.21 | Gan Cao |
| MOL000392 | Formononetin | 69.67 | 0.21 | Gan Cao |
| MOL000422 | Kaempferol | 41.88 | 0.24 | Gan Cao |
| MOL000417 | Calycosin | 47.75 | 0.24 | Gan Cao |
| MOL004991 | 7-Acetoxy-2-methylisoflavone | 38.92 | 0.26 | Gan Cao |
| MOL004990 | 7,2’,4’-Trihydroxy–5-methoxy-3–arylcoumarin | 83.71 | 0.27 | Gan Cao |
| MOL004860 | Licorice glycoside E | 32.89 | 0.27 | Gan Cao |
| MOL000098 | Quercetin | 46.43 | 0.28 | Gan Cao |
| MOL000497 | Licochalcone a | 40.79 | 0.29 | Gan Cao |
| MOL000239 | Jaranol | 50.83 | 0.29 | Gan Cao |
| MOL005016 | Odoratin | 49.95 | 0.3 | Gan Cao |
| MOL000354 | Isorhamnetin | 49.6 | 0.31 | Gan Cao |
| MOL004898 | (E)-3-[3,4-dihydroxy-5-(3-methylbut-2-enyl)phenyl]-1-(2,4-dihydroxyphenyl)prop-2-en-1-one | 46.27 | 0.31 | Gan Cao |
| MOL004910 | Glabranin | 52.9 | 0.31 | Gan Cao |
| MOL004945 | (2S)-7-hydroxy-2-(4-hydroxyphenyl)-8-(3-methylbut-2-enyl)chroman-4-one | 36.57 | 0.32 | Gan Cao |
| MOL004848 | Licochalcone G | 49.25 | 0.32 | Gan Cao |
| MOL004980 | Inflacoumarin A | 39.71 | 0.33 | Gan Cao |
| MOL004961 | Quercetin der. | 46.45 | 0.33 | Gan Cao |
| MOL002565 | Medicarpin | 49.22 | 0.34 | Gan Cao |
| MOL004829 | Glepidotin B | 64.46 | 0.34 | Gan Cao |
| MOL004828 | Glepidotin A | 44.72 | 0.35 | Gan Cao |
| MOL004815 | (E)-1-(2,4-dihydroxyphenyl)-3-(2,2-dimethylchromen-6-yl)prop-2-en-1-one | 39.62 | 0.35 | Gan Cao |
| MOL004907 | Glyzaglabrin | 61.07 | 0.35 | Gan Cao |
| MOL004882 | Licocoumarone | 33.21 | 0.36 | Gan Cao |
| MOL003656 | Lupiwighteone | 51.64 | 0.37 | Gan Cao |
| MOL005020 | Dehydroglyasperins C | 53.82 | 0.37 | Gan Cao |
| MOL004915 | Eurycarpin A | 43.28 | 0.37 | Gan Cao |
| MOL004838 | 8-(6-Hydroxy-2-benzofuranyl)-2,2-dimethyl-5-chromenol | 58.44 | 0.38 | Gan Cao |
| MOL005000 | Gancaonin G | 60.44 | 0.39 | Gan Cao |
| MOL004811 | Glyasperin C | 45.56 | 0.4 | Gan Cao |
| MOL004856 | Gancaonin A | 51.08 | 0.4 | Gan Cao |
| MOL004993 | 8-Prenylated eriodictyol | 53.79 | 0.4 | Gan Cao |
| MOL004864 | 5,7-Dihydroxy-3-(4-methoxyphenyl)-8-(3-methylbut-2-enyl)chromone | 30.49 | 0.41 | Gan Cao |
| MOL004989 | 6-Prenylated eriodictyol | 39.22 | 0.41 | Gan Cao |
| MOL004863 | 3-(3,4-Dihydroxyphenyl)-5,7-dihydroxy-8-(3-methylbut-2-enyl)chromone | 66.37 | 0.41 | Gan Cao |
| MOL004935 | Sigmoidin-B | 34.88 | 0.41 | Gan Cao |
| MOL004866 | 2-(3,4-Dihydroxyphenyl)-5,7-dihydroxy-6-(3-methylbut-2-enyl)chromone | 44.15 | 0.41 | Gan Cao |
| MOL004883 | Licoisoflavone | 41.61 | 0.42 | Gan Cao |
| MOL004949 | Isolicoflavonol | 45.17 | 0.42 | Gan Cao |
| MOL004814 | Isotrifoliol | 31.94 | 0.42 | Gan Cao |
| MOL004913 | 1,3-Dihydroxy-9-methoxy-6-benzofurano[3,2-c]chromenone | 48.14 | 0.43 | Gan Cao |
| MOL004849 | 3-(2,4-Dihydroxyphenyl)-8-(1,1-dimethylprop-2-enyl)-7-hydroxy-5-methoxy-coumarin | 59.62 | 0.43 | Gan Cao |
| MOL004808 | Glyasperin B | 65.22 | 0.44 | Gan Cao |
| MOL004911 | Glabrene | 46.27 | 0.44 | Gan Cao |
| MOL004833 | Phaseolinisoflavan | 32.01 | 0.45 | Gan Cao |
| MOL004857 | Gancaonin B | 48.79 | 0.45 | Gan Cao |
| MOL004908 | Glabridin | 53.25 | 0.47 | Gan Cao |
| MOL004855 | Licoricone | 63.58 | 0.47 | Gan Cao |
| MOL004879 | Glycyrin | 52.61 | 0.47 | Gan Cao |
| MOL005012 | Licoagroisoflavone | 57.28 | 0.49 | Gan Cao |
| MOL004912 | Glabrone | 52.51 | 0.5 | Gan Cao |
| MOL004820 | kanzonols W | 50.48 | 0.52 | Gan Cao |
| MOL004978 | 2-[(3r)-8,8-dimethyl-3,4-dihydro-2h-pyrano[6,5-f]chromen-3-yl]-5-methoxyphenol | 36.21 | 0.52 | Gan Cao |
| MOL004914 | 1,3-Dihydroxy-8,9-dimethoxy-6-benzofurano[3,2-c]chromenone | 62.9 | 0.53 | Gan Cao |
| MOL004810 | Glyasperin F | 75.84 | 0.54 | Gan Cao |
| MOL001484 | Inermine | 75.18 | 0.54 | Gan Cao |
| MOL004885 | Licoisoflavanone | 52.47 | 0.54 | Gan Cao |
| MOL004884 | Licoisoflavone B | 38.93 | 0.55 | Gan Cao |
| MOL004905 | 3,22-Dihydroxy-11-oxo-delta(12)-oleanene-27-alpha-methoxycarbonyl-29-oic acid | 34.32 | 0.55 | Gan Cao |
| MOL004827 | Semilicoisoflavone B | 48.78 | 0.55 | Gan Cao |
| MOL004806 | Euchrenone | 30.29 | 0.57 | Gan Cao |
| MOL004974 | 3’-Methoxyglabridin | 46.16 | 0.57 | Gan Cao |
| MOL004966 | 3’-Hydroxy-4’-o-Methylglabridin | 43.71 | 0.57 | Gan Cao |
| MOL005017 | Phaseol | 78.77 | 0.58 | Gan Cao |
| MOL005003 | Licoagrocarpin | 58.81 | 0.58 | Gan Cao |
| MOL005007 | Glyasperins M | 72.67 | 0.59 | Gan Cao |
| MOL005008 | Glycyrrhiza flavonol A | 41.28 | 0.6 | Gan Cao |
| MOL004824 | (2S)-6-(2,4-dihydroxyphenyl)-2-(2-hydroxypropan-2-yl)-4-methoxy-2,3-dihydrofuro[3,2-g]chromen-7-one | 60.25 | 0.63 | Gan Cao |
| MOL004959 | 1-Methoxyphaseollidin | 69.98 | 0.64 | Gan Cao |
| MOL004904 | Licopyranocoumarin | 80.36 | 0.65 | Gan Cao |
| MOL002311 | Glycyrol | 90.78 | 0.67 | Gan Cao |
| MOL005013 | 18α-Hydroxyglycyrrhetic acid | 41.16 | 0.71 | Gan Cao |
| MOL004805 | (2S)-2-[4-hydroxy-3-(3-methylbut-2-enyl)phenyl]-8,8-dimethyl-2,3-dihydropyrano[2,3-f]chromen-4-one | 31.79 | 0.72 | Gan Cao |
| MOL004891 | Shinpterocarpin | 80.3 | 0.73 | Gan Cao |
| MOL004903 | Liquiritin | 65.69 | 0.74 | Gan Cao |
| MOL000359 | Sitosterol | 36.91 | 0.75 | Gan Cao |
| MOL000211 | Mairin | 55.38 | 0.78 | Gan Cao |
| MOL005001 | Gancaonin H | 50.1 | 0.78 | Gan Cao |
| MOL004917 | Glycyroside | 37.25 | 0.79 | Gan Cao |
| MOL004948 | Isoglycyrol | 44.7 | 0.84 | Gan Cao |
| MOL005018 | Xambioona | 54.85 | 0.87 | Gan Cao |
| MOL004988 | Kanzonol F | 32.47 | 0.89 | Gan Cao |
| MOL004924 | (−)-Medicocarpin | 40.99 | 0.95 | Gan Cao |
## 3.2. Acquisition of intersecting genes
As shown in Figure 2, a total of 11,277 potential targets for T2DM were obtained from the GeneCards database, OMIM database, Pharmacogenomics Knowledge Base database, Therapeutic Target Database database DrugBank database, and the retrieved disease-related genes were combined to draw a Wayne map. In Figure 3, the green and pink circles represent the predicted targets of the compound and T2DM. The intersecting part in the middle indicates the targets shared by both, suggesting that the 210 intersecting targets may be the potential targets of LGZG decoction for the treatment of T2DM.
**Figure 2.:** *Disease-associated genes map.* **Figure 3.:** *Intersectional gene maps.*
## 3.3. Drug-active ingredient-target gene network construction
Figure 4 shows a “drug-active ingredient-target gene” network map with 317 nodes (4 drugs,103 active compounds, and 210 targets) and 1608 edges constructed in an Excel sheet and imported into Cytoscape 3.8.0 software. The red inverted triangle represents Gancao, the golden hexagon represents Fuling, the pink quadrilateral represents Guizhi, the blue diamond represents atractylodes, green rectangle represents a gene. The top active ingredients can be filtered out from the figure, with quercetin (quercetin) targeting up to 137 associations, kaempferol (kaempferol) with 53 associations, naringenin (naringenin) with 34 associations, and 7-methoxy-2-methyl isoflavone (isoflavone) with 34 associations (as shown in Supplemental File S2, Supplemental Digital Content, http://links.lww.com/MD/I645, which illustrates the basic information of network construction and the types of nodes).
**Figure 4.:** *Drug-active ingredient-target network maps. Different shapes and colors represent different information. The red inverted triangle represents Gancao, the golden hexagon represents Fuling, the pink quadrilateral represents Guizhi, the blue diamond represents atractylodes, and the green rectangle represents a gene.*
## 3.4. Construction and analysis of PPI
The above 210 intersecting genes were uploaded to the STRING database to obtain the PPI map, as shown in Figure 5. The PPI network map was imported into Cytosccape 3.8.0 software, and each node was scored using the CytoNCA plug-in. As shown in Figure 6, the nodes with scores less than the median were filtered according to the scores, and the nodes with higher scores were retained to obtain the network core genes.
**Figure 5.:** *Protein interaction network (PPI) maps.* **Figure 6.:** *Network core genetic map.*
## 3.5. GO and KEGG enrichment analysis
The GO analysis was performed in R language for the intersection of LGZG decoction and T2DM (Fig. 7), and the P value was set at 0.05. The GO annotation analysis was performed in 3 aspects: biological process (BP), cellular component, and molecular function (MF). The results of GO annotation analysis showed that the BP mainly involved the response to xenobiotic stimuli, the response to metal ions, the cellular response to chemical stress, the response to lipopolysaccharide, peptide, and extracellular matrix, and the oxidative stress response. Composition mainly involves membrane rafts, membrane microstructure domains, transcriptional regulatory factors, transcriptional regulatory complexes, protein kinase complexes, vesicle lumen, etc. MFs mainly involve DNA-binding transcription factors, cytosolic receptor activity, activated ligands, ligase conjugates, subtilisin conjugates, G protein-coupled amine receptor activity, cytosolic steroid receptor activity, etc.
**Figure 7.:** *GO function enrichment analysis. The top 10 entries are retained separately according to P < .05, where larger and redder bubbles indicate a higher number of enriched targets. GO = Gene Ontology, BP = biological process, CC = cellular component, MF = molecular function.*
The R software was used to analyze the pathways and visualize them by setting P to.05 (KEGG pathway, Fig. 8), and the top 30 signaling pathways were screened. The results indicated that the target genes were mainly involved in these signaling pathways, including lipid and atherosclerosis, advanced glycosylation end products (AGE)-in diabetic complications receptor of AGE (RAGE) signaling pathway, phosphatidylinositol 3 kinase (PI3K)-Akt signaling pathway, interleukin (IL)-17 signaling pathway, HIF-1 signaling pathway, tumor necrosis factor (TNF) signaling pathway, T helper cell 17 cell differentiation, and other pathways (as shown in Supplemental File S3, Supplemental Digital Content, http://links.lww.com/MD/I646, which illustrates GO and KEGG pathway analysis results).
**Figure 8.:** *KEGG pathway enrichment analysis. The y-axis demonstrates the top 18 significantly enriched KEGG pathways, while the x-axis shows the number of enriched genes for these terms (P < .05). The colors and the sizes indicate different P value ranges; the redder and bigger it is, the more significantly enriched it is. KEGG = Kyoto Encyclopedia of Genes and Genomes.*
## 3.6. Active ingredient-target molecular docking
As shown in Table 2, after a comprehensive analysis of their network values, statistical indicators, and count values, the top-ranked quercetin (quercetin), kaempferol (kaempferol), naringenin (naringenin) and licochalcone a (licorice chalcone A) among the obtained active ingredients were molecularly docked with the top-ranked key targets signal transducers and activators of transcription (STAT)-3, hypoxia-inducible factor-1 (HIF1A), STAT1 and AKT serine/threonine kinase 1 (AKT1), respectively. The smaller the binding energy, the greater the affinity and the higher the binding activity. The molecular docking results of licochalcone a with recombinant cyclin D1 (CCND1), quercetin with CCND1, and HIF1A were selected for demonstration according to their binding energy sizes, Figures 9–11.
## 4. Discussion
Our findings indicate that the top-ranked active ingredients in LGZG decoction were quercetin (quercetin), kaempferol (kaempferol), naringenin (naringenin), and licochalcone a (licorice chalcone A). It has been shown that quercetin inhibits NOX expression and ROS production in rat Achilles tendons under hyperglycemic conditions, has antioxidant and anti-inflammatory effects, and can prevent the development of diabetic tendinopathy.[13] Its derivatives have a protective effect on diabetic neuropathy by inhibiting the Wnt/β-catenin signaling pathway.[14] Quercetin intake was negatively correlated with the prevalence of T2DM,[15] and its normalization of pancreatic β-cells was achieved by reducing iron death in T2DM by eliminating oxidative stress.[16] Kaempferol modulates M1/M2 phenotype under hyperglycemia and indirectly protects against podocyte apoptosis by regulating macrophage M1/M2 differentiation[17]; in terms of its hypoglycemic mechanism, it may achieve hypoglycemic efficacy by inhibiting the digestion of carbohydrates.[18] Another scholar found in rat experiments that naringin could prevent hyperglycemia-mediated-induced inflammation and damage to liver and pancreatic tissues by blocking the activation-mediated anti-inflammatory effects of NF-κB and its down-regulated genes, including pro-inflammatory cytokines.[19] In contrast, licochalcone A is capable of modulating insulin sensitivity in modern pharmacology, thus exerting a corresponding hypoglycemic effect.[20] The GO functional enrichment analysis results showed that the treatment of T2DM by LGZG decoction is extremely complex, involving multiple aspects of BPs, cell composition, and MFs. To sum up, the action targets located at the center of the network constructed by Cytoscape include STAT3, HIF1A, STAT1, AKT1, and CCND1.
It has been shown that in studying the effects of T2D risk genes and SNPs on transcriptional binding affinity, the top 4 transcription factors associated with enrichment were found to be Rfx1, Nkx2-5, NR2C2, and MZF15-13, while TF-Rfx1 is in the signal transducer and activator of transcription 3 (STAT3) pathway The expression of Rfx1 is regulated by the IL-6-STAT3 signaling pathway, which also indicates that STAT3 plays an important role in the pathogenesis of diabetes.[21–23] STAT3, under high glucose conditions, after phosphorylation, causes hepatic gluconeogenesis while reducing glycogen synthesis and elevating blood glucose.[24] In experiments in which systemic injection of streptozotocin-induced hyperglycemia in rats, it was found that during streptozotocin-induced diabetic retinopathy, the levels of HIF-1α, as well as the pro-inflammatory cytokines IL-1β, IL-6, and TNF-α, were increased and that HIF-1α led to upregulation of IL-6 and TNF-α and their receptors as well as Caspase-3, and inhibition of HIF-1α decreases the expression of the pro-inflammatory mediators IL-6 and TNF-αin diabetic retinopathy, thereby reducing the incidence of diabetic retinopathy.[25] AKT1 is an important link in the PI3K/AKT/mTOR signaling pathway. Large amounts of AKT1 activate mTOR and enhance SREBP1 efficacy, thereby increasing intracellular triacylglycerol in tissues[26] to achieve energy homeostasis. Akt is essential for insulin and nutrient-mediated regulation of hepatic metabolism in the body.[27] It has also been shown that hepatic CCND1 deficiency leads to increased gluconeogenesis and, consequently, to hyperglycemia.[28] The AGE-RAGE signaling pathway, PI3K-Akt signaling pathway, and HIF-1 signaling pathway in lipids and atherosclerosis, diabetic complications may play an important role in the treatment of T2DM with LGZG decoction as seen in the results of enrichment pathways in KEGG. Studies have shown that a persistent hyperglycemic state will cause plasma protein glycosylation, and insulin glycosylation can distort insulin signaling[29] and reduce insulin sensitivity to adipocyte cell membrane surface receptors,[30] which will cause hyperlipidemia and atherosclerosis manifestations in the long-term[31] and aggravate the risk of T2DM complications. One study showed that the risk of gestational diabetes mellitus increased 18.48-fold for each unit increase in atherosclerotic plasma index, indicating that reasonable lipid control in mid-pregnancy may reduce the incidence of gestational diabetes mellitus, statistically demonstrating an association between lipids and atherosclerosis and diabetes.[32] The PI3K/Akt signaling pathway of key downstream factors is closely related to the regulation of glucose and lipid metabolism,[33] and data suggesting a reduction in the occurrence of diabetic osteoporosis with further upregulation of PI3K/Akt-related protein levels.[34] In turn, activation of the HIF-1 signaling pathway is associated with inflammatory and fibrotic processes in the renal unit and vascular calcification in patients with T2DM.[35,36] In conclusion, the pharmacological process of LGZG decoction for the treatment of T2DM may be a synergistic combination of multiple active ingredients, multiple action targets, and multiple signaling pathways. Various core ingredients in the formula, such as quercetin, kaempferol, naringenin, and licorice chalcone A, may act on STAT3, HIF1A, STAT1, AKT1, CCND1, and other potential key targets and then exert synergistic effects on multiple signaling pathways such as AGE-RAGE signaling pathway, PI3K-Akt signaling pathway and HIF-1 signaling pathway in lipid and atherosclerosis, diabetic complications. In turn, it exerts a holistic and complex regulatory effect on multiple signaling pathways, including the AGE-RAGE signaling pathway, PI3K-Akt signaling pathway, and HIF-1 signaling pathway in lipid and atherosclerosis diabetes complications. According to TCM, T2DM belongs to the category of “thirst” in TCM and is often due to spleen deficiency and water retention. The combination of the 4 herbs in LGZG decoction brings out the effect of “warming and harmonizing” to achieve the function of warming yang, transforming qi, promoting water, and dispelling dampness. This paper predicts and analyzes the possible pharmacological mechanism of LGZG decoction in the treatment of T2DM by using software technology related to network pharmacology. At the same time, the results of the current experimental research are used to prove that the results are still scientific and reasonable. However, from the perspective of scientific rigor, further experimental research data are needed to support the results, so it provides a theoretical basis and direction to explore the pharmacological process of LGZG decoction in the treatment of T2DM. However, from the point of view of scientific rigor, in the study, although we have selected as much data as possible from the database, some important targets may still be missed because the database needs to be updated in time or the research needs to be more comprehensive. Moreover, the study that some of the selected signaling pathways are indeed part of the pathogenesis of diabetes and its complications is not enough. There are differences between in vitro theoretical study and in vivo metabolism. In addition, the study aims to construct a possible active ingredient-target network through relevant software technology. Due to the limitations of the research technology, we need further experimental research data to prop up.
## Author contributions
Conceptualization: Feng Long, Zhe Zhang.
Data curation: Feng Long, Zhe Zhang.
Methodology: Feng Long, Zhe Zhang.
Resources: Feng Long, Chunxiu Luo, Jinlian Guo, Lin An.
Software: Feng Long, Zhe Zhang.
Supervision: Xiao Lei.
Writing – original draft: Feng Long.
Writing – review & editing: Zhe Zhang, Xiao Lei.
## References
1. **Diagnosis and classification of diabetes mellitus.**. *Diabetes Care* (2013) **36** S67-74. PMID: 23264425
2. Cho NH, Shaw JE, Karuranga S. **IDF diabetes atlas: global estimates of diabetes prevalence for 2017 and projections for 2045.**. *Diabetes Res Clin Pract* (2018) **138** 271-81. PMID: 29496507
3. Boughton Charlotte K, Tripyla A, Hartnell S. **Fully automated closed-loop glucose control compared with standard insulin therapy in adults with type 2 diabetes requiring dialysis: an open-label, randomized crossover trial.**. *Nat Med* (2021) **27** 1471-6. PMID: 34349267
4. Turner RM, Kwok CS, Chen-Turner C. **Thiazolidinediones and associated risk of bladder cancer: a systematic review and meta-analysis.**. *Br J Clin Pharmacol* (2014) **78** 258-73. PMID: 24325197
5. Loke YK, Singh S, Furberg CD. **Long-term use of thiazolidinediones and fractures in type 2 diabetes: a meta-analysis.**. *CMAJ* (2009) **180** 32-9. PMID: 19073651
6. Wang H, Shi S, Wang S. **Can highly cited herbs in ancient traditional Chinese medicine formulas and modern publications predict therapeutic targets for diabetes mellitus?**. *J Ethnopharmacol* (2018) **213** 101-10. PMID: 29102765
7. Zhao X, Chen Z, Yin Y. **Effects of polysaccharide from Physalis Alkekengi var. Francheti on liver injury and intestinal microflora in type-2 diabetic mice.**. *Pharm Biol* (2017) **55** 2020-5. PMID: 28832229
8. Zhang BL, Wu MH. *Internal Medicine in Chinese Medicine* (2017) 471-474
9. Ma D, Deng DQ, Zhang YQ. **Effect of Xiaoke Jianpi capsule on islet beta cell function in type 2 diabetes mellitus with spleen deficiency and dampness abundance.**. *World J Integr Tradit West Med* (2019) **14** 1333-1335
10. Ningning M. **Efficacy of Shen-ling-Bai-zhu-powder addition and subtraction in the treatment of obese type 2 diabetes mellitus with evidence of spleen deficiency and dampness.**. *Guiding J Tradit Chin Med Pharm* (2017) **23** 74-6
11. Zhang T, Gai Y, Feng W. **study on the regulation of adiponectin and gastrointestinal hormones in obese type 2 diabetes mellitus with spleen deficiency and dampness obstruction by yun-pi-huo-xue-fang.**. *Liaoning J Tradit Chin Med* (2016) **43** 82-6
12. Luo L, Liu C, Huang Y. **Clinical effect of modified Linggui Zhugan Decoction combined with Met-formin in the treatment of obesity type 2 diabetes mellitus with phlegm-dampness syndrome.**. *China Med Herald* (2021) **18** 127-30
13. Yoshikawa T, Mifune Y, Inui A. **Quercetin treatment protects the Achilles tendons of rats from oxidative stress induced by hyperglycemia.**. *BMC Musculoskelet Disord* (2022) **23** 563. PMID: 35689230
14. Resham K, Khare P, Bishnoi M. **Neuroprotective effects of isoquercitrin in diabetic neuropathy via Wnt/β-catenin signaling pathway inhibition.**. *BioFactors (Oxford, England)* (2020) **46** 411-20. PMID: 31960520
15. Yao Z, Gu Y, Zhang Q. **Estimated daily quercetin intake and association with the prevalence of type 2 diabetes mellitus in Chinese adults.**. *Eur J Nutr* (2019) **58** 819-30. PMID: 29754250
16. Li D, Jiang C, Mei G. **Quercetin alleviates ferroptosis of pancreatic beta cells in type 2 diabetes.**. *Nutrients* (2020) **12** 29542954
17. Li Y, Zheng D, Shen D. **Protective effects of two safflower derived compounds, kaempferol and hydroxysafflor yellow A, on hyperglycaemic stress-induced podocyte apoptosis via modulating of macrophage M1/M2 polarization.**. *J Immunol Res* (2020) **2020** 2462039. PMID: 33102606
18. Varghese GK, Bose LV, Habtemariam S. **Antidiabetic components of Cassia alata leaves: identification through α-glucosidase inhibition studies.**. *Pharm Biol* (2013) **51** 345-9. PMID: 23137344
19. Annadurai T, Thomas PA, Geraldine P. **Ameliorative effect of naringenin on hypergly-cemia-mediated inflammation in hepatic and pancreatic tissues of Wistar rats with streptozotocin-nicotinamide-induced experimental diabetes mellitus.**. *Free Radic Res* (2013) **47** 793-803. PMID: 23841752
20. Luo Z, Fu C, Li T. **Hypoglycemic effects of licochalcone a on the streptozotocin-induced diabetic mice and its mechanism study.**. *J Agric Food Chem* (2021) **8** 2444-56
21. Islam MN, Rabby MG, Hossen MM. **In silico functional and pathway analysis of risk genes and SNPs for type 2 diabetes in Asian population.**. *PLoS One* (2022) **17** e0268826. PMID: 36037214
22. Zhao M, Tan Y, Peng Q. **IL-6/STAT3 pathway induced deficiency of RFX1 contributes to Th17-dependent autoimmune diseases via epigenetic regulation.**. *Nat Commun* (2018) **9** 1-14. PMID: 29317637
23. Abo El-Nasr NME, Saleh DO, Mahmoud SS. **Olmesartan attenuates type 2 diabetes-associated liver injury: cross-talk of AGE/RAGE/JNK, STAT3/SCOS3 and RAS signaling pathways.**. *Eur J Pharmacol* (2020) **874** 10-1
24. Xu L, Li Y, Yin L. **miR-125a-5p ameliorates hepatic glycolipid metabolism disorder in type 2 diabetes mellitus through targeting of STAT3.**. *Theranostics* (2018) **8** 5593-609. PMID: 30555566
25. Gao X, Li Y, Wang H. **Inhibition of HIF-1α decreases expression of pro-inflammatory IL-6 and TNF-α in diabetic retinopathy.**. *Acta Ophthalmol* (2017) **95** e746-50. PMID: 27288252
26. Zhang T, Huang J, Yi Y. **AKT serine/threonine kinase 1 regulates de novo fatty acid synthesis through the mammalian target of rapamycin/sterol regulatory element binding protein 1 axis in dairy goat mammary epithelial cells.**. *J Agric Food Chem* (2018) **66** 1197-205. PMID: 29323924
27. Lu M, Wan M, Leavens KF. **insulin regulates liver metabolism in vivo in the absence of hepatic Akt and Foxo1.**. *Nat Med* (2012) **18** 388-95. PMID: 22344295
28. Lee Y, Dominy JE, Choi YJ. **Cyclin D1-Cdk4 controls glucose metabolism independently of cell cycle progression.**. *Nature* (2014) **510** 547-51. PMID: 24870244
29. Walke PB, Bansode SB, More NP. **Molecular investigation of glycated insulin-induced insulin resistance via insulin signaling and AGE-RAGE axis.**. *Biochim Biophys Acta Mol Basis Dis* (2021) **1867** e166029
30. Kailash P, Manish M. **AGE-RAGE stress, stressors, and antistressors in health and disease.**. *Int J Angiol* (2018) **27** 1-12. PMID: 29483760
31. Mcnair E, Qureshi M, Prasad K. **Atherosclerosis and the hypercholesterolemic AGE-RAGE axis.**. *Int J Angiol* (2016) **25** 110-6. PMID: 27231427
32. Li Y, Wang X, Jiang F. **Serum lipid levels in relation to clinical outcomes in pregnant women with gestational diabetes mellitus: an observational cohort study.**. *Lipids Health Dis* (2021) **20** 125. PMID: 34587947
33. Li ZR, Jia RB, Luo D. **The positive effects and underlying mechanisms of Undaria pinnatifida polysaccharides on type 2 diabetes mellitus in rats.**. *Food Funct* (2021) **12** 11898-912. PMID: 34739010
34. Tian L, Ding L, Wang G. **Qizhi kebitong formula ameliorates streptozocin-induced diabetic osteoporosis through regulating the PI3K/Akt/NF-κ B Pathway.**. *Biomed Res Int* (2022) **2022** 4469766. PMID: 36046447
35. Packer M. **Mechanisms leading to differential hypoxia-inducible factor signaling in the diabetic kidney: modulation by SGLT2 inhibitors and hypoxia mimetics.**. *Am J Kidney Dis* (2021) **77** 280-6. PMID: 32711072
36. Li G, Lu WH, Ai R. **The relationship between serum hypoxia-inducible factor 1alpha and coronary artery calcification in asymptomatic type 2 diabetic patients.**. *Cardiovasc Diabetol* (2014) **13** 52. PMID: 24564828
|
---
title: 'Near-infrared venous imaging may be more useful than ultrasound guidance for
novices to obtain difficult peripheral venous access: A crossover simulation study'
authors:
- Shinichiro Sekiguchi
- Kiyoshi Moriyama
- Joho Tokumine
- Alan Kawarai Lefor
- Harumasa Nakazawa
- Yasuhiko Tomita
- Tomoko Yorozu
journal: Medicine
year: 2023
pmcid: PMC10036034
doi: 10.1097/MD.0000000000033320
license: CC BY 4.0
---
# Near-infrared venous imaging may be more useful than ultrasound guidance for novices to obtain difficult peripheral venous access: A crossover simulation study
## Background:
Difficult peripheral venous access, especially in obese people, is challenging for novices. We conducted a randomized cross-over study to examine whether near-infrared venous imaging or ultrasound guidance is more useful for novice operators to obtain difficult peripheral venous access.
### Methods:
Medical students were recruited as participants. After receiving basic training using commercial simulators, participants were randomly assigned to obtain simulated venous access using a difficult venous access simulator with near-infrared venous imaging or ultrasound guidance in a randomized cross-over design. A difficult venous access simulator was newly developed with deep and narrow vessels to simulate an obese patient. The primary outcome measure of the study was the first-time success rate (%), and the secondary outcome measures included procedure time (seconds) and the number of 3 consecutive successful attempts, to represent proficiency with the procedure. Pearson chi-square test, the Wilcoxon signed-rank test, and generalized estimating equations were used for statistical analysis.
### Results:
Forty-one medical students with no experience performing peripheral venous access were enrolled in this study. The rate of successful first attempts did not differ between the 2 groups ($70\%$ for near-infrared; $65\%$ for ultrasound guidance; $$P \leq .64$$). The duration of the procedure for the first attempt was significantly shorter using near-infrared imaging (median: 14; interquartile range: 12–19) compared to ultrasound guidance (median 46; interquartile range: 26–52; $$P \leq .007$$). The number of attempts until 3 consecutive successes was not significantly different comparing the 2 approaches (near-infrared: 3 (3, 7.25), ultrasound guidance: 3 (3, 6.25), $$P \leq .63$$).
### Conclusion:
There was no difference in success rate of first-time attempts or acquiring proficiency for the 2 methods. However, duration of the first attempt was significantly shorter with near-infrared imaging than with ultrasound guidance. Near-infrared imaging may require less training than ultrasound guidance. Near-infrared venous imaging may be useful for novices to obtain difficult peripheral venous access in obese patients.
## 1. Introduction
Peripheral venous catheters are the most commonly used devices for vascular access in routine clinical practice. Obesity is associated with difficult peripheral venous access.[1] The number of obese people is increasing rapidly worldwide.[2] *If this* trend continues, difficulties obtaining peripheral venous access in clinical practice will also increase. Recently, the usefulness of ultrasound guidance to obtain peripheral venous access has been reported.[3] However, special training is required to master the technique for ultrasound-guided vascular access.[4,5] Another report states that imaging peripheral blood vessels with near-infrared light facilitates peripheral venous access.[6] This study examined whether ultrasound guidance or near-infrared light is more useful for novice operators to obtain difficult peripheral venous access due to obesity.
## 2. Materials and methods
The present study was approved by the Faculty of Medicine Research Ethics Committee, Kyorin University (approval number 1804) and registered in the University Hospital Medical Information Network Center Clinical Registration System (UMIN000045269). The study was conducted in accordance with Consolidated Standards of Reporting Trials guidelines.
The study was designed as a randomized, prospective crossover study. Participants were recruited from among fourth and fifth-year medical students as volunteers. Exclusion criteria included experience of obtaining peripheral venous access in a patient, and refusal to participate. Written informed consent was obtained from all participants. Participant recruitment and data collection were performed from September 2021 to November 2022.
## 2.1. Simulation training
Before starting the study, simulation training was conducted. The Intravenous ARM III (Kyoto Kagaku Co., Japan) simulator was used to train the direct visual and palpation techniques for peripheral venous access for 30 minutes. The peripheral venous catheter was BD Insight 22G, 25 mm (Nippon Becton Dickinson Co., Japan). A brief explanation of the principles of near-infrared venous imaging was given, and each participant searched for a vein in his or her own arm using near-infrared vein imaging. The near-infrared visualization system was Vein Viewer Flex (Terumo Co., Japan).
Next, the principles of ultrasound guidance were briefly explained, followed by a demonstration of ultrasound imaging of veins in the upper extremities. Ultrasound imaging was performed with a Sonotore Linear (linear probe 7.5 MHz, ALFABIO Co., Japan). Out-of-plane (dynamic needle tip positioning[7]) and in-plane approaches were introduced as ultrasound vascular access techniques.[4,5,8] A 1-hour practical training session was held using the UGP GEL (AGL800, Alfabio Co., Japan).[9–11] This simulator uses a simulated blood vessel 5 mm in diameter and 5 mm deep, embedded in an ultrasound-transparent gel. After simulation training, participants were randomly assigned, using a sealed envelope system, to perform simulated vascular access using near-infrared vein imaging or ultrasound guidance using a “difficult venous access simulator.”
## 2.2. Difficult venous access simulator
A simulator was developed especially for this study (Kyoto Kagaku Co., Japan) (Fig. 1). The new simulator used a mixture of urethane gel and epoxy resin (mainly fatty acid esters) to simulate the soft tissues of a limb. In addition, urethane paint was applied to the surface to reduce stickiness. Simulated blood vessels were made with condensed silicone rubber and painted with black pigment on the surface. In this simulator, blood vessels with an inner diameter of 3 mm and a depth of 5 mm are simulated by embedding them in a material that transmits near-infrared light and ultrasound waves. The blood vessels cannot be seen with the naked eye in this simulator. The inner diameter of the simulated blood vessel in this simulator is smaller (3 mm in diameter) than in the simulator used for training, making the puncture more difficult.
**Figure 1.:** *Difficult venous access simulator. (a) Simulated vessels cannot be seen in the simulator with the eye. (b) Near-infrared imaging with the simulator, allowing 1 to see simulated vessels. (c) Ultrasound imaging with the simulator providing a view of simulated vessels.*
Punctures continued until 3 consecutive successful attempts were made. Even if 3 consecutive successful attempts were not achieved, the trial was terminated after 20 attempts. Three consecutive attempts were hypothesized to represent proficiency with the technique. Success was defined when water in the lumen of the simulated vessel was aspirated from the catheter after placement. Failure was defined as failure to place the catheter into the simulated vessel within 3 minutes, or not to achieve 3 consecutive successes within 20 attempts. The procedure time was defined as the time from the start of puncture to confirmation of water backflow from the simulated vessel. Participants performed the puncture using the assigned method (near-infrared or ultrasound), then performed using the other method (crossover study).
The primary outcome measure of the study was the rate of successful first attempts, and the secondary outcome measures included procedure time and the number of 3 consecutive successful attempts. Data were electronically stored in an anonymized manner.
## 2.3. Statistical analysis
Success rates were expressed as percentages (%). Procedure times (seconds) and number of 3 consecutive successes were shown as median (first quartile, third quartile). The presence or absence of carryover effects was evaluated with the Mann–Whitney U test. Pearson chi-square test was used to compare success rates. The Wilcoxon signed-rank test was used to compare procedure times. Three consecutive successes were analyzed with the generalized estimating equations, using a binomial distribution, and logit for link function. Data was analyzed with EZR statistical software (Saitama Medical Center, Jichi Medical University, Saitama, Japan).[12] A P value <.05 was considered statistically significant.
## 2.4. Power analysis
At the time the present study began, there were no comparative studies comparing near-infrared vein imaging and ultrasound guidance in adult patients. The clinical difference between the 2 techniques could not be determined. Therefore, assuming a clinically significant difference, the sample size required for $80\%$ power at ɑ = 0.05 was estimated to be 30 participants. In this study, 40 medical students were enrolled to account for possible drop-outs or exclusions.
## 3. Results
Forty-one medical students participated in the present study. One participant data was excluded due to missing data (refusal to continue participation due to fatigue, Fig. 2). The carryover effect was not statistically significant (success rates: 0.70, procedure time for the first attempt: 0.59, 3 consecutive successful attempts: 0.72). The first attempt success rate comparing near-infrared venous imaging and ultrasound guidance showed no significant difference (near-infrared: $70\%$, ultrasound $65\%$, $$P \leq .64$$). Procedure time for the first attempt was significantly shorter with near-infrared vein imaging compared to ultrasound guidance (near-infrared vein: 14 [12, 19], ultrasound: 45 [26, 52], $$P \leq .007$$). The numbers of 3 consecutive successful attempts were not significantly different between the 2 methods (near-infrared vein imaging: 3 (3, 7.25), ultrasound guidance: 3 (3, 6.25), generalized estimating equations: comparing odds ratios, $$P \leq .63$$).
**Figure 2.:** *CONSORT diagram. CONSORT = consolidated standards of reporting trials.*
## 4. Discussion and conclusions
In the present study, we investigated whether near-infrared venous imaging or ultrasound guidance facilitates novices in obtaining difficult peripheral venous access using a simulated vein. The results show that obtaining difficult vascular access with near-infrared venous imaging was faster than when using ultrasound guidance, with no difference in success rate or proficiency using the technique. The simulator used in this study (a difficult venous access simulator) was designed to mimic the conditions under which veins are invisible to the naked eye, similar to the situation in some obese patients. The authors believe that near-infrared venous imaging may be useful for securing peripheral venous access in obese patients. Near-infrared venous imaging also has the advantage of requiring less training than ultrasound guidance.
Near-infrared venous imaging was expected to improve the success rate for obtaining peripheral venous access in children.[13–21] Most studies show that the effectiveness of near-infrared venous imaging is limited. Curtis et al compared near-infrared venous imaging and ultrasound guidance in children and found no advantage for either method over conventional device placement.[22] Aulagnier et al studied the usefulness of near-infrared venous imaging in adults and found no benefit.[23] Kumar et al also studied adult patients and showed that near-infrared venous imaging increased only the success rate for the initial attempt.[24] Recently, Yalçinli et al compared near-infrared venous imaging, ultrasound guidance, and conventional methods in adult patients and found that ultrasound guidance had a higher initial attempt success rate than near-infrared venous imaging or conventional methods.[25] In that study, participants were knowledgeable and experienced in ultrasound guidance and near-infrared vein imaging before the study. Furthermore, participants were assessed for mastery of skills prior to data collection, and if the skill did not reach a predefined “mastery level,” an additional 4 hours of teaching, including simulation training, was added. The study by Yalçinli et al showed that with adequate clinical and simulation training, ultrasound guidance may have a higher success rate than other methods. However, the present study shows that near-infrared venous imaging is more useful than ultrasound guidance for novices. A limitation of the present study is the difference in fidelity between the simulator and a patient. The influence of simulator fidelity requires further study.
Cannulating veins using ultrasound guidance requires appropriate training and improves with experience. However, no consensus of a clear goal for technical skills acquisition has been established. In the present study, the simulation training given to participating medical students had the same content as preclinical basic training for residents.[9–11] The education is a competency-based modular system designed to allow participants to acquire the basic skills for vascular access, which is constructed in a step-by-step manner.[9] Certain aspects of the effectiveness of this education system have been proven.[10,11] However, this system targets the acquisition of basic skills and does not guarantee success in more difficult vascular access. It was hypothesized that ultrasound guidance after simulation training using this education system would have a higher success rate and improved skills acquisition compared to near-infrared venous imaging, but the results show that this hypothesis was rejected. The development of skill-acquisition programs for ultrasound-guided peripheral venous cannulation in difficult cases may require advanced skills and the need to set higher goals for skill acquisition.
Ultrasound guidance has a great advantage over near-infrared venous imaging in that it can identify deeper veins. Near-infrared venous imaging has a limitation in its ability to locate deep veins, which cannot be observed over 5 mm deep. For this reason, ultrasound guidance is more clinically useful than near-infrared venous imaging. However, appropriate education is critical for this purpose. Near-infrared venous imaging may be useful as a method that requires less education and a useful way for novices to gain experience.
Many studies have evaluated the teaching methods used for ultrasound guided venous access, but there is still no method accepted as the optimal approach.[26] In conclusion, near-infrared venous imaging may be a useful alternative to ultrasound guidance to obtain difficult peripheral venous access in obese patients, especially for novices.
## Acknowledgments
The authors thank Ms. Okada (Laboratory assistant, Division of Biological Function Research) for her assistance.
## Author contributions
Conceptualization: Shinichiro Sekiguchi, Joho Tokumine.
Data curation: Shinichiro Sekiguchi, Joho Tokumine.
Formal analysis: Shinichiro Sekiguchi, Kiyoshi Moriyama, Harumasa Nakazawa.
Investigation: Shinichiro Sekiguchi, Joho Tokumine.
Methodology: Shinichiro Sekiguchi, Kiyoshi Moriyama.
Project administration: Kiyoshi Moriyama, Joho Tokumine, Yasuhiko Tomita.
Supervision: Tomoko Yorozu.
Validation: Tomoko Yorozu.
Writing – original draft: Shinichiro Sekiguchi, Kiyoshi Moriyama, Joho Tokumine.
Writing – review and editing: Alan Kawarai Lefor, Kiyoshi Moriyama.
## References
1. Sebbane M, Claret PG, Lefebvre S. **Predicting peripheral venous access difficulty in the emergency department using body mass index and a clinical evaluation of venous accessibility.**. *J Emerg Med* (2013) **44** 299-305. PMID: 22981661
2. Abarca-Gómez L, Abdeen ZA, Hamid ZA. **Worldwide trends in body-mass index, underweight, overweight, and obesity from 1975 to 2016: a pooled analysis of 2416 population-based measurement studies in 128·9 million children, adolescents, and adults.**. *Lancet* (2017) **390** 2627-42. PMID: 29029897
3. van Loon FHJ, Buise MP, Claassen JJF. **Comparison of ultrasound guidance with palpation and direct visualisation for peripheral vein cannulation in adult patients: a systematic review and meta-analysis.**. *Br J Anaesth* (2018) **121** 358-66. PMID: 30032874
4. Lamperti M, Biasucci DG, Disma N. **European Society of Anaesthesiology guidelines on peri-operative use of ultrasound-guided for vascular access (PERSEUS vascular access).**. *Eur J Anaesthesiol* (2020) **37** 344-76. PMID: 32265391
5. Moureau N, Lamperti M, Kelly LJ. **Evidence-based consensus on the insertion of central venous access devices: definition of minimal requirements for training.**. *Br J Anaesth* (2013) **110** 347-56. PMID: 23361124
6. Zhang Z, Wang X, Zhang L. **Infrared vein imaging for insertion of peripheral intravenous catheter for patients requiring isolation for severe acute respiratory syndrome coronavirus 2 infection: a nonrandomized clinical trial.**. *J Emerg Nurs* (2022) **48** 159-66. PMID: 35115182
7. Bai B, Tian Y, Zhang Y. **Dynamic needle tip positioning versus the angle-distance technique for ultrasound-guided radial artery cannulation in adults: a randomized controlled trial.**. *BMC Anesthesiol* (2020) **20** 231. PMID: 32928119
8. **Practical guide for safe central venous catheterization and management 2017.**. *J Anesth* (2020) **34** 167-86. PMID: 31786676
9. Asao T. *CVC Instructor’s Guide ver. 4* (2018)
10. Kikuchi M, Asao T, Tokumine J. **A novel system for teaching the in-plane vascular access technique: a simulation study.**. *Medicine (Baltim)* (2021) **100** e27201
11. Sugiki D, Matsushima H, Asao T. **A web-based self-learning system for ultrasound-guided vascular access.**. *Medicine (Baltim)* (2022) **101** e31292
12. Kanda Y. **Investigation of the freely available easy-to-use software “EZR” for medical statistics.**. *Bone Marrow Transplant* (2013) **48** 452-8. PMID: 23208313
13. Simhi E, Kachko L, Bruckheimer E. **A vein entry indicator device for facilitating peripheral intravenous cannulation in children: a prospective, randomized, controlled trial.**. *Anesth Analg* (2008) **107** 1531-5. PMID: 18931210
14. Chapman LL, Sullivan B, Pacheco AL. **VeinViewer-assisted Intravenous catheter placement in a pediatric emergency department.**. *Acad Emerg Med* (2011) **18** 966-71. PMID: 21854488
15. Kim MJ, Park JM, Rhee N. **Efficacy of VeinViewer in pediatric peripheral intravenous access: a randomized controlled trial.**. *Eur J Pediatr* (2012) **171** 1121-5. PMID: 22415409
16. Kaddoum RN, Anghelescu DL, Parish ME. **A randomized controlled trial comparing the AccuVein AV300 device to standard insertion technique for intravenous cannulation of anesthetized children.**. *Paediatr Anaesth* (2012) **22** 884-9. PMID: 22694242
17. de Graaff JC, Cuper NJ, Mungra RA. **Near-infrared light to aid peripheral intravenous cannulation in children: a cluster randomised clinical trial of three devices.**. *Anaesthesia* (2013) **68** 835-45. PMID: 23763614
18. Cuper NJ, de Graaff JC, Verdaasdonk RM. **Near-infrared imaging in intravenous cannulation in children: a cluster randomized clinical trial.**. *Pediatrics* (2013) **131** e191-7. PMID: 23230072
19. Rothbart A, Yu P, Müller-Lobeck L. **Peripheral intravenous cannulation with support of infrared laser vein viewing system in a pre-operation setting in pediatric patients.**. *BMC Res Notes* (2015) **8** 463. PMID: 26391665
20. Park JM, Kim MJ, Yim HW. **Utility of near-infrared light devices for pediatric peripheral intravenous cannulation: a systematic review and meta-analysis.**. *Eur J Pediatr* (2016) **175** 1975-88. PMID: 27785562
21. Saju AS, Prasad L, Reghuraman M. **Use of vein-viewing device to assist intravenous cannulation decreases the time and number of attempts for successful cannulation in pediatric patients.**. *Paediatr Neonatal Pain* (2019) **1** 39-44. PMID: 35548377
22. Curtis SJ, Craig WR, Logue E. **Ultrasound or near-infrared vascular imaging to guide peripheral intravenous catheterization in children: a pragmatic randomized controlled trial.**. *CMAJ* (2015) **187** 563-70. PMID: 25897047
23. Aulagnier J, Hoc C, Mathieu E. **Efficacy of AccuVein to facilitate peripheral intravenous placement in adults presenting to an emergency department: a randomized clinical trial.**. *Acad Emerg Med* (2014) **21** 858-63. PMID: 25176152
24. Kumar A, Negi M, Khanka J. **Initial experience with use of infrared assistance for intravenous injection of radiopharmaceuticals.**. *World J Nucl Med* (2020) **20** 172-5. PMID: 34321970
25. Yalçinli S, Karbek Akarca F, Can O. **Comparison of standard technique, ultrasonography, and near-infrared light in difficult peripheral vascular access: a randomized controlled trial.**. *Prehosp Disaster Med* (2022) **37** 65-70. PMID: 34865664
26. Okano H, Mayumi T, Kataoka Y. **Outcomes of simulation-based education for vascular access: a systematic review and meta-analysis.**. *Cureus* (2021) **13** e17188. PMID: 34414052
|
---
title: Long-term weight loss outcome of laparoscopic Roux-en-Y gastric bypass predicted
by weight loss at 6 months in Chinese patients with BMI ≥ 32.5 kg/m2
authors:
- Qiqige Wuyun
- Dezhong Wang
- Chenxu Tian
- Guangzhong Xu
- Buhe Amin
- Dongbo Lian
- Dexiao Du
- Weihua Zhang
- Min Jiang
- Guanyang Chen
- Nengwei Zhang
- Liang Wang
journal: Medicine
year: 2023
pmcid: PMC10036043
doi: 10.1097/MD.0000000000033235
license: CC BY 4.0
---
# Long-term weight loss outcome of laparoscopic Roux-en-Y gastric bypass predicted by weight loss at 6 months in Chinese patients with BMI ≥ 32.5 kg/m2
## Abstract
Laparoscopic Roux-en-Y gastric bypass (LRYGB) is classic bariatric procedure with long-term safety and efficacy. However, no studies have focused on predicting long-term weight loss after LRYGB in Chinese patients with body mass index (BMI) ≥ 32.5 kg/m2. To explore the relationship between initial and long-term weight loss after LRYGB in patients with BMI ≥ 32.5 kg/m2. All patients were followed-up to evaluate BMI, percentage of excess weight loss (%EWL), and comorbidities. Linear and logistic regression were performed to assess the relationship between initial and long-term weight loss. Receiver operating characteristic curve was used to determine optimal cutoff value. We enrolled 104 patients. The median preoperative BMI was 41.44 (37.92–47.53) kg/m2. % EWL ≥ $50\%$ at 5 years was considered as successful weight loss, and $75.00\%$ of the patients successfully lost weight. The cure rates of hypertension, hyperlipidemia, and type 2 diabetes mellitus at 1 year were $84.38\%$, $33.93\%$, and $60.82\%$, respectively. % EWL at 6 months and 5 years were positively correlated and its relationship could be described by following linear equation: %EWL5 years = 43.934 + 0.356 × %EWL6 months ($P \leq .001$; r2 = 0.166). The best cutoff %EWL at 6 months after LRYGB to predict 5-year successful weight loss was $63.93\%$ (sensitivity, $53.85\%$; specificity, $84.62\%$; area under the curve (AUC) = 0.671). In Chinese patients with BMI ≥ 32.5 kg/m2, %EWL at 6 months and 5 years were positively correlated and %EWL at 5 years could be calculated by following linear equation: %EWL5 years = 43.934 + 0.356 × %EWL6 months.
## 1. Introduction
Laparoscopic Roux-en-Y gastric bypass (LRYGB) is a surgical procedure conventionally performed to treat obesity-related complications while reducing the patient’s weight. It has reliable efficacy in patients with type 2 diabetes mellitus (T2DM).[1–7] Although the number of laparoscopic sleeve gastrectomy (LSG) procedures is increasing every year and LSG is accounting for an increasing proportion of all bariatric procedures, LRYGB is widely recommended for patients with body mass index (BMI) ≥ 40 or ≥ 35 kg/m2 with severe obesity-related comorbidities, particularly T2DM.[5] Initial weight loss indicators are used to predict weight loss after LSG.[8–11] Timely prediction of patients who are unlikely to maintain long-term weight loss can help design more targeted lifestyle, medical, and behavioral interventions as early as possible to improve the prognosis and increase surgical satisfaction of these patients. However, in Chinese patients with BMI ≥ 32.5 kg/m2 undergoing LRYGB, weight loss at 5 years is challenging to predict.
The aim of the present study was to explore the association between initial weight loss indicators after LRYGB and weight loss at 5 years in Chinese patients with BMI ≥ 32.5 kg/m2 and establish a predictive model.
## 2.1. Inclusion and exclusion criteria
In this retrospective study, we enrolled 104 patients who underwent LRYGB at our hospital from October 2014 to April 2017.
## 12.2. Exclusion criteria were.
The Ethics Review Board of our hospital approved the study protocol (sjtky11-1x-2022 [076]). Written informed consent was obtained from all participants. All procedures performed in studies involving human participants were in accordance with the ethical standards of the 1964 Helsinki declaration and its later amendments or comparable ethical standards.
## 2.2. Definitions of relative terms
Based on our previous reports,[8,12] the diagnostic criteria and curative outcome were defined for comorbidities, including T2DM, hypertension, and hyperlipidemia.
Hypertension in these patients was defined as SBP ≥ 140 mm Hg or DBP ≥ 90 mm Hg, and remission was defined by blood pressure values below these thresholds without the use of antihypertensive medication at 1-year post-LRYGB. T2DM was defined by a hemoglobin A1c level of ≥ $6.5\%$, with remission being defined by an hemoglobin A1c of < $6.0\%$ without the use of insulin or oral hypoglycemic drugs at 1-year post-LSG. Lipid profiles were analyzed by the determination of fasting total cholesterol and triglyceride levels, together with the cholesterol to high-density lipoprotein ratios. Hyperlipidemic remission was defined by the reduction of these values to fall within normal ranges at 1-year post-LRYGB without the need for lipid-lowering drugs. Postoperative percentage of excess weight loss (%EWL) ≥ $50\%$ was regarded as weight loss success, as is common. The long-term weight loss outcome was defined as the weight loss outcome at 5 years postoperatively.
## 2.3. Operative techniques and postoperative follow-up
The operative techniques and postoperative follow-up were described in our previous studies.[8,12] Three surgeons performed all LRYGB procedures in enrolled patients using a standardized approach. We first constructed a small pouch (approximately 20 mL volume) at the proximal end of the stomach, and the fundus of the stomach was completely isolated. We then transected the proximal jejunum 70 to 120 cm from the Treitz ligament. After anastomosis of the distal end of the jejunum to the small gastric pouch, the afferent biliopancreatic limb was anastomosed to the jejunum 100 to 150 cm distally. The total length of the biliopancreatic branch and the food branch was approximately 250-cm.
Upper gastrointestinal contrast studies were conducted postoperatively on day 1 to 3. When no abnormalities were detected, patients were allowed to freely consume water and were given a liquid diet. The patients can be discharged without any symptoms such as vomiting. The follow-up was completed via phone or WeChat for outpatients and at the hospital for inpatients. Follow-up at 3, 6, 12, 36, and 60 months after LRYGB to assess various physiological parameters of the patients.
## 2.4. Statistical analysis
Nominal data were analyzed using the Fisher exact or chi-squared test. Normally distributed data are expressed as mean ± standard deviation, while non-normally distributed are expressed as median (interquartile range). The independent-samples t test and nonparametric test were performed to compare normally and non-normally distributed data, respectively. Based on initial weight loss, linear and binary logistic regression analyses were employed to evaluate differences in weight loss obtained 5 years after LRYGB. The receiver operating characteristic (ROC) curve was used to determine the optimal cutoff values for initial weight loss indicators. A 2-sided P value <.05 was considered to indicate statistical significance. ROC curves were drawn using MedCalc version 19.2.6 (MedCalc, Inc., Mariakerke, Belgium). Histograms were plotted using Graphpad Prism version 8.4.3 (GraphPad Software, San Diego, CA). Other statistical tests were performed using SPSS version 20.0 (IBM Corp., Armonk, NY).
## 3. Results
We enrolled a total of 104 patients with age and preoperative BMI of 41.71 ± 11.87 years and 41.44 (37.92–47.53) kg/m2, respectively. Preoperatively, $93.27\%$ of the patients had T2DM. Table 1 lists the preoperative patient’s characteristics.
**Table 1**
| Variable | Value |
| --- | --- |
| Age (yr) | 41.71 ± 11.87 * |
| Gender | |
| Male (%) | 51 (49.04%) † |
| Female (%) | 53 (50.96%) |
| The ethnic | |
| Han (%) | 98 (94.23%) |
| Other ethnics (%) | 6 (5.77%) |
| Residence | |
| Urban (%) | 78 (75.00%) |
| Rural (%) | 26 (25.00%) |
| North & South | |
| North (%) | 94 (90.38%) |
| South (%) | 10 (9.62%) |
| Preoperative BMI (kg/m2) | 41.44 (37.92, 47.53) ‡ |
| T2DM (%) | 97 (93.27%) |
| Hypertension (%) | 32 (30.77%) |
| Hyperlipidemia (%) | 56 (53.85%) |
| Hyperuricemia (%) | 79 (75.96%) |
| Fatty liver (%) | 96 (92.31%) |
| OSA (%) | 104 (100.00%) |
| PCOS (%) | 31 (58.49%) |
None of the patients developed serious postoperative complications, such as death, fistula, stenosis, or bleeding. The remission rates of hypertension, hyperlipidemia, and T2DM 1 year postoperatively were $84.38\%$, $33.93\%$, and $90.77\%$, respectively. The preoperative and 3- and 6-month and 1-, 2-, 3-, 4-, and 5-year postoperative BMIs were 41.44 (37.92–47.53), 34.58 (29.93–40.25), 30.44 (26.58–36.40), 28.78 (25.13–32.87), 29.28 (26.58–33.76), 29.62 (26.88–33.06), 29.37 (27.35–32.95), and 29.19 (27.10–33.36) kg/m2, respectively. The %EWLs at 3 and 6 months and 1, 2, 3, 4, and 5 years were 39.51 (27.08–52.00), 57.22 (43.46–75.39), 68.20 (54.21–85.42), 65.37 (52.46–78.90), 62.98 (51.69–78.06), 66.14 (53.26–78.79), and 66.15 (50.17–80.19), respectively. Figure 1 shows the trends in BMI and %EWL over time. At 5 years, 78 ($75.00\%$) patients achieved successful weight loss, whereas 26 ($25.00\%$) patients did not. Weight loss at 5 years was 38.93 (12.5–121.6) kg.
**Figure 1.:** *Trends in the BMI (A) and %EWL (B) over time. (Note: Each point in the figure represents a patient, and the red horizontal line represents the median). %EWL = percentage of excess weight loss.*
Table 2 compares the baseline data between patients with and without successful weight loss. Baseline data did not differ significantly between patients with and without successful weight loss. The binary logistic regression analysis performed with successful weight loss at 5 years as the dependent variable and %EWLs at 3 and 6 months as the independent variables showed that only the 6-month postoperative %EWL was a statistically significant independent variable (Table 3). Therefore, 6-month postoperative %EWL was used for the predictive analysis.
To further clarify the relationship between %EWL at 6 months and long-term weight loss, the linear regression model developed using %EWL at 5 years as the dependent variable and %EWL at 6 months as the independent variable showed a significant positive correlation between %EWLs at 5 years and 6 months ($P \leq .001$; r2 = 0.166). Figure 2 shows the histograms and probability–probability plots of standardized residuals. The following linear equation explained the relationship between initial and long-term postoperative weight loss:
**Figure 2.:** *Histograms (A) and probability–probability plots (B) of standardized residuals for %EWL at 5 years. %EWL = percentage of excess weight loss.*
%EWL5 years = 43.934 + 0.356 × %EWL6 months Postoperative %EWL ≥ $50\%$ was regarded as weight loss success. Weight loss success at 5 years was the categorical variable, and %EWL at 6 months was the variable used to draw the ROC curves. Figure 3 shows the results. The optimal cutoff %EWL at 6 months for predicting the weight loss outcome at 5 years was $63.93\%$ (area under the ROC curve, 0.671; $95\%$ confidence interval: 0.572–0.760; sensitivity, $53.85\%$; specificity, $84.62\%$; $P \leq .001$).
**Figure 3.:** *ROC curve of %EWL at 6 months among patients with successful %EWL at 5 years after LRYGB. %EWL = percentage of excess weight loss, LRYGB = laparoscopic Roux-en-Y gastric bypass, ROC = receiver operating characteristic.*
## 4. Discussion
In this study, the weight loss success rates at 3, 4, and 5 years after LRYGB were $78.85\%$, $80.77\%$, and $75\%$, respectively, and the corresponding BMIs were 29.62 (26.88–33.06), 29.37 (27.35–32.95), and 29.19 (27.10–33.36) kg/m2, respectively. The weight loss success rate suggested that the efficacy of LRYGB was satisfactory. However, only 17 ($16.35\%$) patients had a BMI below the diagnostic criteria for overweight, that is, 25 kg/m2. From this viewpoint, the weight loss seemed to be less than ideal. However, discussing the postoperative BMI alone is inaccurate. The preoperative BMI of the patients in this study exceeded 32.5 kg/m2 at 41.44 (37.92–47.53) kg/m2, and the prevalence rates of T2DM, fatty liver, and obstructive sleep apnea were $93.27\%$, $92.31\%$, and $100\%$, respectively. The patients in this study had morbid obesity, which is more difficult to reduce to normal BMI. Therefore, although only $16.35\%$ of the patients could reduce their BMI to below the overweight criteria, LRYGB could be considered to have a good effect. In many large-scale randomized controlled trials, LRYGB was superior or comparable to LSG in terms of weight loss and management of T2DM.[1,2,13–19] Thus, LRYGB could positively affect weight loss in Chinese patients with BMI ≥ 32.5 kg/m2.
In this study, the prevalence rates of T2DM, hypertension, and hyperlipidemia were $93.27\%$, $30.77\%$, and $53.85\%$, respectively. Obesity-related comorbidities should be treated while reducing weight loss. The remission rates of hypertension, hyperlipidemia, and T2DM 1 year postoperatively were $84.38\%$, $33.93\%$, and $90.77\%$, respectively. Overall, the cure rate of comorbidities with LRYGB was satisfactory. The aforementioned cure rates suggested that LRYGB was significantly effective in treating obesity-related comorbidities. In the past 2 years of clinical practice at our hospital, T2DM was easier to cure in patients with higher preoperative BMI than in those with lower preoperative BMI using surgery, consistent with recent reports.[20,21] Therefore, for patients with low preoperative BMI and T2DM, we have recently gradually adjusted the recommended strategy for bariatric procedures to LRYGB. Du et al[22] reported that the cure rate of T2DM was $80\%$ in patients who underwent LRYGB with preoperative BMI between 27.5 and 32.5 kg/m2. This cure rate of T2DM was less than that in the present study, suggesting that patients with T2DM with low preoperative BMI should undergo LRYGB, which is more effective for T2DM.
In addition, %EWL at 6 months and 5 years were positively correlated (r2 = 0.166), and the relationship could be expressed by the following linear equation: %EWL5 years = 43.934 + 0.356 × %EWL6 months. The prediction effect of the linear regression equation was acceptable. The ROC curve analysis revealed that the optimal cutoff %EWL at 6 months for predicting the weight loss outcome at 5 years was $63.93\%$ (area under the ROC curve, 0.671; sensitivity, $53.85\%$; specificity, $84.62\%$). Although the prediction effect of the ROC curve was unsatisfactory, that of the linear equation was good.
In this study, although the prediction model of the ROC curve was meaningful, it had poor stability, low Youden index, and limited predictive value. There may be 3 reasons. First, the sample size of this study was small, with only 104 cases, which might be insufficient to build a predictive model with a higher Youden index. Second, the total length of the small intestine differed significantly among individuals. After LRYGB was uniformly placed in a 250-cm intestine, the common channel length differed significantly among individuals. This might have resulted in variable weight loss after LRYGB. Third, the size of the gastric pouch constructed by LRYGB was approximately 20 mL, but the gastric pouch size differed among individuals. Further, weight loss in patients with large gastric pouches might be weaker compared to those with small gastric pouches. At 5 years, the specificity of the ROC curve for predicting weight loss success was $84.62\%$. Thus, the predictive value of the ROC curve analysis of %EWL at 6 months to predict successful weight loss at 5 years was satisfactory.
We tend to recommend bariatric procedures to patients at our hospital. First, considering that LRYGB has a more definite effect on T2DM, we recommend LRYGB for patients with T2DM. Second, we strongly recommend LRYGB for patients with a history of T2DM ≥ 5 years, age ≥ 50 years, and poor preoperative pancreatic islet function (fasting C-peptide < 2.0 ng/mL and 2-hour C-peptide not exceedingly thrice the fasting C-peptide level). Furthermore, for patients with a family history of gastric cancer or H pylori-positive or atrophic gastritis, we strongly recommend LSG to prevent carcinogenesis in the stomach. Finally, we recommend LRYGB for patients with ≥ 3 obesity-related comorbidities and LSG for patients with preoperative BMI < 32.5 kg/m2 and T2DM only. We thoroughly educate the patients preoperatively, carefully explain the advantages and disadvantages of each bariatric procedure, and share our recommendations. Subsequently, the patients make the decision themselves.
This study had some limitations. This study is a retrospective study, and selective bias is inevitable. First, we followed-up with 104 Chinese patients with BMI ≥ 32.5 kg/m2 who underwent LRYGB for 5 years. A model for predicting long-term weight loss was obtained by analyzing the follow-up results. However, it was suggested that after LRYGB, the inter-individual differences in the common channel length and gastric pouch size might directly or indirectly affect long-term weight loss. The common channel length and gastric pouch size were large, and confounding factors were difficult to quantify when building predictive models. Second, we only explored the cure rates of obesity-related comorbidities 1 year postoperatively. Whether or not the patients T2DM, hypertension, or hyperlipidemia recurred after 1 year remains unknown.
## 5. Conclusion
In Chinese patients with BMI ≥ 32.5 kg/m2 who underwent LRYGB, %EWL at 6 months was significantly positively correlated with %EWL at 5 years. The relationship could be expressed by the following linear equation: %EWL5 years = 43.934 + 0.356 × %EWL6 months.
## Acknowledgments
The authors would like to thank all the reviewers who participated in the article, as well as MJEditor (www.mjeditor.com) for providing English editing services during the preparation of this manuscript.
## Author contributions
Conceptualization: Qiqige Wuyun, Guangzhong Xu, Nengwei Zhang, Liang Wang.
Data curation: Chenxu Tian, Nengwei Zhang, Liang Wang.
Formal analysis: Qiqige Wuyun, Chenxu Tian.
Funding acquisition: Weihua Zhang, Min Jiang.
Methodology: Dezhong Wang, Dongbo Lian, Nengwei Zhang, Liang Wang.
Project administration: Buhe Amin, Guanyang Chen.
Resources: Qiqige Wuyun, Guangzhong Xu, Dexiao Du, Nengwei Zhang.
Software: Dexiao Du.
Validation: Dezhong Wang.
Visualization: Dezhong Wang.
Writing – original draft: Qiqige Wuyun, Chenxu Tian.
Writing – review & editing: Nengwei Zhang, Liang Wang.
## References
1. Peterli R, Wolnerhanssen BK, Peters T. **Effect of laparoscopic sleeve gastrectomy vs laparoscopic Roux-en-Y gastric bypass on weight loss in patients with morbid obesity: the SM-BOSS randomized clinical trial.**. *JAMA* (2018) **319** 255-65. PMID: 29340679
2. Salminen P, Helmio M, Ovaska J. **Effect of laparoscopic sleeve gastrectomy vs laparoscopic Roux-en-Y gastric bypass on weight loss at 5 years among patients with morbid obesity: the SLEEVEPASS randomized clinical trial.**. *JAMA* (2018) **319** 241-54. PMID: 29340676
3. Han Y, Jia Y, Wang H. **Comparative analysis of weight loss and resolution of comorbidities between laparoscopic sleeve gastrectomy and Roux-en-Y gastric bypass: a systematic review and meta-analysis based on 18 studies.**. *Int J Surg* (2020) **76** 101-10. PMID: 32151750
4. Misra S, Bhattacharya S, Saravana Kumar S. **Long-term outcomes of laparoscopic sleeve gastrectomy from the Indian subcontinent.**. *Obes Surg* (2019) **29** 4043-55. PMID: 31385151
5. Buchwald H, Estok R, Fahrbach K. **Weight and type 2 diabetes after bariatric surgery: systematic review and meta-analysis.**. *Am J Med* (2009) **122** 248-256.e5. PMID: 19272486
6. Magro DO, Ueno M, Coelho-Neto JS. **Long-term weight loss outcomes after banded Roux-en-Y gastric bypass: a prospective 10-year follow-up study.**. *Surg Obes Relat Dis* (2018) **14** 910-7. PMID: 29706496
7. Sjostrom L, Narbro K, Sjostrom CD. **Effects of bariatric surgery on mortality in Swedish obese subjects.**. *N Engl J Med* (2007) **357** 741-52. PMID: 17715408
8. Wang L, Sang Q, Du D. **Early weight loss after laparoscopic sleeve gastrectomy predicts sustained weight maintenance among chinese individuals with a BMI < 35 Kg/m**. *Obes Surg* (2021) **31** 1647-55. PMID: 33392996
9. Chu L, Howell B, Steinberg A. **Early weight loss in adolescents following bariatric surgery predicts weight loss at 12 and 24 months.**. *Pediatr Obes* (2019) **14** e12519. PMID: 30843377
10. Chew CAZ, Tan IJ, Ng HJH. **Early weight loss after laparoscopic sleeve gastrectomy predicts midterm weight loss in morbidly obese Asians.**. *Surg Obes Relat Dis* (2017) **13** 1966-72. PMID: 28764918
11. Obeidat F, Shanti H. **Early weight loss as a predictor of 2-year weight loss and resolution of comorbidities after sleeve gastrectomy.**. *Obes Surg* (2016) **26** 1173-7. PMID: 26428251
12. Sang Q, Wang L, Wuyun Q. **Retrospective comparison of SADI-S versus RYGB in Chinese with diabetes and BMI< 35kg/m**. *Obes Surg* (2021) **31** 5166-75. PMID: 34591261
13. Gronroos S, Helmio M, Juuti A. **Effect of laparoscopic sleeve gastrectomy vs Roux-en-Y gastric bypass on weight loss and quality of life at 7 years in patients with morbid obesity: the SLEEVEPASS randomized clinical trial.**. *JAMA Surg* (2021) **156** 137-46. PMID: 33295955
14. Peterli R, Borbely Y, Kern B. **Early results of the Swiss multi center bypass or sleeve study (SM-BOSS): a prospective randomized trial comparing laparoscopic sleeve gastrectomy and Roux-en-Y gastric bypass.**. *Ann Surg* (2013) **258** 690-95. PMID: 23989054
15. Peterli R, Wolnerhanssen BK, Vetter D. **Laparoscopic sleeve gastrectomy versus Roux-Y-gastric bypass for morbid obesity-3-year outcomes of the prospective randomized Swiss multicenter bypass or sleeve study (SM-BOSS).**. *Ann Surg* (2017) **265** 466-73. PMID: 28170356
16. Shoar S, Saber AA. **Long-term and midterm outcomes of laparoscopic sleeve gastrectomy versus Roux-en-Y gastric bypass: a systematic review and meta-analysis of comparative studies.**. *Surg Obes Relat Dis* (2017) **13** 170-80. PMID: 27720197
17. Wolnerhanssen BK, Peterli R, Hurme S. **Laparoscopic Roux-en-Y gastric bypass versus laparoscopic sleeve gastrectomy: 5-year outcomes of merged data from two randomized clinical trials (SLEEVEPASS and SM-BOSS).**. *Br J Surg* (2021) **108** 49-57. PMID: 33640917
18. Carmona MN, Santos-Sousa H, Lindeza L. **Comparative effectiveness of bariatric surgeries in patients with type 2 diabetes mellitus and BMI ≥25 Kg/m**. *Obes Surg* (2021) **31** 5312-21. PMID: 34611827
19. Chen G, Zhang GX, Peng BQ. **Roux-En-Y gastric bypass versus sleeve gastrectomy plus procedures for treatment of morbid obesity: systematic review and meta-analysis.**. *Obes Surg* (2021) **31** 3303-11. PMID: 33956302
20. Lee WJ, Hur KY, Lakadawala M. **Predicting success of metabolic surgery: age, body mass index, C-peptide, and duration score.**. *Surg Obes Relat Dis* (2013) **9** 379-84. PMID: 22963817
21. Ugale S, Gupta N, Modi KD. **Prediction of remission after metabolic surgery using a novel scoring system in type 2 diabetes - a retrospective cohort study.**. *J Diabetes Metab Disord* (2014) **13** 89. PMID: 25426451
22. Du D, Wang L, Chen W. **Weight loss at six months is the surrogate of long-term treatment outcomes for obese Chinese with a BMI less than 35 kg/m2 undergoing Roux-en-Y gastric bypass.**. *Asian J Surg* (2022) **45:1682–87**
|
---
title: 'Efficacy of traditional Chinese exercises in improving anthropometric and
biochemical indicators in overweight and obese subjects: A systematic review and
meta-analysis'
authors:
- Qianfang Yang
- Fan Wang
- Limin Pan
- Ting Ye
journal: Medicine
year: 2023
pmcid: PMC10036064
doi: 10.1097/MD.0000000000033051
license: CC BY 4.0
---
# Efficacy of traditional Chinese exercises in improving anthropometric and biochemical indicators in overweight and obese subjects: A systematic review and meta-analysis
## Background:
The efficacy of traditional Chinese exercise (TCE)-based intervention in the improvement of anthropometric and biochemical indicators in overweight and obese patients is controversial. In this regard, the aim of this review was to summarize the evidence of TCE interventions to evaluate their effectiveness on the anthropometric and biochemical indicators of overweight and obese patients.
### Method:
Five databases were systematically searched for relevant articles published from inception to October 2022. Randomized controlled trials examining TCE intervention in overweight and obese patients The treatment effects were estimated using a random-effect meta-analysis model with standardized mean differences (Hedges’ g). The categorical and continuous variables were used to conduct moderator analyses. This review was registered in the International Prospective Register of Systematic Reviews (PROSPERO) (identifier CRD42022377632).
### Result:
Nine studies involving a total of 1297 participants were included in the final analysis. In the anthropometric indicators outcomes, the meta-analytic findings revealed large and significant improvements in body mass index ($g = 1.44$, $95\%$ confidence interval [CI] = 1.27–1.61, $$P \leq .000$$, I2 = $99\%$), weight ($g = 1.47$, $95\%$ CI = 1.25–1.68, $$P \leq .000$$, I2 = $95\%$), fat percentage ($g = 1.22$, $95\%$ CI = 0.52–1.93, $$P \leq .000$$, I2 = $93\%$), and small and significant improvements in waist circumference ($g = 0.38$, $95\%$ CI = 0.21–0.54, $$P \leq .000$$, I2 = $99\%$). In the biochemical indicators outcomes, the findings revealed large and significant improvements in low density lipoprotein ($g = 2.08$, $95\%$ CI = 1.80–2.37, $$P \leq .000$$, I2 = $98\%$), moderate and significant improvements in triglyceride ($g = 0.69$, $95\%$ CI = 0.56–0.81, $$P \leq .000$$, I2 = $96\%$), small and significant improvements in total cholesterol ($g = 0.37$, $95\%$ CI = 0.19–0.54, $$P \leq .000$$, I2 = $77\%$), and high-density lipoprotein (g = −0.71, $95\%$ CI = −0.86 to 0.57, $$P \leq .000$$, I2 = $99\%$). The moderator shows that the effects of TCE on anthropometric and biochemical indicators were moderated by frequency of exercise, exercise duration, and type of control group.
### Conclusion:
TCE intervention is a beneficial non-pharmacological approach to improving anthropometric and biochemical indicators in overweight and obese subjects, especially in body mass index, weight, fat percentage, triglyceride, and low-density lipoprotein. The clinical relevance of our findings is pending more extensive trials and more rigorous study designs to strengthen the evidence.
## 1. Introduction
Being overweight or obese is a chronic nutritional disorder that causes excessive accumulation of body fat due to a combination of genetics, environment, and dietary behaviors.[1,2] The World Health Organization (WHO) defines “overweight” as a body mass index (BMI) of 25.0 kg/m2 to 29.9 kg/m2, and “obesity” as a BMI of 30.0 kg/m2. But, there is currently no international standard that applies to all countries or regions; for example, the WHO defines “overweight” as a BMI of 23 kg/m2 to 27.5 kg/m2 for Chinese and a BMI of 27.5 kg/m2 for the “obese” group.[3,4] The global prevalence of overweight or obesity has doubled since 1980, affecting approximately 1 to 3rd of the population.[5] The occurrence and progression of overweight or obesity are closely related to cardiovascular disease, dyslipidemia, and insulin resistance, which in turn lead to a variety of comorbidities and chronic diseases, such as diabetes, stroke, gallstones, fatty liver, etc. In terms of treatment, clinically individualized treatment is adopted, starting from the cause, and includes dietary modification, behavioral intervention, drug treatment, and surgical treatment if necessary.[6] In addition, the active ingredients in many natural plants play an important role in the treatment of obesity. For example, phenolic acids and polyphenols contained in natural plants of Lamiaceae and Rosaceae have been widely used in anti-obesity treatment.[7] However, despite multiple treatment modalities, anthropometric and biochemical markers remain problematic in patients who are overweight or obese, often leading to the development of other comorbidities with serious consequences.[8]Experts recommend exercise and dietary modification therapy as an effective adjunct to the treatment of overweight or obesity because of deficient drug management and practices for overweight or obesity.[9,10]*In a* study investigating the effects of aerobic exercise on body composition in overweight and obese women, significant differences in body composition were found among participants in the training group.[11]Physical activity increases energy expenditure and improves resting metabolic rate, cardiopulmonary function, and the physical component in overweight or obese patients.[12] In recent years, traditional Chinese exercise (TCE) have had significant advantages over medicine and surgery in treating overweight or obesity. Taijiquan and Qigong (Baduanjin, Yijinjing, Wuqinxi, etc), as China’s national intangible cultural heritage,[13] are forms of movement in Eastern culture; they are both internal and external, resilient and soft, and are being welcomed by more and more people.
Studies have found that tai chi exercise can gradually restore the normal activity of abnormally expressed AMPK genes in obese patients and significantly reduce triglycerides, total cholesterol, and low density lipoprotein (LDL) cholesterol indicators. In addition, it can also reduce fasting blood sugar, improve the microinflammatory state caused by obesity, and have a good balancing effect on the body’s energy metabolism.[14–17] However, some randomized controlled trials (RCTs) on the health benefits of TCE in people who are overweight or obese have yielded inconclusive results. In the effect of adding Tai Chi to dietary weight loss programs on lipoprotein hardness in obese older women,[18] for example, neither the Tai Chi nor the diet education groups improved their LDL cholesterol. Another study[19]showed that tai chi is a low-intensity physical exercise that consumes less energy and is detrimental to beneficial changes in body composition. In a study of the combined effects of tai chi, resistance training, and diet on body function and body composition in obese older women, Maris[20] found that the Tai Chi group was more helpful than the control group for improving physical function (e.g., TUG time), but there was no significant increase in body composition or functional indicators.
To date, there have been no meta-analyses to assess whether various types of TCEs (Tai Chi, Qigong, Baduan Jin, Yijin jing, Wuqinxi, etc) and exercise frequency, duration, and number of sessions have an effect on overweight or obese patients. Furthermore, none of them computed effect sizes using Hedges’ g-statistic. Other comprehensive studies are necessary to confirm the effect of TCE on anthropometric and biochemical markers in overweight or obese patients. Therefore, the main objective of this study is to determine the effect of traditional Chinese exercise on anthropometric and biochemical indicators in overweight or obese patients. The second objective was to determine whether any underlying moderating factors (e.g., control type) and TCE dose-related variables (e.g., frequency of exercise, duration of exercise, and number of sessions) affected the effect of the intervention.
## 2. Methods
The results of this meta-analysis are reported in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) guidelines.[21]
## 2.1. Search strategy
First, we searched the Pubmed, Cochrane Library, EMBASE, Web of Science, and Scopus databases from inception to October 2022. Second, the reference lists of the included studies were examined as an additional check for potential studies that could be used in this review. The following keywords were used: “Traditional Chinese exercise” or “Tai Chi” or “Qigong” or “BaduanJin” or “Wuqinxi” or “Yijinjing”; AND; “Obesity” or “obesity disease”; AND; “Randomized controlled trials” or “clinical trials.” The specific search syntax, such as PubMed, can be found in the Supplemental Digital Content (File “Search,” http://links.lww.com/MD/I524).
## 2.2. Eligibility criteria
Eligibility criteria were formulated based on the PICOS framework.
## 2.3. Study selection
Two reviewers (QY and FW) conducted the initial online search independently to avoid selection bias. After excluding duplicate studies, article titles and abstracts were reviewed. If an abstract was considered relevant or ambiguous, the full text was reviewed, and inclusion and exclusion criteria were applied. The Kappa statistic was used to assess the reliability of data selection and selection between 2 reviewers (QY and FW). Cohen suggested the Kappa result be interpreted as follows: values ≤ 0 as indicating no agreement and 0.01 to 0.20 as none to slight, 0.21 to 0.40 as fair, 0.41 to 0.60 as moderate, 0.61 to 0.80 as substantial, and 0.81 to 1.00 as almost perfect agreement.[22]
## 2.4. Data extraction and statistical analysis
Two reviewers (QY and FW) independently evaluated the article and extracted the data. The details of the retrieved articles are summarized in Table 1. Study feature data were extracted for each article (first author; year of publication; country; experimental design; average age; proportion of females; sample size; average BMI; usual care during the experiment; intervention characteristics, including type, frequency, and duration of intervention). Passive intervention was defined as a blank control group in the control group, while the exact total training time in the experimental group was defined as active intervention. During the data extraction process, any conflicts or ambiguities in the reporting method or results will be discussed with a third reviewer (TY) and resolved by consensus.
**Table 1**
| References | Country | Study design | Participant characteristics | Participant characteristics.1 | Participant characteristics.2 | Participant characteristics.3 | Participant characteristics.4 | Intervention protocol | Intervention protocol.1 | Outcome | Adverse effects |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| References | Country | Study design | Age, mean (SD) | %Sex Female, (EG/CG) | N (EG/CG) | BMI, mean (SD) | C | Intervention | Control | Outcome | Adverse effects |
| Liu et al (2015) | Australia | RCT | EG: 52 (12) CG: 53 (11) | 75/74 | 213 (106/107) | EG: 34.8 (6.6) CG: 35.1 (7.1) | Depression | Tai chi 3 × 90 min/wk 24 wk | No intervention | BMI; WC; TG; HDL | Heart failure (n = 1)Depression (n = 1) |
| Chen et al (2010) | China | RCT | EG: 59.1 (6.2) CG: 57.4 (5.8) | 55.4/58.3 | 104 (56/48) | EG: 33.5 (4.7) CG: 33.2 (4.1) | Type 2 diabetes | Tai chi 3 × 60 min/wk 12 wk | Aerobic exercises | BMI; TG; TC; HDL | NR |
| Sun et al (2015) | China | RCT | 45-64 and ≥ 65 | 86/78 | 266 (136/130) | EG: 23.38 (3.05) CG: 23.50 (2.99) | Hypertension | Tai chi 3 × 60 min/wk 2 mo | Reading and learning computer software | BMI; WC; TC; HDL; LDL; TG | NR |
| Beebe et al (2013) | USA | RCT | EG: 60.4 (6.2) CG: 62.6 (5.9) | 100/100 | 26 (13/13) | EG: 33.7 (4.8) CG: 34.8 (2.9) | Coronary heart disease | Tai chi 3 × 60 min/wk 16 wk + diet education | Diet education | BMI; W; WC; HDL; LDL; TC; TG | NR |
| Siu et al (2021) | China | RCT | EG: 62.6 (6.2) CG: 61.0 (5.7) | 77.3/79.0 | 543 (181/181/181) | EG: 25.5 (3.6) CG: 25.5 (3.4) | NR | Tai chi 3 × 60 min/wk 12 wk | UC | BMI; W; WC; HDL; TG | NR |
| Dechamps et al (2009) | France | RCT | 44.4 (11.9) | 100/100 | 21 (11/10) | EG: 37.4 (4.8) CG: 38.5 (7.3) | NR | Tai chi 2 × 60 min/wk 10 wk | Physical activity | BMI; W; FP | NR |
| Leung et al (2019) | China | RCT | EG: 62.19 (5.93) CG: 65.52 (9.34) | 41/56 | 54 (27/27) | EG: 27.40 (4.82) CG: 27.25 (4.34) | Metabolic syndrome | Tai chi 2 × 60 min/wk 12 wk + Tai chi 3 × 30 min/wk 12 wk | Non-exercise recreational class | WC; TC; TG; HDL | NR |
| Soltero et al (2022) | USA | RCT | EG: 49.6 (6.22) CG: 53.2 (9.30) | 100/100 | 20 (10/10) | EG: 29.48 (0.90) CG:31.14 (1.18) | Breast cancer | Qigong/Tai Chi 2 × 60 min/wk8 wk | Latin dance | BMI; FP | NR |
| Li et al (2022) | China | RCT | EG: 23.2 (4.38) CG:22.9 (4.64) | 100/100 | 50 (30/20) | EG: 28.41 (4.03) CG:29.57 (4.47) | Polycystic ovary | Tai chi 3 × 60 min/wk 12 wk | Routine exercise | BMI; W; TG; TC; HDL; LDL | NR |
All data was entered as the mean with standard deviation for the TCE and control groups at baseline and immediately after training. If the data was unsuitable for our analysis, the previous statistical formula was used to convert the data into mean and standard deviation format.[23]All analyses were conducted using comprehensive meta-analysis Version 3.3 software (Biostat Inc., Englewood, NJ). The comprehensive meta-analysis allows for each of these different study outcomes to be flexibly entered into the model. A random effects model was used to correct for variable effect sizes across the studies if these studies showed heterogeneity in their intervention. We used an inter group, pre-to-post-intervention, meta-analysis design based on standardized mean differences (Hedges’ g). Hedges’ g, a variant of Cohen d that corrects for sample size biases, was used to calculate the effect sizes (ESs).[24]We chose the Hedges’ g of the ESs to estimate the efficacy of TCE intervention. Hedge g estimates of < 0.3 were considered as small, ≥0.3 and < 0.6 as moderate, and ≥ 0.6 as large, respectively.[25] Positive effect sizes indicated a more favorable outcome for the experimental group. The I2 statistic estimated heterogeneity among studies and classified it as $25\%$ (low heterogeneity), $50\%$ (moderate heterogeneity), or $75\%$ (high heterogeneity).[24] A sensitivity analysis was also performed to detect the presence of highly influential studies that could skew the results. Studies were deemed influential if their removal significantly modified the summary effect. In addition, a moderator analysis was conducted based on exercise frequency, exercise duration, number of sessions, and type of control group. The significance level was set at P ≤.05.
## 2.5. Risk of bias and study quality assessment
The Cochrane Collaboration’s tool was used to assess the risk of bias in each individual study.[26] The tool contains the domains of sequence generation, allocation concealment, blinding of participants, personnel, and outcome assessors, incomplete outcome data, selective outcome reporting, and other sources of bias. We classified items as “low risk,” “high risk,” or “unclear risk” of bias. The risk of bias is presented in Figure 1.
**Figure 1.:** *Assessment of risk of bias with selected studies.*
We used the PEDro scale (Physiotherapy Evidence Database Rating Scale) to assess the quality of included studies using a score (Table 2).[27] The scale includes 11 items to rate study quality, and the maximum score is 11. Studies that scored 7 or higher were considered high quality, while those that scored 6 or lower were considered low quality. The scoring process was conducted by 2 authors (QY and FW). TY established consensus scores and resolved any disagreements.
**Table 2**
| References | PEDro scores | Methodological quality | PEDro item number | PEDro item number.1 | PEDro item number.2 | PEDro item number.3 | PEDro item number.4 | PEDro item number.5 | PEDro item number.6 | PEDro item number.7 | PEDro item number.8 | PEDro item number.9 | PEDro item number.10 |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| References | PEDro scores | Methodological quality | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 |
| Liu et al (2015) | 8 | H | 1 | 1 | 1 | 1 | 0 | 0 | 1 | 0 | 1 | 1 | 1 |
| Chen et al (2010) | 8 | H | 1 | 1 | 1 | 1 | 0 | 0 | 0 | 1 | 1 | 1 | 1 |
| Sun et al (2015) | 9 | H | 1 | 1 | 1 | 1 | 0 | 0 | 1 | 1 | 1 | 1 | 1 |
| Beebe et al (2013) | 8 | H | 1 | 1 | 1 | 1 | 0 | 0 | 0 | 1 | 1 | 1 | 1 |
| Siu et al (2021) | 9 | H | 1 | 1 | 1 | 1 | 0 | 0 | 1 | 1 | 1 | 1 | 1 |
| Dechamps et al (2009) | 9 | H | 1 | 1 | 1 | 1 | 0 | 0 | 1 | 1 | 1 | 1 | 1 |
| Leung et al (2019) | 8 | H | 1 | 1 | 1 | 1 | 0 | 0 | 1 | 0 | 1 | 1 | 1 |
| Soltero et al (2022) | 6 | L | 1 | 1 | 0 | 1 | 0 | 0 | 0 | 0 | 1 | 1 | 1 |
| Li et al (2022) | 8 | H | 1 | 1 | 1 | 1 | 0 | 0 | 1 | 0 | 1 | 1 | 1 |
## 2.6. Publication bias
We use Stata 17.0 software for Egger test and Begg test. The Egger test and Begg test were used to identify publication bias. If the $95\%$ confidence interval of the intercept of the regression equation is found to contain 0 and $P \leq .05$, it indicates unbiased; otherwise, it is biased.
## 3.1. Search results
Figure 2 shows the flow diagram of study selection. The initial search yielded 863 articles from 7 databases, and 2 additional articles were identified through other sources. After removing duplicates, 641 eligible records were retrieved. After reading the titles and abstracts, 597 articles were excluded. After reading the full text, the remaining 44 articles met the inclusion criteria. Among these 44 articles, 16 were excluded because the study design was not an RCT, 9 because TCE was not used as the intervention, 4 because participants did not meet the inclusion criteria, and 6 because the outcome measures were not the primary outcome. Finally, 9 original research articles were selected for further analysis.[28–36]
**Figure 2.:** *Process of study selection following the preferred reporting items for systematic reviews and meta-analyses (PRISMA).*
## 3.2. Study characteristics
Table 1 lists the characteristics of each of the included studies, which were published between 2009 and 2022. Five studies were conducted in China[29,30,32,34,36]; 2 in USA[31,35]; 1 in France[33]; 1 in Australia.[28] The TCE program was used to treat overweight or obese participants in all experimental groups. Active (routine exercise, diet education, etc) or passive interventions (nonintervention) were employed in the control group. These participants were prescribed 60 to 90 minutes of exercise in each session 2 to 3 times per-week for 8 weeks to 12 months. The outcomes of these 9 studies were as follows: body mass index (BMI), waist circumference (WC), weight (W), fat percentage (FP), total cholesterol (TC), triglyceride (TG), high density lipoprotein (HDL), and LDL. One study[28]reported on suspected side effects such as heart failure ($$n = 1$$) and depression ($$n = 1$$). The remaining studies reported no TCE related side effects.
## 3.3. Synthetic results
Regarding anthropometric indicators outcomes (Fig. 3), the pooled data demonstrated that TCE resulted produced large and significant improvements in BMI ($g = 1.44$, $95\%$ confidence interval [CI] = 1.27–1.61, $$P \leq .000$$, I2 = $99\%$), W ($g = 1.47$, $95\%$ CI = 1.25–1.68, $$P \leq .000$$, I2 = $95\%$) and FP ($g = 1.22$, $95\%$ CI = 0.52–1.93, $$P \leq .000$$, I2 = $93\%$) when compared to the control group. Moreover, pooled analyses from 5 parallel trials[28,30–32,34]revealed that the WC exerted a small and significant increase in effect size ($g = 0.38$, $95\%$ CI = 0.21–0.54, $$P \leq .000$$, I2 = $99\%$) compared with the control group.
**Figure 3.:** *Forest plot showing the effects of TCE vs control group on anthropometric indicators outcomes: BMI, WC, W, FP. BMI = body mass index, FP = fat percentage, TCE = traditional Chinese exercise, W = weight, WC = waist circumference.*
Regarding biochemical indicators outcomes (Fig. 4), the pooled data demonstrated that TCE resulted produced large and significant improvements in LDL ($g = 2.08$, $95\%$ CI = 1.80–2.37, $$P \leq .000$$, I2 = $98\%$) when compared to the control group. Pooled analyses from 7 parallel trials[28–32,34,36]revealed that the TG exerted a moderate and significant increase in effect size ($g = 0.69$, $95\%$ CI = 0.56–0.81, $$P \leq .000$$, I2 = $96\%$) compared with the control group. Moreover, the pooled data demonstrated that TCE resulted produced small and significant improvements in TC ($g = 0.37$, $95\%$ CI = 0.19–0.54, $$P \leq .000$$, I2 = $77\%$) and HDL (g = −0.71, $95\%$ CI = −0.86 to 0.57, $$P \leq .000$$, I2 = $99\%$) when compared to the control group.
**Figure 4.:** *Forest plot showing the effects of TCE vs control group on biochemical indicators outcomes: LDL, TG, TC, and HDL. HDL = high density lipoprotein, LDL = low density lipoprotein, TC = total cholesterol, TCE = traditional Chinese exercise, TG = triglyceride.*
According to the sensitivity analysis, no study significantly impacted the outcomes. No study was deemed insignificant since its removal had no discernible effect on the overall effect.
## 3.4. Moderator analysis
The categorical and continuous variables in Table 3 were used to conduct moderator analyses. In terms of frequency of exercise, more than 3 sessions/week ($g = 1.47$, $95\%$ CI = 1.30–1.64, $$P \leq .000$$) of TCE large and significant improved BMI compared with <3 sessions/week ($g = 0.22$,$95\%$ CI = −0.78 to 1.21, $$P \leq .673$$). Additionally, more than 3 sessions/week RCTs ($g = 0.43$, $95\%$ CI = 0.24–0.62, $$P \leq .000$$) of TCE small and significantly improved TC compared with less than 3 sessions/week RCTs on TCE (g = −0.15, $95\%$ CI = −0.68 to 0.38, $$P \leq .575$$). In terms of type of control group, compared to RCTs on TCE with non-active control group (g = −0.89, $95\%$ CI = −1.17 to 0.61, $$P \leq .000$$), RCTs on TCE with active control group ($g = 2.78$, $95\%$ CI = 2.57–3, $$P \leq .000$$) significantly improved BMI. Moreover, active control group RCTs ($g = 1.00$, $95\%$ CI = 0.85–1.14, $$P \leq .000$$) of TCE large and significantly improved TG compared with non-active control group RCTs on TCE (g = −0.4, $95\%$ CI = −0.67 to 0.13, $$P \leq .004$$).
**Table 3**
| Variables | Anthropometric indicators outcomes | Anthropometric indicators outcomes.1 | Anthropometric indicators outcomes.2 | Anthropometric indicators outcomes.3 | Biochemical indicators outcomes | Biochemical indicators outcomes.1 | Biochemical indicators outcomes.2 | Biochemical indicators outcomes.3 |
| --- | --- | --- | --- | --- | --- | --- | --- | --- |
| Variables | BMI Hedges’ g (95% CI) | WC Hedges’ g (95% CI) | W Hedges’ g (95% CI) | FP Hedges’ g (95% CI) | TC Hedges’ g (95% CI) | TG Hedges’ g (95% CI) | HDL Hedges’ g (95% CI) | LDL Hedges’ g (95% CI) |
| Exercise frequency | | | | | | | | |
| ≥3sessions/wk | 1.47 (1.30 to 1.64) | 0.26 (0.09 to 0.44) | 1.42 (1.2 to 1.64) | – | 0.43 (0.24 to 0.62) | 0.69 (0.56 to 0.82) | −0.64 (−0.79 to −0.5) | 2.08 (1.8–2.37) |
| <3sessions/wk | 0.22 (−0.78 to 1.21) | 1.95 (1.3 to 2.59) | 2.91 (1.71 to 4.12) | 1.22 (0.52–1.93) | −0.15 (−0.68 to 0.38) | 0.69 (0.14 to 1.23) | -6.65 (−8.01 to −5.29) | – |
| Exercise duration | | | | | | | | |
| >12 wk | 0.16 (−0.09 to 0.41) | 0.05 (−0.12 to −0.23) | −1.98 (−2.89 to −1.06) | – | 0.38 (0.15 to 0.61) | 0.02 (−0.15 to 0.20) | 0.27 (0.10 to 0.47) | 1.89 (1.60–2.18) |
| ≤12 wk | 2.52 (2.29 to 2.75) | 4.32 (3.71 to 4.93) | 1.66 (1.44 to 1.88) | 1.22 (0.52–1.93) | 0.34 (0.07 to 0.62) | 1.41 (1.22 to 1.6) | −2.87 (−3.13 to −2.61) | 5.25 (4.08–6.42) |
| Number of sessions | | | | | | | | |
| >36 sessions | 0.16 (−0.09 to 0.41) | 0.18 (0.02 to 0.35) | −1.98 (−2.89 to −1.06) | – | 0.29 (0.08 to 0.51) | 0.09 (−0.08 to 0.26) | 0.16 (−0.02 to 0.33) | 1.89 (1.60–2.18) |
| ≤36 sessions | 2.52 (2.29 to 2.75) | 27.14 (25.15 to 29.13) | 1.66 (1.44 to 1.88) | 1.22 (0.52–1.93) | 0.53 (0.21 to 0.85) | 1.5 (1.30 to 1.70) | −2.72 (−2.99 to −2.46) | 5.25 (4.08–6.42) |
| Type of control group | | | | | | | | |
| Active control | 2.78 (2.57 to 3.00) | 0.46 (0.24 to 0.67) | 1.47 (1.25 to 1.68) | 1.22 (0.52–1.93) | 0.37 (0.19 to 0.54) | 1.00 (0.85 to 1.14) | −1.01 (−1.18 to −0.83) | 2.08 (1.8–2.37) |
| Non-active control | −0.89 (−1.17 to −0.61) | 0.25 (−0.02 to 0.52) | to | – | to | −0.4 (−0.67 to −0.13) | 0.00 (−0.27 to 0.27) | – |
## 3.5. Risk of bias and study quality
Figure 1 show the risk of bias in the 9 included studies. All studies used random sequence generation. In the term of allocation concealment, 8 studies[28–34,36] had a low risk of attrition bias and 1 study had a high risk of attrition bias. In the term of blinding of outcome assessors, 6 studies[28,30,32–34,36] had a low risk of attrition bias and 3 studies[29,31,35] had a unclear risk of attrition bias. Moreover, all studies had a high risk of attrition bias on blinding of therapist and participants. Additionally, all studies had a low risk of attrition bias on incomplete outcome data, incomplete outcome data and other sources of bias. The value of Kappa calculated for the various parameters extracted by the 2 investigators was 0.83 ($P \leq .001$), indicating an excellent degree of inter-investigator agreement.
Table 2 presents the methodology quality of the included studies. The quality of the studies ranged between high quality and low quality (score range: 6–9 points), with 8 studies being classified as high quality[28–34,36]and 1 study as low quality.[35] Eight studies used a concealed allocation procedure[28–34,36] and all reported random assignment. We could not blind patients and therapists because this was an interventional movement study. However, 6 trials blinded the outcome assessors.[28,30,32–34,36] Five studies had a dropout rate of > $85\%$.[29–33] All 9 studies performed intention-to-treat analyses, between-group statistics, and point measurements.
## 3.6. Publication bias
The Egger test and Begg test (Figure S1–S8, Supplemental Digital Content, http://links.lww.com/MD/I525) were used to assess publication bias. The results of the Egger test are as follows: BMI ($$P \leq .3$$>0.05), WC ($$P \leq .18$$>0.05), TC ($$P \leq .27$$>0.05), TG ($$P \leq .76$$>0.05) all showed no publication bias. The results of the Begg test are as follows: BMI ($$P \leq .23$$>0.05), WC ($$P \leq .22$$>0.05), TC ($$P \leq .46$$>0.05), TG ($$P \leq .55$$>0.05) all showed no publication bias.
## 4. Discussion
The current systematic review with meta-analysis showed that TCE significantly improved anthropometric indicator outcomes (BMI, WC, W, and FP), biochemical indicators outcomes (TC, TG and LDL) compared with the control group in individuals with overweight and obese. However, TCE interventions did not have a significant effect on HDL. Data from 9 RCTs involving a total of 1297 participants were analyzed. The moderator analyses showed the effects of TCE on anthropometric and biochemical indicators were moderated by the sample size, type of control group, frequency of exercise, and exercise duration.
## 4.1. Gaps in previous reviews
To our knowledge, this is the first meta-analysis to focus on the effects of TCE interventions on anthropometric and biochemical indicators in overweight or obese patients. The overall conclusions of our study were similar to those reported in other recent reviews, but the analytical methods used in our study differed except for the inclusion of updated evidence. As an example, previous reviews have been limited to qualitative synthesis and have not included quantitative meta-analysis methods.[37,38] Of the studies included in the meta-analysis, 1 meta-analysis was limited to tai chi exercises (no qigong) and the sample size of the study was small, resulting in wide confidence intervals and reduced accuracy.[39] In another meta-analysis, the outcome was too simplistic, assessing only body composition outcomes and not mentioning other measures.[40] Finally, the 2 meta-analyses included in 2020 and 2016 used a broader range of inclusion criteria, combining results from multiple mind-body modalities, including multiple exercise interventions such as martial arts practice, kung fu practice, and tai chi practice.[41,42]
## 4.2. Effects of TCE on anthropometric indicators outcomes
At present, traditional anthropometric indicators related to overweight or obesity include height, body mass, waist circumference, and BMI.[43,44] *It is* well known that BMI can distinguish between overweight or obesity in general, but not body fat content,[45] and WC and FP are effective indicators for measuring visceral fat distribution,[46,47] so WC and FP were used as auxiliary diagnoses in overweight or obese patients. Asians with the same BMI and WC levels may have higher body fat percentages than Western populations due to racial differences in physical characteristics, and BMI or WC values vary by country to race, as in the WHO-recommended diagnostic criteria for overweight or obesity.[48–50] The current study found that TCE could help overweight or obese patients to improve their anthropometric indicators. In 1 study, tai chi exercise was beneficial in regulating physical and mental state in patients with simple obesity, with significant improvements in body fat percentage and body mass index compared with before treatment.[51] *In a* community study of obese elderly women, significant improvements in body fat mass and waist circumference were found in the tai chi group compared with the control group, and that tai chi exercise intervention may be an effective strategy to improve physical function and risk of coronary heart disease in older adults.[52]
## 4.3. Effects of TCE on biochemical indicators outcomes
Dyslipidemia is associated with obesity due to excessive fat accumulation, and the intensity of work stress, unhealthy lifestyle, and diet all directly contribute to excessive fat accumulation.[3,53] Moderate or central obesity subjects have been found to have higher levels of TC, TG, LDL-C, and lower levels of HDL-C compared with nonobese individuals, which may be associated with excess visceral fat in the abdomen.[54,55] *There is* a strong correlation between overweight or obesity and cardiovascular events and dyslipidemia, and effective control of lipid levels is expected to reduce the incidence of metabolic morbidity and death.[56,57] Concerning the biochemical indicators outcomes, our meta-analysis showed large significant improvement in LDL, moderate significant improvement in TG, small significant improvement in TC and HDL of TCE intervention on the biochemical indicators outcomes. According to a meta-analysis of the effect of tai chi exercise on blood lipid profile, Tai chi may be beneficial for lipid profiles in different age groups and populations. Especially for HDL-C with a potential positive effect.[39] According to our meta-analysis, HDL levels in overweight or obese patients did a small improve after TCE intervention, which is differs from the results of the review described above. In contrast, our findings provide more convincing evidence for a variety of reasons. First, all but one of the included studies had high-quality methodologies according to the PEDro scale tool. However, none of Chau literature quality is mentioned. Second, although we included 9 articles, 9 were related to obesity, and Chau included 20 articles, but only 3 were related to obesity. Therefore, more research is needed on changes in HDL levels of TCE interventions in overweight or obese patients.
## 4.4. Effects of TCE on moderator analyses
Our moderator analysis revealed that the effect of TCE on BMI was significantly higher ($g = 1.55$) when only large sample size RCTs were analyzed, but not when small sample size RCTs were analyzed ($g = 0.54$). Additionally, large sample size RCTs ($g = 0.74$) have a greater effect on TG than small sample size RCTs ($g = 0.33$). Moreover, large sample size RCTs ($g = 2.34$) have a greater effect on LDL than small sample size RCTs ($g = 0.91$). As a result, our findings imply that the significant effects of TCE on BMI, TG, and LDL were not based on small sample size RCTs; therefore, future studies must utilize large sample size RCTs to evaluate the effects of TCE on BMI, TG and LDL in overweight or obese patients. As for the frequency of exercise, BMI and TC improved significantly in overweight or obese patients after they performed more than 3 sessions/week (BMI: $g = 1.47$; TC: $g = 0.27$) compared to less than 3 sessions/week (BMI: $g = 0.22$; TC: g = −0.15). Moderator analysis showed that TCE increased the effect on active control group (BMI: $g = 2.78$; TG: $g = 1.00$) compared with non-active control group (BMI: g = −0.89; TG: g = −0.4).
## 4.5. Strengths and limitations
The strength of this meta-analysis is that our systematic review and meta-analysis followed the PRISMA statement to the letter, and our review methodology was registered. Only the RCT design was chosen due to its reliability. Furthermore, we used Hedges’ g to ensure an accurate estimation of the overall effect size. Other potential confounding variables were explored to determine their impact on the effects of TCE. However, there are limitations in this study. First, for the included RCTs, although the literature was searched and screened in strict accordance with the search strategy and inclusion criteria, there was a certain degree of missed detection and bias. Second, the number of RCTs included in this study was 9, and the sample size of the literature was not large enough and the representativeness was not strong enough. Third, the accuracy of meta-analysis results may be reduced because subjective factors from researchers cannot be excluded. Fourth, the accuracy and reliability of the literature results were affected by the failure to blind the assessors or the lack of reporting of adverse effects and recurrence rates in the included studies. Fifth, in terms of safety evaluation, most studies did not report adverse effects and could not determine the safety of treatment.
## 4.6. Implications for future studies
More rigorous randomized controlled trials are needed to provide evidence on the effect of TCE on outcomes in people who are overweight or obese. Interventions are needed for TCE with different frequencies, times, and durations to help compare and decide on the best TCE regimen for overweight or obese patients. In addition, the outcomes of the intervention may include not only post-intervention outcomes, but also medium- and long-term outcomes to give a fuller picture of the effect. There is also a need for prespecified protocols with clear reporting, particularly regarding the randomization process, how to manage missing data, and how to prevent deviations from expected interventions in cases where blinding of participants cannot be made due to the nature of the study.
## 5. Conclusion
Our meta-analysis shows that TCE intervention is a beneficial non-pharmacological approach to improving anthropometric and biochemical indicators in overweight and obese subjects, especially in BMI, W, FP, TG, and LDL. The clinical relevance of our findings is pending more extensive trials and more rigorous study designs to strengthen the evidence.
## Author contributions
Data curation: Qianfang Yang, Limin Pan.
Formal analysis: Qianfang Yang.
Funding acquisition: Ting Ye.
Investigation: Fan Wang.
Methodology: Limin Pan.
Project administration: Ting Ye.
Resources: Limin Pan.
Software: Fan Wang.
Writing – original draft: Qianfang Yang.
Writing – review & editing: Qianfang Yang.
## References
1. Purnell JQ, Feingold KR, Anawalt B, Boyce A. **Definitions, classification, and epidemiology of obesity.**. *Endotext [Internet]* (2000)
2. Gasmi A, Mujawdiya PK, Noor S. **Lifestyle genetics-based reports in the treatment of obesity.**. *Arch Razi Inst* (2021) **76** 707-19. PMID: 35096307
3. Lin X, Li H. **Obesity: epidemiology, pathophysiology, and therapeutics.**. *Front Endocrinol (Lausanne)* (2021) **12** 706978. PMID: 34552557
4. Chen Y, Peng Q, Yang Y. **The prevalence and increasing trends of overweight, general obesity, and abdominal obesity among Chinese adults: a repeated cross-sectional study.**. *BMC Public Health* (2019) **19** 1293. PMID: 31615464
5. Ataey A, Jafarvand E, Adham D. **The relationship between obesity, overweight, and the human development index in World Health Organization Eastern Mediterranean Region Countries.**. *J Prev Med Public Health* (2020) **53** 98-105. PMID: 32268464
6. Panuganti KK, Nguyen M, Kshirsagar RK. **Obesity.**. *StatPearls [Internet]* (2022)
7. Pavlos S, Nikiforou C. **Medicinal plants against obesity: a met-analysis of literature.**. *J Compl Med Res* (2022) **12** 244
8. Li X, Si H, Chen Y. **Effects of fitness qigong and tai chi on middle-aged and elderly patients with type 2 diabetes mellitus.**. *PLoS One* (2020) **15** e0243989. PMID: 33332396
9. Fonseca-Junior SJ, Sá CG, Rodrigues PA. **Physical exercise and morbid obesity: a systematic review.**. *Arq Bras Cir Dig* (2013) **26** 67-73. PMID: 24463903
10. Chao AM, Quigley KM, Wadden TA. **Dietary interventions for obesity: clinical and mechanistic findings.**. *J Clin Invest* (2021) **131** e140065. PMID: 33393504
11. Aghajani R, Nemati N, Hojjati Zidashti Z. **Effect of aerobic program in the morning and afternoon on obestatin and the body composition of overweight and obese women.**. *J Chem Health Risks* (2020) **10** 117-25
12. Berge J, Hjelmesaeth J, Hertel JK. **Effect of aerobic exercise intensity on energy expenditure and weight loss in severe obesity-a randomized controlled trial.**. *Obesity (Silver Spring)* (2021) **29** 359-69. PMID: 33491314
13. Qin Y, Xia W, Huang W. **The beneficial effect of traditional Chinese exercises on the management of obesity.**. *Evid Based Complement Alternat Med* (2020) **2020** 2321679. PMID: 33062003
14. Qian Y, Ke JB. **Effects of Tai Chi exercise on whole genome expression in obese patients.**. *J Milit Sports Training Inst* (2010) **29** 120-2
15. Han Y B. **Effects of tai chi training on growth hormone releasing peptide and adiponectin in high-risk groups with metabolic syndrome.**. (2011) 582-583
16. Zhang J W, Lv S J. **A comparative study on the effect of tai chi exercise on health intervention in patients with type 2 diabetes mellitus of different genders.**. (2019) 1672-3
17. Li XB. **Effects of Tai Chi exercise on oxidative stress and inflammation levels in elderly patients with type 2 diabetes.**. *Chin J Gerontol* (2013) **33** 5465-6
18. Katkowski LA, Benson MC, Magnanti S. **Additive effect of tai chi during dietary weight loss on physical function and body composition in obese older women.**. *J Aging Res Clin Pract* (2012) **2** 78-85
19. Hui SS, Xie YJ, Woo J. **Practicing Tai Chi had lower energy metabolism than walking but similar health benefits in terms of aerobic fitness, resting energy expenditure, body composition and self-perceived physical health.**. *Complement Ther Med* (2016) **27** 43-50. PMID: 27515875
20. Maris SA, Quintanilla D, Taetzsch A. **The combined effects of tai chi, resistance training, and diet on physical function and body composition in obese older women.**. *J Aging Res* (2014) **2014** 657851. PMID: 25614836
21. Page MJ, McKenzie JE, Bossuyt PM. **The PRISMA 2020 statement: an updated guideline for reporting systematic reviews.**. *BMJ* (2021) **372** n71. PMID: 33782057
22. McHugh ML. **Interrater reliability: the kappa statistic.**. *Biochem Med (Zagreb)* (2012) **22** 276-82. PMID: 23092060
23. Hozo SP, Djulbegovic B, Hozo I. **Estimating the mean and variance from the median, range, and the size of a sample.**. *BMC Med Res Methodol* (2005) **5** 13. PMID: 15840177
24. Higgins JP, Thompson SG, Deeks JJ. **Measuring inconsistency in meta-analyses.**. *BMJ* (2003) **327** 557-60. PMID: 12958120
25. Cohen J. *Statistical Power Analysis for The Behavioral Sciences (2nd ed.)* (1988)
26. Higgins JP, Altman DG, Gøtzsche PC. **The cochrane collaboration’s tool for assessing risk of bias in randomised trials.**. *BMJ* (2011) **343** d5928. PMID: 22008217
27. Maher CG, Sherrington C, Herbert RD. **Reliability of the PEDro scale for rating quality of randomized controlled trials.**. *Phys Ther* (2003) **83** 713-21. PMID: 12882612
28. Liu X, Vitetta L, Kostner K. **The effects of tai chi in centrally obese adults with depression symptoms.**. *Evid Based Complement Alternat Med* (2015) **2015** 879712. PMID: 25688280
29. Chen SC, Ueng KC, Lee SH. **Effect of t’ai chi exercise on biochemical profiles and oxidative stress indicators in obese patients with type 2 diabetes.**. *J Altern Complement Med* (2010) **16** 1153-9. PMID: 20973735
30. Sun J, Buys N. **Community-Based Mind-Body Meditative Tai Chi program and its effects on improvement of blood pressure, weight, renal function, serum lipoprotein, and quality of life in Chinese adults with hypertension.**. *Am J Cardiol* (2015) **116** 1076-81. PMID: 26251005
31. Beebe N, Magnanti S, Katkowski L. **Effects of the addition of t’ai chi to a dietary weight loss program on lipoprotein atherogenicity in obese older women.**. *J Altern Complement Med* (2013) **19** 759-66. PMID: 23484524
32. Siu PM, Yu AP, Chin EC. **Effects of Tai Chi or conventional exercise on central obesity in middle-aged and older adults: a three-group randomized controlled trial.**. *Ann Intern Med* (2021) **174** 1050-7. PMID: 34058100
33. Dechamps A, Gatta B, Bourdel-Marchasson I. **Pilot study of a 10-week multidisciplinary Tai Chi intervention in sedentary obese women.**. *Clin J Sport Med* (2009) **19** 49-53. PMID: 19124984
34. Leung LY, Chan AW, Sit JW. **Tai Chi in Chinese adults with metabolic syndrome: a pilot randomized controlled trial.**. *Complement Ther Med* (2019) **46** 54-61. PMID: 31519288
35. Soltero EG, Larkey LK, Kim WS. **Latin dance and Qigong/Tai Chi effects on physical activity and body composition in breast cancer survivors: a pilot study.**. *Complement Ther Clin Pract* (2022) **47** 101554. PMID: 35257993
36. Li Y, Peng C, Zhang M. **Tai Chi for overweight/obese adolescents and young women with polycystic ovary syndrome: a randomized controlled pilot trial.**. *Evid Based Complement Alternat Med* (2022) **2022** 4291477. PMID: 35754690
37. Hamasaki H. **Martial arts and metabolic diseases.**. *Sports (Basel)* (2016) **4** 28. PMID: 29910276
38. Kelly OJ, Gilman JC. **Can unconventional exercise be helpful in the treatment, management and prevention of osteosarcopenic obesity?**. *Curr Aging Sci* (2017) **10** 106-21. PMID: 27156950
39. Chau JPC, Leung LYL, Liu X. **Effects of Tai Chi on health outcomes among community-dwelling adults with or at risk of metabolic syndrome: a systematic review.**. *Complement Ther Clin Pract* (2021) **44** 101445. PMID: 34260997
40. Larkey LK, James D, Belyea M. **Body composition outcomes of Tai Chi and Qigong practice: a systematic review and meta-analysis of randomized controlled trials.**. *Int J Behav Med* (2018) **25** 487-501. PMID: 29856007
41. de Souza F, Lanzendorf FN, de Souza MMM. **Effectiveness of martial arts exercise on anthropometric and body composition parameters of overweight and obese subjects: a systematic review and meta-analysis.**. *BMC Public Health* (2020) **20** 1246. PMID: 32807119
42. Stoner L, Rowlands D, Morrison A. **Efficacy of exercise intervention for weight loss in overweight and obese adolescents: meta-analysis and implications.**. *Sports Med* (2016) **46** 1737-51. PMID: 27139723
43. Ma W, Zhang H, Wu N. **Relationship between obesity-related anthropometric indicators and cognitive function in Chinese suburb-dwelling older adults.**. *PLoS One* (2021) **16** e0258922. PMID: 34705855
44. Jayedi A, Soltani S, Motlagh SZ. **Anthropometric and adiposity indicators and risk of type 2 diabetes: systematic review and dose-response meta-analysis of cohort studies.**. *BMJ* (2022) **376** e067516. PMID: 35042741
45. Dhana K, Koolhaas CM, Schoufour JD. **Association of anthropometric measures with fat and fat-free mass in the elderly: the Rotterdam study.**. *Maturitas* (2016) **88** 96-100. PMID: 27105706
46. Zeng Q, He Y, Dong S. **Optimal cut-off values of BMI, waist circumference and waist:height ratio for defining obesity in Chinese adults.**. *Br J Nutr* (2014) **112** 1735-44. PMID: 25300318
47. See R, Abdullah SM, McGuire DK. **The association of differing measures of overweight and obesity with prevalent atherosclerosis: the Dallas Heart Study.**. *J Am Coll Cardiol* (2007) **50** 752-9. PMID: 17707180
48. Lear SA, Humphries KH, Kohli S. **Visceral adipose tissue accumulation differs according to ethnic background: results of the multicultural community health assessment trial (M-CHAT).**. *Am J Clin Nutr* (2007) **86** 353-9. PMID: 17684205
49. Chan JC, Malik V, Jia W. **Diabetes in Asia: epidemiology, risk factors, and pathophysiology.**. *JAMA* (2009) **301** 2129-40. PMID: 19470990
50. Grundy SM, Brewer HB, Cleeman JI. **Definition of metabolic syndrome: report of the national heart, lung, and blood institute/American heart association conference on scientific issues related to definition.**. *Circulation* (2004) **109** 433-8. PMID: 14744958
51. Song Q, Yuan Y, Jiao C. **Curative effect of Tai Chi exercise in combination with auricular plaster therapy on improving obesity patient with secondary hyperlipidemia.**. *Int J Clin Exp Med* (2015) **8** 21386-92. PMID: 26885081
52. Xu F, Letendre J, Bekke J. **Impact of a program of Tai Chi plus behaviorally based dietary weight loss on physical functioning and coronary heart disease risk factors: a community-based study in obese older women.**. *J Nutr Gerontol Geriatr* (2015) **34** 50-65. PMID: 25803604
53. Lin LY, Hsu CY, Lee HA. **Gender difference in the association of dietary patterns and metabolic parameters with obesity in young and middle-aged adults with dyslipidemia and abnormal fasting plasma glucose in Taiwan.**. *Nutr J* (2019) **18** 75. PMID: 31733645
54. Zhu J, Zhang Y, Wu Y. **Obesity and Dyslipidemia in Chinese adults: a cross-sectional study in Shanghai, China.**. *Nutrients* (2022) **14** 2321. PMID: 35684121
55. Vekic J, Zeljkovic A, Stefanovic A. **Obesity and dyslipidemia.**. *Metabolism* (2019) **92** 71-81. PMID: 30447223
56. Sangrós FJ, Torrecilla J, Giráldez-García C. **Association of general and abdominal obesity with hypertension, dyslipidemia and prediabetes in the PREDAPS Study.**. *Rev Esp Cardiol (Engl Ed)* (2018) **71** 170-7. PMID: 28789915
57. Alenazi AM, Alshehri MM, Hoover JC. **The effect of T’ai Chi exercise on lipid profiles: a systematic review and meta-analysis of randomized clinical trials.**. *J Altern Complement Med* (2018) **24** 220-30. PMID: 28934556
|
---
title: Impact of lead exposure on the thyroid glands of individuals living in high-
or low-lead exposure areas
authors:
- José Estefano Rivera-Buse
- Sheila Jissela Patajalo-Villalta
- Eduardo Antônio Donadi
- Fernando Barbosa
- Patrícia Künzle Ribeiro Magalhães
- Léa Maria Zanini Maciel
journal: Medicine
year: 2023
pmcid: PMC10036069
doi: 10.1097/MD.0000000000033292
license: CC BY 4.0
---
# Impact of lead exposure on the thyroid glands of individuals living in high- or low-lead exposure areas
## Abstract
Ecuador was an endemic area for iodine deficiency; however, due to the population consumption of iodized table salt, the country is nowadays considered iodine sufficient. Despite the population consumption of iodized salt for more than 50 years, the prevalence of hypothyroidism has increased in recent years. A similar increment has been reported for thyroid cancer (TC) becoming the second most common cancer in women and seventh most common cancer in men. High blood lead (BPb) level is a controversial causal factor for impaired thyroid function as well as a debated environmental cause for the increased incidence of TC. To study the association between BPb and thyroid function, anti-thyroid peroxidase (anti-TPO) and anti-thyroglobulin (anti-Tg) antibodies, and the presence of benign and malignant thyroid nodules in Ecuadorian individuals living in high lead exposure (HE) areas compared with those living in low lead exposure (LE) area. We evaluated 197 euthyroid individuals: 70 from Esmeraldas (close to a petrol refinery) and 27 from La Victoria de Pujilí (Pb-glazing ceramics), considered HE areas, and 100 from Quito, considered the LE area. In parallel, we evaluated 187 patients with hypothyroidism (60, 27, and 100 patients from Esmeraldas, Pujilí, and Quito, respectively). BPb was detected using atomic absorption spectroscopy, while thyroid-stimulating hormone (TSH), free-thyroxine (FT4), and autoantibodies were measured using chemiluminescence assays. Thyroid ultrasonography was performed in 300 individuals and fine-needle aspiration biopsy (FNA) was performed only when required based on the guidelines of the American Thyroid Association. The BPb levels (mean ± SD) in the HE areas were increased (8.5 ± 7.4) than those in the LE area (3.2 ± 2.4, $P \leq .001$). No significant associations were observed between BPb and TSH, FT4, or thyroid antibody levels. Enlarged thyroid glands and larger thyroid nodules were primarily observed in HE areas. Just 1 TC was observed. High BPb levels detected in HE areas were not associated with thyroid function or thyroid autoantibodies; however, increased thyroid size and numbers of thyroid nodules were observed, demanding further actions to control lead contamination in these Ecuadorian areas.
## 1. Introduction
Because of the population consumption of iodized table salt, *Ecuador is* nowadays considered an iodine sufficient country.[1] Despite the population consumption of iodized salt for more than 50 years, the prevalence of hypothyroidism has increased in Ecuador.[2,3] A similar increment has been reported for thyroid cancer (TC). According to the Ecuadorian Society to Combat Cancer, in 2017, TC was the second most frequent cancer in women (41 per 100.000 women) after breast cancer, while TC was ranked seventh in men (7.9 per 100.000).[4] In Ecuador, there are populations living in areas highly exposed (HE) to lead (Pb), one of which is the city of Esmeraldas, where the Esmeraldas State *Refinery is* located. According to a report by the General Comptroller of the State, lead tetraethyl is no longer added to gasoline to improve its performance; however, this product has remained in storage for more than 16 years at Esmeraldas State Refinery, constituting an additional permanent pollutant in Esmeraldas.[5] Another Ecuadorian HE population is the parish La Victoria de Pujilí, exhibiting historical exposure to inorganic lead, due to the use of Pb-glazing ceramic cookware industry.[6] The Centers for Disease Control and Prevention has established that the blood lead (BPb) level in adults should be <10 μg/dL.[7] Some studies reported that lead can cause functional deterioration of the pituitary-thyroid axis, modifying the thyroid physiology,[8] and the chronic lead exposure can cause anatomic-histological changes in thyroid gland, such as decrease in the size of thyroid follicles and alteration of the follicle cell nucleus.[9] In recent years, several studies indicate that lead affects the thyroid function,[10] causes a decrease in the production of tetraiodothyronine (T4) and elevation of thyroid-stimulating hormone (TSH).[11] Other authors reported a decrease of T4 without affecting triiodothyronine or TSH,[9,10] apparently, due to the inhibition of the 5 deiodinase type 1 enzyme. These controversial results have been related to different BPb levels and to the mode of lead exposure.[9–11] Additionally, some authors have indicated that there is an inverse relationship between BPb and TSH in hypothyroid women, but not in men[12]; however, other studies have not reported such a relationship.[13] It has also been proposed that lead can increase the risk of developing TC.[14] Notably, most studies associating environmental pollutants and thyroid function were performed in men who are professionally exposed to these pollutants.[15] Therefore, the effect of lead on the thyroid gland remains controversial.
Considering the high incidence rate of TC in Ecuadorian women, in this study, we evaluated the following: the BPb levels, the relationship between BPb and thyroid function, and thyroid autoantibody levels, and the frequency of benign and malignant thyroid nodules in euthyroid and hypothyroid individuals living in Ecuadorian areas exhibiting different lead exposure, that is, high (HE) or low exposure (LE) areas, primarily encompassing women.
## 2.1. Study population
From 2019 to 2021, we studied 197 (181 women) adult healthy euthyroid individuals (mean ± SD = 49.6 ± 14.6 years) and 187 (171 women) hypothyroid patients (51.6 ± 15.3 years) from 2 Ecuadorian areas known to be highly exposed to lead, including individuals living <5 kilometer from the Esmeraldas petrol refinery, which has been installed in the city for more than 50 years, and the parish of La Victoria del Pujilí, where lead, tin, and copper contaminations come from the continuous practice of glazing tiles, handicrafts, and kitchen utensils, using spent and old batteries.[16] In parallel, we evaluated 200 individuals (100 euthyroid aged 47.7 ± 15.0, and 100 hypothyroid aged 50.0 ± 15.5) from the metropolitan area of the Quito District (LE), for whom no report for high lead contamination is reported. The participants were matched according to age and gender. The sample size was defined based on the probability of exposure and OR of $19.27\%$ reported by Pekcici et al[8] obtaining a minimum sample of 43 subjects per group. The demographic and provenance characteristics of participants are presented in Table 1. A venous blood sample of 5 milliliters was collected in free metal tubes, at 8 to 10 am, for the determination of free-thyroxine (FT4), TSH, anti-thyroid peroxidase (anti-TPO), and anti-thyroglobulin (anti-Tg) antibodies. Additionally, a venous blood sample (3 mL) was collected into a tube containing ethylenediaminetetra-acetic acid for lead measurements.
**Table 1**
| Unnamed: 0 | All subjects | All subjects.1 | All subjects.2 | Euthyroid subjects | Euthyroid subjects.1 | Euthyroid subjects.2 | Hypothyroid subjects | Hypothyroid subjects.1 | Hypothyroid subjects.2 |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| | | HE area | LE area | | HE area | LE area | | HE area | LE area |
| No. of patients, n | 384 | 184 | 200 | 197 | 97 | 100 | 187 | 87 | 100 |
| Gender, n (%) | | | | | | | | | |
| Women | 352 (91.7) | 174 (94.6) | 178 (89.0) | 181 (91.9) | 92 (94.8) | 89 (89.0) | 171 (91.4) | 82 (94.3) | 89 (89.0) |
| Men | 32 (8.3) | 10 (5.4) | 22 (11.0) | 16 (8.1) | 5 (5.2) | 11 (11.0) | 16 (8.6) | 5 (5.7) | 11 (11.0) |
| Age, yr (mean ± SD) | 50.6 ± 15.0 | 52.4 ± 14.5 | 48.9 ± 15.3* | 49.6 ± 14.6 | 51.5 ± 14.1 | 47.7 ± 15.0 | 51.6 ± 15.3 | 53.5 ± 15.1 | 50.0 ± 15.5 |
| Exposure time to lead, yr (mean ± SD) | 37.1 ± 18.3 | 40.6 ± 19.6 | 33.9 ± 16.5** | 36.8 ± 18.0 | 39.8 ± 18.7 | 33.9 ± 16.8# | 37.4 ± 18.7 | 41.4 ± 20.5 | 33.9 ± 16.2## |
| Blood lead level, μg/dL, n (%) | 5.5 ± 6.7 | 8.2 ± 8.6 | 3.0 ± 2.2** | 5.8 ± 6.0 | 8.5 ± 7.4 | 3.2 ± 2.4** | 5.1 ± 7.3 | 7.8 ± 9.8 | 2.7 ± 1.9** |
| ≥10 μg/dL | 44 (11.5) | 42 (22.8) | 2 (1.0) | 26 (13,2) | 25 (25,8) | 1 (1.0) | 18 (9.6) | 17 (19.5) | 1 (1.0) |
| <10 μg/dL | 340 (88.5) | 142 (77.2) | 198 (99.0) | 171 (86.8) | 72 (74.2) | 99 (99.0) | 169 (90.4) | 70 (80.5) | 99 (99.0) |
| TSH, mUI/L (mean ± SD) | 3.6 ± 6.8 | 4.2 ± 9.6 | 3.0 ± 2.3 | 2.8 ± 2.0 | 2.7 ± 2.2 | 3.0 ± 1.7 | 4.4 ± 9.5 | 5.9 ± 13.5 | 3.1 ± 2.7 |
| TSH, n (%) | | | | | | | | | |
| >4.5 mUI/L | 78 (20.3) | 36 (19.7) | 42 (21.0) | 36 (18.3) | 15 (15.5) | 21 (21.0) | 42 (22.5) | 21 (24.2) | 21 (21.0) |
| 0.45–4.5 mUI/L | 281 (73.2) | 131 (71.2) | 150 (75.0) | 155 (78.7) | 78 (80.4) | 77 (77.0) | 126 (67.4) | 53 (60.9) | 73 (73.0) |
| <0.45 mUI/L | 25 (6.5) | 17 (9.1) | 8 (4.0) | 6 (3.0) | 4 (4.1) | 2 (2.0) | 19 (10.1) | 13 (14.9) | 6 (6.0) |
| FT4, ng/dL (mean ± SD) | 1.0 ± 0.2 | 1.0 ± 0.2 | 1.0 ± 0.2 | 0.94 ± 0.1 | 0.93 ± 0.2 | 0.96 ± 0.1 | 1.0 ± 0.2 | 1.0 ± 0.2 | 1.0 ± 0.2 |
| FT4, n (%) | | | | | | | | | |
| <0.9 ng/dL | 127 (33.1) | 68 (37.0) | 59 (29.5) | 78 (39.6) | 45 (46.4) | 33 (33.0) | 49 (26.2) | 23 (26.4) | 26 (26.0) |
| 0.9–1.7 ng/dL | 255 (66.4) | 114 (62.0) | 141 (70.5) | 118 (59.9) | 51 (52.6) | 67 (67.0) | 137 (73.3) | 63 (72.4) | 74 (74.0) |
| >1.7 ng/dL | 2 (0.5) | 2 (1.1) | 0 | 1 (0.5) | 1 (1.0) | 0 | 1 (0.1) | 1 (0.2) | 0 |
| Anti-TPO, n (%) | | | | | | | | | |
| ≥35 U/mL | 83 (21.6) | 41 (22.3) | 42 (21.0) | 23 (11.7) | 12 (12.4) | 11 (11.0) | 60 (32.1) | 29 (33.3) | 31 (31.0) |
| <35 U/mL | 301 (78.4) | 143 (77.7) | 158 (79.0) | 174 (88.3) | 85 (87.6) | 89 (89.0) | 127 (67.9) | 58 (66.7) | 89 (89.0) |
| Anti-TG, n (%) | | | | | | | | | |
| ≥40 U/mL | 29 (7.6) | 16 (8.7) | 13 (6.5) | 11 (5.6) | 8 (8.2) | 3 (3.0) | 18 (9.6) | 8 (9.2) | 10 (10.0) |
| <40 U/mL | 355 (92.4) | 168 (91.3) | 187 (93.5) | 186 (94.4) | 89 (91.8) | 97 (97.0) | 169 (90.4) | 79 (90.8) | 90 (90.0) |
Clinical data and blood samples from all individuals were collected after signing an informed consent form, which was approved by the Universidad Central del Ecuador Subcommittee on Ethics in Research in Humans and the Ethics Committee in Research in Human beings at the Eugenio Espejo Specialty Hospital (protocol # MSPCURI0002694).
## 2.2. Lead levels
Pb levels in whole blood were determined at the Department of Clinical, Toxicological, and Bromatological Analysis of the Faculty of Pharmaceutical Sciences of Ribeirão Preto, University of São Paulo. Analyses were performed using inductively coupled plasma mass spectrometry with a quadrupole ion detector (NexION® 2000, PerkinElmer, Waltham, MA).
To verify the accuracy of our data, reference blood samples obtained from the Institut National de Sante` Publique du Quebec in Quebec, Canada were also analyzed as part of the external quality assessment schemes. The obtained values were always in good agreement with the target values (t test,$95\%$). The detection limit was 0.02 µg/L. Values <10 μg/dL were considered normal.
## 2.3. Thyroid evaluation
Thyroid function tests were performed at the University Hospital of Ribeirão Preto Medical School, University of São Paulo (USP), Brazil. Measurements of TSH, FT4, anti-TPO, and anti-Tg antibodies were performed using a chemiluminescence method (IMMULITE ® 2000, DPC, Cirrus Inc., Los Angeles, CA) in a single assay. The sensitivity and intra-assay error were 0.02 µU/mL and $3.13\%$ (for TSH), and 0.15 ng/dL and $5.2\%$ for FT4 (analogous method). Reference values within the normal range were defined as: TSH = 0.4 to 4.5 mU/L, FT4 = 0.8 to 1.78 ng/dL, anti-TPO < 35UI/L, and anti-Tg < 40UI/L.
Subclinical hypo/hyperthyroidism was defined as thyroid hormone levels within the normal range and TSH concentrations above or below the defined normal reference range of 0.4 to 4.5 mU/L, respectively.
Thyroid ultrasonography (US) evaluations were performed at the Naval Hospital in Esmeraldas City, PRO-VIDA Hospital of Latacunga City for individuals from Pujilí, and La Merced Clinic in Quito. All US evaluations were performed using a Philips high-resolution ultrasound apparatus (Diamond Select HD11 XE, Bothell, WA). All the patients were classified according to the American College of Radiology - Thyroid Imaging - Reporting and Data System score,[17] by the same radiologist. Fine-needle aspiration biopsy (FNA) was performed in thyroid nodules ≥10 mm with high and intermediate suspicion sonographic pattern according to the 2015 American Thyroid Association guideline[18] and in nodules ≥15 mm and ≥20 mm in those with low suspicion and very low suspicion, respectively. Patients from HE areas were referred to Quito City. FNA was performed by the same radiologist. Cytological results were obtained by a single pathologist and reported using the 2015 BETHESDA system.[19]
## 2.4. Statistical analysis
Parametric or non-parametric tests were performed according to the distribution of quantitative variables using the Shapiro–Wilk test, Student T or Mann–Whitney test. Qualitative variables were analyzed using the chi-square test. Inferential statistics were calculated using the Spearman correlation coefficient. Statistical analyses were conducted using the SPSS software (version 22.0; IBM, Chicago, IL), and the level of significance was set at $P \leq .05.$
The quantitative variables were analyzed using quantitative methods, and grouped according to the reference value as normal, higher and lower than the reference value, for which a qualitative analysis of this grouping was performed.
## 3. Results
186 and 200 subjects were recruited from the high and low exposure areas, respectively. However, 2 HE patients were excluded because they were under 18 years of age. Thus, we applied the data collection and blood sampling form to 384 participants. Due to the pandemic and the mobility of the patients, we performed 300 thyroid ultrasounds. 29 subjects required FNA, however, only in 18 patients were performed.
## 3.1. BPb levels
First, we compared the total BPb levels among all individuals living in areas considered to be highly exposed to lead (Esmeraldas City and the parish of La Victoria) in relation to Quito. The BPb levels observed in the HE areas were higher than those in the LE area (8.2 ± 8.6 vs 3.0 ± 2.26 µg/dL, $P \leq .001$). In both euthyroid and hypothyroid individuals, the BPb levels were higher in HE when compared to LE areas (8.5 ± 7.4 vs 3.2 ± 2.4 and 7.8 ± 9.8 vs 2.7 ± 1.9, respectively, yielding P values <.001 for both comparisons). Comparisons of BPb concentrations in euthyroid and hypothyroid individuals between the HE areas of Esmeraldas and La Victoria revealed no significant differences.
We stratified all studied individuals according to BPb levels, considering high values those >10 μg/dL and low values with BPb <10 μg/dL. In this context, 44 out of 384 individuals (whole group), 26 out of 197 euthyroid individuals, and 18 out of 187 hypothyroid individuals exhibited high BPb levels. Among the individuals exhibiting BPb >10 µg/dL, 42 were from HE and 2 were from LE areas.
HE individuals lived longer in highly exposed areas (39.8 ± 18.7 years) when compared to individuals living in LE areas (33.9 ± 16.8 years, $$P \leq .02$$). Hypothyroid individuals living in HE areas lived longer in these areas than euthyroid individuals ($$P \leq .02$$). In both hypothyroid and euthyroid individuals, BPb levels correlated positively with lead exposure time; however, this correlation was only significant in hypothyroid patients ($$P \leq .0003$$). The major demographic, clinical, and laboratory characteristics of patients are shown in Table 1.
Evaluation of the influence of BPb levels on thyroid function was performed only in euthyroid individuals, as hypothyroid patients were treated with levothyroxine. Euthyroid individuals exhibiting high BPb levels did not present clinical manifestations indicative of thyroid dysfunction. No significant differences were observed in TSH, FT4, and autoantibody levels. Although healthy individuals did not complain and/or present signs of thyroid dysfunction, in the HE area, 15 subjects had TSH values above 4.5 mU/L and 4 subjects had TSH values lower than 0.4 mU/L, values defined as subclinical hypo and hyperthyroidism, respectively. A similar proportion of individuals was observed in the LE area (21 with subclinical hypothyroidism and 2 with subclinical hyperthyroidism) ($$P \leq .94$$). Therefore, BPb levels did not influence TSH, FT4, or autoantibody levels in the euthroids (Table 2).
**Table 2**
| Variables | All subjects | All subjects.1 | HE area | HE area.1 | LE area | LE area.1 | Crude OR (CI 95%) |
| --- | --- | --- | --- | --- | --- | --- | --- |
| Variables | n = 300 | % | n = 143 | % | n = 157 | % | Crude OR (CI 95%) |
| Ultrasound diagnosis | Ultrasound diagnosis | | | | | | |
| Nodules | 150 | 50 | 69 | 48.2 | 81 | 51.6 | Reference |
| Thyroiditis | 7 | 2.3 | 7 | 4.9 | – | – | 0.97 (0.61–1.53) |
| Normal | 143 | 47.7 | 67 | 46.9 | 76 | 48.4 | 0.97 (0.61–1.53) |
| Nodule number | Nodule number | | | | | | |
| Solitary | 65 | 42.5 | 29 | 40.3 | 36 | 44.4 | 1.19 (0.62–2.26) |
| Multiple | 88 | 57.5 | 43 | 59.7 | 45 | 55.6 | Reference |
| Nodule size (n = 309) | Nodule size (n = 309) | | | | | | |
| ≥1 cm | 79 | 25.6 | 53 | 35.1 | 26 | 16.5 | Reference |
| <1 cm | 230 | 74.4 | 98 | 64.9 | 132 | 83.5 | 2.75 (1.60–4.70)* |
| ACR-TI-RADS (n = 157) | ACR-TI-RADS (n = 157) | | | | | | |
| TI-RADS 1 | 40 | 25.5 | 23 | 30.2 | 17 | 21.0 | 1.12 (0.49–2.58) |
| TI-RADS 2 | 16 | 10.2 | 4 | 5.3 | 12 | 14.8 | 3.25 (0,82–12.88) |
| TI-RADS 3 | 34 | 21.7 | 14 | 18.4 | 20 | 24.7 | 1.55 (0.55–4.38) |
| TI-RADS 4 | 42 | 26.8 | 22 | 28.9 | 20 | 24.7 | 0.98 (0.37–2.65) |
| TI-RADS 5 | 25 | 15.9 | 13 | 17.1 | 12 | 14.8 | Reference |
| Cytological diagnosis (n = 18) | Cytological diagnosis (n = 18) | Cytological diagnosis (n = 18) | | | | | |
| BETHESDA I | 2 | 11.1 | 2 | 16.7 | – | – | – |
| BETHESDA II | 14 | 77.8 | 10 | 83.3 | 4 | 66.7 | – |
| BETHESDA III | – | – | – | – | – | – | – |
| BETHESDA IV | 1 | 5.6 | – | – | 1 | 16.7 | – |
| BETHESDA V | 1 | 5.6 | – | – | 1 | 16.7 | – |
| | x | SD | x | SD | x | SD | P value |
| Thyroid volume | 6.5 | 11.3 | 11.0 | 15.6 | 6.1 | 3.1 | <.001 |
| Nodules per patient | 2.0 | 1.1 | 2.1 | 1.1 | 2.0 | 1.1 | .4 |
| Nodule size | 0.7 | 0.6 | 10 | 0.8 | 0.6 | 0.4 | <.001 |
Thyroid ultrasound was performed in 300 individuals: 143 with HE and 157 with LE. In both euthyroid and hypothyroid individuals: the thyroid volume was greater in the HE zone when compared to LE area (10.2 ± 14.3 vs 6.3 ± 3.1 cm3, and 11.8 ± 16.9 vs 2.0 ± 1.1 cm3, values $P \leq .024$ and $$P \leq .006$$, respectively), the size of thyroid nodule was increased in individuals from HE areas when compared to LE area (0.9 ± 0.7 cm vs 0.4 ± 0.1, and 0.9 ± 01. vs 0.5 ± 0.1 cm, respectively ($P \leq .001$ for both comparisons). The number of nodules per individual was only increased in euthyroid subjects from HE area compared to LE areas (2.4 ± 1.2 vs 1.9 ± 1.0, $$P \leq .04$$).
Eleven euthyroid subjects (8 from HE and 3 from LE) and 14 hypothyroid patients (5 from HE and 9 from LE) exhibited American College of Radiology - Thyroid Imaging - Reporting and Data System-5 (high suspicion of malignancy). Among the ultrasound-guided FNA procedures performed, only 1 patient from the LE area exhibited TC.
## 4. Discussion
Environmental contaminants have been reported as possible causes of the disruption of thyroid function, influencing the development of benign and malignant thyroid nodules.[14] Several studies have reported an association between lead exposure and thyroid structural and functional changes,[9] as a putative environmental factor. In contrast, others studies have not reported such associations.[5,13] We studied 2 Ecuadorian areas, previously reported to have increased lead exposure, including the sea level city of Esmeraldas and the mountainous parish of La Victoria.[6] Currently, approximately 30,000 inhabitants of the Esmeraldas city have lived in legal and illegal settlements 1 to 5 kilometers from the petrol refinery, being exposed to petroleum derivatives and petroleum waste.[5] Although lead contamination in the parish of La Victoria (around 3000 inhabitants) has been associated with adult workers, laboring on ceramic utensil manufacturing, high levels of lead contaminations have been observed both in infants to adults.[6] *In this* study, we confirmed that the areas of Esmeraldas and La Victoria did exhibit increased BPb levels when compared to Quito. Indeed, a previous study evaluating 200 children (5–12 years old) performed by our group revealed that more than half of the children living <5 kilometers from the refinery and one-third of the children living <30 km from the refinery exhibited BPb levels of >5 μg/dL (unpublished results). In contrast, increased BPb levels have been reported for more than 30 years in La Victoria,[6] and recent studies have shown that BPb levels decreased from that time to 2015,[19] probably because of population awareness regarding the source of Pb in this area.
Regarding thyroid function, no relationship was found between BPb levels and thyroid function as evaluated using TSH levels. Our results are in agreement with those of previous studies that reported no relationship between lead exposure and TSH concentration in lead smelter workers.[13,20,21] Similarly, this study showed no association between BPb levels and FT4 levels, corroborating the results reported by other groups in children[22] or adults.[13] Finally, a meta-analysis published in 2016 also found no relationship between BPb, T4, and FT4 in lead-exposed and non-exposed groups.[23] In contrast, a study evaluating car battery factory workers reported that TSH and FT4 levels varied according to BPb levels.[24] Qualitative or quantitative analyses of autoantibodies (anti-TPO and anti-Tg) against thyroid antigens revealed no significant differences among the studied populations, yielding similar percentages among individuals living in HE or LE areas. These results were similar to those observed in healthy North American individuals.[25] Few studies have reported the effects of lead on thyroid nodules, particularly in women,[15] and much needs to be learned about the association between nodular goiters and lead contamination. Considering that the majority of the individuals in this study were women, primarily exposed to lead in the house environment (Esmeraldas) and during household manufacturing (Pujilí); individuals from HE areas exhibited larger thyroid volumes and larger thyroid nodules when compared to LE areas in both euthyroid and hypothyroid individuals; hypothyroid individuals lived longer in these areas than euthyroid individuals; and a positive correlation with the duration of environmental exposure was observed only in hypothyroid patients, suggesting a relationship between nodular goiter, hypothyroidism, and Pb exposure. Although the number of thyroid nodules increases with age,[26] further studies are needed to clarify the association between thyroid goiter and Pb contamination.
Despite the large number of thyroid nodules reported, the majority were smaller than 1 centimeter, exhibiting no indication for FNA.[17] Notwithstanding, only 1 TC was observed in a patient from an LE area, covering a non-representative sample for further conclusions regarding the association between lead exposure and TC.
In conclusion, the present study confirmed high exposure to Pb in the locations of Esmeraldas and La Victoria de Pujilí compared with Quito. Although high BPb levels were observed in these areas, no relationship was observed between TSH and FT4 levels and autoantibody levels. Nevertheless, larger thyroid glands, with a greater number of nodules exhibiting larger dimensions, were found in areas with high Pb exposure. Although there has been an improvement in Pb contamination in La Victoria de Pujilí over the years, further measures to control contamination are yet to be implemented. This is the first report to evaluate Pb contamination in Esmeraldas and deserves attention to prevent environmental complications related to Pb exposure.
## Acknowledgments
The authors thank Ms. Rosemeire de Paula Braz, Siemens Healthiness representative, São Paulo, Brazil, for providing kits for measuring TSH, T4L, anti-peroxidase, and anti-thyroglobulin antibodies using the Immulite system. The authors thank Geraldo Cassio Reis, a biostatistician, for analyzing the data and reviewing the manuscript
## Author contributions
Conceptualization: José Estefano Rivera-Buse, Léa Maria Zanini Maciel.
Data curation: José Estefano Rivera-Buse, Sheila Jissela Patajalo-Villalta, Patrícia Künzle Ribeiro Magalhães.
Formal analysis: José Estefano Rivera-Buse, Sheila Jissela Patajalo-Villalta, Léa Maria Zanini Maciel.
Funding acquisition: Léa Maria Zanini Maciel.
Investigation: José Estefano Rivera-Buse, Sheila Jissela Patajalo-Villalta, Fernando Barbosa Junior, Patrícia Künzle Ribeiro Magalhães.
Methodology: José Estefano Rivera-Buse, Léa Maria Zanini Maciel.
Project administration: José Estefano Rivera-Buse, Sheila Jissela Patajalo-Villalta.
Resources: Léa Maria Zanini Maciel.
Validation: Léa Maria Zanini Maciel.
Visualization: Patrícia Künzle Ribeiro Magalhães.
Writing – original draft: José Estefano Rivera-Buse, Eduardo Antônio Donadi, Léa Maria Zanini Maciel.
Writing – review & editing: Eduardo Antônio Donadi, Léa Maria Zanini Maciel.
## References
1. [1]The Iodine Global Network. Global Iodine Scorecard and Map [Internet]. Zurich: The Iodine Global Network. 2021. Available at: http://www.ign.org/scorecard.htm.. *Global Iodine Scorecard and Map* (2021)
2. Monchas Torres LG. *Prevalencia de hipotiroidismo tanto clínico como subclínico y su efecto sobre el perfil lipídico en pacientes con diabetes mellitus tipo 2, pertenecientes al Club de Diabetes del Hospital de Especialidades de las Fuerzas Armadas en la ciudad de Quito - Ecuador* (2015)
3. Vázquez VM, Rojas J, Bermudez V. **Comportamiento epidemiológico del hipotiroidismo en pacientes con diabetes mellitus tipo 2 en la ciudad de Loja – Ecuador.**. *Rev Latinoam Hipertens* (2013) **8** 95-102
4. **Registro Nacional De Tumores**. (2018)
5. **Auditoría Interna de EP PETROECUADOR.**. (2010)
6. Counter SA, Buchanan LH, Ortega F. **Blood lead levels in andean infants and young children in ecuador: an international comparison.**. *J Toxicol Environ Heal Part A* (2015) **78** 778-87
7. Alarcon WA, Davidson S, Dufour B. **Elevated blood lead levels among employed adults – United States, 1994–2013.**. *MMWR Morb Mortal Wkly Rep* (2016) **63** 59-65. PMID: 27736830
8. Pekcici R, Kavlakoğlu B, Yilmaz S. **Effects of lead on thyroid functions in lead-exposed workers.**. *Cent Eur J Med* (2010) **5** 215-8
9. Yousif A, Ahmed A. **Effects of cadmium (Cd)and lead (Pb)on the structure and function of thyroid gland.**. *Afr J Env Sci Technol* (2009) **3** 78-85
10. Dundar B, Öktem F, Arslan MK. **The effect of long-term low-dose lead exposure on thyroid function in adolescents.**. *Environ Res* (2006) **101** 140-5. PMID: 16360141
11. Gharaibeh MY, Alzoubi KH, Khabour OF. **Lead exposure among five distinct occupational groups: a comparative study.**. *Pak J Pharm Sci* (2014) **27** 39-43. PMID: 24374433
12. Abdelouahab N, Mergler D, Takser L. **Gender differences in the effects of organochlorines, mercury, and lead on thyroid hormone levels in lakeside communities of Quebec (Canada).**. *Environ Res* (2008) **107** 380-92. PMID: 18313043
13. Erfurth EM, Gerhardsson L, Nilsson A. **Effects of lead on the endocrine system in lead smelter workers.**. *Arch Environ Health* (2001) **56** 449-55. PMID: 11777027
14. Rezaei M, Javadmoosavi SY, Mansouri B. **Thyroid dysfunction: how concentration of toxic and essential elements contribute to risk of hypothyroidism, hyperthyroidism, and thyroid cancer.**. *Environ Sci Pollut Res* (2019) **26** 35787-96
15. Li H, Li X, Liu J. **Correlation between serum lead and thyroid diseases: papillary thyroid carcinoma, nodular goiter, and thyroid adenoma.**. *Int J Environ Health Res* (2017) **27** 409-19. PMID: 28891673
16. Counter SA, Buchanan LH, Ortega F. **Environmental lead contamination and pediatric lead intoxication in an Andean Ecuadorian village.**. *Int J Occup Environ Health* (2000) **6** 169-76. PMID: 10926719
17. Tessler FN, Middleton WD, Grant EG. **ACR Thyroid Imaging, Reporting and Data System (TI-RADS): white paper of the ACR TI-RADS committee.**. *J Am Coll Radiol* (2017) **14** 587-95. PMID: 28372962
18. Haugen BR, Alexander EK, Bible KC. **2015 American thyroid association management guidelines for adult patients with thyroid nodules and differentiated thyroid cancer: the American thyroid association guidelines task force on thyroid nodules and differentiated thyroid cancer.**. *Thyroid* (2015) **26** 1-133
19. Cibas ES, Ali SZ. **The 2017 Bethesda system for reporting thyroid cytopathology.**. *J Am Soc Cytopathol* (2017) **6** 217-22. PMID: 31043290
20. Ortega F, Counter SA, Buchanan LH. **Tracking blood lead and zinc protoporphyrin levels in andean adults working in a lead contaminated environment.**. *J Toxicol Environ Heal - A Curr Issues* (2013) **76** 1111-20
21. Chen A, Kim SS, Chung E. **Thyroid hormones in relation to lead, mercury, and cadmium exposure in the national health and nutrition examination survey, 2007-2008.**. *Environ Health Perspect* (2013) **121** 181-6. PMID: 23164649
22. Krieg EF. **The relationships between blood lead levels and serum thyroid stimulating hormone and total thyroxine in the third National Health and Nutrition Examination Survey.**. *J Trace Elem Med Biol* (2019) **51** 130-7. PMID: 30466922
23. Siegel M, Forsyth B, Siegel L. **The effect of lead on thyroid function in children.**. *Environ Res* (1989) **49** 190-6. PMID: 2753005
24. Krieg EF. **A meta-analysis of studies investigating the effects of occupational lead exposure on thyroid hormones.**. *Am J Ind Med* (2016) **59** 583-90. PMID: 27094769
25. López CM, Piñeiro AE, Núñez N. **Thyroid hormone changes in males exposed to lead in the Buenos Aires Area (Argentina).**. *Pharmacol Res* (2000) **42** 599-602. PMID: 11058414
26. Westman KW, Erfurth EM, Hagmar L. **Increased prevalence of thyroid peroxidase antibodies (TPO-Ab) in women with glomerulonephritis.**. *Nephrol Dial Transplant* (1993) **8** 402-6. PMID: 8393543
|
---
title: Modifiers of radiation effects on breast cancer incidence revealed by a reanalysis
of archival data of rat experiments
authors:
- Tatsuhiko Imaoka
- Mayumi Nishimura
- Kazuhiro Daino
- Shizuko Kakinuma
journal: Journal of Radiation Research
year: 2023
pmcid: PMC10036094
doi: 10.1093/jrr/rrac090
license: CC BY 4.0
---
# Modifiers of radiation effects on breast cancer incidence revealed by a reanalysis of archival data of rat experiments
## Abstract
Cancer risk after exposure to ionizing radiation can vary between individuals and populations, but the impact of factors governing those variations is not well understood. We previously conducted a series of carcinogenesis experiments using a rat model of breast cancer, in which 1654 rats born in 2002–2012 were exposed to γ rays at various doses and ages with or without non-radiation factors including high-fat diet, parity and chemical carcinogens. We herein reanalyze the incidence data from these archival experiments to clarify the effect of age at exposure, attained age, radiation dose and non-radiation factors (i.e. fat, parity, chemicals and birth cohorts) on radiation-related mammary cancer incidence. The analysis used excess relative risk (ERR) and excess absolute risk (EAR) models as well as generalized interaction models. Age-at-exposure dependence displayed a peak of susceptibility at puberty in both the ERR and EAR models. Attained age decreased ERR and increased EAR per unit radiation dose. The dose response was concordant with a linear model. Dietary fat exhibited a supra-multiplicative interaction, chemicals represented a multiplicative interaction, and parity and birth cohorts displayed interactions that did not significantly depart from additivity or multiplicativity. Treated as one entity, the four non-radiation factors gave a multiplicative interaction, but separation of the four factors significantly improved the fit of the model. Thus, the present study supports age and dose dependence observed in epidemiology, indicates heterogenous interactions between radiation and various non-radiation factors, and suggests the potential use of more flexible interaction modeling in radiological protection.
## INTRODUCTION
Cancer risk after exposure to ionizing radiation can vary between individuals and between populations, but the factors that govern such variations are not well understood [1]. Risk of radiation-related cancer is generally described by excess relative risk (ERR) and excess absolute risk (EAR) models. In ERR models, the disease rate is expressed as a product of the baseline risk (i.e. risk in the unexposed population) and a relative risk that is a function of radiation dose. In the EAR model, the rate is formulated as a sum of baseline and radiation-related risks. Both baseline and radiation-related risks are subject to modifications by environmental, lifestyle and genetic factors that are characteristic to individuals and populations as well as the age of the subjects at risk and the age at exposure. The ERR model assumes a multiplicative interaction between radiation and the modifiers of the baseline risk, whereas the EAR model assumes an additive interaction. These risk models are representative of a spectrum of interactions that can, more realistically, range from antagonism, through sub-additivity, additivity and multiplicativity, to supra-multiplicativity.
Breast is one of the organs most susceptible to radiation-related carcinogenesis, yet the risk of breast cancer is also influenced by diverse lifestyle and environmental factors [2, 3]. The diversity in lifestyles and environments among populations often complicates the analysis of radiation-related cancer risk of many tissues including breast cancer. For example, Japanese and US populations have different baseline risks of breast cancer [4]. In a pooled analysis of multiple cohorts, the most preferred ERR model suggests different dose-related increases in ERR between Japanese and US populations, whereas the best EAR model infers that the extent of the increase in EAR per unit dose does not differ among populations [5]. The current radiological protection system therefore adopts the EAR model for risk transfer (i.e. prediction of radiation-related risk in non-Japanese populations based on the Japanese atomic bomb survivor data) for breast [6]. On the other hand, the breast cancer incidence has increased over time in the Japanese population itself. As a result, in the studies of the Life Span Study (LSS) cohort of Japanese atomic bomb survivors, the ERR model suggests an identical dose-related increase among different birth cohorts, whereas the EAR model indicates different dose-related increases among the birth cohorts [2, 7]. The birth cohort and the age at exposure are equivalent in the LSS cohort, as the exposure occurred exclusively in the year 1945. Importantly, in both the pooled analysis [5] and the LSS reports [2, 7], the ERR model suggests no significant impact of the age at exposure on radiation-related breast cancer risk, whereas the EAR model indicates a significant effect of the age at exposure. As such, the choice between these risk models greatly complicates the inference regarding application to radiological protection.
Animal experiments provide important information regarding the biological effects of radiation, and they complement epidemiological studies. Reanalysis of archival animal data can produce new important information required in many fields including radiological protection [8–11]. To date, many animal studies have addressed the interaction of radiation with various non-radiation factors [1]. Nevertheless, the interaction in these animal studies has only rarely been assessed quantitatively [12, 13]. We have conducted a series of carcinogenesis experiments using a rat model of breast cancer, in which rats were exposed to 137Cs γ rays and subjected to other non-radiation factors including a high-fat diet, parity (i.e. history of childbirth) and exposure to carcinogenic chemicals [14–21]. One of the rat experiments indicated a significant variation in the baseline cancer risk related to the birth cohort [22]. Thus, these previous data provide an opportunity to test the interaction between radiation dose and various non-radiation factors. As mentioned above, age is also an important modifier of radiation effects. Whereas many animal studies have focused on the influence of age at exposure [1], the effect of attained age has not been as rigorously studied. In our previous studies on rat mammary carcinogenesis [14–22], the weekly palpation data collected were connected with the pathological diagnosis upon autopsy, and thus this information provides an opportunity to retrospectively analyze the influence of attained age in those experiments.
The present study reanalyzes the pooled incidence data from these past experiments (consisting of 1654 rats in total) to quantify the interaction of the effects of high-fat diet, parity, chemicals and birth cohort with radiation. The study also evaluates the modification by age at exposure and attained age as well as the shape of the dose response. The results not only support the trends of age and dose dependence recently suggested by epidemiology but also suggest significant heterogeneity in the modification of radiation-associated risk by non-radiation factors.
## Animal experiments
The data sets from previous experiments are shown in Table 1. In these experiments, female Sprague–Dawley rats (Jcl:SD, CLEA Japan, Tokyo, Japan) were fed with a standard diet (CE-2, CLEA Japan), palpated weekly, and autopsied upon general health deterioration, natural death or predetermined termination endpoints for pathological diagnosis of the palpable mammary tumors [14–22]. Termination occurred at 90 weeks of age [18, 20, 22], 100 weeks of age [21] or was not set (i.e. the rats died naturally or were sacrificed only upon general deterioration) [19]; in a subset of studies, the rats with one or more palpable tumors were autopsied at 50 weeks of age, and those without palpable tumors at that time were checked until a tumor(s) could be palpated, at which time autopsy was done [14–17]. Acute whole-body irradiation with 137Cs γ rays (0.5–0.6 Gy/min) was performed once at either 1, 3, 7, 13 or 15 weeks of age. In some experiments, the rats were fed a high-fat diet ($23.5\%$ corn oil in AIN-76A, CLEA Japan) from 9 weeks of age [14–17], which increased the body weight by $12\%$. In other experiments, they were mated with male Jcl:SD rats and allowed to deliver and nurse a litter [21]. The postpubertally irradiated groups were excluded as this previous study did not identify any modifying effect of parity in those groups [21]. To evaluate the influence of chemical carcinogen exposure, the rats were intraperitoneally injected with 1-methy-1-nitrosourea (MNU) at 20 mg/kg at 3 or 7 weeks of age (denoted MNU$\frac{20}{3}$ and MNU$\frac{20}{7}$, respectively) or at 40 mg/kg at 7 weeks of age (denoted MNU$\frac{40}{7}$), or they were administered 2-amino-1-methyl-6-phenylimidazo[4,5-b]pyridine (PhIP, 40 mg/kg/day) daily for 10 days by gavage from 7 weeks of age with a two-day interval in the middle [14–16].
**Table 1**
| No. | Study focus | Ref. | n a | Birth cohort (date of birth)b | Parity | Fat in diet | Chemicalsc | Radiation (age at exposure) | Termination |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| 1 | Chemicals | (14, 15) | 417 | 1 (10 May 2002–17 Jan 2005) | No | High | None, MNU20/7, MNU40/7, PhIP | 0 Gy; 0.2, 0.5, 1, 2 Gy (7 wk) | Lifetimed |
| 2 | Puberty | (16) | 54 | 1 (18 Oct 2002–16 Jan 2003) | No | High | NoneMNU20/3, MNU20/7 | 2 Gy (3 wk)0 Gy | Lifetimed |
| 3 | Miscellaneous | This study | 159 | 1 (5 Dec 2002–3 Jan 2005)1 (16 Jan 2003) | NoNo | NormalHigh | NoneMNU20/3, MNU20/7 | 0 Gy; 0.2, 0.5, 1, 2 Gy (7 wk)2 Gy (3 wk) | Lifetimed |
| 4 | Carbon ions I | (17) | 49 | 1 (23 Aug 2002–28 Nov 2003) | No | High | | 0 Gy | Lifetimed |
| 5 | Carbon ions II | (18) | 570 | 1–2 (4 Oct 2004–16 Jul 2008) | No | Normal | | 0 Gy; 0.2, 0.5, 1, 2 Gy (1, 3, 7, 15 wk)e | 90 wk |
| 6 | Strain | (19) | 20 | 2 (9 Oct 2006–18 Jan 2007) | No | Normal | | 4 Gy (7 wk) | Lifetime |
| 7 | Neutrons | (20) | 78 | 2–3 (1 Dec 2009–6 Oct 2011) | No | Normal | | 0 Gy | 90 wk |
| 8 | Parity | (21) | 111 | 3 (7 Jul 2011–10 May 2012) | Yes, no | Normal | | 0 Gy; 4 Gy (3 wk) | 100 wk |
| 9 | Dose rate | (22) | 196 | 3 (5 Nov 2010–8 Nov 2012) | No | Normal | | 0 Gy; 0.5, 1, 2, 3, 4 Gy (13 wk) | 90 wk |
The experiment shown in number 3 of Table 1 is reported here for the first time. This experiment was conducted under approval by the Institutional Animal Care and Use Committee of the National Institute of Radiological Sciences (NIRS, approval number 17–1012). Female Jcl:SD rats were purchased from CLEA Japan and fed a standard CE-2 diet and sterile water ad libitum. Experiments were performed as described [14–16]. Briefly, rats were subjected to whole-body γ-irradiation (137Cs, 0.6 Gy/min) at 0, 0.2, 0.5, 1 or 2 Gy at 7 weeks of age. Another set of rats was irradiated at 2 Gy at 3 weeks of age and injected intraperitoneally with MNU (20 mg/kg) after 3 days (at 3 weeks of age) or at 7 weeks of age. These rats were fed a high-fat diet as described above after 9 weeks of age. Animals that showed signs of general deterioration or reached 50 weeks of age with one or more palpable tumors were euthanized and autopsied, animals found dead were also autopsied, and those without a palpable tumor at 50 weeks of age were sacrificed and autopsied upon later detection of the first tumor. During autopsy, palpable tumors were collected, fixed in $10\%$ formalin, embedded in paraffin and processed for hematoxylin and eosin staining for histology [23]. The palpation record was used to determine the age at which tumors first developed.
## Statistical analysis
The first palpation of a mammary tumor that was later diagnosed as adenocarcinoma was treated as an incidence of mammary cancer, whereas any termination of observation before that was treated as censoring of observation. The attained age was categorized into 10-week terms, and the data were organized as the number of incidences in individual terms, the number of animal-weeks (as an equivalent of person-years in epidemiology) in the terms, and covariates (the mid-point of the attained age terms, radiation dose and age at exposure as continuous variables; fat, parity, chemical treatment and birth cohort as categorical variables). Birth cohorts were defined as follows: cohort 1 (born from 10 May 2002 to 14 July 2006, $$n = 900$$), cohort 2 (born from 9 October 2006 to 21 September 2010, $$n = 411$$) and cohort 3 (born from 21 October 2010 to 8 November 2012, $$n = 343$$), so that they included nearly equal numbers of rats without any treatment ($$n = 146$$, 139 and 138, respectively). Animals entered the analysis at the first week after birth if no treatment had been performed or at the first week after the end of the treatments otherwise.
The incidence rate of palpable mammary carcinoma λ (i.e. the number of cases per animal-week) was modeled as follows: where t is attained age, A(t) is the baseline hazard function (for cohort 1 without exposure to any radiation or non-radiation factors), i is an indicator of non-radiation factors (where $i = 1$, 2, … and 8 represent high fat, parity, MNU$\frac{20}{3}$, MNU$\frac{20}{7}$, MNU$\frac{40}{7}$, PhIP, cohort 2 and cohort 3, respectively), Σ denotes sum for all i, βi is a coefficient for risk of non-radiation factor i, *Ii is* a dummy variable for a non-radiation factor i, d is radiation dose, e is age at exposure, R(d, e, t) is the radiation-associated risk, θ is a coefficient describing the interaction between radiation and all non-radiation factors as an aggregate, and θi is a coefficient describing the interaction between radiation and a non-radiation factor i. Herein, equation 4 is a special case of equation 5 in which all θi have identical values, and equations 1 and 3 are special cases of equation 4 in which θ = 1 and 0, respectively. The interaction among non-radiation factors was modeled as additive herein; however, whether this interaction is additive or multiplicative does not matter because, in the experiments above, only one of the non-radiation factors was changed while the others were fixed.
The specific forms of A(t), R(d, e, t) and the set of θi (θ) used herein were: where α0, α1, α2, γ0, γ1, γ2, γ3 and γQ are constant values, I is an indicator which is 1 when the condition specified by the subscript is true and otherwise 0, and θfat, θparity, θchem and θcohort are interaction coefficients for high fat, parity, chemicals and birth cohorts, respectively (identical parameters were used for chemicals [$i = 3$–6] and cohorts [$i = 7$, 8]). Equations 2 and 3 are equivalent under these conditions, but the form of equation 3 allows comparison with the generalized models. Poisson regression, profile likelihood confidence intervals (CIs), Akaike’s information criterion and the likelihood ratio test were used for model fitting and comparison on R [24].
## The non-radiation factors differentially impact baseline incidence
To obtain insights into the modification of radiation-related mammary cancer incidence by age and various non-radiation factors, animal carcinogenesis data from past experiments were pooled and reanalyzed. The non-radiation factors in these experiments included high-fat diet, parity, treatment with chemical carcinogens and birth cohort. First, the impact of the non-radiation factors on cancer incidence in the absence of radiation was visualized with Kaplan–Meier plots (Fig. 1A–D) and quantified (Table 2). Herein, a log linear (equation 6) and log linear-quadratic function (equation 7) of attained age were tested. This analysis indicated a negligible effect of high-fat diet, a significant reduction by parity, a significant increase by chemical treatments except MNU$\frac{20}{7}$, and a significant influence of birth cohort 3 (Table 2). The addition of the log quadratic term for attained age did not improve the fit (Table 2, Fig. 1E). Thus, this analysis clarified the effect of the four factors on baseline incidence, and a log linear function of attained age was adopted for the baseline function in the subsequent analyses.
## Influence of age factors and shape of the dose response in the present data set
Age at exposure and attained age are important modifiers of radiation effects, and conventional ERR and EAR models already include flexibility regarding their interaction. ERR (equation 1) and standard EAR (equation 2) models were thus used to analyze the dependence of mammary cancer incidence on these age factors. Detailed results for the fit to equations are shown in Table 3 (see ‘ERR’ and ‘EAR [standard]’ columns), and representative predictions from the models are shown in Fig. 2. Herein, the categorical estimates of incidence showed a non-monotonic relationship with age at exposure (Fig. 2A–B). In both the ERR and EAR models, the age-at-exposure trend was better modeled as a log linear spline function with a knot at a peripubertal age of 7 weeks (equation 8) than as a simple log linear function ($$P \leq 0.003$$ and 0.002, respectively). This biphasic modeling indicated a significant increasing trend (before 7 weeks) of both ERR and EAR (Table 3, see γ1 values) and a near-significant decreasing trend (after 7 weeks) of ERR and EAR, respectively (Table 3, see γ2 values). Dependence of attained age was modeled as a log linear function of attained age (equation 8) based on the trend of categorical estimates (Fig. 2C–D). Therein, the decreasing trend in ERR and the increasing trend in EAR were significant (Table 3, see γ3 values). A log linear-quadratic function of log attained age in the ERR and EAR models did not significantly improve from a log linear function ($$P \leq 0.2$$ and $$P \leq 0.4$$, respectively). Thus, the effects of age at exposure and attained age were described with a log linear spline function with a knot at 7 weeks and a log linear function, respectively, as in equation 8.
The shape of the dose response was modeled as linear or linear-quadratic in the ERR and standard EAR models. Having hundreds of animals per dose group, the categorical estimates were significant at doses ≥0.5 Gy, but not at 0.2 Gy ($$P \leq 0.33$$), compared with 0 Gy (Fig. 2E–F). These estimates exhibited an apparently linear relationship with dose (Fig. 2E–F), and addition of the quadratic term did not significantly improve the fit from linearity in either the ERR or EAR model ($$P \leq 0.6$$ for both models). Thus, a linear function was used for the dose response hereafter.
## Heterogeneity exists in interactions between radiation and various non-radiation factors
We next set out to elucidate the interaction of non-radiation factors with radiation. For this purpose, the data were analyzed using Generalized Models 1 and 2 (equations 4 and 5). The variant EAR model (equation 3), nested in Generalized Models 1 and 2 and equivalent to the standard EAR model, was used only to make statistical comparisons between models. In Generalized Model 1, the optimal θ value determined by fitting indicates the magnitude of the interaction between radiation and the non-radiation factors (i.e. θ < 0, antagonistic; θ = 0, additive; 0 < θ < 1, supra-additive and sub-multiplicative; θ = 1, multiplicative; θ > 1, supra-multiplicative); the same applies to θi in Generalized Model 2. The details of fitting results are shown in Table 3 (see ‘EAR [variant]’, ‘Generalized 1’ and ‘Generalized 2’), and categorical estimates for representative groups and example predictions from the models are illustrated in Fig. 3. As a result, Generalized Model 1, which treats all non-radiation factors as an aggregate, significantly ($P \leq 0.001$) improved from the variant EAR model (a model assuming additivity) but not from the ERR model (which assumes multiplicativity) (Table 3). Generalized Model 1 gave a θ value of 0.7 ($95\%$ CI [0.3, 1.9]), which suggested a significant departure from additivity and supported a multiplicative interaction (Table 3). Examples shown in Fig. 3 illustrate the general trend of good fit of the ERR model and Generalized Model 1 compared with the EAR model.
**Fig. 3:** *Comparison of four models on the interaction of non-radiation factors with radiation on mammary cancer incidence in a pooled cohort of rats. Dots and vertical bars are shown commonly for all models and indicate categorical estimates and their 95% CIs, respectively, calculated without assuming dose dependence or interactions. In A–D, a single model (boxed) was used to analyze all data, and the results are shown in grey and black lines for representative groups of the individual four categories of non-radiation factors (shown at the top). The EAR model herein refers to both standard and variant EAR models as these are equivalent mathematically. kAW, 103 animal-weeks; MNU20/7, 20 mg/kg of MNU at age 7 weeks.*
Of note, Generalized Model 2, which treats the four non-radiation factors as having different interactions, was significantly improved from the ERR model, variant EAR model and Generalized Model 1 ($$P \leq 0.002$$ or $P \leq 0.001$) (Table 3). This model gave a θfat value of 30 (lower limit of $95\%$ CI, 1.4), which is significantly larger than 1 and hence indicated a supra-multiplicative interaction between radiation and high-fat diet. Regarding parity, this model gave a θparity value of 0.3, with its $95\%$ CI covering both 0 and 1, indicating that the interaction could not be determined as either additive or multiplicative. The θchem value was 0.4 ($95\%$ CI [0.06, 1.3]), which significantly departed from 0 and supported a multiplicative interaction between radiation and chemical treatments. Regarding the birth cohorts, the model yielded a θcohort value of 0.4 ($95\%$ CI [−0.3, 4.4]), indicating an interaction that could not be determined as either additive or multiplicative. Examples in Fig. 3 illustrate the better fit of Generalized Model 2 than the other three models. In addition, use of generalized models did not overtly change the parameter estimates from the ERR and EAR models (Table 3).
Thus, the analysis indicated that the interactions of these four factors with radiation are diverse and should hence be treated separately for the model to better describe the data. Further, when these factors were treated as one entity, the results showed a multiplicative interaction with radiation.
## DISCUSSION
In the present study, the pooled incidence data from past experiments (involving 1654 rats) were reanalyzed to quantify: [1] the modification by age at exposure and attained age; [2] the shape of the dose response; and [3] the interaction of the effects of high-fat diet, parity, chemicals and birth cohorts with radiation. Regarding the age effects, the present analysis supports a peripubertal peak in susceptibility as well as a change associated with attained age, both of which are concordant with a previous report on the LSS cohort [7]. The dose response exhibited linearity, but the incidence in the lowest dose group (0.2 Gy, $$n = 213$$ rats) did not differ significantly from the baseline incidence ($$n = 614$$ rats). Importantly, there was significant heterogeneity—ranging from additivity to supra-multiplicativity—in the modification of radiation-associated risk by non-radiation factors.
Compared with recent findings in epidemiology, the present results support their main conclusions and provide a guide for future research. First, regarding the effect modification by age at exposure, the recent LSS study suggested a non-monotonic trend with a knot around the pubertal period [7], which is supported by the present study. Our analysis showed a peak of ERR and EAR around puberty, which coincides with the results of ERR, rather than EAR, in the LSS cohort [7]. Regarding the modification by attained age, the previous two studies on the LSS cohort support a significant decreasing trend in ERR [2, 7]. The curve for the increase of EAR with attained age was convex upward in both studies. The contribution of the log quadratic term of attained age to the modification of EAR, suggested in the LSS cohort [7], was not reproduced in the present rat cohort; this may reflect the difference in the attained-age dependence of the baseline incidence between the cohorts. Second, regarding the dose response, both studies indicate linear relationships, with no significant departure from linearity. The present study supports a strong linearity up to a high dose of 4 Gy, although the linearity was attenuated over 2 Gy in the LSS cohort [7]. In this sense, the prominent dose rate effect observed in the rats [22] cannot be explained by the curvature of dose response, as the current radiological protection system assumes [25]. Third, regarding the effect modification by non-radiation factors, the LSS study found no significant modification of ERR by body mass index, parity or smoking [7], indicating no significant departure from simple multiplicativity. In contrast, the present study suggests the possibility of more diverse interactions by similar non-radiation factors, i.e. dietary fat, parity and carcinogenic chemicals. Relatively small variations in such factors may exist within the Japanese female population of the LSS cohort, suggesting the need for a pooled analysis of cohorts including non-Japanese populations with information about lifestyle factors. The birth cohort in the LSS and the present studies may have different meanings, as that in the former may represent the rapid Westernization of lifestyles in Japan whereas the latter may include genetic fluctuation (note that Jcl:SD is an outbred closed colony strain), changes in breeding conditions, difference in the palpation skill and variability of the natural components in diet.
The heterogeneity in the interaction of various factors suggests diversity in their mechanisms of interaction. The peripubertal peak of the age-at-exposure dependence suggests its association with the very rapid cell proliferation that occurs during puberty [26]. The genome of radiation-induced rat mammary cancer frequently displays large structural abnormalities [27–29], suggesting the influence of error-prone end-joining machinery. The effect of Brca1 knockout on radiation-induced mammary carcinogenesis was minimal in rats irradiated immediately after, but not before, puberty [30]. These findings imply that the error-free DNA repair by homologous recombination is minimal during puberty, allowing genesis of structural abnormalities by error-prone repair mechanisms. Regarding the interaction with non-radiation factors, it is generally known from biologically based mathematical models that two factors acting on the initiation step of carcinogenesis tend to interact additively, those acting separately on the initiation and promotion/progression steps display multiplicative interactions and those acting on promotion/progression steps show variable interactions [31–33]. The supra-multiplicative modification by dietary fat may thus involve its action on the promotion/progression phase, which could include (i) upregulation of estrogen production via adipocyte aromatase, (ii) elevation of cell proliferation via inflammatory adipokines, and (iii) elevation of protein synthesis and cell proliferation via circulating insulin and leptin [34–36]. The multiplicative interaction of mutagenic chemicals (MNU and PhIP) is odd given that they may act on the initiation step—as would radiation. Nevertheless, previous studies have suggested promotion-like effects of radiation on cells harboring chemically induced mutations [14, 15, 37]. Reduction of such supra-multiplicative and multiplicative interactions would establish a path for deliberately and retrospectively controlling cancer risks from known radiation exposure in the past [38], as these interactions were observed in situations where dietary fat or chemicals were applied subsequent to radiation exposure.
Limitation of the present study may include the insufficient number of animals, resulting in a lack of statistical power for identifying the interaction between radiation and other factors.
Taken together, the present reanalysis of archival rat mammary carcinogenesis data supports major conclusions from epidemiology on the dose response and effect modification by age. It also suggests significant heterogeneity in the interactions of non-radiation factors with radiation. Modeling of the interaction between radiation and non-radiation factors did not affect the shape of dose response or age effects, but its potential impact would be substantial for risk estimation in populations exposed to factors that exhibit supra-multiplicative interactions. Clarification of the effect modification by non-radiation factors will contribute to improvement in risk transfer from epidemiology of atomic bomb survivors to populations worldwide [39] and consideration of radiation risk in the context of overall environmental exposures (or the ‘exposome’) [40]. Further analyses of other archival data—especially on cancer risk of other tissues—are thus warranted.
## CONFLICT OF INTEREST
The authors confirm they have no conflicts of interest to declare.
## FUNDING
This work was supported in part by the Japan Society for the Promotion of *Science via* KAKENHI JP21H03601 to TI.
## PRESENTATION AT A CONFERENCE
64th Annual Meeting of Japanese Radiation Research Society, 81st Annual Meeting of the Japanese Cancer Association.
## DATA AVAILABILITY
The data will be shared on reasonable request to the corresponding author.
## References
1. Applegate KE, Ruhm W, Wojcik A. **Individual response of humans to ionising radiation: governing factors and importance for radiological protection**. *Radiat Environ Biophys* (2020.0) **59** 185-209. PMID: 32146555
2. Preston DL, Ron E, Tokuoka S. **Solid cancer incidence in atomic bomb survivors: 1958-1998**. *Radiat Res* (2007.0) **168** 1-64. PMID: 17722996
3. Tamimi RM, Spiegelman D, Smith-Warner SA. **Population attributable risk of modifiable and nonmodifiable breast cancer risk factors in postmenopausal breast cancer**. *Am J Epidemiol* (2016.0) **184** 884-93. PMID: 27923781
4. Peto J. **Cancer epidemiology in the last century and the next decade**. *Nature* (2001.0) **411** 390-5. PMID: 11357148
5. Preston DL, Mattsson A, Holmberg E. **Radiation effects on breast cancer risk: a pooled analysis of eight cohorts**. *Radiat Res* (2002.0) **158** 220-35. PMID: 12105993
6. **The 2007 Recommendations of the International Commission on Radiological Protection. ICRP Publication 103**. *Ann ICRP* (2007.0) **37** 1-332
7. Brenner AV, Preston DL, Sakata R. **Incidence of breast cancer in the life span study of atomic bomb survivors: 1958-2009**. *Radiat Res* (2018.0) **190** 433-44. PMID: 30044713
8. Morioka T, Blyth BJ, Imaoka T. **Establishing the Japan-Store House of Animal Radiobiology Experiments (J-SHARE), a large-scale necropsy and histopathology archive providing international access to important radiobiology data**. *Int J Radiat Biol* (2019.0) **95** 1372-7. PMID: 31145030
9. Schofield PN, Kulka U, Tapio S. **Big data in radiation biology and epidemiology; an overview of the historical and contemporary landscape of data and biomaterial archives**. *Int J Radiat Biol* (2019.0) **95** 861-78. PMID: 30888231
10. Tran V, Little MP. **Dose and dose rate extrapolation factors for malignant and non-malignant health endpoints after exposure to gamma and neutron radiation**. *Radiat Environ Biophys* (2017.0) **56** 299-328. PMID: 28939964
11. Haley BM, Paunesku T, Grdina DJ. **The increase in animal mortality risk following exposure to sparsely ionizing radiation is not linear quadratic with dose**. *PLoS One* (2015.0) **10** e0140989. PMID: 26649569
12. Mauderly JL, Seilkop SK, Barr EB. **Carcinogenic interactions between a single inhalation of**. *Radiat Res* (2010.0) **173** 665-76. PMID: 20426667
13. Kakinuma S, Nishimura M, Amasaki Y. **Combined exposure to X-irradiation followed by**. *Mutat Res* (2012.0) **737** 43-50. PMID: 22706209
14. Imaoka T, Nishimura M, Teramoto A. **Cooperative induction of rat mammary cancer by radiation and 1-methyl-1-nitrosourea via the oncogenic pathways involving c-Myc activation and H-**. *Int J Cancer* (2005.0) **115** 187-93. PMID: 15688392
15. Imaoka T, Nishimura M, Doi K. **Molecular characterization of cancer reveals interactions between ionizing radiation and chemicals on rat mammary carcinogenesis**. *Int J Cancer* (2014.0) **134** 1529-38. PMID: 24105445
16. Imaoka T, Nishimura M, Iizuka D. **Pre- and postpubertal irradiation induces mammary cancers with distinct expression of hormone receptors, ErbB ligands, and developmental genes in rats**. *Mol Carcinog* (2011.0) **50** 539-52. PMID: 21374731
17. Imaoka T, Nishimura M, Kakinuma S. **High relative biologic effectiveness of carbon ion radiation on induction of rat mammary carcinoma and its lack of H-**. *Int J Radiat Oncol Biol Phys* (2007.0) **69** 194-203. PMID: 17707273
18. Imaoka T, Nishimura M, Daino K. **Influence of age on the relative biological effectiveness of carbon ion radiation for induction of rat mammary carcinoma**. *Int J Radiat Oncol Biol Phys* (2013.0) **85** 1134-40. PMID: 23084769
19. Nishimura M, Daino K, Fukuda M. **Development of mammary cancer in γ-irradiated F**. *PLoS One* (2021.0) **16** e0255968. PMID: 34388197
20. Imaoka T, Nishimura M, Daino K. **Age modifies the effect of 2-MeV fast neutrons on rat mammary carcinogenesis**. *Radiat Res* (2017.0) **188** 419-25. PMID: 28809605
21. Takabatake M, Daino K, Imaoka T. **Differential effect of parity on rat mammary carcinogenesis after pre- or post-pubertal exposure to radiation**. *Sci Rep* (2018.0) **8** 14325. PMID: 30254198
22. Imaoka T, Nishimura M, Daino K. **Prominent dose-rate effect and its age dependence of rat mammary carcinogenesis induced by continuous gamma-ray exposure**. *Radiat Res* (2019.0) **191** 245-54. PMID: 30543491
23. Mohr U. *International Classification of Rodent Tumours. Part I: The Rat. 5. Integumentary Systems* (1993.0)
24. 24.
R Core Team
. R: A Language and Environment for Statistical Computing. https://www.R-project.org/ (12 February 2022, date last accessed). *R: A Language and Environment for Statistical Computing*
25. Clement C, Ruehm W, Harrison JD. **Keeping the ICRP recommendations fit for purpose**. *J Radiol Prot* (2021.0) **41** 1390-409. DOI: 10.1088/361-6498/ac1611
26. Russo J, Russo IH. **Influence of differentiation and cell kinetics on the susceptibility of the rat mammary gland to carcinogenesis**. *Cancer Res* (1980.0) **40** 2677-87. PMID: 7388818
27. Moriyama H, Daino K, Imaoka T. **Neutron-induced rat mammary carcinomas are mainly of luminal subtype and have multiple copy number aberrations**. *Anticancer Res* (2019.0) **39** 1135-42. PMID: 30842142
28. Moriyama H, Daino K, Ishikawa A. **Exome of radiation-induced rat mammary carcinoma shows copy-number losses and mutations in human-relevant cancer genes**. *Anticancer Res* (2021.0) **41** 55-70. PMID: 33419799
29. Iizuka D, Imaoka T, Takabatake T. **DNA copy number aberrations and disruption of the p16Ink4a/Rb pathway in radiation-induced and spontaneous rat mammary carcinomas**. *Radiat Res* (2010.0) **174** 206-15. PMID: 20681787
30. Nakamura Y, Kubota J, Nishimura Y. *Cancer Sci* **113** 3362-75. DOI: 10.1111/cas.15485
31. Kodell RL, Krewski D, Zielinski JM. **Additive and multiplicative relative risk in the two-stage clonal expansion model of carcinogenesis**. *Risk Anal* (1991.0) **11** 483-90. PMID: 1947353
32. Zielinski J, Kodell R, Krewski D. **Interaction between two carcinogens in the two-stage clonal expansion model of carcinogenesis**. *J Epidemiol Biostat* (2001.0) **6** 219-28. PMID: 11434501
33. Kawaguchi I, Doi M, Kakinuma S. **Combined effect of multiple carcinogens and synergy index**. *J Theor Biol* (2006.0) **243** 143-51. PMID: 16860345
34. Ford NA, Devlin KL, Lashinger LM. **Deconvoluting the obesity and breast cancer link: secretome, soil and seed interactions**. *J Mammary Gland Biol Neoplasia* (2013.0) **18** 267-75. PMID: 24091864
35. Belardi V, Gallagher EJ, Novosyadlyy R. **Insulin and IGFs in obesity-related breast cancer**. *J Mammary Gland Biol Neoplasia* (2013.0) **18** 277-89. PMID: 24154546
36. Imaoka T, Nishimura M, Daino K. **A rat model to study the effects of diet-induced obesity on radiation-induced mammary carcinogenesis**. *Radiat Res* (2016.0) **185** 505-15. PMID: 27135968
37. Yamauchi K, Kakinuma S, Sudo S. **Differential effects of low- and high-dose X-rays on**. *Mutat Res* (2008.0) **640** 27-37. PMID: 18242641
38. Imaoka T, Ishii N, Kawaguchi I. **Biological measures to minimize the risk of radiotherapy-associated second cancer: a research perspective**. *Int J Radiat Biol* (2016.0) **92** 289-301. PMID: 26967256
39. Zhang W, Laurier D, Clero E. **Sensitivity analysis of parameters and methodological choices used in calculation of radiation detriment for solid cancer**. *Int J Radiat Biol* (2020.0) **96** 596-605. PMID: 31914349
40. Laurier D, Ruhm W, Paquet F. **Areas of research to support the system of radiological protection**. *Radiat Environ Biophys* (2021.0) **60** 519-30. PMID: 34657188
|
---
title: Nutrigenomic regulation of sensory plasticity
authors:
- Hayeon Sung
- Anoumid Vaziri
- Daniel Wilinski
- Riley KR Woerner
- Lydia Freddolino
- Monica Dus
journal: eLife
year: 2023
pmcid: PMC10036121
doi: 10.7554/eLife.83979
license: CC BY 4.0
---
# Nutrigenomic regulation of sensory plasticity
## Abstract
Diet profoundly influences brain physiology, but how metabolic information is transmuted into neural activity and behavior changes remains elusive. Here, we show that the metabolic enzyme O-GlcNAc Transferase (OGT) moonlights on the chromatin of the D. melanogaster gustatory neurons to instruct changes in chromatin accessibility and transcription that underlie sensory adaptations to a high-sugar diet. OGT works synergistically with the Mitogen Activated Kinase/Extracellular signal Regulated Kinase (MAPK/ERK) rolled and its effector stripe (also known as EGR2 or Krox20) to integrate activity information. OGT also cooperates with the epigenetic silencer Polycomb Repressive Complex 2.1 (PRC2.1) to decrease chromatin accessibility and repress transcription in the high-sugar diet. This integration of nutritional and activity information changes the taste neurons’ responses to sugar and the flies’ ability to sense sweetness. Our findings reveal how nutrigenomic signaling generates neural activity and behavior in response to dietary changes in the sensory neurons.
## Introduction
The levels and types of dietary nutrients play an essential role in cellular processes such as growth, division, and differentiation by providing fuel and biomass. However, nutrients can also affect these aspects of cell physiology by influencing, and often orchestrating, gene expression programs (Vaziri and Dus, 2021; Dai et al., 2020). These effects are mediated through nutrient-sensitive modifications to DNA, RNA, and proteins, as well as changes to the activity, binding, and localization of enzymes and signaling factors (Huang et al., 2015; Katada et al., 2012). These nutrigenomic signaling pathways – nutrigenomics is the field that studies food–genes interactions – could explain how the food environment affects the risk of non-communicable diseases such as diabetes, cancer, and neurodegeneration. They also hold the potential to uncover new interventions and treatments for these debilitating diseases. While the effects of nutrients on gene expression are well established, relatively little is known about the molecular mechanisms at the food–gene interface. A significant challenge of the field has been to explain how global variations in the nutrient environment lead to changes in cell physiology and behavior, especially in neuroscience. To overcome these challenges, we have developed an experimental system where the contributions of nutrients to physiology can be studied mechanistically and in vivo (Vaziri and Dus, 2021). Here, we use this model to characterize how changes in metabolic signaling due to diet are integrated with cellular context to diet nutrient adaptations.
Taste sensation changes depending on diet composition. In animals, the levels of bitter, sweet, and salty foods influence how these taste stimuli are perceived, with a general inverse relationship between the amount of a particular food in the diet and the responses of the sensory system to it (May and Dus, 2021; Sarangi and Dus, 2021; Reed et al., 2020). For example, in humans and rodents, the dietary concentration of sugars affects sweetness intensity or the electrophysiological responses of the sensory nerves to sucrose (Wise et al., 2016; McCluskey et al., 2020; Sung et al., 2022; Sartor et al., 2011; May and Dus, 2021). A similar phenomenon occurs in flies, where diets supplemented with 15–$30\%$ sucrose, glucose, or fructose decrease the responses of the sensory neurons to sucrose and the transmission of the sweetness signal to higher brain areas (May et al., 2019; Vaziri et al., 2020; Wang et al., 2020; Ganguly et al., 2021; May et al., 2020). In rats and flies, the dulling of the sensory system to sugar occurs even without weight gain, suggesting that diet exposure is sufficient to drive sweet-taste plasticity (Sung et al., 2022; May et al., 2019). Our previous work in flies implicated metabolic signaling through the Hexosamine Biosynthesis Pathway (HBP) enzyme O-GlcNAc Transferase (OGT) in this phenomenon (May et al., 2019). Specifically, knockdown of OGT exclusively in the fly sweet-taste cells prevented the neural and behavioral decrease in sugar responses observed with a high-sugar diet (May et al., 2019). OGT uses the metabolic end-product of the HBP, UDP-GlcNAc, to post-translationally modify proteins and change their stability or activity (Hart, 2019). OGT activity is sensitive to all cellular levels of UDP-GlcNAc without substrate inhibition, but it is enhanced by high dietary sugar due to a higher flux through the HBP (Hanover et al., 2010; Bouché et al., 2004; Hawkins et al., 1997; Marshall et al., 2004; Wang et al., 1998; Olivier-Van Stichelen et al., 2017; May et al., 2019; Wilinski et al., 2019; Na et al., 2015). OGT is also a nucleocytoplasmic protein that interacts with many chromatin- and DNA-modifying complexes; as such, it is thought to function as a nutrigenomic sensor, bridging diet and genes (Olivier-Van Stichelen et al., 2017; Olivier-Van Stichelen and Hanover, 2015; Hart, 2019; Hardivillé and Hart, 2014). Despite global changes in HBP flux with high dietary sugar, the consequences of OGT activity differ among cell types. Understanding how this occurs would provide an opportunity to study how nutrigenomic signaling is integrated with cell-specific contexts, like activity, to generate unique adaptations. Here, we exploited the effects of OGT on Drosophila sensory neurons and the exquisite genetic tools of this organism to investigate this question. Our experiments reveal that nutrigenomic signaling synergizes metabolic state with ongoing cellular physiology to integrate cellular signals. In the sensory neurons, OGT decorates nutrient-sensitive loci also occupied by the epigenetic silencer PRC2.1 and the activity-dependent ERK effector Stripe (Sr). This cooperation leads to changes in chromatin accessibility and transcription that drive sensory plasticity, and the catalytic activity of OGT plays an instructional role in this process. Thus, our results uncover mechanistic insights into how nutrigenomic signaling translates nutritional information into dietary adaptations in the sensory neurons.
## The nutrient sensor OGT decorates the chromatin of sweet sensory cells
Since transcriptional changes have been implicated in sugar diet-induced taste plasticity (Vaziri et al., 2020; May et al., 2019; Wang et al., 2020) and OGT associated with chromatin-binding factors (Vella et al., 2013; Gao et al., 2018; Hart et al., 2011; Gambetta and Müller, 2015), we asked whether this metabolic enzyme moonlights on the chromatin of sweet-taste neurons. We used DNA adenosine methyltransferase Identification (Dam-ID or TaDA) to measure the association of OGT with DNA (Marshall et al., 2016; van Steensel and Henikoff, 2000) and Chromatin Accessibility profiling using Targeted DamID (CaTaDA) to assess chromatin accessibility (Sen et al., 2019). Transgenic UAS-LT3-Dam::OGT or UAS-LT3-Dam flies were crossed with Gustatory Receptor 5a GAL4 (Gr5a) flies (Chyb et al., 2003) to drive expression exclusively in the ~60 sweet-taste cells of the fly mouthpart, and Tubulin-GAL80ts to control the timing of transgene induction. Gr5a>LT3-Dam::OGT; tubulin-GAL80ts (green) and Gr5a>LT3-Dam; tubulin-GAL80ts (yellow) transgenic flies were kept at the permissive temperature and fed a control (CD, $5\%$ sucrose) or sugar (SD, $30\%$ sucrose) diet for 3 days (Figure 1A). Dam::OGT and Dam were then induced by heat shocking the animals at 28°C for 18 hr on day 4, as in our prior experimental design (Figure 1A; Vaziri et al., 2020). The normalized Dam::OGT replicates clustered together by diet (Figure 1—figure supplement 1A), and the chromatin accessibility profile of Dam at the Gr5a sweet-taste receptor gene promoter was high, while at the bitter Gustatory Receptor 66a (Gr66a) promoter – only expressed in adjacent cells – accessibility was low (Figure 1—figure supplement 1B), suggesting that these transgenes were targeted to the correct cells.
**Figure 1.:** *O-GlcNAc Transferase (OGT) decorates the chromatin of the sweet-taste cells.(A) Design of Targeted Dam-ID for OGT occupancy (Dam::OGT) and Dam accessibility (CATaDa) experiments. Age-matched Gr5a;tubulin-GAL80ts>UAS-LT3-Dam::OGT and Gr5a;tubulin-GAL80ts>UAS-LT3-Dam flies were placed on a CD or SD for 3 days at 20–21°C and then switched to 28°C between days 3 and 4 to induce expression of the transgenes. (B) Annotation of OGT chromatin occupied regions (all peaks) using HOMER. (C) The proportion of observed Dam::OGT consensus peaks allocated to their respective chromatin domains normalized to the expected proportions across the whole genome. Heterochromatin: black, high in H3K27m; green, bound by HP-1, high in H3K9me2; blue, bound by Polycomb Group Proteins, high in H3K27m. Yellow and red euchromatin are high in H3K4me2 and H3K79m3; yellow is also enriched for H3K36me3. (D) The distribution in normalized reads (Transcript Per Million, TPM + 1) for genes occupied by OGT (green). Two-tailed t test, ****p<0.0001 (E) Overlap of log2(Dam::OGT/Dam) chromatin-binding peaks of CD (light green) and SD (dark green) (find_peaks, q < 0.01). (F) Average CATaDa signal on CD (light yellow) and SD (dark yellow) centered at OGT peaks. (G) iPAGE summary plots for OGT peaks on a CD (top left), SD (bottom left), and the difference of SD/CD (right). Text in blue represents neural GO terms, orange represents metabolic GO terms, and green represents regulatory GO terms.*
Dam::OGT was associated with chromatin at introns ($51\%$) and transcriptional start sites (TSSs) and promoters ($30\%$) (Figure 1B, Supplementary file 1; all peaks); these patterns are similar to those observed in the only other study that measured OGT occupancy on the chromatin of mouse embryonic stem cells (Vella et al., 2013). In flies, chromatin has been classified into five types according to the histone modifications present and the proteins bound (Filion et al., 2010). In this chromatin characterization, there are three types of heterochromatin, developmentally regulated ‘black’ chromatin (high in Histone (H) 3 Lysine (K) methylation, H3K27m), Heterochromatin-protein 1 (HP1) associated ‘green’ chromatin (high in H3K9me2), and Polycomb group proteins-bound ‘blue’ chromatin, and two types of euchromatin (high in H3K4me2 and H3K79m3), the actively elongating H3K36me3 ‘yellow’ chromatin enriched in nucleic acid metabolism genes and the ‘red’ chromatin enriched in other cellular processes (Filion et al., 2010). Our analysis found that OGT was enriched in transcriptionally active yellow euchromatin (453 genes), consistent with its role in splicing, and at ‘blue’ Polycomb heterochromatin (415 genes), consistent with the known associations between Polycomb Group proteins and this metabolic enzyme (Gambetta et al., 2009; Hart, 2019; Figure 1C). As expected, the accessibility at yellow chromatin intervals was higher than that of blue chromatin regions (Figure 1—figure supplement 1C).
We next examined the differential binding of OGT between the two diets. Although the majority of intervals were shared between a CD and SD (Figure 1E, find_peaks False Discovery Rete (FDR) <0.01), a few hundred loci were uniquely associated with OGT in either the CD ($36\%$) or SD ($10\%$) only conditions. However, the chromatin accessibility at OGT-bound peaks decreased in the high-sugar diet condition (Figure 1F, both at blue and yellow regions, Figure 1—figure supplement 1C). To characterize the function of the genes occupied by OGT, we performed pathway enrichment analysis using iPAGE (Goodarzi et al., 2009). On CD only, OGT-decorated genes were involved in signal transduction, membrane potential, and calmodulin-dependent protein kinase activity (Figure 1G, left). Instead, genes targeted by OGT in the SD-only condition were enriched in G-protein-coupled receptor activity, synaptic target attraction, and transcription (Figure 1G, left). Finally, genes with differential OGT binding between SD/CD were enriched for regulatory/signaling and neural GO terms, including dendrite morphogenesis, neuron projection membrane, synaptic target attraction, signal transduction, pattern formation, and asymmetric cell division (Figure 1G right, for full iPAGE, GO term analysis see Figure 1—figure supplement 2). Interestingly, when we examined the pathways associated with genes found in OGT-associated blue and yellow chromatin intervals, only the blue genes revealed strongly significant enrichment in GO terms. These blue Polycomb chromatin genes were involved in GO terms such as dendrite morphogenesis (8.9E−08), axon guidance (7.88E−04), actin filament organization (2.45E−07), transcription factor activity (3.8E−07), MAPK kinase signaling (9.18E−05), and Transforming Growth Factor β pathway (0.0042). In contrast, the yellow genes only showed a small enrichment for plasma membrane, transcription factor activity (2.9E−01), basolateral plasma membrane (2.4E−01), and phosphonate metabolism (2.0E−01). Together, these experiments show that OGT resides on the chromatin of the sweet taste at open domains characterized by a small but significant diet sensitivity; genes associated with neural functions are abundant among the set with diet-dependent OGT binding but only enriched in the blue H3K27 Polycomb chromatin.
## OGT and PRC2.1 share diet-sensitive chromatin sites
Our previous work showed that the epigenetic silencer PRC2.1 – specifically its H3K27m activity – was necessary and sufficient to drive sweet-taste plasticity in response to the nutrient environment (Vaziri et al., 2020). In the presence of high dietary sugar, PRC2.1 decreased chromatin accessibility and expression of transcription factors involved in synaptic function and signaling; these genes were located in blue-Polycomb H3K27m chromatin. Silencing these genes and their regulons lowered neural and behavioral responses to sweetness in high-sugar diet flies (Vaziri et al., 2020). Since OGT and PRC2.1 play a role in sweet-taste plasticity and OGT occupancy was enriched at blue Polycomb chromatin for neural functions, we asked whether there was an overlap in their occupancy.
A comparison of the peaks occupied by both Dam::Pcl (pink, *Pcl is* the recruiter for PRC2.1) and Dam::OGT (green) revealed a small number of shared intervals (Figure 2A, ~$10\%$; Supplementary file 1). These 162 loci were enriched in the blue ‘Polycomb’ chromatin ($p \leq 0.001$, permutation test) and had lower expression levels in the Gr5a+ neurons (from TRAP experiment in Vaziri et al., 2020) compared to those bound by OGT alone (Figure 2B, purple vs. green), which include both Polycomb ‘blue’ and actively transcribed ‘yellow’ chromatin regions (Figure 1C; Filion et al., 2010). OGT × Pcl intervals had higher expression than those occupied by PRC2.1 alone, suggesting they could represent a subtype of Polycomb blue chromatin (Figure 2B, purple vs. pink). We next asked whether the dietary environment changed the association of OGT and Pcl at these loci. There was more OGT and Pcl at the OGT × Pcl shared sites in the SD condition compared to CD, and more OGT than Pcl was present at these sites in both diets (Figure 2C). Strikingly, chromatin accessibility at OGT × Pcl was markedly ($50\%$) decreased on SD compared to CD (Figure 2D). This nutrient-dependent shift in accessibility was threefold higher at the shared loci compared to those bound by OGT alone (compare Figures 1F and 2D; also comparatively higher than those bound by PRC2.1 alone, Vaziri et al., 2020).
**Figure 2.:** *O-GlcNAc Transferase (OGT) and Polycomb Repressive Complex 2.1 (PRC2.1) mark nutrient-sensitive chromatin in the sweet-taste cells.(A) Diagram of the Targeted Dam-ID (TaDa occupancy, Dam::OGT green, and Dam::Pcl pink) and (CATaDa, accessibility, yellow) experiments analyzed in this figure. Overlap of log2(Dam::Pcl/Dam, pink) and log2(Dam::OGT/Dam, green) chromatin occupancy peaks (all peaks, peak calling: find_peaks, q<0.01). (B) The distribution in normalized reads (Transcript Per Million, TPM +1) for genes occupied by OGT (green), Pcl (pink), and OGT and Pcl (purple). Two-tailed t test, ****p<0.0001. (C) Average log2(Dam::OGT/Dam; left) and log2(Dam::Pcl/Dam) (right) signal on a CD (lighter shades) and SD (darker shades) diet centered at OGT + Pcl co-occupied peaks. (D) Average CATaDa signal on CD (lighter shade) and SD (darker shade) centered at OGT + Pcl co-occupied peaks. (E) (left) iPAGE pathway analysis of genes co-occupied by OGT and Pcl and (right) STRING interaction network of genes co-occupied by OGT + Pcl, colors represent GO terms from the pathway enrichment analysis.*
GO term analysis of genes shared by OGT/Pcl targets revealed enrichment in regulatory pathways involved in sequence-specific DNA binding, including those implicated in neural differentiation, sodium channel regulator activity, Transforming Growth Factor β and Activin receptor signaling, and dendrite development (Figure 2E, left). $30\%$ of the OGT × Pcl sites corresponded to genes encoding DNA-binding and regulatory factors, including two Homeobox transcription factors known to play a role in sweet-taste function and plasticity, cad and Ptx1 (Figure 2E, right) (Vaziri et al., 2020). Analysis of protein interactions between OGT × PRC2.1 genes (Szklarczyk et al., 2020) uncovered a Protein–Protein Interaction network enrichment ($p \leq 1.0$e−16) among DNA-binding factors (pink, $$p \leq 2.08$$e−09), Mitogen-Activated Protein Kinase (MAPK, blue, $$p \leq 0.00059$$), signal transduction (Transforming Growth Factor, TGF-β/Activin signaling, yellow), neuron projection (red outline, $$p \leq 4.95$$e−7), and response to stimuli ($$p \leq 7.15$$e−0.5). Consistent with OGT/Pcl targets being ~$40\%$ of OGT-associated peaks, the GO terms for the shared intervals were a subset of those enriched in the OGT-bound blue chromatin.
## The catalytic activity of OGT is required for diet-induced taste plasticity
Our data show that OGT occupies the chromatin of the sensory neurons and that its binding is diet dependent at loci also bound by PRC2.1. To understand more about the mechanisms of OGT function and, thus, nutrigenomic signaling, we examined the role of OGT activity on taste plasticity using the Proboscis Extension Response (PER). As shown in Figure 3A, the fly proboscis houses the cell bodies and dendrites of the sensory neurons. When the taste sensilla in the labellum are stimulated with sucrose, the fly extends its proboscis to reach the sweet solution. The amount of proboscis extension for each concentration tested – 1 is a full extension, 0.5 a half, and 0 none – corresponds to the fly’s ability to taste and can be compared across genotypes and diets. As previously shown, consumption of SD for 7 days results in a decrease in PER for high ($30\%$) and low ($5\%$) concentrations of sucrose compared to animals that ate a control diet (Figure 3B, circles vs. squares, gray shades). However, knocking down OGT in the Gr5a+ sweet-sensing neurons resulted in flies with similar sweet sensitivity between the two diets (Figure 3B). Thus, OGT is required for diet-dependent sweet-taste plasticity. To ask if the catalytic activity of OGT was required for this taste phenotype, we compared the ability of protein null (OGT1) and catalytically dead mutants (OGTK872M) to rescue taste plasticity; both of these alleles are homozygous lethal and thus were tested in combination with w1118CS control flies. Neither mutant affected sweet-taste responses on a control diet, but both prevented the lower PER to sucrose observed in SD-fed control flies (Figure 3C). This argues that the catalytic activity of OGT is required for the effects of this enzyme on taste plasticity. Consistent with this, knocking down the antagonistic enzyme O-GlcNAcase (OGA), which removes the GlcNAc moiety from proteins, resulted in lower sweet-taste responses on CD (Figure 3D). We next asked if increasing the levels of OGT was sufficient to induce sweet-taste changes. OGT activity is linear across all levels of cellular UDP-GlcNAc, so increasing its levels also increases its activity (Hart, 2019). Overexpression of OGT in the Gr5a+ neurons resulted in sucrose responses on CD comparable to those observed in sugar diet-fed flies (Figure 3E). However, inhibiting the activity of OGT with the specific OGT Small Molecule Inhibitor-1 (OSMI) (Ortiz-Meoz et al., 2015; May et al., 2020) blocked the effects of OGT overexpression on sweet-taste responses (Figure 3E, right); this drug treatment had no effect on survival (Figure 4—figure supplement 1A). To finally test if the effects of SD on taste plasticity were dependent on OGT activity, we supplemented the CD and SD with OSMI during the entire duration of the diet exposure (7 days) and then tested PER to sucrose. Control (vehicle, Dimethyl Sulfoxide (DMSO)) flies exhibited a dulling of sweet-taste responses on SD (squares), but this decrease was entirely blocked by OSMI (Figure 3F); no effects were observed on a CD (circles). Thus, decreasing the activity of OGT, either with genetics or pharmacological tools, resulted in similar effects on taste plasticity, arguing that the activity of this enzyme plays an essential role in taste changes in response to the dietary environment.
**Figure 3.:** *O-GlcNAc Transferase (OGT) activity is necessary for taste plasticity in response to the sugar diet environment.(A) (top) Anatomy of the sensory system showing the cell bodies, dendrites, and axons of the sweet-sensing Gr5a+ neurons; (bottom) Diagram of the Proboscis Extension Response (PER). (B) Taste responses (y-axis) to stimulation of the labellum with 30, 10, and 5% sucrose (x-axis) in flies with knockdown of OGT (green) or controls (shades of gray) in flies fed a CD (circles) or SD (squares); n = 18–51. Two-way repeated measure analysis of variance (ANOVA), main effect of genotype: Gr5a>wcs p < 0.0001 (Tukey multiple comparison 30% p = 0.0008, 10% p = 0.0047, 5% p < 0.0001), Gr5a>OGT-RNAi p = 0.2657 (Sidak multiple comparison 30% p = 0.2792, 10% p = 0.9756, 5% p = 0.4883), OGT-RNAi>wcs (Sidak multiple comparison 30% p = 0.5923, 10% p = 0.0381, 5% p < 0.0001). (C) Taste responses (y-axis) to stimulation of the labellum with 30, 10, and 5% sucrose (x-axis) in flies with mutations in OGT (green) or controls (black) while on a control diet (CD, n = 16–26) or SD, n = 26–29. Two-way repeated measure ANOVA, main effect of genotype compared to wcs controls: ****p < 0.0001. Tukey multiple comparisons test, ****p < 0.0001, ***p < 0.001, **p < 0.01, *p < 0.05. (right) Diagram of OGT alleles: OGT1, protein null, OGTK872M catalytically dead. (D) Taste responses (y-axis) to stimulation of the labellum with 30, 10, and 5% sucrose (x-axis) in flies with knowdown of OGA (green) or controls (shades of gray) while on a control diet (CD), n = 31–47. Two-way repeated measure ANOVA, main effect of Gr5a>OGARNAi genotype compared to each control genotype: ****p < 0.0001. Tukey multiple comparisons test, ****p < 0.0001. (E) (left) Diagram of experiments with 10 μM OGT Small Molecule Inhibitor-1 (OSMI-1) in E and F; (right) taste responses (y-axis) to stimulation of the labellum with 30, 10, and 5% sucrose (x-axis) in flies with overexpression of OGT (green) or controls (shades of gray) while on a control diet (CD) supplemented with OSMI or vehicle, n = 19–22. Two-way repeated measure ANOVA, main effect of Gr5a>OGT genotype compared to each control genotype: ****p < 0.0001. Tukey multiple comparisons test, ****p < 0.0001 and *p < 0.05. (F) Taste responses (y-axis) to stimulation of the labellum with 30, 10, and 5% sucrose (x-axis) of age-matched male w1118cs flies on a CD (circle) or SD (square) diet with vehicle (DMSO) or OSMI. n = 14–17. Two-way repeated measure ANOVA, main effect of OSMI treatment p = 0.0089; Tukey multiple comparisons test for 30, 10, and 5% sucrose: (1) CD vs. SD (DMSO) p < 0.05, p = 0.0090, p = 0.0034 and (2) CD vs. SD (+OSMI-1), p > 0.05 at all concentrations. Data are shown as mean ± standard error of the mean (SEM).*
## OGT and PRC2.1 genetically interact to drive taste plasticity
To determine the effects of OGT catalytic activity on chromatin accessibility and PRC2.1 occupancy, we fed Gr5a>LT3-Dam; tubulin-GAL80ts and Gr5a>LT3-Dam::Pcl; tubulin-Gal80ts (Figure 4—figure supplement 1A) flies a CD or an SD supplemented with OSMI. Strikingly, OSMI treatment completely abolished the changes in chromatin accessibility observed with SD at OGT × Pcl sites (Figure 4A, compared to Figure 2D; Supplementary file 1), suggesting that OGT activity is necessary for diet-dependent dynamics at these loci. However, inhibition of OGT activity did not affect Pcl occupancy at these peaks, indicating that recruitment of PRC2.1 to these sites is largely independent of this metabolic enzyme (pink, Figure 4A). OSMI also had a mild effect on the occupancy of Dam::Pcl genomewide since the number (~1800), and identity ($80\%$) of Dam::Pcl peaks were mainly the same with or without OSMI (Figure 4—figure supplement 1C). Only a smaller fraction of new PRC2.1-only peaks emerged with OSMI treatment, and the genes in these intervals were enriched in GO terms such as detection of chemical stimuli, DNA binding, and protein kinase activation (Figure 4—figure supplement 2). Thus, OGT activity is required for the diet-dependent decrease in chromatin accessibility but not PRC2.1 recruitment or occupancy, suggesting that other factors or events mediate these. However, we found that the catalytic activity of OGT was necessary for PRC2.1-mediated taste modulation. Overexpression of Pcl in the Gr5a+ neurons mimics the effects of SD on taste by decreasing responses to sucrose in flies fed a CD (Figure 4B, left) – a result dependent on the H3K27 methylation activity of this complex (Vaziri et al., 2020). However, OSMI blocked the effects of Pcl overexpression on sucrose responses compared to vehicle-fed flies (Figure 4B, right). These results place OGT upstream of PRC2.1 at both the molecular and behavioral levels. Since OGT is an enzyme known to modify proteins, we also asked whether its effects on taste plasticity were mediated, at least in part, through PRC2.1. Pclc429 mutations blocked diet-induced taste plasticity on SD (Figure 4C, pink vs. black, compare squares vs. circles), while Gr5a>OGT overexpression promoted a decrease in sucrose PER even on CD (Figure 4C, green vs. black, compare circles and squares). However, when Pclc429 mutants were combined with Gr5a>OGT, these flies failed to develop taste plasticity in response to SD without any effects on CD (Figure 4C purple). Thus, Gr5a>OGT;Pclc429 phenocopied Pclc429 mutants (compare pink and purple), suggesting that the effects of OGT act largely through PRC2.1. To further confirm these results and link them to the catalytic H3K27 methylation activity of PRC2.1, we treated Gr5a>OGT and control flies with a vehicle or the specific inhibitor of PRC2 (EEDi) while on CD (Qi et al., 2017). This manipulation restored normal taste responses to control levels in Gr5a>OGT flies (Figure 4D, compare right vs. left), consistent with what we observed with Gr5a>OGT; Pclc429 flies (Figure 4C). Together these results argue for a strong genetic interaction between OGT and PRC2.1.
**Figure 4.:** *O-GlcNAc Transferase (OGT) activity is necessary for chromatin and transcriptional dynamics in response to the sugar diet environment.(A) (left) Average CATaDa signal on a CD (light yellow) and SD (dark yellow) with OSMI centered at OGT + Pcl peaks (compare to Figure 2D); (right) Average log2 Dam::Pcl/Dam signal on a CD (light pink) and SD (dark pink) with OSMI centered at OGT + Pcl peaks (compare to Figure 2). (B) Taste responses (y-axis) to stimulation of the labellum with 30, 10, and 5% sucrose (x-axis) of age-matched male Gr5a>Pcl (pink) and transgenic controls (shades of gray) on CD supplemented with vehicle (DMSO, n = 14–23) or 10 µM OSMI (n = 20–42). Two-way repeated measure analysis of variance (ANOVA): (1) DMSO, main effect of genotype (****p < 0.0001) and genotype x concentration (*p < 0.05); Tukey multiple comparisons test for 30, 10, and 5% sucrose concentrations: Gr5a>wcs vs. Gr5a>Pcl p = 0.0165, p = 0.0056, p = 0.0025; Pcl>wcs vs. Gr5a>wcs, ns. (2) OSMI: main effect of genotype p = 0.3194 and genotype × concentration p = 0.6893. (C) Taste responses (y-axis) to stimulation of the labellum with 30, 10, and 5% sucrose (x-axis) in Gr5a>OGT;Pclc429 (purple), Gr5a>OGT (green), Pclc429/+ (pink), and transgenic controls (gray) on CD or SD, n = 20–37. Two way repeated measure ANOVA, main effect of diet: ****p < 0.0001. Tukey multiple comparisons test, ****p < 0.0001 and *p < 0.05. (D) Taste responses (y-axis) to stimulation of the labellum with 30, 10, and 5% sucrose (x-axis) in Gr5a>OGT (green) and transgenic controls (shades of gray) on a CD supplemented with vehicle or 8 µM EEDi, n = 20–22. Two-way repeated measure ANOVA, main effect of Gr5a>OGT genotype compared to each control genotype: ****p < 0.0001. Tukey multiple comparisons test, ****p < 0.0001. (E) Log2fold (l2fc) of differentially expressed genes (DEGs) between SD/CD in w1118cs ± OSMI and Pclc429 SD/CD. (F) GO term analysis of the DEGs measured in the Gr5a+ neurons of flies fed a CD and SD + OSMI. (G) Taste responses (y-axis) to stimulation of the labellum with 30, 10, and 5% sucrose (x-axis) for a subset of DEGs in (E, purple circle) that show dependence on OGT and Polycomb Repressive Complex 2.1 (PRC2.1). n = 14–49. Purple, knockdown; red, overexpression; bold, direct OGT/PRC2.1 targets. Two-way repeated measure ANOVA, main effect of GAL4>wcs control genotype compared to each control genotype: ****p <0.0001. Tukey multiple comparisons test, ****p < 0.0001, ***p < 0.001, **p < 0.01, *p < 0.05. ACL, ATP Citrate Lyase; Irk1, inwardly rectifier potassium channel 1; daw, dawdle; cbt, cabut; exex, extra extra; syt-a, synaptotagmin alpha. Data are shown as mean ± standard error of the mean (SEM).*
To understand the consequences of the observed OGT-dependent shifts in chromatin accessibility, we isolated mRNAs associated with the ribosomes of the Gr5a+ cells using Translating mRNA Affinity Purification (TRAP) (Chen and Dickman, 2017) in flies fed a CD + OSMI and SD + OSMI. Principal component analysis revealed that most of the variation between samples was due to diet (Figure 4—figure supplement 3A); mRNAs specifically expressed in the Gr5a+ cells, such as the sweet-taste receptor genes (Gr5a, Gr64f, and Gr64a) and the fatty acids taste receptor Ir56D, were enriched in the Gr5a+ fraction compared to the input, while bitter receptor genes (Gr66a and Gr32a) were depleted (Figure 4—figure supplement 3B), indicating that the selection of Gr5a+ mRNAs was successful and comparable to our prior experiments (Vaziri et al., 2020). However, compared to the marked negative skew in gene expression we previously observed with a high-sugar diet, where $90\%$ of genes had negative log2 fold changes, OSMI-differentially expressed genes (DEGs) showed a similar distribution in positive and negative changes (Figure 4—figure supplement 3C; Supplementary file 1; Vaziri et al., 2020). Indeed, further analyses revealed that OSMI treatment reverted (i.e., showed the opposite direction of change; q < 0.1, Wald test) or restored (practical equivalence test using a null hypothesis of a change of at least 1.5-fold and q < 0.05) the expression of $52\%$ of the DEGs with SD/CD change (Figure 4E, gray are downregulated and red are upregulated), and that most of the genes changed with OSMI treatment [367] were also similarly affected by a loss of function Pclc429 mutation (Figure 4E). *These* genes were enriched in metabolic and neural processes, such as chemical synapse transmission, synaptic target attraction, cell differentiation, glucose metabolism, and detection of chemical stimuli (Figure 4G and Figure 4—figure supplement 4). Notable among these were the homeobox cad and Ptx1 and their regulons, which have been implicated in taste function and plasticity (Figure 4—figure supplement 3D; Vaziri et al., 2020), but also many whose effects of sweet-taste sensation are not known, like the Activin ligand dawdle (daw) and the transcription factor cabut (cbt). To functionally validate some of these new genes, we knocked down or overexpressed them (depending on their SD/CD log2 fold) in the Gr5a+ neurons and assayed sweet-taste function by PER (Figure 4F). This revealed a mild-to-moderate effect of these genes on sucrose sensation; interestingly, daw and Irk1 are direct targets of Pcl and OGT.
Together, these experiments place OGT and PRC2.1 in the same genetic pathway that directs diet-induced taste plasticity at the chromatin, transcriptional, and behavioral levels. They also argue that the activity of the metabolic sensor OGT may provide the nutrient-dependent context for PRC2.1-mediated changes in chromatin accessibility (but not PRC2.1 recruitment), while PRC2.1 may instead function as an effector of these chromatin and transcriptional changes.
## The transcription factor Sr is part of the OGT–PRC2.1 nutrigenomic pathway
Our data show that OGT orchestrates responses to the dietary environment in the sensory neurons. To understand which other cellular context factors may cooperate with OGT and PRC2.1 to mediate taste plasticity, we examined the regulatory regions of OGT × Pcl and OGT × PREs (Polycomb Responsive Elements, DNA motifs to which Polycomb Proteins bind) loci for enriched cis-regulatory motifs (Figure 5—figure supplement 1A, B and Supplementary file 1; also, note that a TF may appear multiple times as dots in this graph because of different binding motifs). In this analysis, the highest log2 fold enrichments were for Polycomb Group proteins like PhoRC (1.018, $$p \leq 0.0099$$) and Trx-recruiter Trithorax-like (Trl, 1.972, $$p \leq 0.0099$$ and 0.803, $$p \leq 0.0099$$; *Trx is* antagonistic to PRC2.1), as well as transcription factors, such as the Zn-finger immediate early gene Stripe (Sr, homolog of activity-dependent human Early Growth Response 2, EGR2, alias Krox20), and the nutrient-sensitive factor Sterol-Responsive-Element Binding Protein (SREBP) (Supplementary file 1). To determine if they affected sweet taste, we measured the proboscis extension in response to sucrose when these genes were overexpressed or knocked down in the Gr5a+ neurons (Figure 5—figure supplement 1C). The only factor that affected sweet-taste responses across low and high sucrose concentrations was Sr, which also showed higher mRNA abundance in the sensory neurons of SD-flies (Figure 5—figure supplement 1D).
Sr is a conserved transcription factor induced by neural activity via the MAPK/ERK pathway (Chen et al., 2016; Gonzales et al., 2020; Beckmann and Wilce, 1997) and is essential for sensory nerve development and plasticity (Murphy et al., 1989; Duclot and Kabbaj, 2017). The MAPK pathway is sensitive to neural activity and nutrients (Robles-Flores et al., 2021; Papa et al., 2019); it is stimulated by mitogens, such as TGF-β/Activin signaling, which increase with high-sugar levels, eating, and neural activity in flies and mammals (Lavoie et al., 2020; Liu and Chen, 2022; Wilinski et al., 2019). Of note, OGT × Pcl co-occupied loci were enriched in MAPK/ERK targets (Figure 2E, blue). Enrichment for Sr cis-regulatory motifs was modest at OGT or Pcl-only loci (l2fc = 0.196 and 0.398, respectively; $p \leq 0.001$ in both cases via an approximate permutation test) but strong at genes bound by both factors (OGT × Pcl = 0.429 and OGT × PREs = 0.976, $p \leq 0.001$, permutation test) (Figure 5A; Supplementary file 1). When we compared the distribution of Sr sites around the regulatory regions of OGT × PRC2.1 genes (either together or separately), we found a marked bias around the TSS (Figure 5B, top), with enrichment of Sr-binding sites in the 500 bp window immediately preceding the TSS. In contrast, genes that OGT or PRC2.1 did not occupy, were depleted for Sr-binding sites at the TSS (Figure 5B, bottom). This suggests that *Sr is* enriched at the TSS of genes that are bound by OGT and PRC2.1. To this end, when we examined the expression of Sr-targets in the Gr5a+ neurons of flies on the two diets, we noticed that these genes had negative log2 fold changes on the sugar diet; this repression, however, was abolished by mutations in Pcl and by inhibition of OGT activity (Figure 5C, compare gray vs. pink and green, respectively), which hints toward functional cooperation between Sr and OGT/PRC2.1. To characterize the effects of higher Sr levels on neural activity and behavior, we used the UAS/GAL4 system to overexpress this gene in the Gr5a+ neurons of adult flies. Overexpression of Sr resulted in lower electrophysiological responses of the gustatory neurons to sucrose (Figure 5D, Mann–Whitney test, $$p \leq 0.001$$) as well as lower PER at both high and low sucrose concentrations (Figure 5E). This effect, however, was dependent on the catalytic activities of PRC2 and OGT. Indeed, overexpression of Sr in flies treated with OSMI (green) or with an inhibitor of PRC2 (pink, EEDi Vaziri et al., 2020) resulted in sucrose responses comparable to those of control flies (Figure 5F). Overall, these results argue for a role of Sr in taste plasticity in coordination with OGT and PRC2.1. Importantly, since OGT and PRC2.1 are not always found at TSS, the bias in Sr distribution at the TSS of ‘OGT/PRC2.1 regulated genes’ indicates that coordination between Sr and OGT/PRC2.1 may arise not from direct interactions but rather from two distinct paths of information flow, such as metabolism and neural activity.
**Figure 5.:** *The immediate early gene Sr is found at PRC2 × OGT genes and is involved in sweet-taste sensation.(A) Log2fold (l2f) enrichment for Sr motifs at sites occupied by O-GlcNAc Transferase (OGT; green), Pcl (pink), or OGT + Pcl and OGT + PREs (purple), p < 0.0001. (B) Normalized distribution of Sr motifs along the regulatory regions 2500 bp up and downstream the transcriptional start site (TSS) for the genes in (A); the 500 bp immediately before each TSS are shaded in grey. Counts are normalized relative to the genome-wide expectation (derived by multiplying the number of potential target genes by the fractional coverage of Sr motifs on the genome); a score of 1.0 indicates the hypothetical genome-wide average overlap with Sr motifs and is shown as a red dashed line. (C) The distribution of RNA l2fc for genes that have Sr sites and are expressed in the Gr5a+ neurons of flies on a CD and SD in control flies or flies with mutations in Pcl or inhibition of OGT. q > 0.01. (D) Representative traces (left) and averaged neuronal responses to 25 mM sucrose stimulation from the L-type sensilla of Sr overexpression flies (blue) and controls (gray). n = 11. Mann–Whitney test: **p = 0.001. (E) Taste responses (y-axis) to stimulation of the labellum with 30, 10, and 5% sucrose (x-axis) in Sr overexpression flies (blue) and controls (gray). n = 22–38. Two-way repeated measure analysis of variance (ANOVA), main effect of genotype ****p < 0.0001 and genotype × concentration ****p < 0.0001; Tukey post-test for multiple comparisons: 30%: ***p = 0.0002 for Gr5a>Sr compared to each control, 10%: Gr5a>Sr vs. Gr5a>wcs p = 0.0025 and Gr5a>Sr vs. Sr >wcs ****p < 0.0001; 5%: ****p < 0.0001 for Gr5a>Sr compared to each control. Gr5a>wcs vs. Sr>wcs p > 0.05 at all concentrations. (F) Taste responses (y-axis) to stimulation of the labellum with 30, 10, and 5% sucrose (x-axis) in Sr overexpression flies (blue) and controls (gray) treated with the OGT inhibitor OSMI (green) or the PRC2 inhibitor EEDi (pink). n = 30. Two-way repeated measure ANOVA, main effect of genotype p = 0.2993 and p = 0.9146 and genotype × concentration p = 0.9293 and p = 0.9146, respectively. Data are shown as mean ± standard error of the mean (SEM).*
## The ERK pathway modulates taste adaptations in response to diet
Sr is the downstream transcriptional effector for the Extracellular-signal Regulated Kinase (ERK), a pathway stimulated by neural activity that plays a role in plasticity (Lavoie et al., 2020; Miningou and Blackwell, 2020; Thomas and Huganir, 2004; Figure 6A). We reasoned that ERK/EGR2 might provide sensory neurons with a specific context to drive dietary adaptations. To test this hypothesis, we examined the role of the kinase rolled (rl) – the ERK homolog in D. melanogaster – in sweet-taste and diet-induced taste plasticity.
**Figure 6.:** *The effect of the kinase rl/ERK on sweet taste depends on O-GlcNAc Transferase (OGT) activity.(A) Diagram of the rl/ERK > Sr pathway, red sparks represent neural activity, and red outline represents activation. (B) Diagram of the two types of rl/ERK transgenes used. (C) Representative traces (left) and averaged neuronal responses to 25 mM sucrose of L-type sensilla in overexpression of wild-type (rlWT) or constitutively active (rlSem) rl/ERK in the Gr5a+ neurons (blue) and control (gray) on a CD. n = 11–23. One-way analysis of variance (ANOVA); Tukey’s multiple comparison test: ****p < 0.0001 for Gr5a/+ vs. Gr5a>rlSem, p = 0.279 for Gr5a/+ vs. Gr5a>rlWT, and *p = 0.018 for Gr5a>rlSem vs. Gr5a>rlWT. (D) Taste responses (y-axis) to stimulation of the labellum with 30, 10, and 5% sucrose (x-axis) in flies with overexpression of wild-type (rlWT) or constitutively active (rlSem) rl/ERK in the Gr5a+neurons (blue) and control (gray) flies on a CD + vehicle (DMSO). n = 24–27. Two-way repeated measure ANOVA, main effect of genotype p < 0.0001 and concentration × genotype p < 0.0001. Tukey multiple comparisons tests: ****p < 0.0001 for Gr5a>rlSem vs. all other genotypes at 30, 10, and 5% and p > 0.05 for all other comparisons at all concentrations. (E) Representative traces (left) and averaged neuronal responses to 25 mM sucrose of L-type sensilla in overexpression of wild-type (rlWT) or constitutively active (rlSem) rl/ERK in the Gr5a+ neurons (blue) and control (gray) on a CD + OSMI. n = 11–23. One-way ANOVA; Tukey’s multiple comparison test: p = 0.172 for Gr5a/+ vs. Gr5a>rlSem, p = 0.603 for Gr5a/+ vs. Gr5a>rlWT, and p = 0.034 for Gr5a>rlSem vs. Gr5a>rlWT. (F) Taste responses (y-axis) to stimulation of the labellum with 30, 10, and 5% sucrose (x-axis) in flies with overexpression of wild-type (rlWT) or constitutively active (rlSem) rl/ERK in the Gr5a+ neurons (blue) and control (gray) in flies fed a CD + OSMI. n = 26–33. Two-way repeated measure ANOVA, main effect of genotype p = 0.005; Tukey multiple comparisons tests: p > 0.05 for all other comparisons at all concentrations except for p < 0.0001 for Gr5a>rlSem vs. rlSem/wcs at 10% p = 0.0216. Effect of OSMI vs. vehicle: Gr5a>rlSem 30% p = 0.0012, 10% p = 0.0030, 5% p < 0.0001, and p < 0.05 for all other genotypes. Data are shown as mean ± standard error of the mean (SEM).*
First, we observed that, as with Sr, the mRNA abundance of rl was higher in the Gr5a+ neurons of flies on SD, but this gene was not a direct target of OGT or PRC2.1 (Figure 6—figure supplement 1A). We tried several available antibodies against rl and activated (phosphorylated) rl to establish whether increased transcript levels also resulted in higher activation of this kinase; however, none of them resulted in a reliable signal in our hands. We thus turned to genetic tools to investigate whether higher rl expression or activity played a role in sweet-taste plasticity. To differentiate between these two possibilities, we expressed either a wild-type rl (rlWT, Figure 6B, top, Biggs et al., 1994) or constitutively active form of the kinase (rlSem Figure 6B, bottom, Oellers and Hafen, 1996) in the Gr5a+ neurons and tested neural and taste responses to sucrose. Overexpression of rlWT with Gr5a-GAL4 did not affect the electrophysiological responses of the sensory neurons to sucrose (Figure 6C, dark blue); however, expression of the active rlSem decreased neuronal responses to sucrose (Figure 6C, light blue). These activity phenotypes were reflected in the behavioral taste responses to sucrose, with rlWT flies having identical PER to sucrose as controls and rlSem showing reduced PER across high and low sucrose concentrations (Figure 6D, left vs. right; note that controls are shared here, plotted separately for clarity). Thus, rl activity, but not higher levels alone, was sufficient to affect sweet-taste plasticity. Not surprisingly, given the known function of rl/ERK in neural activity, we found that this kinase was also necessary for sweet-taste responses, as loss of function mutants and RNAi for rl had lower electrophysiological and behavioral responses to sucrose (Figure 6—figure supplement 1B, C). To test whether there was a synergetic interaction between rl and OGT, we repeated the same experiments in the presence of the OGT inhibitor OSMI. Strikingly, OSMI treatment almost entirely blocked the effects of rlSem on both neural and behavioral responses to sucrose while having no effect on rlWT flies (Figure 6E, F).
To characterize the function of the rl/Sr pathway on taste plasticity, we used Trametinib, a drug that inhibits ERK activation in animals (currently used for treating melanoma). At concentrations previously used in flies (15.6 μM) (Castillo-Quan et al., 2019; Slack et al., 2015), Trametinib blocked the effects of rlSem expression on sweet-taste responses (Figure 7—figure supplement 1A); this had no effect on survival (Figure 4—figure supplement 1A). Treatment with this ERK inhibitor also negated the effects of Sr overexpression on sucrose responses, resulting in flies with PER comparable to controls (Figure 7—figure supplement 1B), placing Sr downstream of ERK activation. Thus, Trametinib treatment efficiently blocks ERK signaling. To determine if the activity of the rl/Sr (ERK/EGR2) pathway was necessary for taste plasticity in response to the sugar diet environment, we fed flies a control or sugar diet with or without Trametinib for 7 days, then measured their neural and behavioral responses to sucrose. Exposure to a high-sugar diet decreased the electrophysiological (Figure 7A) and behavioral (Figure 7B) responses to sucrose. However, when rl activity was blocked with Trametinib, there was no decrease in neural responses or PER (Figure 7C and D). Of note, Trametinib had a minor but significant effect on sweet-taste activity (compare CD of Figure 7C with CD of Figure 7A), consistent with the observation that rl is necessary for normal sweet-taste function (Figure 6—figure supplement 1B, C). Together, these data indicate that the ERK pathway plays a critical role in the development of taste adaptations in response to diet and place its function upstream of OGT.
**Figure 7.:** *The rl > Sr pathway is important for taste adaptations in response to diet.Representative traces (left) and averaged responses to 25 mM sucrose from L-type sensilla of flies fed a CD and SD (A), gray or Trametinib (C), blue. n = 11–14. Unpaired t-test: ***p = 0.0001 for CD vs. SD, and p = 0.486 for CD Trametinib vs. SD Trametinib. Taste responses to stimulation of the proboscis with sucrose in flies fed a CD and SD + vehicle (B), DMSO, gray or Trametinib (D), blue. PER, n = 20–33. PER: two-way repeated measure analysis of variance (ANOVA), main effect of diet, vehicle p < 0.0001 and Trametinib p = 0.4701; Tukey multiple comparison test: vehicle CD vs. SD 30% p = 0.421, 10% **p = 0.0017, and 5% ***p = 0.0002 and Trametinib CD vs. SD 30% p = 0.9702, 10% p = 0.4470, and 5% p = 0.9575. Effect of Trametinib: CD vehicle vs. CD Trametinib, 30% p = 0.1745, 10% p = 0.0108, 5% p = 0.0015; SD vehicle vs. SD Trametinib, 30% p = 0.4837, 10% p = 0.0228, 5% p = 0.2339. (E) A model for how O-GlcNAc Transferase (OGT), PRC2, and ERK orchestrate taste plasticity in response to a changing food environment. Boxes in pink describe the different steps of ‘information processing’ (see discussion). Data are shown as mean ± standard error of the mean (SEM).*
## Discussion
Nutrigenomic signaling plays a role in health and disease by bridging the dietary environment with physiological adaptations. However, the molecular mechanisms and consequences of this type of nutrient sensing are still poorly understood. In particular, how nutrigenomic signals are integrated with cellular contexts has remained hard to define due to the lack of mechanistic nutrigenomic models (Müller and Kersten, 2003; Vaziri and Dus, 2021). In this work, we exploited the conserved phenomenon of diet-induced taste plasticity and the genetic tools of the D. melanogaster fly to answer these questions.
Here, we report that the metabolic enzyme OGT is associated with neural chromatin at introns and TSSs. While OGT-associated genes showed subtle but significant changes in chromatin accessibility in response to diet, these dynamics were much more robust at loci co-occupied by both OGT and the epigenetic silencer PRC2.1. *At* genes decorated by both factors, we observed sizable diet-dependent chromatin variations that were critically dependent on the catalytic activity of OGT. OGT activity was also necessary for the differential transcriptional and taste responses to the high-sugar diet. The OGT- and PRC2.1-bound nutrient-dependent loci were enriched for binding motifs for the activity-dependent transcription factor Sr at TSS, the effector of the ERK pathway. We show that this signaling pathway functions upstream of OGT/PRC2.1 to shape neural and behavioral taste responses to the dietary environment. We thus propose a model where a nutrigenomic pathway composed of OGT and PRC2.1 integrates information from the nutrient and cellular environment via ERK signaling to orchestrate sensory responses to diet (Figure 7E). Our data suggest that this integration occurs at the level of chromatin and modulates the expression of transcription factors and signaling regulators that further amplify and extend the reach of nutrigenomic signaling. Our findings thus shed light on how nutrigenomics contributes to neural plasticity and behavior.
## OGT and chromatin dynamics
OGT is a conserved enzyme that catalyzes the transfer of UDP-GlcNAc to the serine and threonine residues of proteins (Hart, 2019; Olivier-Van Stichelen et al., 2017). Because UDP-GlcNAc synthesis by the HBP combines sugar, amino acid, nucleotide, and fatty acid metabolism, the levels of this metabolite, as well as the activity of the enzymes in this pathway, are inextricably linked to cellular metabolism and diet (Hart, 2019; Olivier-Van Stichelen et al., 2017). Higher HBP flux directly impacts OGT activity because the function of this enzyme is linear across a vast range of physiological UDP-GlcNAc concentrations. Because of this, OGT is recognized as a critical nutrient sensor in animal physiology, particularly in development, cancer, and metabolic disease (Hart, 2019; Olivier-Van Stichelen et al., 2017). More recently, its importance for neural function and plasticity has also been recognized, with studies implicating it in synapse maturation, neural excitability, activity, and plasticity (Butler et al., 2019; Ardiel et al., 2018; Lagerlöf et al., 2016; Su and Schwarz, 2017; Giles et al., 2019; Lagerlöf et al., 2017; Ruan et al., 2014; Hwang and Rhim, 2019; Li et al., 2019; May et al., 2019). Our group showed that a high-sugar diet acutely and chronically increased HPB activity in flies and played a role in diet-induced sensory plasticity (Wilinski et al., 2019; May et al., 2019).
OGT is a nucleocytoplasmic enzyme, and the GlcNAc modification is enriched in nuclear and synaptic proteins, which extends its reach on cellular physiology (Hart, 2019; Olivier-Van Stichelen et al., 2017). Although OGT is thought to play a role in gene regulation, only one study has shown its direct association with chromatin in murine embryonic stem cells (Vella et al., 2013). Here, we report that OGT also decorates neural chromatin in Drosophila melanogaster. Like in murine embryonic stem cells, OGT was enriched at introns and TSSs and primarily associated with transcriptionally active chromatin. However, we found that half of OGT intervals were also enriched at Polycomb repressive chromatin, consistent with previous reports that the GlcNAc modification is found at PREs, as well as on many Polycomb Group (PcG) proteins; OGT is also associated with PcG complexes to mediate Hox-gene repression (Hart, 2019; Olivier-Van Stichelen et al., 2017; Schuettengruber et al., 2017). On a high-sugar diet, there was a higher association of OGT with DNA but lower chromatin accessibility. However, the magnitude of these changes depended on what other regulatory and DNA-binding factors were found at OGT loci. At loci with PRC2.1 binding and Sr/EGR2 motifs, chromatin openness was markedly reduced in response to the high-sugar diet environment. This is the first study to show that OGT-associated chromatin is nutrient sensitive to the best of our knowledge. Importantly, this nutrient sensitivity was entirely dependent on the catalytic activity of OGT because it was abolished in the presence of the inhibitor OSMI. Interestingly, OGT activity had no effect on PRC2.1 association with co-occupied loci (and only a small effect at non-OGT loci, data not shown). The H3K27 methylation activity of PRC2.1 is required for changes in chromatin accessibility, including those that depend on diet in the sensory neurons (Schuettengruber et al., 2017; Vaziri et al., 2020). Thus, our data suggest that OGT activity affects the repressive action but not the recruitment of PRC2.1; we also demonstrate that the catalytic activity of PRC2.1 is required for the effects of OGT on taste plasticity.
These findings raise several important questions about the biochemical mechanisms of OGT action that our genetic system is poorly suited to address, but that will be important to define in future studies. First, what are the targets of OGT at nutrient-sensitive loci? Is OGT directly GlcNAcylating PRC2.1 to modify its repressive drive? Several studies have linked OGT activity with the stability, chromatin occupancy, or catalytic function of Polycomb Group Proteins (Hart, 2019; Olivier-Van Stichelen et al., 2017; Chu et al., 2014; Sakabe and Hart, 2010; Forma et al., 2018; Sui et al., 2020; Jiang et al., 2019; You et al., 2021; Decourcelle et al., 2020; Gambetta and Müller, 2014). Thus, converging evidence suggests that OGT impacts different aspects of PRC2.1 and PcG biology and is broadly consistent with our data. Connections between OGT and ERK have also been uncovered in the context of cancer and cell division, with studies showing that inhibition of ERK signaling decreases O-GlcNAcylation and vice versa (Zhang et al., 2015; Jiang et al., 2016) and that GlcNAcylation promotes ERK effects while OGT inhibition blocks them (Cork et al., 2018; Weiss et al., 2021; Lei et al., 2020). These findings are consistent with the effect and direction of the genetic interactions we observed between OGT and PRC2.1 and OGT and ERK, the direction of ‘information flow’ within the cell (Figure 7E), and the effects of our genetic manipulations. OGT could also affect chromatin accessibility by GlcNAcylating histones, although the function and effects of these histone modifications are still poorly understood (Gambetta and Müller, 2015; Hirosawa et al., 2018; Konzman et al., 2020; Olivier-Van Stichelen et al., 2017). Another outstanding question is how OGT is targeted to chromatin and whether this recruitment is dynamic and related to nutrient availability. For example, are there different local pools of OGT and GlcNAc in the nucleus vs. cytoplasm (or mitochondria) where OGT has been described? Because of the lack of functional fly OGT antibodies, we could not ask this question, but it is possible that the levels of OGT in the nucleus and cytoplasm change between diets. Finding answers related to the cellular compartmentalization of this metabolic enzyme and its targets will be an essential step in understanding nutrient signaling.
## Sensors and effector mechanisms of nutrigenomic signaling in neural plasticity
In the case of the sweet-taste neurons, sugar directly activates the cells via receptor-dependent mechanisms and enhances OGT’s metabolic activity. Our data support the idea that integrating these two signals at the level of chromatin – a synergy almost entirely unique to these cells – is key for developing sensory plasticity. First, binding sites for the ERK effector Sr were among the most enriched at OGT × Pcl/PRE loci, consistent with our finding that OGT/Pcl/PRE loci were enriched in MAPK signaling (Figures 2 and 5); second, diet-driven changes in Sr regulons depended on the activity or presence of OGT and PRC2.1, and most importantly, the effects of rl/Sr (ERK/EGR2) on taste plasticity had strong epistatic interactions with OGT. Together, these molecular and functional data support the idea that the activity-dependent ERK pathway provides a relevant cellular context (likely neural activity) for adaptations to the nutrient environment. However, these sensing mechanisms must be turned into action to be effective. This is likely the role of PRC2.1 since the effects of ERK/EGR2 and OGT manipulations were dependent on the PRC2.1 function. Of note, only a small portion of the genes occupied by PRC2.1 is sensitive to diet and OGT activity, and our data show that PRC2.1 is not binding to new loci but instead tuning the output of those it is already bound to, likely via OGT instruction.
In the model we propose, metabolic and activity sensors integrate cellular information to promote changes in gene expression. But how are these actualized into physiological (in this case, neural) adaptations that underlie behavior or disease? *This is* one of the central and unresolved questions in nutrigenomics. Our model’s genetic and neural tractability provides a unique opportunity to get some answers. The 162 co-occupied loci identified were enriched for transcription and regulatory factors involved in cell proliferation, differentiation, signaling, and neural activity, as well as pathways implicated in neural plasticity. Many of these DNA-binding factors play essential roles during the development of sensory neurons to set their biophysical properties, such as Ptx1 and cad, but also affect adult taste plasticity. The regulons of these TFs include genes known to affect pre- and postsynaptic branching and structure, as well as synaptic physiology. Thus, our collective data indicate that this nutrigenomic pathway promotes taste adaptations, most likely by re-engaging developmental gene batteries, a mechanism that has been hypothesized to play a role in neural plasticity (Hobert, 2011; Marder and Prinz, 2002; Parrish et al., 2014; Vaziri et al., 2020). Whether this is a general rule of nutrigenomic signaling or something specific to these cells or neurons is yet to be determined; however, it is interesting to note that this is similar to how cancer cells exploit developmental networks for uncontrolled growth (Faubert et al., 2020; DeBerardinis and Chandel, 2016). On this note, many neural and psychiatric conditions show associations and connections with diet and metabolic states, including epilepsy, schizophrenia, bipolar, depression, Alzheimer’s, and Parkinson’s (Sarangi and Dus, 2021; Grigolon et al., 2020). Thus, uncovering nutrigenomic mechanisms in the brain could shed light on the etiology of these conditions and help design nutritional strategies to support people suffering from them; this is similar to how metabolic disorders like diabetes and cardiovascular disease are treated with a combination of drugs and nutrition.
## Limitations
Although using sensory plasticity and fly gustatory neurons as a model to study nutrigenomic signaling brings unique advantages, it also has significant limitations. These primarily arise from the small number of cells [60] and the in vivo nature of our model. First, we cannot probe whether OGT, PRC2.1, and Sr/EGR2 physically interact or modify each other in these cells. We also could not probe whether SD enhances the presence of OGT protein in the nucleus compared to a control diet due to the lack of functional antibodies. Thus, evidence for our model arises from the combination of cell-specific molecular, genetic, and physiological data. Second, we only inferred that the loci with Sr/EGR2 motifs integrate activity due to the well-established function of the ERK pathway in activity-dependent plasticity; future studies should address this directly and compare the effects of acute vs. chronic nutrient influx.
Further, while inhibitors have allowed us to establish critical epistatic interactions and conduct dietary manipulations while bypassing developmental effects and genetic challenges, we cannot exclude that some of these effects may be non-cell autonomous. Integrating this model with biochemical approaches that preserve the appropriate activity and nutrient context would help address these critical questions. Finally, pathways beyond OGT, ERK, and PRC2.1 may also play a role in sensory plasticity.
## Conclusions
In summary, we show that activity and nutrient-sensing mechanisms are integrated at the genomic level to promote neural adaptations to the food environment. In particular, our data reveal a central and instructional role for OGT and meaningful epistatic interactions with sensors (ERK) and effectors (PRC2.1). *More* generally, we put forth a model where cell and context specificity transforms ‘nutritional data’ – that is, variations in nutrient and metabolite levels – into nutritional information (Floridi, 2005), as shown in Figure 7E (pink boxes). This information is processed and interpreted by gene regulatory processes to make ‘decisions’ about responding to environmental challenges and carrying out physiological, neural, and behavioral changes. Thus, nutrigenomic mechanisms could provide a critical path for information flow in biological systems (Shannon, 1948; Reinagel, 2000; Smith, 2000; Fabris, 2009). A clear advantage could reside in their ability to amplify transient, and often minor, variations in nutrient and activity levels into strong reactions, which can be used to orchestrate responses to current and future environmental challenges. Future studies in this field will no doubt uncover fascinating insights about the rules of nutrigenomic communication: these discoveries will illuminate how nutrition and gene expression converge to shape cell physiology and provide us with new tools to promote wellness and diminish the burden of disease.
## Materials and methods
**Key resources table**
| Reagent type (species) or resource | Designation | Source or reference | Identifiers | Additional information |
| --- | --- | --- | --- | --- |
| Genetic reagent (D. melanogaster) | w[1118]CS | Other | | Gift from A Simon |
| Genetic reagent (D. melanogaster) | UAST-sxc(Ogt)RNAiCLb38 | PMID:24706800 | | Gift from C Lehner |
| Genetic reagent (D. melanogaster) | Gr5a-GAL4 | Bloomington Drosophila Stock Center | BDSC: 57592 | |
| Genetic reagent (D. melanogaster) | Tubulin-GAL80ts | Bloomington Drosophila Stock Center | BDSC: 7018 | |
| Genetic reagent (D. melanogaster) | UAS-Pcl | FlyORF | FlyORF: F001897 | |
| Genetic reagent (D. melanogaster) | Pclc429 | Other | | Gift from N Liu |
| Genetic reagent (D. melanogaster) | UAS-Rpl3-3XFLAG | PMID:29194454 | | Gift from D Dickman |
| Genetic reagent (D. melanogaster) | UAS-LT3-Dam | Other | | Gift from AH Brand |
| Genetic reagent (D. melanogaster) | UAS-LT3-Dam::Pcl | PMID:33177090 | | |
| Genetic reagent (D. melanogaster) | UAS-LT3-Dam::OGT | This paper | | See the Materials and methods |
| Genetic reagent (D. melanogaster) | Gr64f-GAL4 | Bloomington Drosophila Stock Center | BDSC: 57669 | |
| Genetic reagent (D. melanogaster) | UAS-Sr | Bloomington Drosophila Stock Center | BDSC: 26553 | |
| Genetic reagent (D. melanogaster) | UAS-rlWT | Bloomington Drosophila Stock Center | BDSC: 36270 | |
| Genetic reagent (D. melanogaster) | UAS-rlSem | Bloomington Drosophila Stock Center | BDSC: 59006 | |
| Genetic reagent (D. melanogaster) | rl1 | Bloomington Drosophila Stock Center | BDSC: 386 | |
| Genetic reagent (D. melanogaster) | rl RNAi | Bloomington Drosophila Stock Center | BDSC: 34855 | |
| Genetic reagent (D. melanogaster) | OGA RNAi | Bloomington Drosophila Stock Center | BDSC: 41882 | |
| Genetic reagent (D. melanogaster) | OGT1 | PMID:26348912 | | Gift from D van Aalten |
| Genetic reagent (D. melanogaster) | OGTK872M | PMID:26348912 | | Gift from D van Aalten |
| Antibody | Mouse monoclonal anti-Flag | Sigma | Cat#: F1804, RRID:AB_262044 | 3:50 |
| Peptide, recombinant protein | Dynabeads Protein G | Thermo Fisher Scientific | Cat#: 10004D | |
| Peptide, recombinant protein | T4 DNA ligase | New England Biolabs | Cat#: M0202S | |
| Commercial assay or kit | NEBuilder HiFi DNA Assembly kit | New England Biolabs | Cat#: E5520S | |
| Commercial assay or kit | ThruPLEX Kit | Takara | Cat#: 022818 | |
| Chemical compound, drug | OSMI-1 | Sigma | Cat#: SML1621 | |
| Chemical compound, drug | EED226 | Axon Medchem | Cat#: 2701 | |
| Chemical compound, drug | Trametinib | LC labs | Cat#: T-8123 | |
| Chemical compound, drug | TRIzol LS Reagent | Thermo Fisher Scientific | Cat#: 10296010 | |
| Software, algorithm | Autospike3.9 | Syntech | | http://www.ockenfels-syntech.com/products/signal-acquisition-systems-2/ |
| Software, algorithm | Prism 9 | GraphPad | RRID:SCR_002798 | |
| Software, algorithm | Python | Python | RRID:SCR_008394 | |
## Fly husbandry, strains, and diets
All flies were grown and fed cornmeal food (Bloomington Food B recipe) at 25°C and 45–$55\%$ humidity under a 12 hr light/12 hr dark cycle (Zeitgeber time 0 at 9:00 AM) unless otherwise stated. Male flies were collected under CO2 anesthesia 1–3 days after eclosion and maintained in a vial that housed 35–40 flies. Flies were acclimated to their new vial environment for an additional 2 days and were moved to fresh food vials every other day. The GAL4/UAS system was used to express transgenes of interest using the Gustatory receptor 5a Gr5a-GAL4 transgene. For each GAL4/UAS cross, transgenic controls were made by crossing the w1118CS (gift from A. Simon, CS and w1118 lines from the Benzer laboratory) to GAL4 or UAS flies, sex-matched to those used in the GAL4/UAS cross. The fly lines used for this paper are listed in Supplementary file 1.
For all dietary manipulations, the following compounds were mixed into standard cornmeal food (Bloomington Food B recipe) (0.58 calories/g) by melting, mixing, and pouring new vials as in Musselman and Kühnlein, 2018 and Na et al., 2013. For the $30\%$ sugar diet (1.41 calories/g), Domino granulated sugar (wt/vol) was added. Inhibitors were solubilized in $10\%$ DMSO and added to the control o sugar diet at a total concentration of 10 μM for OSMI (Ortiz-Meoz et al., 2015; May et al., 2020), 8 μM for EEDi (Vaziri et al., 2020), and 15.6 μM for Trametinib (Castillo-Quan et al., 2019; Slack et al., 2015). Animals were assigned randomly to dietary groups. The sample sizes were determined based on standards in the field. No animal was excluded from any of the analyses.
## Proboscis extension response
Male flies were food deprived for 18–20 hr in a vial with a Kimwipe dampened with Milli-Q filtered deionized water. PER was carried out as described in Shiraiwa and Carlson, 2007. Extension responses were recorded manually, and experimenters were blinded whenever possible. Experiments were replicated two to three times by two different experimenters.
## Affinity purification of ribosome-associated mRNA (TRAP)
Male fly heads (300 per replicate, ~10,000 Gr5a+ cells) were collected using sieves chilled in liquid nitrogen and dry ice. Frozen tissue was then lysed as previously described (Chen and Dickman, 2017; Vaziri et al., 2020). From $10\%$ of the total lysate, total RNA was extracted by TRIzol LS Reagent (Thermo Fisher Scientific, 10296010) for input. The remainder of the lysate was precleared by incubation with Dynabeads Protein G (Thermo Fisher Scientific, 10004D) for 2 hr and subsequently incubated with Dynabeads Protein G and an anti-Flag antibody (Sigma-Aldrich, F1804) at 4°C with rotation for 2 hr, then RNA was extracted from ribosomes bound to beads by TRIzol Reagent (Chen and Dickman, 2017).
## Targeted DNA adenine methyltransferase identification (TaDa) and chromatin accessibility TaDa (CATada)
*To* generate the UAS-LT3-Dam::OGT construct, the coding region of the OGT gene was amplified from w1118CS animals with the primers listed below and assembled into the UAS-LT3-DAM plasmid (gift from A. Brand, University of Cambridge) using the NEBuilder HiFi DNA Assembly kit based on the manufacturer’s instructions (New England Biolabs). Transgenic animals were validated by reverse transcription PCR targeting the correct insert. UAS-LT3-Dam::Pcl was as previously described in Vaziri et al., 2020. The UAS-LT3-Dam::OGT, UAS-LT3-Dam::Pcl, and UAS-LT3-Dam line were crossed to the Gr5a-GAL4; tubulin-GAL80ts. All animals were raised and maintained at 20°C. Expression of Dam::OGT/Pcl and Dam was induced at 28°C for 18 hr. For all experiments, 300 heads of male and female flies were collected per replicate on dry ice by sieving. DNA was extracted following kit instructions (Invitrogen). To identify methylated regions, purified DNA was digested by Dpn I, followed by PCR purification of digested sequences. TaDa adaptors were ligated by T4 DNA ligase (NEB). Adapter ligated DNA was PCR amplified and purified according to the protocol (Marshall et al., 2016). Purified DNA was digested with Dpn II, followed by sonication to yield fragments averaging 300 base pairs. TaDa adaptors were removed from sonicated DNA by digestion followed by PCR purification, and purified sonicated DNA was used for library preparation (Vaziri et al., 2020; Marshall et al., 2016).
pUAST-Sxc. Forward gatctgGCCGGCGCaATGCATGTTGAACAAACACGAATAAATATG, pUAST-Sxc. Reverse gttccttcacaaagatcctTTATACTGCTGAAATGTGGTCCGGAAG.
## Library preparation
Generation of RNA sequencing (RNA-seq) libraries was with the Ovation SoLo RNA-seq System for Drosophila (NUGEN, 0502-96). All reactions included integrated Heat-Labile Double-Strand Specific DNase treatment (ArcticZymes, catalog no. 70800-201). The DNA-sequencing libraries for TaDa were generated using the Takara ThruPLEX Kit (catalog no. 022818). For rat RNA-seq, libraries were prepared using the Nugen Ovation Model organism (Rat #0349-32) with $\frac{1}{10}$th ERCC spike-in mix. These libraries were run on a NextSeq instrument using a HO 150 cycle cit (75 × 75 bp paired-end reads). All Drosophila libraries were sequenced on the Illumina NextSeq platform (High-output kit v2 75 cycles) at the University of Michigan Genomics Core facility.
## High-throughput RNA-seq analysis
Fastq files were assessed for quality using FastQC (Andrews, Simon, and Others, 2010). Reads with a quality score below 30 were discarded. Sequencing reads were aligned by STAR (Dobin et al., 2013) to dmel-all-chromosomes of the dm6 genome downloaded from Ensembl genomes. Counting was conducted by HTSeq (Anders et al., 2015). Gene counts were used to call differential RNA abundance by DESeq2 (Love et al., 2014). A pipeline was generated from Wilinski et al., 2019. To determine the efficiency and cell specificity of the TRAP, pairwise comparisons were made between the Gr5a+-specific fraction and the input. For comparisons between dietary conditions, DESeq2 was only applied to the Gr5a+-specific IP condition. SD7 and Pclc429 datasets were analyzed from and described in Vaziri et al., 2020. A cutoff of q < 0.1 was used to call DEGs. To identify overlap between datasets GeneOverlap was used (Shen and Sinai, 2013).
## High-throughput TaDa and CATaDa analysis
Fastq files were assessed for quality using FastQC (Andrews, Simon, and Others, 2010). Reads with a quality score below 30 were discarded. The damidseq_pipeline was used to align, extend, and generate log2 ratio files (Dam::OGT/Dam and Dam::Pcl/Dam) in GATC resolution as described previously (Marshall and Brand, 2015). Reads were mapped by Bowtie2 (Langmead and Salzberg, 2012) to dmel-all-chromosomes of the dm6 genome downloaded from Ensembl genomes, followed by read extension to 300 bp (or to the closest GATC, whichever is first). Bam output is used to generate the ratio file in bedgraph format. Bedgraph files were converted to bigwig and visualized in the UCSC Genome Browser. Principal components analysis plots between biological replicates were computed by multibigwigSummary and plotCorrelation in deepTools (Ramírez et al., 2016). Peaks were identified from ratio files using find_peaks (FDR <0.01) (Marshall and Brand, 2015) and as in Vaziri et al., 2020. Overlapping intervals or nearby intervals (up to 50 bp) were merged into a single interval using mergeBed in BEDtools (Quinlan and Hall, 2010). Intervals common in at least two replicate peak files were identified by Multiple Intersect in BEDtools and used to generate the consensus peaks (Quinlan and Hall, 2010). For CATaDa experiments, all analyses were performed similarly to those of TaDa with the exception that Dam only profiles were not normalized as ratios but shown as normalized binding profiles generated by converting bam files to bigwig files normalized to 1× dm6 genome as reads per genome coverage (Sequencing depth is defined as the total number of mapped reads times the fragment length divided by the effective genome size). Binding intensity metaplots were made by computing a matrix for specified regions (Ramírez et al., 2016). To determine the proportion of genes that fit within the various chromatin domain subtypes, we first matched Dam::OGT/Dam targets to coordinates identified by Filion et al., 2010 and then determined their gene count in each chromatin subtype (observed) compared to the whole genome (expected). Peak annotations were conducted using the HOMER annotatePeaks tool (Heinz et al., 2010) with the dm6 reference genome. In TaDa analysis, genes were considered targets of the factor being investigated if a peak existed anywhere on their length.
## Pathway enrichment analysis
For all fly experiments, GO term enrichment analysis was performed using the iPAGE package (Goodarzi et al., 2009), using gene-GO term associations extracted from the Flybase dmel 6.08 2015_05 release. For all analyses, iPAGE was run in discrete mode. Independence filtering was deactivated for all discrete calculations. All other iPAGE settings default values. All shown GO terms pass the significance tests for overall information described in Goodarzi et al., 2009. For each term, bins that are outlined show especially strong contributions [p values such that a Benjamini–Hochberg FDR (Benjamini and Hochberg, 1995) calculated across that row yields q < 0.05].
## Analysis of cis-regulatory enrichments
For each D. melanogaster DNA-binding protein motif available from the CIS-BP database (Weirauch et al., 2014), we scanned the D. melanogaster genome (dmel 6.08 2015_05 release) using the FIMO-binding site discovery tool [cite:doi:10.1093/bioinformatics/btr614]. Hits for each motif were retained as potential binding sites and used to calculate overlaps with other features (e.g., OGT of Pcl sites), as noted. Permutation tests to assess significance were performed through repeated application of the bedtools shuffle [cite: https://doi.org/10.1093/bioinformatics/btq033] command to obtain 100 resamplings (Supplementary file 1) or 1000 resamplings (Figure 4) of the feature location of interest, requiring non-overlap of the randomly placed features. For the analysis in Supplementary file 1, we separately considered each potential motif for each transcription factor extracted from the CIS-BP database (separate motifs for the same factor are denoted by the gene name followed by a ‘_#’ suffix, with # and integer). In the case of analysis of Sr-binding sites in Figure 4, we obtained a merged set of potential Sr-binding sites by filtering potential binding sites at a q value threshold of 0.1 (acting separately for each motif) and combining all of the locations that were counted as a potential binding site for any of the Sr motifs available from CIS-BP. Enrichments of overlaps with OGT, Pcl, and PRE sites were calculated by comparing the actual observed count of overlapping features with the mean overlap observed across 1,000 random samplings of the Sr motif locations (preserving the chromosome on which each motif is located during shuffling). For comparison of Sr motif locations with TSSs, we first identified the (strandedness-aware) start location of all ‘gene’, ‘mobile_genetic_element’, or ‘pseudogene’ features from the dmel6 Genbank annotations and then categorized all of these locations as ‘OGT/PRC2’ or ‘Not OGT/PRC2’ based on whether or not the gene was associated with an OGT, Pcl, or PRE location (see Supplementary file 1 for gene lists for each feature type). The density of Sr motif hits (as defined above) was then calculated as a function of position relative to the TSS.
## Electrophysiology
Extracellular recording on labellar sensilla was performed using the tip recording method (Delventhal et al., 2014). Ten- to thirteen-day-old flies were anesthetized by short ice exposure. The reference electrode containing the Beadle–Ephrussi Ringer solution was inserted through the thorax into the labellum to immobilize the proboscis. The neuronal firing in L-type sensilla was recorded with a recording electrode (10–20 µm diameter) containing 25 mM sucrose dissolved in 30 mM tricholine citrate as an electrolyte. The recording electrode was connected to TastePROBE (Syntech), and electrical signals were obtained using the IDAC acquisition controller (Syntech). The signals were amplified (10×), band-pass-filtered (100–3000 Hz), and sampled at 12 kHz. Neuronal firing rates were analyzed by counting the number of spikes for a 500-ms period starting from 200 ms after contact using the Autospike 3.9 software. Experimenters were blinded in the initial characterization of the phenotypes and experiments were independently performed at least three times.
## Data analysis and statistics
Statistical tests, sample size, and p or q values are listed in each figure legend. One- or two-way repeated measure analysis of variance with post hoc tests were used for all PER experiments. All behavioral data were tested for normality, and the appropriate statistical tests were applied if data were not normally distributed. For the RNA-seq expression datasets, we coupled our standard differential expression with a test for whether each gene could be flagged as ‘significantly not different’ – that is, a gene for which we can confidently state that no substantial change in expression occurred (rather than just a lack of evidence for change, as would be inferred from a large p-value on the differential expression test). Defining a region of practical equivalence as a change of no more than 1.5-fold in either direction, we tested the null hypothesis of a change larger than 1.5-fold using the gene-wise estimates of the SE in log2fold change (reported by Deseq2) and the assumption that the actual l2fcs are normally distributed. Rejection of the null hypothesis is evidence that the gene’s expression is not changed substantially between the conditions of interest. Python code for the practical equivalence test is in Source code 1. All data in the figures are shown as means ± standard error of the mean, ****$p \leq 0.0001$, ***$p \leq 0.001$, **$p \leq 0.01$, and *$p \leq 0.05$, unless otherwise indicated.
## Data and material availability statement
All high-throughput data are available at the GEO repository: GSE188757 and GSE146245. LT3-Dam::OGT and Pcl flies are available upon request; all other fly lines are available in the BDSC database as shown in Supplementary file 1.
## Funding Information
This paper was supported by the following grants:
## Data availability
Sequencing data have been deposited in GEO under accession codes: GSE188757 and GSE146245. Analyzed omics data are included in Supplementary file 1.
The following dataset was generated: Sung and Vaziri 2022A nutrient information pathway links diet to tasteNCBI Gene Expression OmnibusGSE188757 The following previously published dataset was used: VaziriA KhabiriM GenawBT MayCE 2020Persistent Epigenetic Reprogramming of Sweet Taste by DietNCBI Gene Expression OmnibusGSE146245
## References
1. Anders S, Pyl PT, Huber W. **HTSeq -- a python framework to work with high-throughput sequencing data**. *Bioinformatics* (2015) **31** 166-169. DOI: 10.1093/bioinformatics/btu638
2. Andrews, Simon, and Others
2010FastQC: A Quality Control Tool for High Throughput Sequence DataCambridge, United KingdomBabraham Bioinformatics, Babraham Institute. *FastQC: A Quality Control Tool for High Throughput Sequence Data* (2010)
3. Ardiel EL, McDiarmid TA, Timbers TA, Lee KCY, Safaei J, Pelech SL, Rankin CH. **CaMK (CMK-1) and O-GlcNAc Transferase (OGT-1) Modulate Mechanosensory Responding and Habituation in an Interstimulus Interval-Dependent Manner in**. *bioRxiv* (2018). DOI: 10.1101/115972
4. Beckmann AM, Wilce PA. **Egr transcription factors in the nervous system**. *Neurochemistry International* (1997) **31** 477-510. DOI: 10.1016/s0197-0186(96)00136-2
5. Benjamini Y, Hochberg Y. **Controlling the false discovery rate: a practical and powerful approach to multiple testing**. *Journal of the Royal Statistical Society* (1995) **57** 289-300. DOI: 10.1111/j.2517-6161.1995.tb02031.x
6. Biggs WH, Zavitz KH, Dickson B, van der Straten A, Brunner D, Hafen E, Zipursky SL. **The**. *The EMBO Journal* (1994) **13** 1628-1635. DOI: 10.1002/j.1460-2075.1994.tb06426.x
7. Bouché C, Serdy S, Kahn CR, Goldfine AB. **The cellular fate of glucose and its relevance in type 2 diabetes**. *Endocrine Reviews* (2004) **25** 807-830. DOI: 10.1210/er.2003-0026
8. Butler AA, Sanchez RG, Jarome TJ, Webb WM, Lubin FD. **O-Glcnac and EZH2-mediated epigenetic regulation of gene expression during consolidation of fear memories**. *Learning & Memory* (2019) **26** 373-379. DOI: 10.1101/lm.049023.118
9. Castillo-Quan JI, Tain LS, Kinghorn KJ, Li L, Grönke S, Hinze Y, Blackwell TK, Bjedov I, Partridge L. **A triple drug combination targeting components of the nutrient-sensing network maximizes longevity**. *PNAS* (2019) **116** 20817-20819. DOI: 10.1073/pnas.1913212116
10. Chen X, Rahman R, Guo F, Rosbash M. **Genome-wide identification of neuronal activity-regulated genes in**. *eLife* (2016) **5**. DOI: 10.7554/eLife.19942
11. Chen X, Dickman D. **Development of a tissue-specific ribosome profiling approach in**. *PLOS Genetics* (2017) **13**. DOI: 10.1371/journal.pgen.1007117
12. Chu CS, Lo PW, Yeh YH, Hsu PH, Peng SH, Teng YC, Kang ML, Wong CH, Juan LJ. **O-glcnacylation regulates EZH2 protein stability and function**. *PNAS* (2014) **111** 1355-1360. DOI: 10.1073/pnas.1323226111
13. Chyb S, Dahanukar A, Wickens A, Carlson JR. *PNAS* (2003) **100 Suppl 2** 14526-14530. DOI: 10.1073/pnas.2135339100
14. Cork GK, Tan EP, Thompson J, McGreal S, Machecek M, Kelsh J, Peterson K, Apte U, Slawson C. **Sustained O‐glcnacylation amplifies ERK signaling**. *The FASEB Journal* (2018) **32**. DOI: 10.1096/fasebj.2018.32.1_supplement.673.20
15. Dai Z, Ramesh V, Locasale JW. **The evolving metabolic landscape of chromatin biology and epigenetics**. *Nature Reviews. Genetics* (2020) **21** 737-753. DOI: 10.1038/s41576-020-0270-8
16. DeBerardinis RJ, Chandel NS. **Fundamentals of cancer metabolism**. *Science Advances* (2016) **2**. DOI: 10.1126/sciadv.1600200
17. Decourcelle A, Very N, Djouina M, Loison I, Thévenet J, Body-Malapel M, Lelièvre E, Coqueret O, Leprince D, Yazidi-Belkoura IE, Dehennaut V. **O-glcnacylation links nutrition to the epigenetic downregulation of UNC5A during colon carcinogenesis**. *Cancers* (2020) **12**. DOI: 10.3390/cancers12113168
18. Delventhal R, Kiely A, Carlson JR. **Electrophysiological recording from**. *Journal of Visualized Experiments* (2014) **1**. DOI: 10.3791/51355
19. Dobin A, Davis CA, Schlesinger F, Drenkow J, Zaleski C, Jha S, Batut P, Chaisson M, Gingeras TR. **Star: ultrafast universal RNA-seq aligner**. *Bioinformatics* (2013) **29** 15-21. DOI: 10.1093/bioinformatics/bts635
20. Duclot F, Kabbaj M. **The role of early growth response 1 (EGR1) in brain plasticity and neuropsychiatric disorders**. *Frontiers in Behavioral Neuroscience* (2017) **11**. DOI: 10.3389/fnbeh.2017.00035
21. Fabris F. **Shannon information theory and molecular biology**. *Journal of Interdisciplinary Mathematics* (2009) **12** 41-87. DOI: 10.1080/09720502.2009.10700611
22. Faubert B, Solmonson A, DeBerardinis RJ. **Metabolic reprogramming and cancer progression**. *Science* (2020) **368**. DOI: 10.1126/science.aaw5473
23. Filion GJ, van Bemmel JG, Braunschweig U, Talhout W, Kind J, Ward LD, Brugman W, de Castro IJ, Kerkhoven RM, Bussemaker HJ, van Steensel B. **Systematic protein location mapping reveals five principal chromatin types in**. *Cell* (2010) **143** 212-224. DOI: 10.1016/j.cell.2010.09.009
24. Floridi L. **Is semantic information meaningful data?**. *Philosophy and Phenomenological Research* (2005) **70** 351-370. DOI: 10.1111/j.1933-1592.2005.tb00531.x
25. Forma E, Jóźwiak P, Ciesielski P, Zaczek A, Starska K, Bryś M, Krześlak A. **Impact of OGT deregulation on EZH2 target genes FOXA1 and FOXC1 expression in breast cancer cells**. *PLOS ONE* (2018) **13**. DOI: 10.1371/journal.pone.0198351
26. Gambetta MC, Oktaba K, Müller J. **Essential role of the glycosyltransferase sxc/ogt in polycomb repression**. *Science* (2009) **325** 93-96. DOI: 10.1126/science.1169727
27. Gambetta MC, Müller J. **O-Glcnacylation prevents aggregation of the polycomb group repressor polyhomeotic**. *Developmental Cell* (2014) **31** 629-639. DOI: 10.1016/j.devcel.2014.10.020
28. Gambetta MC, Müller J. **A critical perspective of the diverse roles of O-GlcNAc transferase in chromatin**. *Chromosoma* (2015) **124** 429-442. DOI: 10.1007/s00412-015-0513-1
29. Ganguly A, Dey M, Scott C, Duong VK, Arun Dahanukar A. **Dietary macronutrient imbalances lead to compensatory changes in peripheral taste via independent signaling pathways**. *The Journal of Neuroscience* (2021) **41** 10222-10246. DOI: 10.1523/JNEUROSCI.2154-20.2021
30. Gao J, Yang Y, Qiu R, Zhang K, Teng X, Liu R, Wang Y. **Proteomic analysis of the OGT interactome: novel links to epithelial-mesenchymal transition and metastasis of cervical cancer**. *Carcinogenesis* (2018) **39** 1222-1234. DOI: 10.1093/carcin/bgy097
31. Giles AC, Desbois M, Opperman KJ, Tavora R, Maroni MJ, Grill B. **A complex containing the O-GlcNAc transferase OGT-1 and the ubiquitin ligase EEL-1 regulates GABA neuron function**. *The Journal of Biological Chemistry* (2019) **294** 6843-6856. DOI: 10.1074/jbc.RA119.007406
32. Gonzales BJ, Mukherjee D, Ashwal-Fluss R, Loewenstein Y, Citri A. **Subregion-specific rules govern the distribution of neuronal immediate-early gene induction**. *PNAS* (2020) **117** 23304-23310. DOI: 10.1073/pnas.1913658116
33. Goodarzi H, Elemento O, Tavazoie S. **Revealing global regulatory perturbations across human cancers**. *Molecular Cell* (2009) **36** 900-911. DOI: 10.1016/j.molcel.2009.11.016
34. Grigolon RB, Gerchman F, Schöffel AC, Hawken ER, Gill H, Vazquez GH, Mansur RB, McIntyre RS, Brietzke E. **Mental, emotional, and behavioral effects of ketogenic diet for non-epileptic neuropsychiatric conditions**. *Progress in Neuro-Psychopharmacology & Biological Psychiatry* (2020) **102**. DOI: 10.1016/j.pnpbp.2020.109947
35. Hanover JA, Krause MW, Love DC. **The hexosamine signaling pathway: O-glcnac cycling in feast or famine**. *Biochimica et Biophysica Acta* (2010) **1800** 80-95. DOI: 10.1016/j.bbagen.2009.07.017
36. Hardivillé S, Hart GW. **Nutrient regulation of signaling, transcription, and cell physiology by O-GlcNAcylation**. *Cell Metabolism* (2014) **20** 208-213. DOI: 10.1016/j.cmet.2014.07.014
37. Hart GW, Slawson C, Ramirez-Correa G, Lagerlof O. **Cross talk between O-GlcNAcylation and phosphorylation: roles in signaling, transcription, and chronic disease**. *Annual Review of Biochemistry* (2011) **80** 825-858. DOI: 10.1146/annurev-biochem-060608-102511
38. Hart GW. **Nutrient regulation of signaling and transcription**. *The Journal of Biological Chemistry* (2019) **294** 2211-2231. DOI: 10.1074/jbc.AW119.003226
39. Hawkins M, Barzilai N, Liu R, Hu M, Chen W, Rossetti L. **Role of the glucosamine pathway in fat-induced insulin resistance**. *The Journal of Clinical Investigation* (1997) **99** 2173-2182. DOI: 10.1172/JCI119390
40. Heinz S, Benner C, Spann N, Bertolino E, Lin YC, Laslo P, Cheng JX, Murre C, Singh H, Glass CK. **Simple combinations of lineage-determining transcription factors prime cis-regulatory elements required for macrophage and B cell identities**. *Molecular Cell* (2010) **38** 576-589. DOI: 10.1016/j.molcel.2010.05.004
41. Hirosawa M, Hayakawa K, Shiota K, Tanaka S. **Histone O-glcnacylation and potential biological functions**. *OBM Genetics* (2018) **2**. DOI: 10.21926/obm.genet.1803036
42. Hobert O. **Regulation of terminal differentiation programs in the nervous system**. *Annual Review of Cell and Developmental Biology* (2011) **27** 681-696. DOI: 10.1146/annurev-cellbio-092910-154226
43. Huang Z, Cai L, Tu BP. **Dietary control of chromatin**. *Current Opinion in Cell Biology* (2015) **34** 69-74. DOI: 10.1016/j.ceb.2015.05.004
44. Hwang H, Rhim H. **Acutely elevated O-GlcNAcylation suppresses hippocampal activity by modulating both intrinsic and synaptic excitability factors**. *Scientific Reports* (2019) **9**. DOI: 10.1038/s41598-019-43017-9
45. Jiang M, Qiu Z, Zhang S, Fan X, Cai X, Xu B, Li X, Zhou J, Zhang X, Chu Y, Wang W, Liang J, Horvath T, Yang X, Wu K, Nie Y, Fan D. **Elevated O-GlcNAcylation promotes gastric cancer cells proliferation by modulating cell cycle related proteins and ERK 1/2 signaling**. *Oncotarget* (2016) **7** 61390-61402. DOI: 10.18632/oncotarget.11359
46. Jiang M, Xu B, Li X, Shang Y, Chu Y, Wang W, Chen D, Wu N, Hu S, Zhang S, Li M, Wu K, Yang X, Liang J, Nie Y, Fan D. **O-Glcnacylation promotes colorectal cancer metastasis via the mir-101-O-glcnac/EZH2 regulatory feedback circuit**. *Oncogene* (2019) **38** 301-316. DOI: 10.1038/s41388-018-0435-5
47. Katada S, Imhof A, Sassone-Corsi P. **Connecting threads: epigenetics and metabolism**. *Cell* (2012) **148** 24-28. DOI: 10.1016/j.cell.2012.01.001
48. Konzman D, Abramowitz LK, Steenackers A, Mukherjee MM, Na H-J, Hanover JA. **O-Glcnac: regulator of signaling and epigenetics linked to X-linked intellectual disability**. *Frontiers in Genetics* (2020) **11**. DOI: 10.3389/fgene.2020.605263
49. Lagerlöf O, Slocomb JE, Hong I, Aponte Y, Blackshaw S, Hart GW, Huganir RL. **The nutrient sensor OGT in PVN neurons regulates feeding**. *Science* (2016) **351** 1293-1296. DOI: 10.1126/science.aad5494
50. Lagerlöf O, Hart GW, Huganir RL. **O-glcnac transferase regulates excitatory synapse maturity**. *PNAS* (2017) **114** 1684-1689. DOI: 10.1073/pnas.1621367114
51. Langmead B, Salzberg SL. **Fast gapped-read alignment with bowtie 2**. *Nature Methods* (2012) **9** 357-359. DOI: 10.1038/nmeth.1923
52. Lavoie H, Gagnon J, Therrien M. **Erk signalling: a master regulator of cell behaviour, life and fate**. *Nature Reviews. Molecular Cell Biology* (2020) **21** 607-632. DOI: 10.1038/s41580-020-0255-7
53. Lei Y, Chen T, Li Y, Shang M, Zhang Y, Jin Y, Yu Q, Guo F, Wang T. **O-Glcnacylation of PFKFB3 is required for tumor cell proliferation under hypoxia**. *Oncogenesis* (2020) **9**. DOI: 10.1038/s41389-020-0208-1
54. Li YH, Liu X, Vanselow JT, Zheng H, Schlosser A, Chiu JC. **O-Glcnacylation of period regulates its interaction with clock and timing of circadian transcriptional repression**. *PLOS Genetics* (2019) **15**. DOI: 10.1371/journal.pgen.1007953
55. Liu H, Chen YG. **The interplay between TGF-β signaling and cell metabolism**. *Frontiers in Cell and Developmental Biology* (2022) **10**. DOI: 10.3389/fcell.2022.846723
56. Love MI, Huber W, Anders S. **Moderated estimation of fold change and dispersion for RNA-Seq data with deseq2**. *Genome Biology* (2014) **15**. DOI: 10.1186/s13059-014-0550-8
57. Marder E, Prinz AA. **Modeling stability in neuron and network function: the role of activity in homeostasis**. *BioEssays* (2002) **24** 1145-1154. DOI: 10.1002/bies.10185
58. Marshall S, Nadeau O, Yamasaki K. **Dynamic actions of glucose and glucosamine on hexosamine biosynthesis in isolated adipocytes: differential effects on glucosamine 6-phosphate, UDP-N-acetylglucosamine, and ATP levels**. *The Journal of Biological Chemistry* (2004) **279** 35313-35319. DOI: 10.1074/jbc.M404133200
59. Marshall OJ, Brand AH. **Damidseq_pipeline: an automated pipeline for processing DamID sequencing datasets**. *Bioinformatics* (2015) **31** 3371-3373. DOI: 10.1093/bioinformatics/btv386
60. Marshall OJ, Southall TD, Cheetham SW, Brand AH. **Cell-Type-Specific profiling of protein–DNA interactions without cell isolation using targeted DamID with next-generation sequencing**. *Nature Protocols* (2016) **11** 1586-1598. DOI: 10.1038/nprot.2016.084
61. May CE, Vaziri A, Lin YQ, Grushko O, Khabiri M, Wang QP, Holme KJ, Pletcher SD, Freddolino PL, Neely GG, Dus M. **High dietary sugar reshapes sweet taste to promote feeding behavior in**. *Cell Reports* (2019) **27** 1675-1685. DOI: 10.1016/j.celrep.2019.04.027
62. May CE, Rosander J, Gottfried J, Dennis E, Dus M. **Dietary sugar inhibits satiation by decreasing the central processing of sweet taste**. *eLife* (2020) **9**. DOI: 10.7554/eLife.54530
63. May CE, Dus M. **Confection confusion: interplay between diet, taste, and nutrition**. *Trends in Endocrinology and Metabolism* (2021) **32** 95-105. DOI: 10.1016/j.tem.2020.11.011
64. McCluskey LP, He L, Dong G, Harris R. **Chronic exposure to liquid sucrose and dry sucrose diet have differential effects on peripheral taste responses in female rats**. *Appetite* (2020) **145**. DOI: 10.1016/j.appet.2019.104499
65. Miningou N, Blackwell KT. **The road to ERK activation: do neurons take alternate routes?**. *Cellular Signalling* (2020) **68**. DOI: 10.1016/j.cellsig.2020.109541
66. Müller M, Kersten S. **Nutrigenomics: goals and strategies**. *Nature Reviews. Genetics* (2003) **4** 315-322. DOI: 10.1038/nrg1047
67. Murphy P, Davidson DR, Hill RE. **Segment-Specific expression of a homoeobox-containing gene in the mouse hindbrain**. *Nature* (1989) **341** 156-159. DOI: 10.1038/341156a0
68. Musselman LP, Kühnlein RP. *The Journal of Experimental Biology* (2018) **221**. DOI: 10.1242/jeb.163881
69. Na J, Musselman LP, Pendse J, Baranski TJ, Bodmer R, Ocorr K, Cagan R. **A**. *PLOS Genetics* (2013) **9**. DOI: 10.1371/journal.pgen.1003175
70. Na J, Sweetwyne MT, Park ASD, Susztak K, Cagan RL. **Diet-Induced podocyte dysfunction in**. *Cell Reports* (2015) **12** 636-647. DOI: 10.1016/j.celrep.2015.06.056
71. Oellers N, Hafen E. **Biochemical characterization of rolledSem, an activated form of**. *The Journal of Biological Chemistry* (1996) **271** 24939-24944. DOI: 10.1074/jbc.271.40.24939
72. Olivier-Van Stichelen S, Hanover JA. **You are what you eat: O-linked N-acetylglucosamine in disease, development and epigenetics**. *Current Opinion in Clinical Nutrition and Metabolic Care* (2015) **18** 339-345. DOI: 10.1097/MCO.0000000000000188
73. Olivier-Van Stichelen S, Wang P, Comly M, Love DC, Hanover JA. **Nutrient-driven O-linked N-acetylglucosamine (O-glcnac) cycling impacts neurodevelopmental timing and metabolism**. *Journal of Biological Chemistry* (2017) **292** 6076-6085. DOI: 10.1074/jbc.M116.774042
74. Ortiz-Meoz RF, Jiang J, Lazarus MB, Orman M, Janetzko J, Fan C, Duveau DY, Tan Z-W, Thomas CJ, Walker S. **A small molecule that inhibits OGT activity in cells**. *ACS Chemical Biology* (2015) **10** 1392-1397. DOI: 10.1021/acschembio.5b00004
75. Papa S, Choy PM, Bubici C. **The ERK and JNK pathways in the regulation of metabolic reprogramming**. *Oncogene* (2019) **38** 2223-2240. DOI: 10.1038/s41388-018-0582-8
76. Parrish JZ, Kim CC, Tang L, Bergquist S, Wang T, Derisi JL, Jan LY, Jan YN, Davis GW. **Krüppel mediates the selective rebalancing of ion channel expression**. *Neuron* (2014) **82** 537-544. DOI: 10.1016/j.neuron.2014.03.015
77. Qi W, Zhao K, Gu J, Huang Y, Wang Y, Zhang H, Zhang M, Zhang J, Yu Z, Li L, Teng L, Chuai S, Zhang C, Zhao M, Chan H, Chen Z, Fang D, Fei Q, Feng L, Feng L, Gao Y, Ge H, Ge X, Li G, Lingel A, Lin Y, Liu Y, Luo F, Shi M, Wang L, Wang Z, Yu Y, Zeng J, Zeng C, Zhang L, Zhang Q, Zhou S, Oyang C, Atadja P, Li E. **An allosteric PRC2 inhibitor targeting the H3K27me3 binding pocket of Eed**. *Nature Chemical Biology* (2017) **13** 381-388. DOI: 10.1038/nchembio.2304
78. Quinlan AR, Hall IM. **BEDTools: a flexible suite of utilities for comparing genomic features**. *Bioinformatics* (2010) **26** 841-842. DOI: 10.1093/bioinformatics/btq033
79. Ramírez F, Ryan DP, Grüning B, Bhardwaj V, Kilpert F, Richter AS, Heyne S, Dündar F, Manke T. **DeepTools2: a next generation web server for deep-sequencing data analysis**. *Nucleic Acids Research* (2016) **44** W160-W165. DOI: 10.1093/nar/gkw257
80. Reed DR, Alhadeff AL, Beauchamp GK, Chaudhari N, Duffy VB, Dus M, Fontanini A, Glendinning JI, Green BG, Joseph PV, Kyriazis GA, Lyte M, Maruvada P, McGann JP, McLaughlin JT, Moran TH, Murphy C, Noble EE, Pepino MY, Pluznick JL, Rother KI, Saez E, Spector AC, Sternini C, Mattes RD. **Nih workshop report: sensory nutrition and disease**. *The American Journal of Clinical Nutrition* (2020) **113** 232-245. DOI: 10.1093/ajcn/nqaa302
81. Reinagel P. **Information theory in the brain**. *Current Biology* (2000) **10** R542-R544. DOI: 10.1016/s0960-9822(00)00609-6
82. Robles-Flores M, Moreno-Londoño AP, Castañeda-Patlán MC. **Signaling pathways involved in nutrient sensing control in cancer stem cells: an overview**. *Frontiers in Endocrinology* (2021) **12**. DOI: 10.3389/fendo.2021.627745
83. Ruan H-B, Dietrich MO, Liu Z-W, Zimmer MR, Li M-D, Singh JP, Zhang K, Yin R, Wu J, Horvath TL, Yang X. **O-Glcnac transferase enables AGRP neurons to suppress browning of white fat**. *Cell* (2014) **159** 306-317. DOI: 10.1016/j.cell.2014.09.010
84. Sakabe K, Hart GW. **O-Glcnac transferase regulates mitotic chromatin dynamics**. *The Journal of Biological Chemistry* (2010) **285** 34460-34468. DOI: 10.1074/jbc.M110.158170
85. Sarangi M, Dus M. **Crème de la créature: dietary influences on behavior in animal models**. *Frontiers in Behavioral Neuroscience* (2021) **15**. DOI: 10.3389/fnbeh.2021.746299
86. Sartor F, Donaldson LF, Markland DA, Loveday H, Jackson MJ, Kubis H-P. **Taste perception and implicit attitude toward sweet related to body mass index and soft drink supplementation**. *Appetite* (2011) **57** 237-246. DOI: 10.1016/j.appet.2011.05.107
87. Schuettengruber B, Bourbon H-M, Di Croce L, Cavalli G. **Genome regulation by polycomb and trithorax: 70 years and counting**. *Cell* (2017) **171** 34-57. DOI: 10.1016/j.cell.2017.08.002
88. Sen SQ, Chanchani S, Southall TD, Doe CQ. **Neuroblast-specific open chromatin allows the temporal transcription factor, hunchback, to bind neuroblast-specific loci**. *eLife* (2019) **8**. DOI: 10.7554/eLife.44036
89. Shannon CE. **A mathematical theory of communication**. *Bell System Technical Journal* (1948) **27** 379-423. DOI: 10.1002/j.1538-7305.1948.tb01338.x
90. Shen L, Sinai M. *R Package Version* (2013)
91. Shiraiwa T, Carlson JR. **Proboscis extension response (PER) assay in**. *Journal of Visualized Experiments* (2007) **1**. DOI: 10.3791/193
92. Slack C, Alic N, Foley A, Cabecinha M, Hoddinott MP, Partridge L. **The ras-erk-ETS-signaling pathway is a drug target for longevity**. *Cell* (2015) **162** 72-83. DOI: 10.1016/j.cell.2015.06.023
93. Smith Jm. **The concept of information in biology**. *Philosophy of Science* (2000) **67** 177-194. DOI: 10.1086/392768
94. Su C, Schwarz TL. **O-Glcnac transferase is essential for sensory neuron survival and maintenance**. *The Journal of Neuroscience* (2017) **37** 2125-2136. DOI: 10.1523/JNEUROSCI.3384-16.2017
95. Sui B, Chen D, Liu W, Wu Q, Tian B, Li Y, Hou J, Liu S, Xie J, Jiang H, Luo Z, Lv L, Huang F, Li R, Zhang C, Tian Y, Cui M, Zhou M, Chen H, Fu ZF, Zhang Y, Zhao L. **A novel antiviral lncRNA, EDAL, shields a T309 O-GlcNAcylation site to promote EZH2 lysosomal degradation**. *Genome Biology* (2020) **21**. DOI: 10.1186/s13059-020-02150-9
96. Sung H, Vesela I, Driks H, Ferrario CR, Mistretta CM, Bradley RM, Dus M. **High-sucrose diet exposure is associated with selective and reversible alterations in the rat peripheral taste system**. *Current Biology* (2022) **32** 4103-4113. DOI: 10.1016/j.cub.2022.07.063
97. Szklarczyk D, Gable AL, Nastou KC, Lyon D, Kirsch R, Pyysalo S, Doncheva NT. **The STRING database in 2021: customizable protein–protein networks, and functional characterization of user-uploaded gene/measurement sets**. *Nucleic Acids Research* (2020) **49** D605-D612. DOI: 10.1093/nar/gkaa1074
98. Thomas GM, Huganir RL. **Mapk cascade signalling and synaptic plasticity**. *Nature Reviews. Neuroscience* (2004) **5** 173-183. DOI: 10.1038/nrn1346
99. van Steensel B, Henikoff S. **Identification of in vivo DNA targets of chromatin proteins using tethered dam methyltransferase**. *Nature Biotechnology* (2000) **18** 424-428. DOI: 10.1038/74487
100. Vaziri A, Khabiri M, Genaw BT, May CE, Freddolino PL, Dus M. **Persistent epigenetic reprogramming of sweet taste by diet**. *Science Advances* (2020) **6**. DOI: 10.1126/sciadv.abc8492
101. Vaziri A, Dus M. **Brain on food: the neuroepigenetics of nutrition**. *Neurochemistry International* (2021) **149**. DOI: 10.1016/j.neuint.2021.105099
102. Vella P, Scelfo A, Jammula S, Chiacchiera F, Williams K, Cuomo A, Roberto A, Christensen J, Bonaldi T, Helin K, Pasini D. **Tet proteins connect the O-linked N-acetylglucosamine transferase OGT to chromatin in embryonic stem cells**. *Molecular Cell* (2013) **49** 645-656. DOI: 10.1016/j.molcel.2012.12.019
103. Wang J, Liu R, Hawkins M, Barzilai N, Rossetti L. **A nutrient-sensing pathway regulates leptin gene expression in muscle and fat**. *Nature* (1998) **393** 684-688. DOI: 10.1038/31474
104. Wang QP, Lin YQ, Lai ML, Su Z, Oyston LJ, Clark T, Park SJ, Khuong TM, Lau MT, Shenton V, Shi YC, James DE, Ja WW, Herzog H, Simpson SJ, Neely GG. **Pgc1Α controls sucrose taste sensitization in**. *Cell Reports* (2020) **31**. DOI: 10.1016/j.celrep.2020.03.044
105. Weirauch MT, Yang A, Albu M, Cote AG, Montenegro-Montero A, Drewe P, Najafabadi HS, Lambert SA, Mann I, Cook K, Zheng H, Goity A, van Bakel H, Lozano J-C, Galli M, Lewsey MG, Huang E, Mukherjee T, Chen X, Reece-Hoyes JS, Govindarajan S, Shaulsky G, Walhout AJM, Bouget F-Y, Ratsch G, Larrondo LF, Ecker JR, Hughes TR. **Determination and inference of eukaryotic transcription factor sequence specificity**. *Cell* (2014) **158** 1431-1443. DOI: 10.1016/j.cell.2014.08.009
106. Weiss M, Anderluh M, Gobec M. **Inhibition of O-glcnac transferase alters the differentiation and maturation process of human monocyte derived dendritic cells**. *Cells* (2021) **10**. DOI: 10.3390/cells10123312
107. Wilinski D, Winzeler J, Duren W, Persons JL, Holme KJ, Mosquera J, Khabiri M, Kinchen JM, Freddolino PL, Karnovsky A, Dus M. **Rapid metabolic shifts occur during the transition between hunger and satiety in**. *Nature Communications* (2019) **10**. DOI: 10.1038/s41467-019-11933-z
108. Wise PM, Nattress L, Flammer LJ, Beauchamp GK. **Reduced dietary intake of simple sugars alters perceived sweet taste intensity but not perceived pleasantness**. *The American Journal of Clinical Nutrition* (2016) **103** 50-60. DOI: 10.3945/ajcn.115.112300
109. You Z, Peng D, Cao Y, Zhu Y, Yin J, Zhang G, Peng X. **P53 suppresses the progression of hepatocellular carcinoma via mir-15a by decreasing OGT expression and EZH2 stabilization**. *Journal of Cellular and Molecular Medicine* (2021) **25** 9168-9182. DOI: 10.1111/jcmm.16792
110. Zhang X, Ma L, Qi J, Shan H, Yu W, Gu Y. **Mapk/Erk signaling pathway-induced hyper-O-GlcNAcylation enhances cancer malignancy**. *Molecular and Cellular Biochemistry* (2015) **410** 101-110. DOI: 10.1007/s11010-015-2542-8
|
---
title: Investigation of Vitamin D Levels and the Effects of Being an Agricultural
Worker on Etiology and Night Pain in Children and Adolescents With Chronic Low Back
Pain
journal: Cureus
year: 2023
pmcid: PMC10036143
doi: 10.7759/cureus.36601
license: CC BY 3.0
---
# Investigation of Vitamin D Levels and the Effects of Being an Agricultural Worker on Etiology and Night Pain in Children and Adolescents With Chronic Low Back Pain
## Abstract
Objective: Chronic low back pain in children is a condition that should be investigated. In this study, we examined the effects of agricultural work on imaging results, risk factors, night pain, and vitamin D levels in children and adolescents with chronic low back pain.
Material and methods: The study included 133 patients who presented to the Physical Medicine and Rehabilitation and Neurosurgery outpatient clinics with low back pain that had lasted more than three months. The patients were evaluated based on the duration of their low back pain, the presence of night pain, a family history of low back pain, their employment status, local or radicular pain, and their body mass index (BMI). A physical examination was carried out to look into the etiologies of low back pain. Appropriate imaging, such as x-ray radiography, magnetic resonance imaging (MRI), and computed tomography (CT), was performed for the patients. Blood samples were collected from patients to assess inflammatory pathologies and vitamin D levels.
Results: The 133 patients in the study ranged in age from seven to 16 years, with a mean age of 14.3 + 1.9 years. Further, $60.2\%$ ($$n = 80$$) of the cases were male, while $39.8\%$ ($$n = 53$$) were female. Imaging revealed findings in $59.4\%$ of the patients. In $97.7\%$ of the participants, D hypovitaminosis was detected. There was no significant relationship between the patients’ imaging findings and vitamin D deficiency, family history, BMI, and employment status ($$p \leq 0.441$$, 0.147, 0.082, 0.605). The relationship between family history, employment status, and night pain was statistically significant ($p \leq 0.001$). There was no statistically significant relationship between night pain and vitamin D deficiency ($$p \leq 0.667$$).
Conclusion: Mechanical strain due to agricultural work and family history was found to be associated with night pain in patients with chronic low back pain in our study. The most important finding of this study is that night pain, which is considered a red flag, can occur in both inflammatory pathologies and situations causing mechanical low back pain, and risk factors should be thoroughly investigated. Studies with patients who have sufficient vitamin D will help to clarify the relationship between chronic low back pain and vitamin D.
## Introduction
Low back pain is a common condition in children that worsens with age. While it occurs at a rate of $1\%$ in early childhood, it can reach $50\%$ in adolescence. Moreover, $80\%$ of people have at least one occurrence before the age of 20. Low back pain in childhood also increases the likelihood of having low back pain in adulthood [1-3]. Further, it is more common in girls [4].
According to studies, obesity, female gender, increased screen exposure, sedentary lifestyle, depression, catastrophe, low socioeconomic level, child labor, accompanying systemic diseases, postural disorders, non-ergonomic study and sitting positions, sleep disorders, and hypermobility have all been linked to chronic low back pain [5].
Llow back pain ($80\%$) in children is caused by muscle spasms, which are easily treated and are referred to as non-specific low back pain. However, in pediatric patients, conditions such as disc pathologies, spondylolisthesis, spondylolysis, scoliosis, vertebral alignment disorders, Scheuermann's disease, juvenile fibromyalgia, infections, tumoral formations, and ankylosing spondylitis, which can be diagnosed and treated, can be classified as specific low back pain [5-7]. Chronic low back pain is defined in the literature as low back pain lasting longer than three months. Specific low back pain is prone to becoming chronic and recurring [6-8].
When evaluating a pediatric patient presenting with low back pain, conditions such as night pain, pain lasting more than four weeks, constant pain, weight loss, incontinence, immunosuppression, trauma, and neurologic deficits, which we call red flags, should be questioned. As a rule, applications from patients under the age of seven should be evaluated as a red flag and a detailed examination should be performed [5,9].
During the patient examination, which begins with an inspection, conditions such as posture, gait disturbance, and skin pigmentation changes should be examined. A thorough neurological examination should be performed on the patient, and the range of motion of the joint should be measured in all directions. Starting with the painless area, palpation should be performed. The Adam test should be used to assess patients for scoliosis. Sciatic and femoral nerve stretching tests should be performed on eligible patients [5,10].
In cases where non-specific low back pain is considered, imaging is not required in the initial evaluation. Patients who have red flags should begin with x-rays and, if necessary, advanced examination techniques such as magnetic resonance imaging (MRI) for soft tissue and disc pathologies, computed tomography (CT) and scintigraphy for bone pathologies. Blood tests should be performed on patients who require them, and treatment should begin immediately [2,5,6,11,12].
Vitamin D deficiency is a common condition in society and among children. The 25-(OH)D3 form is frequently measured in routine laboratory tests. 25-(OH)D3 values are evaluated as ≤20 ng/mL deficiency, 21-29 ng/mL deficiency, and ≥30 ng/mL adequate. Moreover, vitamin D deficiency in children has been linked to low back pain, and chronic low back pain has been associated with D hypovitaminosis [13-16]. In this study, we prospectively investigated the relationship between pediatric patients with chronic low back pain, etiology, risk factors, pain duration, night pain, and vitamin D.
## Materials and methods
This study included 133 children and adolescents who presented to Şanlıurfa Training and Research Hospital with a complaint of chronic low back pain and who presented to the Physical Medicine and Rehabilitation and Neurosurgery outpatient clinics with low back pain lasting more than three months between April 1, 2020 and October 31, 2020, and whose informed consent form was obtained from their parents.
The patients were evaluated based on the duration of their low back pain, the presence of night pain, a family history of low back pain, their employment status, local or radicular pain, and their body mass index (BMI). For the etiology of the patients, appropriate imaging such as x-ray, MRI, and CT was performed. Blood samples were collected from patients to assess inflammatory pathologies and vitamin D levels. After fasting for 8 h, blood was drawn from the cubital vein. The 25-(OH)D3 chemiluminescence immunoassay method was used to determine vitamin D levels. Since only three of the patients had sufficient vitamin D levels in the examinations, the patients were divided into two groups: those with vitamin D deficiency and those without (insufficient and sufficient). At the end of the study, it was determined whether the patients’ night pain, low back pain in the family, employment status, and etiology were related to vitamin D.
The Harran University Faculty of Medicine Clinical Research Ethics Committee approved this study (Decision no: HRU 20.06.14, date: March 30, 2020). The study was planned in accordance with the Helsinki Declaration.
To define continuous variables in statistical analysis, descriptive statistics are mean (mean), standard deviation (sd), minimum (min), median (med), and maximum (max). The descriptive statistics of categorical variables were calculated using frequency and percentage values. The Mann-Whitney U test was used to compare independent and non-normally distributed continuous variables. Chi-square or, where appropriate, Yates correction for continuity was used to compare categorical variables. The statistical significance level was set at $p \leq 0.005.$ MedCalc Statistical Software version 12.7.7 (MedCalc Software bvba, Ostend, Belgium; http://www.medcalc.org; 2013) was used for the analyses.
## Results
The 133 patients in the study ranged in age from seven to 16 years, with a mean age of 14.3 + 1.9 years. Further, $60.2\%$ ($$n = 80$$) of the cases were male, while $39.8\%$ ($$n = 53$$) were female. Table 1 shows the participants' general demographic data, etiology, pain type, special conditions on physical examination, vitamin D status, and treatment recommendations.
**Table 1**
| Unnamed: 0 | n | % |
| --- | --- | --- |
| Sex | | |
| Male | 80 | 60.2 |
| Female | 53 | 39.8 |
| Age (years ) | Mean+Sd | Med(min-max) |
| | 14.3+1.9 | 15(7-16) |
| Employment Status | | |
| Employed | 59 | 44.4 |
| Not employed | 74 | 55.6 |
| Night Pain | | |
| Yes | 66 | 49.6 |
| No | 67 | 50.4 |
| Family History of Low Back Pain | | |
| Yes | 68 | 51.1 |
| No | 65 | 48.9 |
| Adam Test | | |
| Negative | 126 | 94.7 |
| Positive | 7 | 5.3 |
| Neurological Examination | | |
| Sciatic nerve stretching | 2 | 1.5 |
| Normal | 131 | 98.5 |
| Etiology and Imaging | | |
| Findings With Imaging | 54 | 40.6 |
| Findings Without Imaging | 79 | 59.4 |
| Disc pathologies | 62 | 46.6 |
| Bulging | 56 | 42.1 |
| Protrusion | 6 | 4.5 |
| Ssacroiliitis | 5 | 3.8 |
| L1 fracture | 1 | 0.8 |
| Structural pathologies | 11 | 8.3 |
| Scoliosis | 7 | 5.3 |
| Spondylolisthesis | 4 | 3.0 |
| Vitamin D Level | | |
| Deficiency | 90 | 67.7 |
| Insufficient | 40 | 30 |
| Sufficient | 3 | 2.3 |
| Pain Type | | |
| Local | 129 | 97.0 |
| Radicular | 4 | 3.0 |
| Treatment | | |
| Surgical | 5 | 3.8 |
| Conservative | 128 | 96.2 |
| BMI | Mean+Sd | Med(min-max) |
| | 22.0+3.2 | 21.5(14.3-32.9) |
| Pain Duration (months) | Mean+Sd | Med(min-max) |
| | 10.3+7.4 | 7(3-36) |
The patients were classified based on the etiological diagnosis they received following imaging. Findings were found in $59.4\%$ ($$n = 79$$) of the patients who underwent imaging. Disc pathology was detected in $78.5\%$ ($$n = 62$$) of the patients with imaging findings. Patients with disc pathologies account for $46.6\%$ of all patients. Disc bulging was detected in $90.3\%$ ($$n = 56$$) of patients with disc pathology, while disc protrusion was detected in $9.7\%$ ($$n = 6$$). The levels of patients with bulging were L4-5 and L5-S1. After disc pathologies, structural pathologies are the second most common imaging finding, and scoliosis was found in seven patients after a positive Adam test, and spondylolisthesis was found in four patients. Three of the patients with spondylolisthesis were female, with their lumbar levels ranging from L5 to S1. The etiology of low back pain was found to be sacroiliitis in five patients, accounting for $3.8\%$ of all patients. After a five-month trauma, a fracture in the L1 vertebra was discovered in one patient. Immunosuppression, cancer, and infection were not found in any of the patients.
Moreover, $44.4\%$ ($$n = 59$$) of the patients were employed. In $97.7\%$ ($$n = 129$$) of the patients, the pain was localized to the lumbar region. In four patients, the pain was radicular. There were no motor deficits among the patients in the study. In two patients, the sciatic nerve stretch test (Straight Leg Raise Test and Lasegue) was positive.
A family history of low back pain was discovered in $51.1\%$ ($$n = 68$$) of the study participants. Night pain was detected in 66 patients, accounting for $49.6\%$ of the total. According to BMI, three patients were obese, 12 were overweight, 10 were underweight, and 108 were within normal weight limits.
In three of the patients, the vitamin D level was ≥30 ng/mL, which was sufficient. Vitamin D levels were insufficient (between 21 and 29 ng/mL) in 90 patients and deficiency (≤20 ng/mL) in 40 patients [13-16]. Vitamin D hypovitaminosis affected $97.7\%$ of the patients.
While medical therapy, physical therapy, and rest are recommended for $96.2\%$ of patients, surgery is recommended for five patients due to spondylolisthesis and fracture.
There was no significant difference between the groups when the patients were examined as those with and without imaging findings, based on their working status ($$p \leq 0.605$$). There was no statistically significant correlation between imaging findings and whether there was a family history of low back pain ($$p \leq 0.147$$) (Table 2).
**Table 2**
| N/% | Employment | Not employment | p* |
| --- | --- | --- | --- |
| Findings With Imaging | 37/46.8 | 42/53.2 | 0.605 |
| Findings Without Imaging | 22/40.7 | 32/59.3 | |
| Whole Group | 59/44.4 | 74/55.6 | |
| N/% | Family history Yes | Family history No | |
| Findings With Imaging | 45/66,2 | 34/52,3 | 0.104 |
| Findings Without Imaging | 23/33,8 | 31/47,7 | |
| Whole Group | 68/100 | 65/100 | |
Between the patients' vitamin D levels and imaging was no statistically significant correlation ($$p \leq 0.441$$, $$p \leq 0.667$$). When the patients were divided into groups based on their vitamin D status, there was no significant difference in terms of night pain ($$p \leq 0.667$$).
While $63.6\%$ of those suffering from night pain work, $36.4\%$ do not. This ratio differs statistically significantly ($p \leq 0.001$) Night pain affects $73.5\%$ of those with a family history of low back pain, but only $24.6\%$ of those without a family history. There was a statistically significant link between family history and night pain ($p \leq 0.001$) (Tables 3, 4).
The relationship between night pain and gender of the patients was not statistically significant ($$p \leq 0.915$$) as well as between the patients' vitamin D status and gender ($$p \leq 0.169$$). While $72.9\%$ of working patients are male, $50\%$ are non-working patients. Male employees make up a larger proportion of the workforce. The difference is statistically significant ($$p \leq 0.012$$).
## Discussion
Low back pain is a common condition in children that can lead to serious complications. Since chronic pain is one of the red flags in pediatric patients, its etiology should be determined quickly and treated with a multidisciplinary approach if necessary [4,5,7,8].
Having a family history of low back pain, being under emotional stress, and leading a sedentary lifestyle all increase the risk of low back pain becoming chronic [17,18]. A family history of low back pain was found in $51.1\%$ of our patients. Having a family history of low back pain has been shown in studies to be a risk factor for low back pain, and low back pain is more common in patients with a family history of low back pain [19,20]. The results of our study support the literature. One of the interesting results of our study was that patients with a family history had a higher rate of night pain.
Imaging findings were discovered in $59.4\%$ of the patients in our study. Moreover, the most common etiologies were disc pathologies and scoliosis. In a study by Yang et al., the most common diagnoses in adolescents with spasms and strain in the low back muscles were disc pathology and scoliosis [7]. Our findings were found to be etiologically consistent with the literature. According to studies, three of the spondylolisthesis cases in our cases were female and had anterolisthesis, and the level was L5-S1 [6,8].
Although studies have found that low back pain is more common in females in children and adolescents, other studies have concluded that there is no relationship with gender [4]. There was no gender difference in our study. The reason for this could be that the patients in our study had a working status, which was more common in the male gender.
The patients in our study were younger than 18 years old, the pain lasted longer than three months, and night pain was all red flags. In the literature, night pain, in particular, has been shown to occur in serious conditions such as malignancy, metastatic malignancy, infection, and spondyloarthropathy [5,9,21]. In our study, we discovered inflammatory pathology in five patients with night pain, which was determined to be sacroiliitis. One of our patients was experiencing night pain as a result of previous trauma and a fracture of the L1 vertebra. Our study found a statistically significant relationship between night pain and working status, indicating that mechanical strain may be associated with night pain. The regions where the patients participating in the study lived were those with low socioeconomic status. Further, $44.4\%$ ($$n = 59$$) of the participants were agricultural workers, performing tasks such as cotton picking, onion picking, and hoeing. During these procedures, the patients stated that they did activities such as bending and straightening, carrying loads on occasion, and squatting. Musculoskeletal pain is frequently encountered in child workers, according to a study conducted on potato worker children in India, and these pains are mostly in the lumbar region. According to the same study, some of the activities performed during potato crafting paved the way for this pain. Among these are people standing while watering and planting seeds, as well as activities like carrying potatoes and bending over. The study found that the most basic reason for children to do agricultural work is to help the family’s livelihood due to the poor socioeconomic conditions of the families [22]. This is also why our pediatric patients who work as agricultural workers in our study work. It has been reported that families in poor socioeconomic conditions have an adverse effect on children’s health and are especially associated with spinal pain, and having poor socioeconomic conditions may be a risk factor for musculoskeletal problems [17,20]. In various studies in the literature, it has been shown that the musculoskeletal system, particularly low back pain, is common in agricultural workers [23,24]. In a study with farmers, chronic pain was found to be more common in farmers than in healthy individuals, and it was shown that chronic pain resulted in disc herniation over time [25].
Vitamin D hypovitaminosis affects $97.7\%$ of our patients. Moreover, $67.7\%$ of the patients were vitamin D deficient. In the literature, due to differences in data from countries, the level of vitamin D in children and adolescents varies between $19\%$ and $60\%$, and the level of deficiency varies between $7\%$ and $68\%$ [26]. Studies have shown that patients with musculoskeletal pain may have D hypovitaminosis, and hypovitaminosis D may cause widespread chronic musculoskeletal pain, particularly chronic low back pain [13,27-29]. Vitamin D is a hormone that regulates calcium and bone metabolism and is important in the inflammatory and immune response of the body. Inflammatory parameters such as interleukin 6, tumor necrosis factor, and C-reactive protein (CRP) are high in patients with low back pain, which may be related to D hypovitaminosis, and patients with CRP > 3 are more likely to have low back pain [7-9]. Vitamin D deficiency has been linked to increased oxidative stress, muscle atrophy, and a decrease in mitochondrial functions in the multifidus muscle [30].
The limitation of our study was the small number of participants who had adequate vitamin D levels.
## Conclusions
When the results of our study were examined, it was discovered that night pain was associated with work status and family history, implying that mechanical strain and family history in our patients could be risk factors for nocturnal low back pain. It is important to remember that the underlying cause of night pain, which is one of the red flags, could be an inflammatory pathology. However, we have determined in our study that night pain can also occur in conditions such as disc degeneration and structural disorders. Due to the small number of participants with adequate vitamin D levels, the statistical difference between vitamin D status and night pain may have been insignificant. Studies with a large number of participants and a patient group with adequate vitamin D levels will help to shed light on these issues.
## References
1. Akbar F, AlBesharah M, Al-Baghli J, Bulbul F, Mohammad D, Qadoura B, Al-Taiar A. **Prevalence of low back pain among adolescents in relation to the weight of school bags**. *BMC Musculoskelet Disord* (2019) **20** 37. PMID: 30670005
2. Booth TN, Iyer RS, Falcone RA Jr. **ACR appropriateness criteria(®) back pain-child**. *J Am Coll Radiol* (2017) **14** 0-24
3. Hwang J, Louie PK, Phillips FM, An HS, Samartzis D. **Low back pain in children: a rising concern**. *Eur Spine J* (2019) **28** 211-213. PMID: 30506290
4. Beynon AM, Hebert JJ, Lebouef-Yde C, Walker BF. **Potential risk factors and triggers for back pain in children and young adults. A scoping review, part II: unclear or mixed types of back pain**. *Chiropr Man Therap* (2019) **27** 61
5. Kordi R, Rostami M. **Low back pain in children and adolescents: an algorithmic clinical approach**. *Iran J Pediatr* (2011) **21** 259-270. PMID: 23056800
6. Kjaer P, Leboeuf-Yde C, Sorensen JS, Bendix T. **An epidemiologic study of MRI and low back pain in 13-year-old children**. *Spine (Phila Pa 1976)* (2005) **30** 798-806. PMID: 15803084
7. Yang S, Werner BC, Singla A, Abel MF. **Low back pain in adolescents: a 1-year analysis of eventual diagnoses**. *J Pediatr Orthop* (2017) **37** 344-347. PMID: 26368854
8. MacDonald J, Stuart E, Rodenberg R. **Musculoskeletal low back pain in school-aged children: a review**. *JAMA Pediatr* (2017) **171** 280-287. PMID: 28135365
9. DePalma MG. **Red flags of low back pain**. *JAAPA* (2020) **33** 8-11
10. Taxter AJ, Chauvin NA, Weiss PF. **Diagnosis and treatment of low back pain in the pediatric population**. *Phys Sportsmed* (2014) **42** 94-104. PMID: 24565826
11. Feldman DS, Straight JJ, Badra MI, Mohaideen A, Madan SS. **Evaluation of an algorithmic approach to pediatric back pain**. *J Pediatr Orthop* (2006) **26** 353-357. PMID: 16670548
12. Sheehan NJ. **Magnetic resonance imaging for low back pain: indications and limitations**. *Postgrad Med J* (2010) **86** 374-378. PMID: 20547606
13. Alghadir AH, Gabr SA, Al-Eisa ES. **Mechanical factors and vitamin D deficiency in schoolchildren with low back pain: biochemical and cross-sectional survey analysis**. *J Pain Res* (2017) **10** 855-865. PMID: 28442927
14. Briggs MS, Givens DL, Schmitt LC, Taylor CA. **Relations of C-reactive protein and obesity to the prevalence and the odds of reporting low back pain**. *Arch Phys Med Rehabil* (2013) **94** 745-752. PMID: 23187041
15. Wang H, Schiltenwolf M, Buchner M. **The role of TNF-alpha in patients with chronic low back pain-a prospective comparative longitudinal study**. *Clin J Pain* (2008) **24** 273-278. PMID: 18287835
16. Xu HW, Zhang SB, Yi YY, Chen H, Hu T, Wang SJ, Wu DS. **Relationship between vitamin D and nonspecific low back pain may be mediated by inflammatory markers**. *Pain Physician* (2021) **24** 0-23
17. King S, Chambers CT, Huguet A, MacNevin RC, McGrath PJ, Parker L, MacDonald AJ. **The epidemiology of chronic pain in children and adolescents revisited: a systematic review**. *Pain* (2011) **152** 2729-2738. PMID: 22078064
18. Lynch AM, Kashikar-Zuck S, Goldschneider KR, Jones BA. **Psychosocial risks for disability in children with chronic back pain**. *J Pain* (2006) **7** 244-251. PMID: 16618468
19. Amorim AB, Ferreira PH, Ferreira ML. **Influence of family history on prognosis of spinal pain and the role of leisure time physical activity and body mass index: a prospective study using family-linkage data from the Norwegian HUNT study**. *BMJ Open* (2018) **8** 0
20. Joergensen AC, Hestbaek L, Andersen PK, Nybo Andersen AM. **Epidemiology of spinal pain in children: a study within the Danish National Birth Cohort**. *Eur J Pediatr* (2019) **178** 695-706. PMID: 30788593
21. Houghton KM. **Review for the generalist: evaluation of low back pain in children and adolescents**. *Pediatr Rheumatol Online J* (2010) **8** 28. PMID: 21092217
22. Das B, Ghosh T, Gangopadhyay S. **Child work in agriculture in West Bengal, India: assessment of musculoskeletal disorders and occupational health problems**. *J Occup Health* (2013) **55** 244-258. PMID: 23748207
23. Ganesh S, Chhabra D, Kumari N. **The effectiveness of rehabilitation on pain-free farming in agriculture workers with low back pain in India**. *Work* (2016) **55** 399-411. PMID: 27689586
24. Gangopadhyay S, Das B, Das T, Ghoshal G. **An ergonomic study on posture-related discomfort among preadolescent agricultural workers of West Bengal, India**. *Int J Occup Saf Ergon* (2005) **11** 315-322. PMID: 16219159
25. Tomczyszyn D, Solecki L, Pańczuk A. **Assessment of the type of farmers' low back pain**. *Med Pr* (2018) **69** 355-364. PMID: 29946196
26. Saggese G, Vierucci F, Boot AM. **Vitamin D in childhood and adolescence: an expert position statement**. *Eur J Pediatr* (2015) **174** 565-576. PMID: 25833762
27. Beck JJ, Mahan ST, Nowicki P, Schreiber VM, Minkowitz B. **What is new in pediatric bone health**. *J Pediatr Orthop* (2021) **41** 0-9
28. Çalık Y, Aygün Ü. **Evaluation of vitamin D levels in patients with chronic low back-leg pain**. *Acta Orthop Traumatol Turc* (2017) **51** 243-247. PMID: 28438384
29. Panwar A, Valupadas C, Veeramalla M, Vishwas HN. **Prevalence of vitamin D deficiency in chronic and subacute low back pain patients in India: a triple-arm controlled study**. *Clin Rheumatol* (2018) **37** 1367-1374. PMID: 28842760
30. Dzik KP, Skrobot W, Kaczor KB. **Vitamin D Deficiency Is Associated with Muscle Atrophy and Reduced Mitochondrial Function in Patients with Chronic Low Back Pain**. *Oxid Med Cell Longev* (2019) **2019** 6835341. PMID: 31281588
|
---
title: 'Intensive Blood Pressure Control and Diabetes Mellitus Incidence for Patients
with Impaired Fasting Glucose: A Secondary Analysis of SPRINT'
authors:
- Beiru Lin
- Xiaochuan Liu
- Sichen Yao
- Zhigang Pan
journal: International Journal of Hypertension
year: 2023
pmcid: PMC10036175
doi: 10.1155/2023/7533353
license: CC BY 4.0
---
# Intensive Blood Pressure Control and Diabetes Mellitus Incidence for Patients with Impaired Fasting Glucose: A Secondary Analysis of SPRINT
## Abstract
### Background
Previous studies indicated that intensive blood pressure (BP) control (systolic BP < 120 mm·Hg) compared with standard BP control (<140 mm·Hg) was associated with an increased risk of type 2 diabetes (T2D) and impaired fasting glucose (IFG) among hypertensive patients with normoglycemia. However, the impact of intensive BP control on the incidence of T2D for those with IFG is still unknown.
### Methods
This was a secondary analysis of the SPRINT (Systolic Blood Pressure Intervention Trial) of the study. We included participants with IFG at randomization, which was defined as fasting blood glucose (FBG) between 100 and 125 mg/dL. The primary outcome was incident T2D, defined as events of reaching FBG ≥ 126 mg/dL, participant self-report T2D at annual examination, or a record of hypoglycemic medications at follow-up. The secondary outcome was incident IFG reversion (IFGR), defined as the time to first FBG back to normoglycemia (<100 mg/dl) among participants without incident T2D. Cox proportional hazards models were used to compare the cumulative incidence of outcomes between the two BP control groups. Hazard ratios (HRs) with $95\%$ confidence intervals (CIs) were calculated.
### Results
A total of 3310 participants were included in our primary outcome analysis (median age 67 years, $29\%$ female). There were 293 participants who developed T2D among the intensive BP control group and 256 participants who developed T2D among the standard BP control group, resulting in 56.87 (50.36–63.39) versus 49.33 (43.29–55.37) events per 1000 person-years of treatment (HR 1.18 [$95\%$ CI, 1.00–1.40], $$P \leq 0.052$$). After excluding 549 participants who developed T2D, 2761 participants were included in our secondary outcome analysis with 559 participants who developed IFGR among the intensive BP control group and 632 participants who developed IFGR among the standard BP control group, resulting in 141.20 (129.50–152.91) versus 158.20 (145.86,170.53) events per 1000 person-years of treatment (HR 0.9 [$95\%$ CI, 0.8–1.01], $$P \leq 0.067$$).
### Conclusions
Our study found that in comparison to the standard BP control for hypertensive patients with IFG, intensive BP control was associated with a small increased risk of new-onset T2D, though it did not reach statistical significance. This kind of impact should be considered when implementing the strategy, especially for those with high risks of developing T2D. This trial is registered with NCT01206062.
## 1. Introduction
Hypertension and type 2 diabetes (T2D) are two common chronic conditions that often coexist in the same individual [1–3]. They have many shared risk factors such as obesity, diet, and insulin resistance. Due to shared risk factors and diverse classes of antihypertensive medications, a complex relationship has been reported between hypertension and T2D. Several studies have indicated that higher blood pressure is associated with an increased risk of T2D [4, 5]. Nearly $80\%$ higher risk of T2D was observed with each 20 mm·Hg elevated systolic blood pressure (SBP) in a meta-analysis of cohort studies [5]. However, commonly used antihypertensive agents such as thiazide diuretics and β-blockers have been reported to be associated with increased risk of new-onset T2D, while angiotensin-converting enzyme inhibitors and angiotensin II receptor blockers were on the contrary [6–9]. Antihypertensive drug-drug interactions have been proposed as the underlying causes of the conflicting results.
With the publication of several positive clinical trials investigating the beneficial effect of intensive blood pressure (BP) control, a lower SBP target seems to be plausible among those who meet the criteria [10–12]. Strict SBP target often requires more antihypertensive agents, which would increase the risk of drug-drug interactions. A secondary analysis of SPRINT (Systolic Blood Pressure Intervention Trial) showed us that intensive BP control (SBP < 120 mm·Hg) compared with standard BP control (SBP < 140 mm·Hg) was associated with increased risk of T2D and impaired fasting glucose (IFG) among hypertensive patients with normoglycemia at the baseline [13]. Up to now, the impact of intensive BP control on the incidence of T2D in those with IFG at the baseline is still unknown. A secondary analysis of the SPRINT population might be able to fill this gap. In light of previous findings, our hypotheses in this study were that [1] intensive BP control was associated with increased risk of T2D for those with IFG at the baseline and [2] intensive BP control was associated with a lower rate of IFG reversion (IFGR) to normoglycemia.
## 2.1. Data Reproducibility Statement
SPRINT anonymized data are available at the National Heart, Lung, and Blood Institute (NHLBI) Biologic Specimen and Data Repository (https://biolincc.nhlbi.nih.gov/home/).
## 2.2. Study Design and Population
Our study was a secondary analysis of SPRINT, which is a multicenter randomized controlled trial that was conducted in the United States between November 2010 and March 2013. The major finding of the trial was that intensive BP treatment (SBP target <120 mm·Hg) compared with standard BP treatment (SBP target <140 mm·Hg) was more effective in preventing cardiovascular outcomes. Details of the trial have been reported elsewhere [14, 15]. Briefly, participants with screened SBP 130 to 180 mm·Hg and an increased risk of cardiovascular events were included. Participants were excluded if they had diabetes mellitus, severe heart failure, stroke, or dementia.
In this study, we further excluded participants who may have had diabetes mellitus at the baseline. Those with a missing record of blood glucose or with normoglycemia (<100 mg/dl), who had a fasting glucose ≥126 mg/dL, or who were on a glucose-lowering medication at randomization were also excluded. IFG was defined as fasting blood glucose (FBG) between 100 and 125 mg/dL.
This study was approved by the institutional review board (IRB) of each clinical facility, and all participants provided written informed consent.
## 2.3. Baseline Characteristics
Demographic and clinical data were collected through randomization. Medical histories were collected annually and included messages from hypoglycemic agents and self-reported T2D. BP was obtained by calculating the mean of 3 automated cuff readings with an automated device (Omron-HEM-907 XL) following standardized procedures. Laboratory data were collected at the baseline and at the 24 and 48 months or closeout visits. The value of blood glucose was considered missing if the sample was marked as nonfasting. Blood glucose was measured in serum using the hexokinase method on a Roche analyzer. Study covariates included age, gender, race, body mass index (BMI), number of antihypertensive drugs prescribed prior to randomization, and the Framingham risk score (FRS). Baseline estimated glomerular filtration rates (eGFR) were obtained by using the modification of diet in renal disease 4-component equation [16]. Previous cardiovascular disease (CVD) was defined as a history of clinical or subclinical CVD. Chronic kidney disease was defined as eGFR <60 mL/min/1.73 m2.
## 2.4. Outcome Definitions
The primary outcome of our study was incident T2D, defined as events of reaching FBG ≥ 126 mg/dL, participant self-report of T2D at annual examination, or a record of hypoglycemic medications at follow-up. The secondary outcome was incident IFGR, defined as the time to first FBG back to normoglycemia (<100 mg/dl) among participants without incident T2D.
## 2.5. Study Power Consideration
First of all, our study was a secondary analysis of SPRINT participants, so like many other secondary analyses, the power of our finding might be insufficient because the number of participants that fulfilled our study purpose was fixed. In our primary outcome analysis, 1,659 patients were allocated to the intensive BP control group and 1,651 patients were allocated to the standard BP control group. By using PASS 15.0 software, group sample sizes of 1,650 in group 1 and 1,650 in group 2 will achieve $57.151\%$ power to detect a ratio of the group proportions of 1.18, which indicates that the power of our findings is insufficient. To achieve a power higher than $90\%$, the sample size needs to be about 7,000 (3,500 in each group).
## 2.6. Statistical Analysis
All statistical analyses were conducted in R version 3.6.2. where a P value <0.05 was considered statistically significant. Baseline characteristics were compared between participants with IFG randomized in the intensive BP control group and standard BP control group. Continuous variables were compared with the Wilcoxon rank sum test, and categorical variables were compared with Pearson's chi-squared test or Fisher's exact test. Cox proportional hazards models were used to compare the cumulative incidence of new-onset T2D between the intensive and standard BP control groups. For the comparison of the cumulative incidence of IFGR, we excluded participants who developed T2D after randomization. Hazard ratios (HRs) with $95\%$ confidence intervals (CIs) were calculated, with the standard BP control group as the reference group. The follow-up time was censored at the end of the trial (August 20, 2015), upon death, failure to follow-up, or reaching the outcomes (T2D, IFGR). The proportional hazards assumptions were verified through checking Schoenfeld residuals. Subgroup analyses were conducted to test the interaction effect (treatment arm∗ subgroup) for our primary outcome among the following groups: age (≥75 and <75 years), race (black and nonblack), FRS (≥$15\%$ and <$15\%$), number of antihypertensive drugs (≥2 and <2), gender (male and female), previous CVD (yes and no), SBP tertile at randomization (<132 mm·Hg; 132 to 145 mm·Hg; >145 mm·Hg), and baseline CKD (eGFR ≥60 and <60 mL/min/1.73 m2).
## 2.7. Sensitivity Analysis
First, we excluded participants who developed incident T2D within the first year of follow-up to explore the potential impact of reverse causality on our primary outcome. For the secondary outcome, we excluded participants who developed IFGR within 24 months of follow-up. Second, we excluded participants who withdrew their consent or failed to follow up after randomization to explore the potential impact of drop-out. Third, we compared the incidence of T2D and IFGR at different time points of follow-up between intensive and standard BP control groups.
## 3.1. Characteristics of Study Participants
Figure 1 shows us the study flowchart. There were 9,361 participants enrolled in SPRINT. For the main analysis of incident T2D, we further excluded 6,051 participants: 153 participants had T2D at randomization; 611 participants had a baseline blood glucose sample marked as nonfasting; 5027 participants with the baseline blood glucose level <100 mg/dl; 257 participants had a baseline blood glucose level ≥126 mg/dl; and 3 participants with self-reported use of a hypoglycemic medicine. For the secondary analysis of the IFGR, we further excluded 549 participants who developed T2D after randomization. Baseline characteristics were well balanced between intensive and standard BP control groups, no matter in our primary analysis or in our secondary analysis (Table 1 and Supplemental Table s1). The median duration of follow-up was 3.22 years in our primary outcome analysis and 2.98 years in our secondary outcome analysis.
## 3.2. Impact of Intensive BP Control on Incident T2D
Out of the 3,310 participants included in our primary outcome analysis, 1,659 were allocated to the intensive BP control group and 1,651 were allocated to the standard BP control group (Figure 1). There were 293 participants who developed T2D among the intensive BP control group and 256 participants who developed T2D among the standard BP control group, resulting in 56.87 (50.36–63.39) versus 49.33 (43.29–55.37) events per 1,000 person-years of treatment (HR 1.18 [$95\%$ CI, 1.00–1.40], $$P \leq 0.052$$, Table 2).
## 3.3. Impact of Intensive BP Control on Incident IFGR
After excluding 549 participants who developed T2D, 2,761 participants were included in our secondary outcome analysis (Figure 1). Of them, 1366 participants were allocated to the intensive BP control group. There were 559 participants who developed IFGR among the intensive BP control group and 632 participants who developed IFGR among the standard BP control group, resulting in 141.20 (129.50–152.91) versus 158.20 (145.86, 170.53) events per 1000 person-years of treatment, respectively (HR 0.9 [$95\%$ CI, 0.8–1.01], $$P \leq 0.067$$, Table 2).
## 3.4. Subgroup Analysis of the Effect of Intensive BP Control on Incident T2D
The interactive effect was tested between the treatment strategy and prespecified subgroups (Figure 2). Overall, no significant interactive effect was observed among all subgroups (P for interaction >0.05). Increased risk of T2D with intensive BP control was observed only among participants with eGFR ≥ 60 mL/min/1.73 m2 (HR 1.23 ($95\%$ CI, 1.02–1.49)), those whose SBP ranged from 132 to 145 mm·Hg at randomization (HR 1.42 ($95\%$ CI, 1.06–1.89)), and those with a self-reported race of nonblack (HR 1.22 [$95\%$ CI, 1.00–1.50]).
## 3.5. Sensitivity Analyses
After we excluded participants who developed incident T2D within the first year of follow-up and participants who developed IFGR within the 24 months of follow-up, the risks of incident T2D and IFGR were 1.16 ($95\%$ CI 0.98–1.38, $$P \leq 0.09$$) and 0.89 ($95\%$ CI 0.73–1.09, $$P \leq 0.05$$, Table s2), respectively, for the intensive BP control group. In addition, results were consistent as we excluded participants who failed to follow up or withdrew their consent (Table s3). As for the incidence of T2D and IFGR at different time points of follow-up between the two treatment groups, no significant differences were observed except for the incidence of T2D at 24 months ($5.3\%$ versus $7.2\%$, $$P \leq 0.033$$, Table s4).
## 4. Discussion
Our study is an extension of the findings revealed by Roumie et al. [ 13] who found that intensive BP control compared with standard BP control was associated with increased risk of T2D and IFG among participants with normoglycemia at randomization. Through our study, risk of T2D was also found to be increased among participants with IFG at randomization treated with intensive BP control, though it did not reach statistical significance. It is worth noting that the confidence interval for the risk of T2D was also wide in the study conducted by Roumie et al. Subgroup analysis indicated intensive BP control was associated with increased risk of T2D among participants with eGFR ≥ 60 mL/min, those with SBP ranging from 132 to 145 mm·Hg at randomization, and those with self-reported race of nonblack. As for the capability of reverting the progression of IFG, intensive BP control was not associated with a higher percentage of participants with IFGR. In contrast, it might have a negative impact on IFGR.
Recently, several large randomized controlled trials have shown us that intensive BP control to a lower SBP target can reduce the risk of cardiovascular events for hypertensive patients with high cardiovascular risk. SPRINT indicated that intensive SBP control of <120 mm·Hg can achieve impressive cardiovascular benefits in comparison to standard SBP control of <140 mm·Hg [11]. The BP measurement method of SPRINT (automated Omron-HEM-907 XL) was different from other BP control trials (office BP). Although the procedures for BP measurement in SPRINT were consistent with other trials, some questioned whether the BP readings may be misinterpreted in SPRINT due to the absence of staff in the room, resulting in lower BP values than those obtained in other trials or clinical practice. However, the benefit observed in SPRINT was similar to that in other trials. STEP (Strategy of Blood Pressure Intervention in the Elderly Hypertensive Patients) found that intensive SBP treatment of <130 mm·Hg resulted in $26\%$ lower incidence of cardiovascular events than standard SBP treatment of <150 mm·Hg in Chinese hypertensive patients [10]. Secondary analysis of participants who received standard glycemic therapy in the ACCORD (Action to Control Cardiovascular Risk in Diabetes Blood Pressure trial) identified benefits similar to those seen in SPRINT treated with intensive BP control [12, 17]. In light of these findings, the recommended SBP target has a tendency to be lower than previously recommended in many hypertension management guidelines [18–20]. Achieving a lower SBP target is administered with caution in daily practice. Concerns over adopting this intensive therapy mainly arise from its adverse events. However, the impact of intensive BP control on the metabolism of blood glucose should also be considered. There was a $19\%$ higher risk of T2D and a $17\%$ higher risk of IFG among those who received intensive BP control with normoglycemia at randomization [13]. To our surprise, there was only $18\%$ higher risk of T2D among those who received intensive BP control with IFG at randomization since IFG is commonly considered as a prediabetes condition. This observation needs to be investigated in the future. Physicians should have a thorough discussion with patients about risks and benefits of pursuing intensive BP target, especially for those with high risks of developing T2D [21].
As for the reasons why intensive BP control can have an impact on the metabolism of blood glucose is still unknown. On one hand, different classes of antihypertensive drugs have been reported to be associated with the risk of T2D [6–9, 22]. Due to the limited sample size and observational nature of previous studies, a causal relationship cannot be achieved. On the other hand, the interaction of different classes of antihypertensive drugs was also likely to be associated with the risk of T2D. In SPRINT, the use of multiple antihypertensive drugs among both the intensive and standard BP control groups makes it difficult to investigate the relationship between drug-drug interaction and the risk of T2D. A well-designed meta-analysis of individual patient data from clinical trials or a mendelian randomization study might be the promised ways for future research studies to answer the remaining questions.
## 4.1. Limitations
Some limitations should be considered when interpreting the results of our study. First, this was a secondary analysis of SPRINT, so the power was not enough to detect differences in risk of T2D between the two treatment groups, as suggested in our study power consideration. The impact may be stronger in our subgroup analysis. For future research studies, pooled individual data of several finished trials may have sufficient power to detect differences between the two groups regarding the incidence of T2D. Although baseline characteristics were well balanced between the two treatment groups, residual confounders such as insulin levels and markers of insulin resistance may have an impact on our results. Second, the definitions of IFG and IFGR were based on a single fasting glucose test at the baseline and time-updated measures of glycemic control. We cannot exclude the possibility of differing risks of T2D if IFG had been based on HbA1c levels or an oral glucose tolerance test; however, this kind of data is not available in SPRINT. To test the robustness of our findings, we excluded participants with IFGR that happened within 24 months after randomization, and results were consistent as shown in our primary analysis. Third, the definition of incident T2D included participants with self-reported T2D at the annual examination or a record of hypoglycemic medications at follow-up. Because the diagnosis was not adjudicated by a physician, this raised concerns over misreporting. Finally, as we know the median follow-up time of SPRINT was only about 3 years, the overall risk of T2D might change with a longer duration of follow-up.
## 5. Conclusion
Our study found that, in comparison to standard BP control for hypertensive patients with IFG, intensive BP control was associated with a small increased risk of new-onset T2D, though it did not reach statistical significance. This kind of impact should be considered when implementing the strategy, especially for those with high risks of developing T2D.
## Data Availability
SPRINT anonymized data are available at the National Heart, Lung and Blood Institute (NHLBI) as the biological specimen and Data Repository (https://biolincc.nhlbi.nih.gov/home/).
## Ethical Approval
This study has been registered at https://www.clinicaltrials.gov, unique identifier: NCT01206062.
## Conflicts of Interest
The authors declare that they have no conflicts of interest.
## Authors' Contributions
Beiru Lin was involved the study's concept, writing of the first draft, and revision of the manuscript. Xiaochuan Liu was involved data curation, writing of the first draft, and revision of the manuscript. Sichen Yao performed data curation and analysis. Zhigang Pan was involved the study's concept, supervision of the study, and revision of the manuscript. Beiru Lin and Xiaochuan Liu contributed equally to this work.
## References
1. Reaven G. M.. **Insulin resistance, hyperinsulinemia, hypertriglyceridemia, and hypertension. Parallels between human disease and rodent models**. (1991) **14** 195-202. DOI: 10.2337/diacare.14.3.195
2. Reaven G. M., Lithell H., Landsberg L.. **Hypertension and associated metabolic abnormalities--the role of insulin resistance and the sympathoadrenal system**. (1996) **334** 374-382. DOI: 10.1056/NEJM199602083340607
3. Reaven G. M.. **Relationship between insulin resistance and hypertension**. (1991) **14** 33-38. DOI: 10.2337/diacare.14.4.33
4. Fletcher A., Amery A., Birkenhager W.. **Risks and benefits in the trial of the European working party on high blood pressure in the elderly**. (1991) **9** 225-230. DOI: 10.1097/00004872-199103000-00005
5. Emdin C. A., Anderson S. G., Woodward M., Rahimi K.. **Usual blood pressure and risk of new-onset diabetes: evidence from 4.1 million adults and a meta-analysis of prospective studies**. (2015) **66** 1552-1562. DOI: 10.1016/j.jacc.2015.07.059
6. Elliott W. J., Meyer P. M.. **Incident diabetes in clinical trials of antihypertensive drugs: a network meta-analysis**. (2007) **369** 201-207. DOI: 10.1016/S0140-6736(07)60108-1
7. Abuissa H., Jones P. G., Marso S. P., O’Keefe J. H.. **Angiotensin-converting enzyme inhibitors or angiotensin receptor blockers for prevention of type 2 diabetes: a meta-analysis of randomized clinical trials**. (2005) **46** 821-826. DOI: 10.1016/j.jacc.2005.05.051
8. Gillespie E. L., White C. M., Kardas M., Lindberg M., Coleman C. I.. **The impact of ACE inhibitors or angiotensin II type 1 receptor blockers on the development of new-onset type 2 diabetes**. (2005) **28** 2261-2266. DOI: 10.2337/diacare.28.9.2261
9. Jandeleit-Dahm K. A., Tikellis C., Reid C. M., Johnston C. I., Cooper M. E.. **Why blockade of the renin-angiotensin system reduces the incidence of new-onset diabetes**. (2005) **23** 463-473. DOI: 10.1097/01.hjh.0000160198.05416.72
10. Zhang W., Zhang S., Deng Y.. **Trial of intensive blood-pressure control in older patients with hypertension**. (2021) **385** 1268-1279. DOI: 10.1056/NEJMoa2111437
11. The Sprint Research Group C. E., Lewis L. J.. **Final report of a trial of intensive versus standard blood-pressure control**. (2021) **384** 1921-1930. DOI: 10.1056/NEJMoa1901281
12. Beddhu S., Chertow G. M., Greene T.. **Effects of intensive systolic blood pressure lowering on cardiovascular events and mortality in patients with type 2 diabetes mellitus on standard glycemic control and in those without diabetes mellitus: reconciling results from ACCORD BP and SPRINT**. (2018) **7** p. 9326. DOI: 10.1161/JAHA.118.009326
13. Roumie C. L., Hung A. M., Russell G. B.. **Blood pressure control and the association with diabetes mellitus incidence: results from SPRINT randomized trial**. (2020) **75** 331-338. DOI: 10.1161/HYPERTENSIONAHA.118.12572
14. Ambrosius W. T., Sink K. M., Foy C. G.. **The design and rationale of a multicenter clinical trial comparing two strategies for control of systolic blood pressure: the Systolic Blood Pressure Intervention Trial (SPRINT)**. (2014) **11** 532-546. DOI: 10.1177/1740774514537404
15. The Sprint Research Group J. T., Wright J. D.. **A randomized trial of intensive versus standard blood-pressure control**. (2015) **373** 2103-2116. DOI: 10.1056/NEJMoa1511939
16. Levey A. S., Bosch J. P., Lewis J. B., Greene T., Rogers N., Roth D.. **A more accurate method to estimate glomerular filtration rate from serum creatinine: a new prediction equation**. (1999) **130** 461-470. DOI: 10.7326/0003-4819-130-6-199903160-00002
17. Buckley L. F., Dixon D. L., Wohlford G. F., Wijesinghe D. S., Baker W. L., Van Tassell B. W.. **Intensive versus standard blood pressure control in SPRINT-eligible participants of ACCORD-BP**. (2017) **40** 1733-1738. DOI: 10.2337/dc17-1366
18. **2018 Chinese guidelines for prevention and treatment of hypertension-A report of the revision committee of Chinese guidelines for prevention and treatment of hypertension**. (2019) **16** 182-241. DOI: 10.11909/j.issn.1671-5411.2019.03.014
19. Whelton P. K., Carey R. M., Aronow W. S.. **2017 ACC/AHA/AAPA/ABC/ACPM/AGS/APhA/ASH/ASPC/NMA/PCNA guideline for the prevention, detection, evaluation, and management of high blood pressure in adults A report of the American college of cardiology/American heart association task force on clinical practice guidelines**. (2018) **71** e127-e248. DOI: 10.1016/j.jacc.2017.11.006
20. Williams B., Mancia G., Spiering W.. **2018 ESC/ESH Guidelines for the management of arterial hypertension**. (2018) **39** 3021-3104. DOI: 10.1093/eurheartj/ehy339
21. Nazarzadeh M., Rahimi K.. **Intensive blood pressure lowering and risk of diabetes: friend or foe?**. (2020) **75** 293-294. DOI: 10.1161/HYPERTENSIONAHA.119.14050
22. Nazarzadeh M., Bidel Z., Canoy D.. **Blood pressure lowering and risk of new-onset type 2 diabetes: an individual participant data meta-analysis**. (2021) **398** 1803-1810. DOI: 10.1016/S0140-6736(21)01920-6
|
---
title: End to End Multitask Joint Learning Model for Osteoporosis Classification in
CT Images
authors:
- Kun Zhang
- Pengcheng Lin
- Jing Pan
- Peixia Xu
- Xuechen Qiu
- Danny Crookes
- Liang Hua
- Lin Wang
journal: Computational Intelligence and Neuroscience
year: 2023
pmcid: PMC10036193
doi: 10.1155/2023/3018320
license: CC BY 4.0
---
# End to End Multitask Joint Learning Model for Osteoporosis Classification in CT Images
## Abstract
Osteoporosis is a significant global health concern that can be difficult to detect early due to a lack of symptoms. At present, the examination of osteoporosis depends mainly on methods containing dual-energyX-ray, quantitative CT, etc., which are high costs in terms of equipment and human time. Therefore, a more efficient and economical method is urgently needed for diagnosing osteoporosis. With the development of deep learning, automatic diagnosis models for various diseases have been proposed. However, the establishment of these models generally requires images with only lesion areas, and annotating the lesion areas is time-consuming. To address this challenge, we propose a joint learning framework for osteoporosis diagnosis that combines localization, segmentation, and classification to enhance diagnostic accuracy. Our method includes a boundary heat map regression branch for thinning segmentation and a gated convolution module for adjusting context features in the classification module. We also integrate segmentation and classification features and propose a feature fusion module to adjust the weight of different levels of vertebrae. We trained our model on a self-built dataset and achieved an overall accuracy rate of $93.3\%$ for the three label categories (normal, osteopenia, and osteoporosis) in the testing datasets. The area under the curve for the normal category is 0.973; for the osteopenia category, it is 0.965; and for the osteoporosis category, it is 0.985. Our method provides a promising alternative for the diagnosis of osteoporosis at present.
## 1. Introduction
Osteoporosis (OP) is a disease characterized by impaired bone microstructure and decreased bone mineral density (BMD). With the acceleration of population aging, OP has become an increasingly serious global health problem [1]. Fragile fracture is the most serious complication of OP [2]. OP causes more than 8.9 million brittle fractures each year worldwide [3]. In the US, fragile fractures are more than four times more common than stroke, acute myocardial infarction, and breast cancer [4]. In several developed countries, osteoporotic fractures account for longer hospitalization time than these diseases according to a meeting of the World Health Organization [5]. By 2025, the number of fragility fractures is expected to increase from 3.5 million in 2010 to 4.5 million, a $28\%$ increase [6]. Therefore, reliable technology for the early detection and prevention of OP is urgently needed.
Currently, although dual-energyX-ray absorptiometry (DXA) is the gold standard for measuring bone mineral density for the diagnosis of OP, it is not widely used as a screening tool for OP owing to its high cost and limited availability of equipment [7]. To overcome these limitations, a variety of osteoporosis screening tools have emerged. Quantitative ultrasound (QUS) is one of them, which has developed into an alternative method for DXA screening of osteoporosis. Its benefits include being portable and economical; however, it may be unavailable in all primary medical settings [8]. In addition, a variety of clinical risk assessment tools have been developed to predict osteoporosis, including the fracture risk assessment tool (FRAX), the QFracture algorithm, the Garvan Fracture Risk Calculator, and the osteoporosis self-assessment tool [9]. Unfortunately, these tools are based on a combination of known risks to calculate the risk of fracture in patients and have poor efficiency.
Artificial intelligence and machine learning algorithms have recently been used in the diagnosis and prediction of osteoporosis [10]. The existing methods have achieved some success in solving the problem of binary classification (osteoporosis and nonosteoporosis) of which the main purpose is to identify whether the patient has osteoporosis [11]. However, these methods also have some obvious shortcomings: [1] the existing artificial intelligence algorithms treat segmentation and classification as two separate tasks, ignoring the information fusion and complementarity between the two tasks; [2] taking the average of two lumbar cancellous bone mineral density measurements (commonly the first and second lumbar) is widely acknowledged as the best diagnostic criterion for osteoporosis in lumbar QCT [12]. In current models, these data inputs tend to be CT images of a single vertebral body, disregarding the information fusion and complementarity between multiple vertebral images; [3] the problem of class imbalance in the collected data is prevalent due to the lack of standard public datasets; [4] most methods treat osteoporosis as a binary problem, regardless of the urgent need and a strong incentive to turn the binary into a trinomial (osteoporosis, osteopenia, and normal) problem. Although the three classifications are more difficult, osteopenia can bring some predictability to the prevention and treatment of osteoporosis. In this paper, we address the challenges above in the diagnosis of osteoporosis to facilitate the timely detection of the condition and propose an instance-based and class-based multilevel joint learning framework for bone state classification. The innovation of this method lies in the following steps. Firstly, we locate a vertebral body and remove redundant information from the image. Secondly, by constructing the boundary heat map regression auxiliary branch, the vertebral edge is refined, and the segmentation performance is improved on the segmentation branch of the shared encoder. In addition, low-level and high-level features from the segmentation branch and the auxiliary branch, including the shape and boundary of the vertebral body, are fused with feature layers from the diagnostic classifier. Finally, considering the different effects of different vertebral bodies on the classification results of bone state, we design a feature fusion module to adaptively learn feature fusion weights. The proposed method is novel because it solves the challenges of high dimensionality, multimodality, and multiclassification associated with osteoporosis diagnosis, and these challenges have not been resolved in earlier methods. The contributions of the research are as follows:A joint learning framework is proposed to segment vertebral bodies from CT images and classify bone states (normal, osteopenia, and osteoporosis)An instance-based and class-based data sampling balancing strategy is introduced to solve the problem of poor model prediction caused by imbalanced data between training datasetsA boundary heat map regression branch is proposed, which uses the Gaussian function to do “soft labeling,” accelerating network convergence and improving the performance of vertebral segmentation in joint learning and single-task learning environmentsThe effectiveness of segmentation features in guiding a deep classification network is verified by hierarchically fusing the features of the decoder and classifier related to two segmentation tasksA feature fusion module is proposed to adaptively learn the feature weights of vertebrae 1 and 2 and balance the influence of two vertebrae images on classification results To our knowledge, there are many studies [13–16] on the classification of bone status using vertebral images, but there are few studies on multitask joint learning and detection of bone status based on soft tissue window images at the central level of lumbar 1 and lumbar 2 vertebrae. Experimental results show that multitask joint learning can improve the accuracy of disease classification.
## 2. Related Works
In this section, we briefly review the related research on bone state classification, categorizing them into three subareas to introduce the current research on the bone state in the medical image, i.e., vertebral positioning, vertebral CT image segmentation, and vertebral medical CT image classification.
## 2.1. Vertebral Positioning
With the development of deep learning, convolutional neural networks are increasingly used for positioning tasks. However, most of these works describe vertebral recognition as a centroid point detection task. Chen et al. used the advanced features of convolutional neural networks to represent vertebrae from 3D CT volume and eliminated the detection of misplaced centroids based on a random forest classifier [17]. Dong et al. iterated the centroid probability map of a convolutional neural network using a message-passing scheme according to the relationship between the centroids of the vertebrae and used sparse regularization to optimize the localization results to obtain a pixel-level probability of each vertebral centroid [18]. However, it may be more meaningful to directly identify the labels and bounding boxes of vertebrae (rather than the probability map of the centroid point). Zhao et al. proposed a category-consistentself-calibration recognition system to accurately predict the bounding boxes of all vertebrae, improving the discrimination ability of vertebrae categories and the self-awareness of false positive detection [19]. All of these methods identify the vertebrae from the coronal plane, whereas what we want is to get a small image from the transverse view that only contains the vertebrae.
## 2.2. Vertebral Segmentation
Recently, machine learning is increasingly used in the recognition and segmentation of vertebral bodies. Michael Kelm et al. used iterative variants of edge-space learning to find the bounding boxes of intervertebral discs and utilized Markov-based random fields and graphical cutting to initialize and guide the segmentation of the vertebrae [20]. Zukić et al. employed the AdaBoost-based Viola–Jones object detection framework to find the bounding boxes of the vertebrae and then split them by expanding the mesh from the center of each vertebra [21]. Chu et al. applied random forest regression to detect the vertebral center and used these to define target regions for the segmentation of the vertebrae with random forest voxel classification [22]. Although these methods can find certain vertebral bodies with specific appearances, they still need to set some parameters empirically and fail to deal with complex pathological cases. However, many recent segmentation methods are based on deep learning, using convolutional neural networks instead of the traditional explicit modeling of spine shape and appearance. For example, Sekuboyina et al. used a multiclass convolutional neural network for pixel labeling, segmented the lumbar spine on a 2D facet slice, and estimated the bounding boxes of the waist region using a simple multilayer perceptron to identify regions of interest in the graph [23]. Janssens et al. depended on two continuous networks to realize this task. First, they used a regression convolutional neural network to estimate the bounding box of the lumbar region and then used a classification convolutional neural network to perform voxel labeling in the bounding box to segment the vertebral body [24]. Mushtaq et al. used ResNet-UNet to semantically segment the lost vertebral body, achieving 0.97 DSC and 0.86 IOU [25].
## 2.3. Vertebral Medical Image Classification
In the study of establishing the osteoporosis model, Yoo et al. established a support vector machine model using age, height, weight, body mass index, hypertension, hyperlipidemia, and other factors to identify osteoporosis in postmenopausal women. Compared with traditional osteoporosis self-assessment tools, they found that the support vector machine model is more accurate [26]. Pedrassani de Lira et al. established a J48 decision number model to identify osteoporosis through multiple indicators such as age, previous fracture, number of previous fractures, and previous spinal fractures [27]. Tafraouti et al. extracted features from X-ray images and used a support vector machine model to identify osteoporosis, which can well distinguish osteoporosis patients from normal people [28]. Kilic and Hosgormez studied the identification of osteoporosis based on a random subspace method and random forest ensemble model. Jang et al. used a deep learning method to identify osteoporosis [29]. In the internal and external test sets, the area under curve (AUC) of osteoporosis screening was 0.91 ($95\%$ confidence interval (CI), 0.90–0.92) and 0.88 ($95\%$ confidence interval (CI), 0.85–0.90), respectively. The experimental results illustrate that the use of chest radiographs based on deep learning models may be used for opportunistic automatic screening of osteoporosis patients in the clinical environment [30]. In the latest study, Xue et al. conducted a study in which they labeled the L1–L4 vertebral body in CT images and divided it into three categories based on bone mineral density: osteoporosis, osteopenia, and normal. The study achieved a high level of accuracy, with a prediction accuracy of $83.4\%$ and a recall rate of $90.0\%$ [31]. Dzierżak and Omiotek have developed a novel method for diagnosing osteoporosis through the use of spine CT imaging and deep convolutional neural networks. To address the issue of a small sample size, they utilized a large dataset to pretrain their model, which resulted in the successful classification of osteoporosis and normal cases. This approach showed promising results for the accurate diagnosis of osteoporosis using CT scans [32]. In these methods, both the traditional machine learning algorithm and the current popular deep learning algorithm use the image containing only the region of interest as the data source. The step-by-step preprocessing process is tedious, time-consuming, and inefficient. Therefore, the integration of positioning, segmentation, and classification into a network should help to improve efficiency, and no research has shown that explicit or implicit features related to the first $\frac{3}{4}$ of the vertebral body can be effectively and interpretably used in deep classification networks.
## 3.1. Overview
Our proposed method aims to classify vertebral images within a joint framework to enable a more flexible diagnosis of osteoporotic lesions. To achieve this goal, as shown in Figure 1, we propose an instance-based and class-based end-to-end multitask joint learning framework. It mainly has a strategy to solve class imbalance and four deep learning modules, including vertebral positioning module, vertebral segmentation module, cascade feature extraction module combined with gated attention, and feature fusion module. As shown in Figure 2, a new multilayer and multilevel joint learning framework is introduced, which integrates positioning, segmentation, and classification. Firstly, realizing the accurate location of the target lesion (coronal vertebral body), removing the redundant information of the image through the reduction of resolution (from 512 × 512 to 224 × 224). Secondly, the boundary heat map auxiliary branch is employed to refine the edge to improve the performance of segmentation; meanwhile, segmentation features are cascaded with the classification features to improve the accuracy of classification. Finally, we propose a feature fusion module, which adaptively assigns feature weights to fuse the features of lumbar L1 and lumbar L2. Different magnitudes of losses in multitask learning tend to bring about negative effects on other tasks when the model tends to fit a certain task; to balance this problem, we use the gradient update method to assign weights to each loss, exploiting neural networks to update the weight parameters.
## 3.2. Instance and Class-Based Sampling Methods
In the actual clinical scene, the data collected by image acquisition will be unbalanced owing to the inherent difficulty of collecting labels of rare diseases or other unusual cases. Therefore, when training on extremely unbalanced data, the model may have a high probability of being affected by the number of different categories, resulting in the underfitting of some categories which may be ignored. At present, the methods to solve the data imbalance include data resampling [33], adaptive loss function [34], and curriculum learning [35]. Inspired by the paper [36, 37], methods are introduced to solve the problem of extreme imbalance of our category images. It combines unbalanced (instance-based) and balanced (class-based) sampling of data, where we extend the method to our three-category practical problems.
We define the training set as D={(xi, yi), $i = 1$,2,…, N}, where xi is the sample, yi is the sample category. Assuming that for multiclassification problems with K categories, each category has Mk samples, and N represents the total number of samples, where ∑$K = 1$KMk=N, the general sampling strategies can be described as[1]pj=Mjn∑$k = 1$kMjn,where pj is the probability of sampling from the j th category. If we set $$n = 0$$, the probability of sampling from each category is equal to 1/K. This is the class-based sampling method.
If we set $$n = 1$$, then it is equivalent to selecting the sample by the proportion of a category of samples to all samples, which is instance-based sampling. Here, we introduce a mixed sampling method based on instance and class, which is suitable for data imbalance. We denote the training dataset and sampling strategy by the symbol (D, S). Instance-based sampling and class-based sampling are represented by SI and SC, respectively, so this mixture can be described as[2]x∧=λxI+1−λxC,y∧=λyI+1−λyC,where λ ~ beta(α, β), α > 0, β > 0, λ ∈ [0,1], (xI, yI) ∈ (D, SI), (xC, yC) ∈ (D, SC). x∧ and y∧ represent random convex combinations of data and label inputs. Here, we set β=1. As shown in Figure 3, as α grows, examples from minority classes are combined with a greater weight to avoid overfitting of minority classes. Here, we set α=0.1 to induce a more balanced distribution of training samples by creating synthetic data points around spatial regions where minority classes provide fewer data density.
## 3.3. Vertebral Positioning Module Based on YOLOv3
The basic step of vertebral CT image classification is to extract robust features from CT images, given W and H of the original images are 512 pixels. To remove redundant features, we use the YOLOv3 [38] to locate the vertebral body in the image with size 512 × 512 × 3 as input to YOLOv3. The image feature is extracted by DarkNet-53, and then the target classification and position regression are performed on the acquired feature map with the help of the FPNs (feature pyramid networks) structure.
In this study, we will obtain the position of the prediction box in the original image px, py, pw, ph, in YOLOv3, a set of anchor frames is composed of nine initial frames of different sizes. Assuming that the center coordinates, width, and height of an anchor frame are expressed as ax, ay, aw, ah, px, py, pw, ph can be obtained by reverse calculation of the regression parameter tx, ty, tw, th by the output network. Details of the calculation formula are as follows:[3]px=σtx+ax,py=σty+ay,pw=awetw,ph=aheth,where σ(·) represents the sigmoid transformation of the variable, aiming at controlling the offset of the center point between 0 and 1.
The main purpose of employing YOLOv3 is to obtain the center coordinates px and py of the prediction box and utilize this position as the center cutting position of the vertebral body to obtain a 224 × 224 image containing the complete vertebral body as the input of the subsequent convolution module. In this way, we can remove tens of thousands of useless features and improve the efficiency of the model.
## 3.4. Boundary Regression Auxiliary Branches
We suggest dividing the segmentation task into two tasks: vertebral segmentation and contour determination. Thus, our network is mainly composed of a weight-sharing encoder and two decoders composed of the segmentation branch and boundary regression branch. In the encoder, we improve the original U-Net [39] by applying residual blocks to replace the original two effective convolutions. In the decoder stage, we cascade the penultimate features from the boundary regression branch with the penultimate features of the segmentation branch, helping the network to better perceive and refine the vertebral contour. Since vertebrae in CT images may show up hyperosteogeny or other conditions, it is necessary to reconstruct edges by constructing auxiliary tasks, which provide more explicit and implicit topological priors for the coding layer and enable them to assist with the segmentation branches to obtain more accurate target masks.
The problem of boundary inaccuracy is rooted in the similarity of information in the corresponding receptive field of pixels. When similar features belong to the interior or exterior of the segmented region, this similarity will be advantageous, inversely similar information lies in the segmented boundary will undoubtedly increase the uncertainty of the edge. In terms of the boundary regression auxiliary branch in the segmentation module, we propose to divide the edge based on the region and graph from the whole image, combining it with the spatial proximity and pixel value similarity. In this paper, the accurate boundary of vertebral segmentation should be the inner boundary. We combine the convolutional neural network with the level set, taking the segmentation result obtained by the neural network as the prior knowledge of level set segmentation; then we construct a gray level constraint term on the original level set function and improve the edge indicator function to deal with uneven intensity in the image.
## 3.4.1. Improve the Edge Indicator Function
Getreuer [40] proposed the famous Chan–Vese (CV) model in 2001. This method uses a region-based segmentation strategy to divide the image into two homogeneous regions, the inner and outer regions, using active contoured lines to find the image to be segmented and the original image with the minimum difference to minimize the energy function.
Given the input image I(x, y), the energy function based on the CV is shown as follows:[4]EC,C1,C2=μ∫Ωgδϕ∇ϕdx dy+υ∫ΩgH−ϕdx dy+λ1∫Ω1I−C12dx dy+λ2∫Ω2I−C22dx dy,where C1 and C2 describe the average gray levels of equivalent parts inside and outside the contour, respectively, Ω1 and Ω2 represent the inner and outer regions of the contour, λ1, λ2, μ, υ are constants, g=($\frac{1}{1}$+|∇G(x, y, σ)∗I(x, y)|) is the edge indicator function which can be used to prevent the curve from exceeding the target area, G is the Gaussian calculation sub, σ is the standard deviation, and δ and H represent Dirac and Heaviside functions, respectively. The position of contour C and unknowns C1(ϕ) and C2(ϕ) are finally obtained through optimization formula [4].
The evolution of the CV model is constrained by global gray-level information. However, most images, especially medical images, have uneven intensity. To solve this problem, we improve the function g and construct gray-level information constraint terms to constrain the evolution direction. Bilateral filtering is a method that combines the spatial proximity of images with the similarity of pixel values. Based on Gaussian filtering, bilateral filtering introduces the gray value of pixels for the local weighted average. When smoothing the speckle noise of images, bilateral filtering can better maintain the edge features.
In the first step, the Gaussian function Gsr(x, y, σ) is used to construct bilateral filters to obtain smooth images:[5]Gsrx,y,σ=Gσs∗Gσr,Gσs=e−x−k2+y−l$\frac{2}{2}$σs2,Gσr=e−Ix,y−fk,l$\frac{2}{2}$σr2.
Image I(x, y) is filtered using bilateral filter operator g(x, y)=Gsr(x, y, σ)·I(x, y), where σr is the standard deviation used to control the smoothness, i, j, k, l are the weight coefficients.
In the second step, the optimal threshold T is calculated based on the filtered image using the adaptive threshold principle. The maximized interclass variance value of T is shown in the following equation:[6]ν2=w0×w1×u0−u1,where w0 represents the ratio of pixels in the target area to the image, u0 represents the corresponding average gray level, w1 is the proportion of background pixels, and u1 is the average gray level of background pixels. Then, the new edge indicator function gr can be described as[7]gT=11+ν2∇Gsrx,y,σ∗Ix,y.
## 3.4.2. Auxiliary Branch
We advocate the segmentation results of convolutional neural networks as prior knowledge, namely, the initial contour of the level set, and the curve contour evolved through the level set is used to guide the neural network to optimize toward the edge of the vertebral body.
The specific expression of the gray level constraint Q is described as[8]Q=γ1+Γ2−1−Γ2Hϕ,Γ=−1,I∈Ilow,Ihigh,1,I∉Ilow,Ihigh,Ilow=η−w·σ,Ihigh=η+w·σ,where *Ihigh is* the upper limit of the vertebral gray value obtained by using the convolutional neural network model, *Ilow is* the lower limit of vertebral gray value, σ is the average of vertebral gray value, η is the variance of vertebral gray value, and w is a constant.
The function of the gray level information constraint term is to make the level set curve evolve inside the vertebral body to approximate the inner edge contour. When the gray value of the pixel is within the upper and lower limits of the initial vertebral gray value, the energy value of the point is negative, otherwise positive. The edge result obtained by the neural network is used to replace x and y on the initial contour plane. Gradient descent is used to minimize the energy function, and the formula form of the final evolution equation after adding the gray constraint function is shown as follows:[9]∂ϕ∂t=δϕμdivgT∇ϕ∇ϕ−gTυ−λ1Ix,y−C12+γ1+Γ2−1−Γ2+λ2Ix,y−C22,C1ϕ=∫ΩIx,yHϕdx dy∫ΩHϕdx dy,C2ϕ=∫ΩIx,y1−Hϕdx dy∫Ω1−Hϕdx dy,ϕ0=ϕ0,Ix,y.
In the label aspect of the auxiliary branch, we use the Canny operator to detect the edge of the binary image label. Canny is built on a two-dimensional convolution. To improve the calculation speed of the Canny operator, two-dimensional convolution can be decomposed into one-dimensional filters, and then a convolution operation with the image A(x, y) is carried out, respectively: Ex=(∂G/∂x)·A(x, y), Ey=(∂G/∂y)·A(x, y). Then, the gradient amplitude A(x, y) and gradient a(x, y) direction can be expressed as[10]Ax,y=Ex2x,y+Ey2x,y,ax,y=arctanEyx,yExx,y.
The size of the Gaussian window is adjusted by changing the standard deviation σ of the Gaussian function, that is Ax,y=maxEx2+Ey2. We first apply nonmaximum suppression, and then segment images through the dual-threshold method. When the gradient of some pixel is greater than the limit threshold, it will be considered as an edge pixel.
Then, we construct a soft label heat map in the form of Heatsum based on the processed images:[11]HeatsumGx1,y1,σ,Gx2,y2,σ=1−1−Gx1,y1,σ1−Gx2,y2,σ,Gbd=Gaussheat∂G,=HeatsumGx1,y1,σ,…Gxn,yn,σ,∀Gxn,yn,σ∈∂G,where ○ represents the Hadamard product; it is noted that *Gbd is* normalized between [0, 1].
Here, the boundary regression branch is utilized to refine the segmented edges. We treat this branch as a regression task through mean square error rather than a whole work consisting of a boundary segmentation task together with the segmentation branch.
## 3.5. Cascading Classification Module
In the classification module, we use ResNet-101 as a basic feature extractor. ResNet [41] is a traditional deep convolutional neural network where the residual structure is used in the shallow network. The corresponding structure is illustrated in Figure 4(b). By adding the input value x with the output unit, the residual gains better performance in convergence after the operation of ReLU active. These steps can be approximated as an identical mapping of equal input and output, which effectively solves the problems of network learning ability decline, gradient disappearance, and gradient explosion when the number of convolutional neural network layers increases.
Inspired by the gating attention [42] and residual structure, we designed a gating residual module as shown in Figure 4 to replace the first convolution module in ResNet-101 from conv2_x to conv5_x. The specific network parameters can be found in Figure 5. The gated residual model can be described as follows.
Assuming that x ∈ ℝC×H×W is the activation feature of the convolutional neural network, where H and W are the height and width of the image, and C is the number of channels of the image, in general, the gating attention performs the following transformation.[12]x∧=Fxα,γ,β,α,γ,β∈RC.
Among them, a, β, and γ are trainable parameters. The embedding weight a is mainly responsible for adjusting the embedding output, and the gating weight γ and the bias weight β are responsible for adjusting the gating activation.
They determine the behavior of gated attention in each channel.
For the specific process, assuming the given embedding weight as α=[α1, α2 …, αc], modules can be defined as[13]sc=αcxc2=αc∑$i = 1$H∑$j = 1$Wxci,j2+∈$\frac{1}{2}$,[14]sc∧=CscS2=Csc∑$c = 1$Csc2+n$\frac{1}{2}$,where ∈ is a small constant, which is mainly used to avoid the derivation of zeros. Equation [14] is used to normalize channels, and n represents a small constant. C is used for normalization the ratio of sc, preventing the condition of small sc when C is too large, αc is a trainable parameter used for controlling the weight of each channel. When αc is close to 0, the channel will not participate in channel normalization.
Then, we suppose the selection weight γ=[γ1, γ2 …, γc] and the gating offset β=[β1, β2 …, βc], the gating function can be depicted as follows:[15]xc∧=xc1+tanh γcsc∧+βc.
Each primitive channel xc is adapted by the corresponding gate, γ and β are trainable weights and deviations which is used to control the activation of the gate. Finally,xc∧=x1∧,x2∧…,xc∧ will be entered into the residual module to obtain the feature map y=[y1, y2 …, yc] of the gating attention after the convolution operation. Supposing the feature map concatenated from the segmentation module S=[S1, S2 …, Sc], we can perform the following operations on the classification network feature y=[y1, y2 …, yc] and the segmentation module feature to obtain the final feature map yc∧.[16]yc∧=Conv1×1Concatyc,Sci=1,2…c.
Two 1 × 2048-dimensional feature vectors of vertebrae can be obtained by flattening the feature map.
## 3.6. Feature Fusion Module
As mentioned above, the detection of bone status is based on the average of lumbar L1 and lumbar L2. To explain the different effects of different lumbar vertebrae on classification, we learn W1 and W2 adaptively for each vertebra, which satisfies W1+W2=1; W1 and W2 represent the fusion weights, respectively.[17]Xfuse=ConcatW1×X1,W2×X2.
Specifically, we calculate W1 and W2 (W1+W2=1) by Ffuse(X1) and Ffuse(X2), respectively, where F represents the perception of two layers, that is, two fully connected layers. The following softmax layer can be used to eliminate the influence of different feature dimensions. After gaining the feature Xfuse, the prediction of bone state P(M|IN) can be given by the fully connected layer and softmax function.[18]PMIN=softmaxfcXfuse,num−classes.
## 3.7. Cascading Classification Models
To balance the impact of different dimensions of multiple tasks in the training process we introduce the trade-off parameters λ1, λ2, λ3, λ4 and λ5 to balance these four tasks. The total loss function of multitask learning can be defined as[19]Lmul=λ1LIOC+λ2Lcla+λ3Lconf+λ4Lseg+λ5Lseg+λ6Lcla=λ1LIOCpiI,t+λ2Lsobjpic1,qcla1+λ3Lobjpic2,q+λ4LdiceSpis,G+λ5Lmsepib,Gbdn+λ6Lcrossentropypic3,qcla2,where piI, pic2, pis, pis, pib, pic3, respectively, represent the predicted results of the positioning branch, category branch, confidence branch, and segmentation branch of the positioning module for a given input image, the boundary heatmap regression branch, and the classification network. S represents the Sigmoid function, t represents the prediction box result, and qcla1 is the result of the category in the positioning module. q represents the probability that a vertebral body exists, Gbdn represents the normalized result of Gbd, and qcla2 is the expected result of the classification network.
## 4.1. Dataset and Preprocessing
To assess the effectiveness and benefit of the joint learning framework in bone state classification, we conducted experiments in a dataset obtained from the Nantong First People's Hospital from May 2021 to May 2022, consisting of CT images of 1048 routine-dose cases. All images were collected by Ingenuity Core 128 CT (Philips Health Care, Holland), the tube voltage was 120 kV, the inpatient tube current modulation technique was used, and the iDose 4 was used to reconstruct the cross-sectional image of the mediastinal window (standard B standard reconstruction algorithm). The reconstruction layer thickness and layer interval were both 2 mm. The longitudinal window images of the lumbar 1 and lumbar 2 center planes of each subject were selected for BMD measurement and deep learning model construction. The QCT pro4 software (Mindways, CA, USA) was used to set the same size of the region of interest (ROI) in the central cancellous bone area of the lumbar 1 and lumbar 2 vertebral bodies, avoiding the cortical bone and the visible vascular area. The software automatically calculated the BMD values of the lumbar 1 and lumbar 2 vertebral bodies and used their mean values as the BMD values of the individual subjects (BMD individuals). According to the standard recommended by the “expert consensus on imaging and bone mineral density diagnosis of osteoporosis” BMD individuals > 120 mg/cm3 are normal bone mass, 80 mg/cm3 ≤ BMD individuals ≤ 120 mg/cm3 are osteopenia, and BMD individuals < 80 mg/cm3 are osteoporosis.
We divide the dataset into training data ($50\%$), validation data ($10\%$), and test data ($40\%$); the class distribution of training, validation, and testing datasets is shown in Figure 6. These three datasets do not have any overlapping images, and the CT images of each category in the three datasets are placed in strict proportions. Then, all images are resized to 512 × 512 and each image is normalized from [0, 255] to [0, 1] before being fed into the network.
To increase the amount of training data and improve the generalization ability and robustness of the model, we enhance the image data employing flipping, rotating, and scaling on the basis of the original data balancing strategy based on an instance and actual class.
## 4.2. Implementation of Framework
To implement the joint learning framework, we implemented the model based on Python 3.6.12, using the PyTorch framework and two NVIDIA GeForce 3090Ti GPUs. We apply the SGD optimizer to train the joint learning framework for 300 epochs with a learning rate of (10e − 1–10e − 5) and add six adaptive parameters to the SGD optimizer to weigh the loss of multitask learning.
## 4.3.1. Measurements
Based on previous work[49–52], accuracy, sensitivity, specificity, and F1-score were used to evaluate the performance of classification. The accuracy rate is the ratio of the number of samples correctly classified by the classifier to the total number of samples. The sensitivity reflects the proportion of positive cases correctly judged by the classifier to the total positive samples. The specificity indicates the proportion of negative cases correctly judged by the classifier to the total negative samples. F1-score is the sum of accuracy and sensitivity. In this paper, the three-category problem is transformed into a two-category problem to evaluate; that is, the category studied at this time is a positive sample and the other categories are negative samples. Accuracy: Acc = (TP + FN/TP + TN + FP + FN)Sensitivity: Pre = (TP/TP + FN)Specificity: Spe = (TN/FP + TN)F1-score: F1 = (2 × P × R/P + R), P = (TP/TP + FP), R = (TP/TP + FN) P denotes the model prediction and T denotes the true label. Positive samples are predicted as positive samples (true positive, TP), positive samples are predicted as negative samples (false negative, FN), negative samples are predicted as positive samples (false positive, FP), and negative samples are predicted as negative samples (true negative, TN).
Based on previous works [53–55], we use the intersection over union (IOU) and dice coefficient (Dice) to evaluate the effectiveness of our model segmentation task and use the average precision (AP) to evaluate the effectiveness of the positioning task. Intersection over union: IOU=(TP/TP+FP+FN)Dice coefficient: Dice=(2TP/2TP+FP+FN)
## 4.3.2. Baselines
To demonstrate the performance of our federated framework model, we compared our work with popular machine learning and deep learning methods, including AlexNet [43], VGG-19 [44], GoogLeNet [45], ResNet [41], DenseNet [46], ShuffleNet [47], and EfficientNet [48].
## 4.4. Results
We use ten-foldcross-validation to calculate the average results and show the performance of the joint framework in Table 1. We set the learning rate of 10e − 1–10e − 5 to evaluate the classification performance of the joint framework in different situations. We used normal (osteopenia and osteoporosis) as a positive sample and other categories as negative samples, achieving an accuracy of 0.971, a sensitivity of 0.964, a specificity of 0.976, and an F1-score of 0.964. We achieved 0.933 in accuracy, 0.970 in sensitivity, 0.836 in specificity, and 0.954 F1-score when osteopenia was used as a positive sample and other categories (normal and osteoporosis) as a negative sample. When we used osteoporosis as a positive sample and other categories (normal and osteopenia) as negative samples, we achieved an accuracy of 0.957, a sensitivity of 0.962, a specificity of 0.922, and an F1-score of 0.975. The best performance is obtained by the learning rate of 10e − 3, indicating that the classification problem of bone state CT images can be effectively solved by adjusting the hyperparameters.
In addition, we compare the best results of joint learning with the most advanced baselines. The comparison results are reported in Table 2, where the best comparable performance is represented in bold. For the input images of other classification methods, we use CT images (512 × 512) generated by labels manually drawn by physicians that contain only regions of interest. To better intuitively compare the classification performance of the model, we use the confusion matrix for visual analysis. As shown in Figure 7, joint learning in dealing with the task of identifying low-dose achieves good performance with only 5 cases misclassified as normal, 2 cases misclassified as osteoporosis, and 8 cases misclassified as low doses; in the task of identifying osteoporosis, only 3 cases were misclassified as low dose. This result fully indicates the nonexistence of overfitting and underfitting states; this result further illustrates that there is no bias to a certain category which increases accuracy results.
The histogram of accuracy and F1-score can be found in Figure 8. Intuitively, the accuracy rate has increased. Compared with the highest accuracy rate among advanced baseline methods, the accuracy rate of joint learning has increased by $6.2\%$ in the osteopenia category, $3.3\%$ in the normal category, and $0.1\%$ in the osteoporosis category. Notably, when compared to the overall accuracy of advanced baseline methods, the overall accuracy of joint learning was improved by $3.8\%$ which proved the effectiveness of joint learning strategies once again.
## 4.5.1. Roc Curve
To better demonstrate the classification ability of our proposed joint learning framework, we use the operating characteristic curve (ROC) and the area under curve (AUC) of receivers as further evaluation indicators. Taking the experimental results with a learning rate of 0.01 as an example, we draw the ROC curves of three categories in Figure 9, AUC for each category is also depicted in the figure. It can be found that the AUC in the osteopenia state is 0.965, the AUC value in the Normal state is 0.973, and the AUC value in the osteoporosis state is 0.985. These values prove the effectiveness of joint learning in bone CT image classification tasks.
## 4.5.2. Training Convergence
For model training, we use the accuracy and loss curve and the training process to imply the training trend of accuracy and model cost. The accuracy and loss curves of the joint learning framework with a learning rate of 0.01 are shown in Figure 10, which reflects that the model's performance achieved satisfactory results at the 150th epoch and became stable. These two curves show the convergence of the model and assess its stability in bone CT image classification. In addition, the total training time of the joint learning framework on our dataset is about 10 hours, and each epoch takes 2 minutes. In short, training convergence and time reveal the computational efficiency of our network.
## 4.5.3. Model Visualization
We further use gradient weighted class activation mapping (Grad-CAM) to visualize the decision information of the feature extraction module. Figure 11 shows that the feature extraction modules for different categories (normal, osteopenia, and osteoporosis) focus on different regions, and the model automatically focuses on the corresponding regions. Compared with the correctly classified decision information, we also list some cases of misclassification in Figure 12. The focus area of the wrong case has changed significantly compared with the correct case in Figure 11, which may be used as an explanation for the neural network decision error.
Meanwhile, we calculated that the AP value of all testing datasets in the positioning task is $95\%$, the average IOU in the segmentation task is 0.972 ± 0.125, and the average *Dice is* 0.983 ± 0.036, which shows that we have good efficiency in selecting features in the positioning and segmentation tasks, but in some cases, these features have no good effect on classification.
## 4.5.4. Ablation Experiments
In this section, we conduct an ablation study (learning rate is 10e − 3) of our method to prove the effective impact of segmentation feature and classification feature layered fusion (LF), gated convolution (GC) module, and feature fusion module (FF).
We use the three modules separately and combine them randomly and calculate the overall accuracy of each experiment to evaluate whether the model is improved. The quantitative result can be found in Table 3. In Figure 13, it can be clearly seen that the accuracy of the model has been greatly improved. When we calculate without using the method of three modules; it is unfortunate to find that the accuracy of the model is only $82.1\%$. However, when we perform a hierarchical fusion of segmentation features and classification features, the overall accuracy rate rises to $85.6\%$, an increase of $3.5\%$. When we use the gated convolution module, we find that the accuracy rate has increased by $2.8\%$. When we use feature fusion of vertebral bodies at different levels, the overall accuracy rate has increased by $3.3\%$. When we select any two of them, we find that the overall accuracy rate has increased by $4\%$, $8.1\%$, and $10.5\%$, respectively. The seven additional experiments prove the feasibility and effectiveness of our proposed modular methods in improving classification accuracy.
## 5. Conclusion
Machine learning can help a great deal in accurately identifying osteoporosis from CT images. In this study, we propose a joint learning framework for bone state detection, where we integrate positioning, segmentation, and classification into an end-to-end multitask joint learning framework. The framework processes from the original input to the final output, increasing the overall fit of the model. The accuracy of classification has been improved by modular task fusion, global feature association, and fusion of different vertebral features. We used a CT image database containing three categories of vertebrae to evaluate this method. A large number of experiments confirm this method improves the overall accuracy from $82.1\%$ to $93.3\%$, which shows the effectiveness of joint learning in bone state image classification and contributes to solving the problem of clinical diagnosis of osteoporosis.
## Data Availability
The data used to support the study are included within the article.
## Ethical Approval
This retrospective study was approved by the Ethics Committee of Nantong First People's Hospital (No.: 2021KT028), who waived the need for informed consent. The study protocol was implemented according to the Good Clinical Practice guidelines defined by the Helsinki Declaration and the International Conference on Harmonisation (ICH).
## Conflicts of Interest
The authors declare that they have no conflicts of interest.
## References
1. Lorentzon M., Cummings S. R.. **Osteoporosis: the evolution of a diagnosis**. (2015.0) **277** 650-661. DOI: 10.1111/joim.12369
2. Ebeling P. R., Nguyen H. H., Aleksova J., Vincent A. J., Wong P., Milat F.. **Secondary osteoporosis**. (2022.0) **43** 240-313. DOI: 10.1210/endrev/bnab028
3. **Total hip arthroplasty or hemiarthroplasty for hip fracture**. (2019.0) **381** 2199-2208. DOI: 10.1056/nejmoa1906190
4. Gharib H., Papini E., Paschke R.. **American association of clinical endocrinologists, associazione medici endocrinologi, and European thyroid association medical guidelines for clinical practice for the diagnosis and management of thyroid nodules: executive summary of recommendations**. (2010.0) **33** 287-291. DOI: 10.1007/bf03346587
5. Piccoli A., Cannata F., Strollo R.. **Sclerostin regulation, microarchitecture, and advanced glycation end‐products in the bone of elderly women with type 2 diabetes**. (2020.0) **35** 2415-2422. DOI: 10.1002/jbmr.4153
6. Beaudart C., Lengelé L., Leclercq V.. **Symptomatic efficacy of pharmacological treatments for knee osteoarthritis: a systematic review and a network meta-analysis with a 6-month time horizon**. (2020.0) **80** 1947-1959. DOI: 10.1007/s40265-020-01423-8
7. Kimball J. S., Johnson J. P., Carlson D. A.. **Oxidative stress and osteoporosis**. (2021.0) **103** 1451-1461. DOI: 10.2106/jbjs.20.00989
8. Carey J. J., Chih-Hsing Wu P., Bergin D.. **Risk assessment tools for osteoporosis and fractures in 2022**. (2022.0) **36**. DOI: 10.1016/j.berh.2022.101775
9. Chavda S., Chavda B., Dube R.. **Osteoporosis screening and fracture risk assessment tool: its scope and role in general clinical practice**. (2022.0) **14**. DOI: 10.7759/cureus.26518
10. Yamamoto N., Sukegawa S., Kitamura A.. **Deep learning for osteoporosis classification using hip radiographs and patient clinical covariates**. (2020.0) **10** p. 1534. DOI: 10.3390/biom10111534
11. Jang R., Choi J. H., Kim N., Chang J. S., Yoon P. W., Kim C. H.. **Prediction of osteoporosis from simple hip radiography using deep learning algorithm**. (2021.0) **11** 19997-19999. DOI: 10.1038/s41598-021-99549-6
12. Cheng X., Zhao K., Zha X.. **Opportunistic screening using low‐dose CT and the prevalence of osteoporosis in China: a nationwide, multicenter study**. (2021.0) **36** 427-435. DOI: 10.1002/jbmr.4187
13. Zhang B., Yu K., Ning Z.. **Deep learning of lumbar spine X-ray for osteopenia and osteoporosis screening: a multicenter retrospective cohort study**. (2020.0) **140**. DOI: 10.1016/j.bone.2020.115561
14. Tecle N., Teitel J., Morris M. R., Sani N., Mitten D., Hammert W. C.. **Convolutional neural network for second metacarpal radiographic osteoporosis screening**. (2020.0) **45** 175-181. DOI: 10.1016/j.jhsa.2019.11.019
15. Areeckal A. S., Jayasheelan N., Kamath J., Zawadynski S., Kocher M., David S.. **Early diagnosis of osteoporosis using radiogrammetry and texture analysis from hand and wrist radiographs in Indian population**. (2018.0) **29** 665-673. DOI: 10.1007/s00198-017-4328-1
16. Smets J., Shevroja E., Hugle T., Leslie W. D., Hans D.. **Machine learning solutions for osteoporosis—a review**. (2021.0) **36** 833-851. DOI: 10.1002/jbmr.4292
17. Chen H., Shen C., Qin J.. **Automatic localization and identification of vertebrae in spine CT via a joint learning model with deep neural networks**. (2015.0) 515-522
18. Dong Y., Xiong T., Xu D.. **Automatic vertebra labeling in large-scale 3D CT using deep image-to-image network with message passing and sparsity regularization**. (2017.0) 633-644
19. Zhao S., Wu X., Chen B., Li S.. **Automatic vertebrae recognition from arbitrary spine MRI images by a category-Consistentself-calibration detection framework**. (2021.0) **67**. DOI: 10.1016/j.media.2020.101826
20. Michael Kelm B., Wels M., Kevin Zhou S.. **Spine detection in CT and MR using iterated marginal space learning**. (2013.0) **17** 1283-1292. DOI: 10.1016/j.media.2012.09.007
21. Zukić D., Vlasák A., Egger J., Hořínek D., Nimsky C., Kolb A.. **Robust detection and segmentation for diagnosis of vertebral diseases using routine MR images**. (2014.0) **33** 190-204. DOI: 10.1111/cgf.12343
22. Chu C., Belavy D. L., Armbrecht G., Bansmann M., Felsenberg D., Zheng G.. **Fully automatic localization and segmentation of 3D vertebral bodies from CT/MR images via a learning-based method**. (2015.0) **10**. DOI: 10.1371/journal.pone.0143327
23. Sekuboyina A., Jan K., Kirschke J. S., Menze B. H., Alexander V.. **Attention-driven deep learning for pathological spine segmentation**. (2017.0) 108-119
24. Janssens R., Zeng G., Zheng G.. **Fully automatic segmentation of lumbar vertebrae from CT images using cascaded 3D fully convolutional networks**. 893-897
25. Mushtaq M., Akram M. U., Alghamdi N. S., Fatima J., Masood R. F.. **Localization and edge-based segmentation of lumbar spine vertebrae to identify the deformities using deep learning models**. (2022.0) **22** p. 1547. DOI: 10.3390/s22041547
26. Yoo T. K., Kim S. K., Kim D. W.. **Risk prediction of femoral neck osteoporosis using machine learning and conventional methods**. (2013.0) 181-188
27. Pedrassani de Lira C., Toniazzo de Abreu L. L., Veiga Silva A. C.. **Use of data mining to predict the risk factors associated with osteoporosis and osteopenia in women**. (2016.0) **34** 369-375. DOI: 10.1097/cin.0000000000000253
28. Tafraouti A., El Hassouni M., Toumi H., Lespessailles E., Jennane R.. **Osteoporosis diagnosis using fractal analysis and support vector machine**. 73-77
29. Kilic N., Hosgormez E.. **Automatic estimation of osteoporotic fracture cases by using ensemble learning approaches**. (2016.0) **40** 61-10. DOI: 10.1007/s10916-015-0413-1
30. Jang M., Kim M., Bae S. J., Lee S. H., Koh J. M., Kim N.. **Opportunistic osteoporosis screening using chest radiographs with deep learning: development and external validation with a cohort dataset**. (2022.0) **37** 369-377. DOI: 10.1002/jbmr.4477
31. Xue L., Hou Y., Wang S.. **A dual-selective channel attention network for osteoporosis prediction in computed tomography images of lumbar spine**. (2022.0) **1** 30-39. DOI: 10.56578/ataiml010105
32. Dzierżak R., Omiotek Z.. **Application of deep convolutional neural networks in the diagnosis of osteoporosis**. (2022.0) **22** p. 8189. DOI: 10.3390/s22218189
33. Buda M., Maki A., Mazurowski M. A.. **A systematic study of the class imbalance problem in convolutional neural networks**. (2018.0) **106** 249-259. DOI: 10.1016/j.neunet.2018.07.011
34. Cui Y., Jia M., Lin T.-Y., Song Y., Belongie S.. **Class-balanced loss based on effective number of samples**. 9268-9277
35. Jiménez-Sánchez A., Mateus D., Kirchhoff S.. **Medical-based deep curriculum learning for improved fracture classification**. (2019.0) 694-702
36. Galdran A., Carneiro G., Miguel A., Ballester G.. **Balanced-mixup for highly imbalanced medical image classification**. (2021.0) 323-333
37. Lei K., Mardani M., Pauly J. M., Vasanawala S. S.. **Wasserstein GANs for MR imaging: from paired to unpaired training**. (2021.0) **40** 105-115. DOI: 10.1109/tmi.2020.3022968
38. Tan M., Pang R., Quoc V., Efficientdet Le.. **Scalable and efficient object detection**. 10781-10790
39. Shorten C., Khoshgoftaar T. M.. **A survey on image data augmentation for deep learning**. (2019.0) **6** 60-48. DOI: 10.1186/s40537-019-0197-0
40. Getreuer P.. **Chan-vese segmentation**. (2012.0) **2** 214-224. DOI: 10.5201/ipol.2012.g-cv
41. He K., Zhang X., Ren S., Sun J.. **Deep residual learning for image recognition**. 770-778
42. Yang Z., Zhu L., Wu Y., Yang Y.. **Gated channel transformation for visual recognition**. 11794-11803
43. Krizhevsky A., Sutskever I., Hinton G. E.. **Imagenet classification with deep convolutional neural networks**. (2012.0) **25**
44. Simonyan K., Zisserman A.. **Very deep convolutional networks for large-scale image recognition**. (2014.0)
45. Szegedy C., Liu W., Jia Y.. **Going deeper with convolutions**. 1-9
46. Huang G., Liu Z., van der Maaten L., Kilian Q.. **Weinberger. Densely connected convolutional networks**. 4700-4708
47. Zhang X., Zhou X., Lin M., Sun J.. **Shufflenet: an extremely efficient convolutional neural network for mobile devices**. 6848-6856
48. Tan M., Le Q.. **Efficientnet: rethinking model scaling for convolutional neural networks**. 6105-6114
49. Tang C., Zhang W., Li H.. **CNN-based qualitative detection of bone mineral density via diagnostic CT slices for osteoporosis screening**. (2021.0) **32** 971-979. DOI: 10.1007/s00198-020-05673-w
50. Wei Z., Yuhan P., Jianhang J.. **RMSDSC‐Net: a robust multiscale feature extraction with depthwise separable convolution network for optic disc and cup segmentation**. (2022.0) **37** 11482-11505. DOI: 10.1002/int.23051
51. Wang W., Zhou W., Ji J.. **Deep sparse autoencoder integrated with three‐stage framework for glaucoma diagnosis**. (2022.0) **37** 7944-7967. DOI: 10.1002/int.22911
52. Lei K., Syed A. B., Zhu X., Pauly J. M., Vasanawala S. S.. **Artifact-andcontent-specific quality assessment for MRI with image rulers**. (2022.0) **77**. DOI: 10.1016/j.media.2021.102344
53. Yi Y., Guo C., Hu Y., Zhou W., Wang W.. **BCR-UNet: Bi-directional ConvLSTM residual U-Net for retinal blood vessel segmentation**. (2022.0) **10** 1056226-1056313. DOI: 10.3389/fpubh.2022.1056226
54. Lei K., Syed A. B., Zhu X., Pauly J. M., Vasanawala S. V.. **Automated MRI field of view prescription from region of interest prediction by intra-stack attention neural network**. (2023.0) **10** p. 92. DOI: 10.3390/bioengineering10010092
55. Zeng N., Wu P., Wang Z., Li H., Liu W., Liu X.. **A small-sized object detection oriented multi-scale feature fusion approach with application to defect detection**. (2022.0) **71** 1-14. DOI: 10.1109/tim.2022.3153997
|
---
title: Methionine-Mediated Regulation of Intestinal Lipid Transportation Induced by
High-Fat Diet in Rice Field Eel (Monopterus Albus)
authors:
- Yajun Hu
- Junzhi Zhang
- Minglang Cai
- Wuying Chu
- Yi Hu
journal: Aquaculture Nutrition
year: 2023
pmcid: PMC10036194
doi: 10.1155/2023/5533414
license: CC BY 4.0
---
# Methionine-Mediated Regulation of Intestinal Lipid Transportation Induced by High-Fat Diet in Rice Field Eel (Monopterus Albus)
## Abstract
An eight-week feeding trial explored the mechanism that supplemented methionine (0 g/kg, 4 g/kg, 8 g/kg, and 12 g/kg) in a high-fat diet (120 g/kg fat) on intestinal lipid transportation and gut microbiota of M. Albus (initial weight 25.03 ± 0.13 g) based on the diet (60 g/kg fat), named as Con, HFD+M0, HFD+M4, HFD+M8, and HFD+M12, respectively. Compared with Con, gastric amylase, lipase, trypsin ($P \leq 0.05$), and intestinal lipase, amylase, trypsin, Na+/K+ -Adenosinetriphosphatase, depth of gastric fovea, and the number of intestinal villus goblet cells of HFD+M0 were markedly declined ($P \leq 0.05$), while intestinal high-density lipoprotein-cholesterol, very low-density lipoprotein-cholesterol and microsomal triglyceride transfer protein of HFD+M0 were markedly enhanced ($P \leq 0.05$); compared with HFD+M0, gastric lipase, amylase, trypsin, and intestinal lipase, trypsin, Na+/K+ -Adenosinetriphosphatase, microsomal triglyceride transfer protein, very low-density lipoprotein-cholesterol, and apolipoprotein -A, the height of intestinal villus and the number of intestinal villus goblet cells of HFD+M8 were remarkably enhanced ($P \leq 0.05$). Compared with Con, intestinal occ, cl12, cl15, zo-1, zo-2 of HFD + M0 were markedly down-regulated ($P \leq 0.05$), while intestinal vldlr, npc1l1, cd36, fatp1, fatp2, fatp6, fatp7, apo, apoa, apob, apof, apoo, mct1, mct2, mct4, mct7, mct12, lpl, mttp, moat2, dgat2 of HFD M0 were remarkably upregulated ($P \leq 0.05$); compared with HFD+M0, intestinal gcn2 and eif2α of HFD+M8 were remarkably downregulated ($P \leq 0.05$), intestinal occ, cl12, cl15, zo-1, zo-2, hdlbp, ldlrap, vldlr, cd36, fatp1, fatp2, fatp6, apo, apoa, apob, apof, apoo, mct1, mct2, mct8, mct12, lpl, mttp, moat2, and dgat2 were remarkably upregulated ($P \leq 0.05$). Compared with Con, the diversity of gut microbiota of HFD+M0 was significantly declined ($P \leq 0.05$), while the diversity of gut microbiota in HFD+M8 was significantly higher than that in HFD+M0 ($P \leq 0.05$). In conclusion, a high-fat methionine deficiency diet destroyed the intestinal barrier, reduced the capacity of intestinal digestion and absorption, and disrupted the balance of gut microbiota; supplemented methionine promoted the digestion and absorption of lipids, and also improved the balance of gut microbiota.
## 1. Introduction
Soybean protein is a high-quality plant protein and, with a short growth cycle, was widely used in aquatic feed to replace a fish meal [1]. However, methionine is the most lacking amino acid in soybean protein, it is also essential amino acid for fish [2]. There have been lots of studies reported that a higher proportion of soybean protein used in aquatic feed replacing fish meal inhibits growth performance and the body's metabolism of fish [3–5]. Additionally, methionine also regulates the body's metabolism with metabolites of methionine through transsulfuration [6], transamination [7], and transmethylation [8]. Besides, methionine can also be considered as a signal molecule induce various metabolism for aquatic animals [9–11]. Methionine restriction not only inhibits protein synthesis but also damages the healthy state of fish [12–14] and decreases intestinal digestive and antioxidant enzymes and immunomodulatory of rohu (Labeo rohita) [15].
The gastrointestinal tract is mainly involved in digestion and absorption. Gastrointestinal tract health and functional integrity have significant effects on an animal's overall healthy statement and utilization of nutrients, and gastrointestinal function includes digestion and absorption of nutrients by epithelial and goblet cells, secretion of immunoglobulins and mucins, also forming a barrier against harmful antigens and pathogens [16]. A study reported that branched-chain amino acid could promote intestinal morphology and cell breeding and enhance intestinal amino acid absorption by inducing intestinal amino acid transporters and promoting intestinal protein turnover [17]. Methionine also benefits intestinal morphology; besides, gut bacteria are related to extensive catabolism of dietary methionine in the intestine [18]. A previous study on nursery pigs indicated that dietary methionine improves small intestinal morphology by enhancing villous height and reducing bacteria fermentation via improving nutrient digestion and absorption [19]. What is more, supplement methionine could assist in producing glutathione and enhance the morphology of the duodenum of the nursery pig [20].
Gut microbiota could affect a host's lipid metabolism through multiple direct and indirect biological mechanisms [21], even considered as an endocrine organ [22]. Semova et al. [ 23] used fluorescent markers to image zebrafish (Barchydanio rerio var) and found that gut microbiota could stimulate the uptake of fatty acids and formation of lipid droplets by intestinal epithelium and liver, sclerenchyma enhances the absorption capacity of fatty acids by host's intestinal cells, thus promote to increase the amount of lipid droplets. Meanwhile, the size of lipid droplets was increased with the increasing abundance of bacteria, they demonstrated that the golden rod bacteria (Chrysobacterium hispanicum) and pseudomonas (Pseudomonas adaceae) are the main bacteria regulating the size of host lipid droplets. [ 24] reported that gut bacteria could ferment the indigestible carbohydrates, and digest the indigestible fiber of gut contents into short-chain fatty acids. Additionally, gut microbiota could promote the absorption of fatty acids by activating the absorption capacity of intestinal epithelial cells. Our previous studies showed that a high-fat (fish oil) diet disturbance the homeostasis of intestinal microbiota, which includes microbial population, an abundance of main microbiota, and an intestinal environment [25]. Meanwhile, dysbiosis of gut microbiota induced by dietary oxidized fish oil, while supplemented taurine could recover gut microbiota homeostasis and restore intestinal function [26]. In addition, the abundance of firmicutes of gut microbiota decreased with the supplementation of black soldier fly (Hermetia illucens L.) larvae meal [27].
Rice field eel (Monopterus albus, M. albus) is subtropical freshwater benthic fish [28], and prey on frog eggs, insects, earthworms, and water earthworms in nature [17]. M. albus has a gastrointestinal system, that needs better quality and higher levels of protein, and optimum protein/lipid ratio [29]. We found that when fish meal is replaced by soybean meal [30], soy protein concentrate [31] declines the growth performance of M. albus. Our earlier study showed that methionine deficiency inhibits the growth performance of M. albus [32, 33], also reduces bodies and hepatic lipid deposition, and mainly declined fatty acid synthesis [34]. Additionally, a deficiency of methionine damages gastric and intestinal structure reduces the function of the intestinal barrier and inhibits the ability of intestinal lipid and fatty acid transportation of M. albus (Hu et al., [ 32, 33]). What is more, a study reported that suitable methionine restriction improved gut function by regulating gut microbiota and its metabolite profiles in a high-fat diet on mice [35]. However, how gut microbiota adaptively change and regulate the host's metabolism under the condition of methionine restriction and methionine affect lipid transportation by gut microbiota remains unknown. Here, we made methionine deficiency experimental feed according to our previous study [32, 33]; additionally, made a high-fat diet based on our previous study that the high-fat model on M. albus [36]. We explored methionine and how to regulate intestinal lipid transportation and the gut microbiome of M. albus.
## 2.1. Experimental Feed
The basic feed (60 g/kg fat) was made according to our previous paper [32, 33], named Con; and the high-fat basic feed (120 g/kg fat) was based on our study on M. Albus [36], different levels of methionine (0, 4 g/kg, 8 g/kg, and 12 g/kg) were supplemented into high-fat diet obeyed the principle of equal nitrogen referenced our study [32, 33], named as, HFD+M0, HFD+M4, HFD+M8, and HFD+M12, respectively. Experimental diets and levels of nutrition is shown in Table 1.
The method of proximate analysis (moisture, crude protein, crude lipid, and ash) of experimental feed was referenced in our paper [37]. Amino acids were determined by an automatic amino acid analyzer (Agilent-1100, Agilent Technologies Co., Ltd., Santa Clara, CA, USA) referenced by the method reported by Wiriduge et al. [ 38], fatty acids were analyzed by GC-MS (Agilent 7890B-5977A, Agilent Technologies Co., Ltd., Santa Clara, CA, USA) referenced the method reported by Jin et al. [ 39], as shown in Tables 2 and 3.
## 2.2. Feeding Trial
The uniform size of M. albus (25.03 ± 0.13 g) was randomly raised in 15 cages (2.0 × 1.5 × 1.5 m), and every group contained triplicates with 60 fish, referenced to our study [32, 33]. Each cage was covered by $95\%$ of the fresh alternanthera philoxeroides (Mart.) Griseb to simulate the natural environment for M. albus, water temperature 25 ± 6°C, dissolved O2 ≥ 6.0 mg/L, NH4+-N < 0.5 mg/L, respectively). The feeding rate was 3-$4\%$ of body weight and adjusted every two weeks. The feeding trial was lasting for 8 weeks.
## 2.3. Ethics Statement
Our study was approved by the Committee of Laboratory Animal Management and Animal Welfare of Hunan Agricultural University (Changsha, China) No. 094.
## 2.4. Sample Collection and Analyses
Stomach and intestine were obtained from six fish per cage, stored at -80°C until use. A gastric digestive enzyme (amylase, lipase, and trypsin) and intestinal biochemical indices (amylase, lipase, trypsin, Na+/K+ -Adenosinetriphosphatase, high-density lipoprotein cholesterol, and low-density lipoprotein cholesterol) were determined by a kit of NanJing JianCheng Bioengineering (Nanjing, China). Intestinal very-low-density lipoprotein cholesterol, microsomal triglyceride transfer protein, and Apolipoprotein-A were determined by a kit of Shanghai Enzyme-linked Biotechnology Co., Ltd (Shanghai, China).
The stomach and intestine of five fish in each cage were taken for observing organizational structure. Intestinal H&E-stained sections were referenced in our paper [40], and intestinal sections were made by cryostat microtome, stained with Oil Red O [32, 33], and observed by CaseViewer.
Intestinal total RNA was obtained from 6 fish per cage by Monzol reagent (Monad, Shanghai, China). RNA was reversed by MonScript (Monad, Shanghai, China) and obtained cDNA. Primers were synthesized by Biosune Biotechnology, Inc. (Shanghai, China), as shown in Table 4. Quantitative real-time PCR (qPCR) referenced our earlier paper [41]. The amplification efficiency was between 0.96 and 1.10, calculated by formula $E = 10$∗(−1/slope) − 1, and 5-fold serial dilutions of cDNA (triplicate) were used to generate the standard curve. Relative mRNA expression was calculated by 2-△△Ct [42].
## 2.5. Gut Microbiota Analysis
Gut bacterial DNA was obtained from the gut content of three groups (Con, HFD+M0, and HFD+M8) by Power Fecal DNA Isolation Kit (MoBio Laboratories, Inc), and gut bacterial high-throughput sequencing by Illumina MiSeq platform, all sequences were classified into operational taxonomic units (OTUs) above the level of $97\%$ similarity by quantitative insights into microbial ecology (QIIME) after removing low-quality scores (Q score, 20) with FASTX-Toolkit (Hannon Lab, USA).
## 2.6. Statistical Analysis
Data among groups (Con, HFD+M0, HFD+M4, HFD+M8, and HFD+M12) were analyzed by one-way analysis of variance (ANOVA), and remarkable differences among all groups were assessed by Turkey's multiple-range test. Values among groups (Con and HFD+M0; HFD+M0 and HFD+M8, respectively) were calculated by independent T-test by SPSS 22 software. Values were indicated as means ± SEM (standard error of the mean), and a significant difference was considered at $P \leq 0.05.$
## 3.1. Gastric and Intestinal Biochemical Indices
Compared with Con, gastric amylase and trypsin were remarkably decreased in HFD+M0 ($P \leq 0.05$); intestinal lipase, amylase, trypsin, and Na+/K+ -Adenosinetriphosphatase were markedly decreased in HFD+M0 ($P \leq 0.05$), while intestinal high-density lipoprotein cholesterol, very-low-density lipoprotein cholesterol, and microsomal triglyceride transfer protein were significantly enhanced ($P \leq 0.05$). Compared with HFD+M0, dietary 8 g/kg methionine remarkably increased gastric lipase, amylase, and trypsin ($P \leq 0.05$), also significantly increased intestinal lipase, trypsin and Na+/K+ -Adenosinetriphosphatase, very-low-density lipoprotein cholesterol, microsomal triglyceride transfer protein, and Apolipoprotein-A ($P \leq 0.05$) (Table 5).
## 3.2. Gastric and Intestinal Sections
Compared with Con, gastric fovea remarkably decreased in HFD+M0 ($P \leq 0.01$), while dietary methionine increased gastric fovea ($P \leq 0.05$) (Figure 1, Table 6). Compared with Con, intestinal villus height and intestinal muscular thickness decreased in HFD+M0 ($P \leq 0.05$), and amounts of intestinal goblet cells per root in HFD+M0 were markedly decreased ($P \leq 0.05$), while dietary 8 g/kg methionine markedly improved intestinal villus height and goblet cells per root, remarkably decreased crypt depth ($P \leq 0.05$) (Figure 2, Table 7). Additionally, intestinal Oil Red O-stained sections showed that the content of lipid droplets of intestinal villi and the mucosal layer was as follows: Con < HFD+M0 < HFD+M8 (Figure 3).
## 3.3. Intestinal mRNA Expression
Compared with Con, the intestinal occ, cl12, cl15, zo-1, and zo-2 markedly downregulated in HFD+M0 ($P \leq 0.001$, $P \leq 0.001$, $P \leq 0.001$, $P \leq 0.001$, and $P \leq 0.01$, respectively), while intestinal vldlr, npc1l1, cd36, fatp1, fatp2, fatp6, fatp7, apo, apoa, apob, apof, apoo, mct1, mct2, mct4, mct7, mct12, lpl, mttp, moat2, and dgat2 were markedly upregulated in HFD+M0 ($P \leq 0.001$, $P \leq 0.001$, $P \leq 0.001$, $P \leq 0.05$, $P \leq 0.001$, $P \leq 0.01$, $P \leq 0.01$, $P \leq 0.001$, $P \leq 0.01$, $P \leq 0.001$, $P \leq 0.01$, $P \leq 0.001$, $P \leq 0.05$, $P \leq 0.01$, $P \leq 0.01$, $P \leq 0.05$, $P \leq 0.001$, $P \leq 0.01$, $P \leq 0.01$, $P \leq 0.01$, and $P \leq 0.001$). Compared with HFD+M0, intestinal gcn2 and eif2α downregulated in HFD+M8 ($P \leq 0.001$, $P \leq 0.05$), while intestinal occ, cl12, cl15, zo-1, zo-2, hdlbp, ldlrap, vldlr, cd36, fatp1, fatp2, fatp6, apo, apoa, apob, apof, apoo, mct1, mct2, mct8, mct12, lpl, mttp, moat2, and dgat2 significantly upregulated ($P \leq 0.001$, $P \leq 0.001$, $P \leq 0.001$, $P \leq 0.001$, $P \leq 0.01$, $P \leq 0.01$, $P \leq 0.001$, $P \leq 0.001$, $P \leq 0.01$, $P \leq 0.05$, $P \leq 0.001$, $P \leq 0.01$, $P \leq 0.01$, $P \leq 0.01$, $P \leq 0.001$, $P \leq 0.05$, $P \leq 0.01$, $P \leq 0.01$, $P \leq 0.01$, $P \leq 0.01$, $P \leq 0.01$, $P \leq 0.01$, $P \leq 0.01$, $P \leq 0.05$, and $P \leq 0.001$) (Figure 4).
## 3.4. Gut Bacterial Diversity Indices
Compared with Con, the gut bacterial Chao1, OTUs, Shannon, Simpson, and Faith_pd in HFD+M0 were markedly declined ($P \leq 0.001$, $P \leq 0.001$, $P \leq 0.001$, $P \leq 0.01$, and $P \leq 0.001$). Compared with HFD+M0, gut bacterial Chao1, OTUs, Shannon, Simpson, and Faith_pd in HFD+M8 were significantly enhanced ($P \leq 0.05$, $P \leq 0.05$, $P \leq 0.01$, $P \leq 0.05$, and $P \leq 0.001$) (Table 8).
## 3.5. Gut Bacterial Levels of Phylum and Genus
Firmicutes, Fusobacteria, and Proteobacteria were the main populations at the phylum level, and Clostridium, Cetobacterium, and Plesiomonas were the main populations at the genus level. Compared with Con, the proportions of Firmicutes and Fusobacteria of the phylum and *Clostridium and* Cetobacterium of the genus were increased in HFD+M0, and the proportion of Proteobacteria of phylum and Plesiomonas of genus decreased. Compared with HFD+M0, Firmicutes of the phylum and *Clostridium of* the genus were decreased in HFD+M8, and the proportion of Fusobacteria and Proteobacteria of the phylum and Cetobacterium and Plesiomonas of the genus were increased in HFD+M8 (Figure 5).
## 3.6. Gut Bacterial Functional Classification of KEGG
The gut bacterial predictive main function was fatty acid biosynthesis, glutamine, glutamate, and alanine metabolism; valine, leucine, and isoleucine biosynthesis; and also including starch and sucrose metabolism (Figure 6).
## 4. Discussions
Recently, we found that dietary methionine restriction inhibits muscle fiber growth, development, and differentiation in M. albus, declines the growth performance in M. albus [32, 33], and induces hepatic lipid metabolism disorder and declined lipid content of M. Albus [34]. What is more, methionine deficiency also affected gastric and intestinal structure, damaged the intestinal barrier, and declined intestinal lipid and fatty acid transportation of M. albus [32, 33]. In this study, compared with the Con, gastric amylase and trypsin were remarkably decreased in the high-fat and methionine deficiency diet; also intestinal lipase, amylase, and trypsin were remarkably decreased in HFD+M0, while supplemented 8 g/kg methionine markedly increased gastric lipase, amylase, and trypsin, also markedly increased intestinal lipase and trypsin. From intestinal H&E-stained images, compared with Con, gastric fovea remarkably decreased in HFD+M0, while dietary methionine increased gastric fovea. Compared with Con, intestinal villus height and muscular thickness decreased in HFD+M0, and the number of intestinal goblet cells of root in HFD + M0 was markedly decreased, while dietary 8 g/kg methionine markedly improved intestinal villus height and goblet cells per root, remarkably decreased crypt depth. We inferred that a high-fat diet inhibited gastric-intestinal digestive capacity. Additionally, methionine sufficiency improved the gastric-intestinal main digestive enzymes (amylase, lipase, and trypsin) of M. albus, which is consistent with the study on grass carp (Ctenopharyngodon idella) [43].
MTTP promotes the transportation of fat by assisting triglyceride-rich apolipoprotein [44], such as ApoA1 [45]. HDL, LDL, and VLDL are major lipoproteins that carry cholesterol [46]. Na+/K+ -*Adenosinetriphosphatase is* the main enzyme involved in assisting the intestinal absorption and transportation of nutrients (such as amino acids, lipids, and glucose) [47]. Here, compared with Con, intestinal Na+/K+ -Adenosinetriphosphatase was remarkably decreased in HFD+M0, while intestinal HDL-C, VLDL-C, and MTTP were remarkably increased. Meanwhile, supplemented 8 g/kg methionine markedly increased intestinal Na+/K+ -Adenosinetriphosphatase, VLDL-C, MTTP, and ApoA. This indicated that an intestinal high concentration of fatty acids stimulated lipoprotein secretion [48], and dietary methionine could promote lipid transportation. This meant that the capacity of intestinal absorption declined, and the intestinal barrier was damaged [49].
Furtherly, we chose Con, HFD+M0, and HFD+M8 to explain the molecular mechanism of how methionine regulates gastrointestinal lipid digestion and absorption of M. albus. gcn2 and eif2a can sense essential amino acid deprivation and regulates their downstream relative genes to adapt to nutrient deficiency [50]. In this study, compared with HFD+M0, intestinal gcn2 and eif2α downregulated in HFD+M8, which meant that methionine deficiency could be sensed by M. albus. Occludens, claudin, and zonula are intestinal tight junction proteins, also important in protecting barrier integrity and preventing infiltration [51]. Here, compared with Con, the intestinal occ, cl12, cl15, and zo-1, zo-2 remarkably downregulated in HFD+M0, dietary methionine markedly upregulated intestinal occ, cl12, cl15, zo-1, and zo-2 expression. This explained that high-fat diet-induced free fatty acids damaged the intestine causes by which termed “intestinal lipotoxicity” [52], and methionine repaired the gastric and intestinal structure, also improved the intestinal barrier as we reported [32, 33]. MTTP promotes fatty transportation by assisting the secretion of lipoproteins [44]. lpl is involved in lipolysis [53], while mogat2 and dgat2 are the main enzymes involved in lipogenesis [54, 55]. In this study, compared with Con, the intestinal lpl, mttp, moat2, and dgat2 were upregulated markedly in HFD+M0 upregulated, while dietary methionine significantly upregulated intestinal lpl, mttp, moat2, and dgat2 expressions, which indirectly explained that why supplementing methionine enhanced gastric and intestinal digestive enzyme, also promoted the capacity of digestive. dlbp plays a pivotal role in the regulation of lipids and cholesterol [56], ldlra pathway is related to reducing circulating cholesterol [57]. npc1l1 is mainly charged in cholesterol absorption [58]. cd36 is related to fatty acid absorption and transportation [59, 60]. Apolipoprotein can bind and transport lipoproteins [61]. mct mainly transports short-chain monocarboxylates, including lactate, pyruvate, and ketone bodies [62]. Here, compared with Con, the intestinal vldlr, npc1l1, cd36, fatp1, fatp2, fatp6, fatp7, apo, apoa, apob, apof, apoo, mct1, mct2, mct4, mct7, mct12, lpl, mttp, moat2, and dgat2 were significantly upregulated in HFD+M0. Compared with HFD+M0, intestinal hdlbp, ldlrap, vldlr, cd36, fatp1, fatp2, fatp6, apo, apoa, apob, apof, apoo, mct1, mct2, mct8, mct12, lpl, mttp, moat2, and dgat2 remarkably upregulated. In addition, intestinal oil red O stained sections showed that the content of lipid droplets in intestinal villi, and the mucosal layer was as follows: Con < HFD+M0 < HFD+M8. We inferred that a high-fat diet damaged intestinal mucosal growth and reduces intestinal epithelial cells breeding, damaged the intestinal barrier [49, 63], then, increased intestinal permeability, intestinal lipid, and fatty acids transportation. While supplemented sufficiency methionine (8 g/kg) restores the intestinal barrier and improved intestinal function.
Compared with Con, the gut bacterial Chao1, OTUs, Shannon, Simpson, and Faith_pd in HFD+M0 were significantly decreased; compared with HFD + M0, the gut bacterial Chao1, OTUs, Shannon, Simpson, and Faith_pd in HFD+M8 were significantly increased. In addition, Firmicutes, Fusobacteria, and Proteobacteria were the main populations at the phylum level, and Clostridium, Cetobacterium, and Plesiomonas were the main populations at the genus level. What is more, compared with Con, the proportion of Firmicutes and Fusobacteria of the phylum and *Clostridium and* Cetobacterium of the genus were increased in HFD+M0, and the proportion of Proteobacteria of the phylum and Plesiomonas of the genus were decreased; compared with HFD+M0, Firmicutes of the phylum and *Clostridium of* the genus were decreased in HFD+M8, and the proportion of Fusobacteria and Proteobacteria of the phylum and Cetobacterium and Plesiomonas of the genus were increased in HFD+M8. We considered that a high-fat diet disturbed the balance of the bacterial community and induced microbial dysfunctions, the phenomenon was similar to our earlier study [25, 26], while supplemented methionine improved the gut microbiota homeostasis, and also promoted lipid metabolism. Meanwhile, the gut bacteria also had the predictive main function was fatty acid biosynthesis, glutamine, glutamate, and alanine metabolism; valine, leucine, and isoleucine biosynthesis; also including starch and sucrose metabolism.
In conclusion, the high-fat methionine deficiency diet affected the gastric and intestinal function of M. albus, damaged the intestinal barrier, reduced the capacity of intestinal digestion and absorption, and disrupted the balance of gut microbiota; supplemented methionine could improve the intestinal function, promoting the digestion and absorption of lipids, and also improving the gut microbiota balance.
## Data Availability
The datasets used and/or analyzed during the current study are available from the corresponding author upon reasonable request.
## Conflicts of Interest
There are no conflicts of interest in this manuscript.
## Authors' Contributions
Yajun Hu was in charge of methodology, data curation, and writing the original draft. Junzhi Zhang was in charge of formal analysis, reviewing, and writing the editing. Minglang Cai was in charge of data curation and software. Wuying Chu was in charge of formal analysis, software, and writing—review and editing. Yi Hu was in charge of funding acquisition, writing—review and editing, and validation.
## References
1. Jannathulla R., Rajaram V., Kalanjiam R., Ambasankar K., Muralidhar M., Dayal J. S.. **Fishmeal availability in the scenarios of climate change: inevitability of fishmeal replacement in aquafeeds and approaches for the utilization of plant protein sources**. (2019) **50** 3493-3506. DOI: 10.1111/are.14324
2. Wang P., Zhu J., Feng J., He J., Lou Y., Zhou Q.. **Effects of dietary soy protein concentrate meal on growth, immunity, enzyme activity and protein metabolism in relation to gene expression in large yellow croaker**. (2017) **477** 15-22. DOI: 10.1016/j.aquaculture.2017.04.030
3. Bruce T. J., Luke P. O., Jie M.. **An initial evaluation of fishmeal replacement with soy protein sources on growth and immune responses of burbot (Lota lota maculosa)**. (2021) **545, article 737157**. DOI: 10.1016/j.aquaculture.2021.737157
4. Liu X., Han B., Xu J.. **Replacement of fishmeal with soybean meal affects the growth performance, digestive enzymes, intestinal microbiota and immunity of**. (2020) **18, article 100472**. DOI: 10.1016/j.aqrep.2020.100472
5. Wang L., Jian W., Kangle L.. **Total replacement of fish meal with soybean meal in diets for bullfrog (**. (2020) **524, article 735236**. DOI: 10.1016/j.aquaculture.2020.735236
6. Sbodio J. I., Snyder S. H., Paul B. D.. **Regulators of the transsulfuration pathway**. (2019) **176** 583-593. DOI: 10.1111/bph.14446
7. Tavares C. D. J., Sharabi K., Dominy J. E.. **The methionine transamination pathway controls hepatic glucose metabolism through regulation of the GCN5 acetyltransferase and the PGC-1**. (2016) **291** 10635-10645. DOI: 10.1074/jbc.M115.706200
8. Stoccoro A., Tannorella P., Migliore L., Coppede F.. **Polymorphisms of genes required for methionine synthesis and DNA methylation influence mitochondrial DNA methylation**. (2020) **12** 1003-1012. DOI: 10.2217/epi-2020-0041
9. He Y., Cui X., Chi S.. **Changes in the PI3K/Akt/TOR signalling pathway after methionine treatment in the primary muscle cells of cobia (rachycentron canadum)**. (2021) **52** 2783-2790. DOI: 10.1111/are.15130
10. Song Y., Gao Y., Hogstrand C., Li D., Pan Y., Luo Z.. **Upstream regulators of apoptosis mediates methionine-induced changes of lipid metabolism**. (2018) **51** 176-190. DOI: 10.1016/j.cellsig.2018.08.005
11. Wang Z., Mai K., Xu W., Zhang Y., Liu Y., Ai Q.. **Dietary methionine level influences growth and lipid metabolism via GCN2 pathway in cobia (**. (2016) **454** 148-156. DOI: 10.1016/j.aquaculture.2015.12.019
12. Chen J., Li X., Xu H., Sun W., Leng X.. **Substitute of soy protein concentrate for fish meal in diets of white shrimp (Litopenaeus vannamei Boone)**. (2017) **25** 1303-1315. DOI: 10.1007/s10499-017-0115-4
13. Craig P. M., Moon T. W.. **Methionine restriction affects the phenotypic and transcriptional response of rainbow trout (Oncorhynchus mykiss) to carbohydrate-enriched diets**. (2013) **109** 402-412. DOI: 10.1017/S0007114512001663
14. Zhang Y., Ji W., Wu Y., Han H., Qin J., Wang Y.. **Replacement of dietary fish meal by soybean meal supplemented with crystalline methionine for Japanese seabass (Lateolabrax japonicus)**. (2016) **47** 243-252. DOI: 10.1111/are.12486
15. Noor Z., Noor M., Khan S. A.. **Dietary supplementations of methionine improve growth performances, innate immunity, digestive enzymes, and antioxidant activities of rohu (Labeo rohita)**. (2021) **47** 451-464. DOI: 10.1007/s10695-021-00924-x
16. Yang Z., Liao S. F.. **Physiological effects of dietary amino acids on gut health and functions of swine**. (2019) **6** 169-182. DOI: 10.3389/fvets.2019.00169
17. Duan Y., Tan B., Li J.. **Optimal branched-chain amino acid ratio improves cell proliferation and protein metabolism of porcine enterocytesin in vivo and in vitro**. (2018) **54** 173-181. DOI: 10.1016/j.nut.2018.03.057
18. Wang W. W., Qiao S. Y., Li D. F.. **Amino acids and gut function**. (2009) **37** 105-110. DOI: 10.1007/s00726-008-0152-4
19. Kaewtapee C., Krutthai N., Poosuwan K., Poeikhampha T., Koonawootrittriron S., Bunchasak C.. **Effects of adding liquid DL-methionine hydroxy analogue-free acid to drinking water on growth performance and small intestinal morphology of nursery pigs**. (2010) **94** 395-404. PMID: 19663980
20. Shen Y. B., Weaver A. C., Kim S. W.. **Effect of feed grade L-methionine on growth performance and gut health in nursery pigs compared with conventional DL-methionine**. (2014) **92** 5530-5539. DOI: 10.2527/jas.2014-7830
21. Ghazalpour A., Cespedes I., Bennett B. J., Allayee H.. **Expanding role of gut microbiota in lipid metabolism**. (2016) **27** 141-147. DOI: 10.1097/MOL.0000000000000278
22. Clarke G., Stilling R. M., Kennedy P. J., Stanton C., Cryan J. F., Dinan T. G.. **Minireview: gut microbiota: the neglected endocrine organ**. (2014) **28** 1221-1238. DOI: 10.1210/me.2014-1108
23. Semova I., Carten J. D., Stombaugh J.. **Microbiota regulate intestinal absorption and metabolism of fatty acids in the zebrafish**. (2012) **12** 277-288. PMID: 22980325
24. Almeida A., Mitchell A. L., Miguel B.. **A new genomic blueprint of the human gut microbiota**. (2019) **568** 499-504. DOI: 10.1038/s41586-019-0965-1
25. Peng M., Xue J., Hu Y.. **Disturbance in the homeostasis of intestinal microbiota by a high-fat diet in the rice field eel (**. (2019) **502** 347-355. DOI: 10.1016/j.aquaculture.2018.12.062
26. Peng M., Luo H., Kumar V., Kajbaf K., Hu Y., Yang G.. **Dysbiosis of intestinal microbiota induced by dietary oxidized fish oil and recovery of diet-induced dysbiosis via taurine supplementation in rice field eel (**. (2019) **512, article 734288**. DOI: 10.1016/j.aquaculture.2019.734288
27. Hu Y., Huang Y., Tang T.. **Effect of partial black soldier fly (**. (2020) **17, article 100345**. DOI: 10.1016/j.aqrep.2020.100345
28. Hu Y., Zhang J., He L.. **Effects of dietary vitamin C on growth, antioxidant activity, and immunity in ricefield eel (**. (2020) **51** 159-170. DOI: 10.1111/jwas.12636
29. Ma X., Hu Y., Wang X.. **Effects of practical dietary protein to lipid levels on growth, digestive enzyme activities and body composition of juvenile rice field eel (**. (2014) **22** 749-760. DOI: 10.1007/s10499-013-9703-0
30. Hu Y., Yang G., Li Z.. **Effect of dietary taurine supplementation on growth, digestive enzyme, immunity and resistant to dry stress of rice field eel (**. (2018) **49** 2108-2118. DOI: 10.1111/are.13665
31. Zhang J., Zhong L., Peng M.. **Replacement of fish meal with soy protein concentrate in diet of juvenile rice field eel**. (2019) **15, article 100235**. DOI: 10.1016/j.aqrep.2019.100235
32. Hu Y., Cai M., Zhong H., Chu W., Hu Y.. **Methionine-mediated regulation of intestinal structure and lipid transport in the rice field eel (**. (2022) **2022, article 2293484** 1-11. DOI: 10.1155/2022/2293484
33. Hu Y., Zhang J., Zhong L., Chu W., Hu Y.. **A study on methionine-mediated regulation of muscle fiber growth, development and differentiation in the rice field eel (**. (2022) **547, article 737430**. DOI: 10.1016/j.aquaculture.2021.737430
34. Hu Y., Cai M., Zhong H., Chu W., Hu Y.. **A study on how methionine restriction decreases the body's hepatic and lipid deposition in Rice field eel (Monopterus albus)**. (2021) **22** p. 13379. DOI: 10.3390/ijms222413379
35. Yang Y., Zhang Y., Xu Y.. **Dietary methionine restriction improves the gut microbiota and reduces intestinal permeability and inflammation in high-fat-fed mice**. (2019) **10** 5952-5968. DOI: 10.1039/C9FO00766K
36. Shi Y., Zhong L., Zhong H.. **Taurine supplements in high-fat diets improve survival of juvenile**. (2022) **547, article 737431**. DOI: 10.1016/j.aquaculture.2021.737431
37. Hu Y., Chu W., Xue J., Hu Y., Zhong L., Zhong H.. **Using unessential sulfur amino acids to overcome methionine deficient diets on rice field eel (**. (2021) **533, article 736196**. DOI: 10.1016/j.aquaculture.2020.736196
38. Wijerath Wiriduge H. A. S., Zhang Y., Liu J., Yang M., Zhang W., Mai K.. **Dietary taurine improves muscle growth and texture characteristics in juvenile turbot (**. (2020) **17, article 100305**. DOI: 10.1016/j.aqrep.2020.100305
39. Jin M., Monroig Ó., Lu Y.. **Dietary DHA/EPA ratio affected tissue fatty acid profiles, antioxidant capacity, hematological characteristics and expression of lipid-related genes but not growth in juvenile black seabream (Acanthopagrus schlegelii)**. (2017) **12**. DOI: 10.1371/journal.pone.0176216
40. Hu Y., Zhang J., Xue J., Chu W., Hu Y.. **Effects of dietary soy isoflavone and soy saponin on growth performance, intestinal structure, intestinal immunity and gut microbiota community on rice field eel (**. (2021) **537, article 736506**. DOI: 10.1016/j.aquaculture.2021.736506
41. Hu Y., Hu Y., Wu T., Chu W.. **Effects of high dietary levels of cottonseed meal and rapeseed meal on growth performance, muscle texture, and expression of muscle-related genes in grass carp**. (2019) **81** 235-241. DOI: 10.1002/naaq.10091
42. Livak K. J., Schmittgen T. D.. **Analysis of relative gene expression data using real-time quantitative PCR and the 2**. (2001) **25** 402-408. DOI: 10.1006/meth.2001.1262
43. Jiang J., Shi D., Zhou X. Q.. **Effects of lysine and methionine supplementation on growth, body composition and digestive function of grass carp (Ctenopharyngodon idella) fed plant protein diets using high-level canola meal**. (2015) **22**
44. Hussain M. M., Nijstad N., Franceschini L.. **Regulation of microsomal triglyceride transfer protein**. (2011) **6** 293-303. DOI: 10.2217/clp.11.21
45. Brouillette C. G., Anantharamaiah G. M., Engler J. A., Borhani D. W.. **Structural models of human apolipoprotein A-I**. (1995) **1256** 103-129. DOI: 10.1016/0005-2760(95)00018-8
46. Kwiterovich P. J.. **The metabolic pathways of high-density lipoprotein, low-density lipoprotein, and triglycerides: a current review**. (2000) **86** 5-10. DOI: 10.1016/S0002-9149(00)01461-2
47. Maulu S., Hualiang L., Ke J.. **Dietary clostridium autoethanogenum protein modulates intestinal absorption, antioxidant status, and immune response in GIFT (**. (2021) **52** 5787-5799. DOI: 10.1111/are.15454
48. Kawamoto K., Sakuma M., Tanaka S.. **High-fat diets provoke phosphorus absorption from the small intestine in rats**. (2020) **72, article 110694**. DOI: 10.1016/j.nut.2019.110694
49. Shen Y. B., Ferket P., Park I., Malheiros R. D., Kim S. W.. **Effects of feed grade l-methionine on intestinal redox status, intestinal development, and growth performance of young chickens compared with conventional dl-methionine**. (2015) **93** 2977-2986. DOI: 10.2527/jas.2015-8898
50. Efeyan A., Comb W. C., Sabatini D. M.. **Nutrient-sensing mechanisms and pathways**. (2015) **517** 302-310. DOI: 10.1038/nature14190
51. Ulluwishewa D., Anderson R. C., McNabb W. C., Moughan P. J., Wells J. M., Roy N. C.. **Regulation of tight junction permeability by intestinal bacteria and dietary components**. (2011) **141** 769-776. DOI: 10.3945/jn.110.135657
52. Tanaka S., Yasuhiro N., Yuria T.. **High-fat diet-derived free fatty acids impair the intestinal immune system and increase sensitivity to intestinal epithelial damage**. (2020) **522** 971-977. DOI: 10.1016/j.bbrc.2019.11.158
53. Tian J., Wen H., Zeng L.. **Changes in the activities and mRNA expression levels of lipoprotein lipase (LPL), hormone-sensitive lipase (HSL) and fatty acid synthetase (FAS) of Nile tilapia (**. (2013) **400-401** 29-35. DOI: 10.1016/j.aquaculture.2013.01.032
54. Choi C. S., Savage D. B., Kulkarni A.. **Suppression of Diacylglycerol Acyltransferase-2 (DGAT2), but Not DGAT1, with Antisense Oligonucleotides Reverses Diet-induced Hepatic Steatosis and Insulin Resistance**. (2007) **282** 22678-22688. DOI: 10.1074/jbc.M704213200
55. Take K., Mochida T., Maki T.. **Pharmacological inhibition of Monoacylglycerol O-acyltransferase 2 improves hyperlipidemia, obesity, and diabetes by change in intestinal fat utilization**. (2016) **11**. DOI: 10.1371/journal.pone.0150976
56. Chen J., Chen J., Wu J.. **Molecular cloning and functional analysis of zebrafish high-density lipoprotein-binding protein**. (2003) **136** 117-130. DOI: 10.1016/S1096-4959(03)00181-7
57. Wijers M., Kuivenhoven J. A., van de Sluis B.. **The life cycle of the low-density lipoprotein receptor**. (2015) **26** 82-87. DOI: 10.1097/MOL.0000000000000157
58. Li D., Cui Y., Wang X., Liu F., Li X.. **Apple polyphenol extract improves high-fat diet-induced hepatic steatosis by regulating bile acid synthesis and gut microbiota in C57BL/6 male mice**. (2021) **69** 6829-6841. DOI: 10.1021/acs.jafc.1c02532
59. Buttet M., Traynard V., Tran T. T., Besnard P., Poirier H., Niot I.. **From fatty-acid sensing to chylomicron synthesis: role of intestinal lipid- binding proteins**. (2014) **96** 37-47. DOI: 10.1016/j.biochi.2013.08.011
60. Huang K., Chen K., Hsu L.. **Decreased PEDF promotes hepatic fatty acid uptake and lipid droplet formation in the pathogenesis of NAFLD**. (2020) **12** 270-283. DOI: 10.3390/nu12010270
61. Liu T., Chen J., Zhang D.. **ApoPred: identification of apolipoproteins and their subfamilies with multifarious features**. (2021) **8** 621144-621156. DOI: 10.3389/fcell.2020.621144
62. Droździk M., Pieniek S. S., Grzegółkowska J.. **Monocarboxylate transporter 1 (MCT1) in liver pathology**. (2020) **21** 1606-1618. DOI: 10.3390/ijms21051606
63. Bauchart-Thevret C., Stoll A., Burrin D. G.. **Intestinal metabolism of sulfur amino acids**. (2009) **22** 175-187. PMID: 19835653
|
---
title: Factors Determining Not Returning to Full-Time Work 12 Months After Mild Ischemic
Stroke
authors:
- Georgios Vlachos
- Hege Ihle-Hansen
- Torgeir Bruun Wyller
- Anne Brækhus
- Margrete Mangset
- Charlotta Hamre
- Brynjar Fure
journal: Archives of Rehabilitation Research and Clinical Translation
year: 2022
pmcid: PMC10036226
doi: 10.1016/j.arrct.2022.100245
license: CC BY 4.0
---
# Factors Determining Not Returning to Full-Time Work 12 Months After Mild Ischemic Stroke
## Body
Unemployment is associated with poor physical health1 and reduced quality of life.2 For some patients with stroke, work can be a cause of stress and therefore potentially a risk, whereas for others it is a way of demonstrating recovery.3 It has previously been found that $53\%$ of patients returned to work 1 year after stroke,4 with the proportion being lower in women than in men.5 Another study6 indicated that $47\%$ of patients working fulltime prestroke were still working fulltime 1 year later, while $27\%$ were no longer in work, and $24\%$ were working fewer hours a week. Predictors of returning to work were less severe neurologic deficits, better cognitive ability, greater independency in daily life activities, age younger than 51 years, and male sex.4,7 Psychiatric morbidity, depressive symptoms, and poststroke fatigue might hinder younger patients from returning to professional activity for as long as 2 years after first-ever mild stroke.6,8, 9, 10 Functional outcome after stroke has been improved during the last years because of both intravenous thrombolysis and mechanical thrombectomy.11, 12, 13 Thrombolysis for first-ever ischemic stroke was independently associated with a reduced rate of dementia in one study.14 The relationship between thrombolytic treatment and return to paid work has not been much investigated,15 but a recent article16 reported a high proportion of patients treated with thrombolysis returning to work ($64\%$) 1 year after stroke. Age 41-60 years, high stroke severity expressed as National Institutes of Health Stroke Scale (NIHSS)17 score ≥5, and female sex were associated with lower odds of workforce attachment.16 The effect of endovascular stroke treatment on work ability is largely unknown.
One incentive to initiate this study was our own experience from the stroke outpatient clinic, where we observed that several young working patients with minimal or no neurologic sequelae after a mild stroke reported difficulties involving cognitive functions. In addition, some patients struggled with fatigue, indifference or apathy, anxiety, depression, or mood swings. It was previously shown that hidden outcomes can hinder the patients from returning to normal daily activities and affect both family and work life.18 *Unemployment is* often associated with poorer physical1 and mental health19 and reduced quality of life.2 *In a* Dutch study, one-third of the participants who had not gone back to work 1 year after a mild-to-moderate stroke were unsatisfied with their occupational situation.6 In addition, the work satisfaction of working patients increased with the daily number of hours spent in a paid work.
Most studies about returning to work after stroke have included patients with mild-to-moderate ischemic and hemorrhagic strokes defined as NIHSS score ≤15,20 while our study is one of the few studies that has included patients with only mild ischemic strokes defined as NIHSS score ≤3. Other authors21 have found that stroke size, localization, and etiology, i.e. large anterior strokes, and stroke caused by large artery atherosclerosis and cardioembolism, are associated with not returning to work 1 year after ischemic stroke. We previously22 found that younger age at stroke onset and finding of multiple cerebral infarctions were associated with cognitive impairment 1 year after stroke.
Overall, the aim of this prospective cohort study was to identify the prevalence of not returning to full-time work and to study whether factors such as demographic characteristics, stroke etiology, localization, and reperfusion therapy were associated with not returning to full-time work 12 months after first-ever mild stroke in patients 70 years or younger.
## Highlights
•A total of $40\%$ of the patients did not return to full-time work 12 months after a first-ever mild ischemic stroke.•Low functional level expressed as modified Rankin scale score >1 at 12-month follow-up was significantly associated with not returning to full-time work.•*Diabetes mellitus* was a borderline significant factor of not returning to full-time work.
## Abstract
### Objective
To evaluate prevalence and factors determining not returning to full-time work 1 year after first-ever mild ischemic stroke.
### Design
Prospective, observational cohort study with 12-month follow-up.
### Setting
Stroke units and outpatient clinics at 2 Norwegian hospitals.
### Participants
We included 84 ($$n = 84$$) full-time working, cognitively healthy patients aged 70 years or younger who suffered an acute first-ever mild ischemic stroke, defined as National Institutes of Health Stroke Scale (NIHSS) score ≤3 points.
### Interventions
Not applicable.
### Main Outcome Measures
Vascular risk factors, sociodemographic factors, stroke localization, and etiology were recorded at inclusion. Cognitive impairment, anxiety, depression, fatigue, and apathy 12 months after stroke were assessed with validated instruments. Logistic regression analyses were performed to find correlates of not returning to full-time employment.
### Results
Of 78 patients assessed 1 year after stroke, 63 ($81\%$) had returned to work, 47 ($60\%$) to full-time employment status. Modified Rankin scale score >1 (adjusted odds ratio, 12.44 [$95\%$ confidence interval, 2.37-65.43], $$P \leq .003$$) at follow-up was significantly associated, and diabetes (adjusted odds ratio, 10.56 [$95\%$ confidence interval, 0.98-113.47], $$P \leq .052$$) was borderline significantly associated with not returning to full-time work. Female sex, NIHSS at discharge, anxiety per point on the anxiety scale, depression per point on the depression scale, and fatigue per point on the fatigue scale were significantly associated with not returning to full-time work after 1 year, but these associations were only seen in the unadjusted models.
### Conclusions
Low functional level that persists 12 months after stroke is related to not returning to full-time work. Patients with diabetes mellitus, female patients, and patients with a higher score on fatigue, anxiety, and depression scales may also be at risk of not returning to full-time work post stroke. Working patients should be followed up with a particular focus on factors determining participation in work and social life.
## Population
The included patients were admitted with an acute stroke to the stroke units at Oslo University Hospital, Norway and Bærum Hospital, Norway in the period from December 2014 until December 2016. Details regarding the study design have previously been published.23 *In this* substudy, we included patients employed fulltime, prescribed by law in Norway as working at least 37.5 hours a week, at stroke onset. Inclusion criteria were age 18-70 years and that the participants had a mild first-ever ischemic stroke defined as a NIHSS17 score ≤3 points at discharge.24 Patients who did not speak Norwegian, patients with diagnosed dementia or cognitive impairment defined as a score > 3.2 on the short form of the Informant Questionnaire on Cognitive Decline in the Elderly,25 and patients with known psychiatric disease were not eligible. Patients who had a new stroke during the first year after stroke were not invited to follow-up.
## Assessments
We recorded age, sex, educational level in years, and marital state at stroke onset in addition to vascular risk factors as treated hypertension, hyperlipidemia, coronary heart disease, atrial fibrillation, cigarette smoking, diabetes mellitus, body mass index, and presence of any apolipoprotein E-epsilon 4 alleles.
The diagnosis of ischemic stroke was based on the medical history, findings on the neurologic examination, and findings of an acute infarction on cerebral computed tomography (CT) or magnetic resonance imaging (MRI) scans. Patients presenting with neurologic deficits lasting more than 24 hours, but with no acute lesions on CT or MRI, were also considered to have an ischemic stroke.26 The ischemic strokes were classified by etiology according to the Trial of ORG 10172 in Acute Stroke Treatment (TOAST) classification.27 The Barthel Index score and the modified Rankin scale (mRS) score28,29 was used to assess the general functional level at discharge and at 12-month follow-up.
## Twelve-month follow-up
Study participants were invited to follow-up in the outpatient clinic. The cognitive and emotional assessments were performed by either a stroke physician, an occupational therapist, or a research nurse. These professionals had been trained by experienced physicians with high competence and interest in the field.
## Cognitive and emotional assessments
Cognitive and emotional functions were evaluated using an extended battery of screening tests.23 The Mini-Mental Status Examination Norwegian Revision 2 (MMSE-NR2) is considered to be a screening test of global cognitive functioning.30 To examine particular cognitive domains, we used the Clock Drawing Test31 (executive and visuospatial function), the Trail Making Test (TMT) A32 (focused attention, psychomotor speed) and B (divided attention, executive function, psychomotor speed), the Controlled Oral World Association (COWA) Verbal Fluency Test33 (attention, executive function, psychomotor speed, language ability), the Rey-Osterrieth Complex Figure Test (ROCF)34 (executive and visuospatial function, memory), and the California Verbal Learning Test II (CVLT II)35 (learning ability, memory). Because of the participants’ high educational level and low mean age (52.4 years), we chose a higher score (<$\frac{27}{30}$ points)36 than usual as a cutoff for the MMSE-NR2.30 For the Clock Drawing Test, the standard cutoff of 4 of 5 points was used,31 while the correct administrations within 60 seconds for the TMT A and 120 seconds for the TMT B were chosen as cutoffs.32 Because of the sparse evidence for differences related to age and sex on CVLT II scores, to age on ROCF, and to sex and education level on COWA and Trail Making Tests, raw scores were converted into standardized scores (mean±SD, 50±10) based on appropriate normative sample according to test manuals used by the Norwegian Registry of Persons Assessed for Cognitive Symptoms in Specialist Health Care services. In accordance with the criteria of the International Society for Vascular Behavioral and Cognitive Disorders,37 mild cognitive impairment can be present when the performance on validated cognitive tests in 1 or more of the cognitive domains of attention, processing speed, executive function, learning and memory, language, visuoconstructional-perceptual ability, praxis-gnosis-body schema, and social cognition is 1-2 SD below appropriate norms. Accordingly, we defined a score of at least −1 SD or any score outside the reference range on at least 1 of the cognitive tools as indicative for impairment of that particular cognitive domain.23 The presence of depression and anxiety symptoms, fatigue, and apathy was defined using well-established cutoffs of the Hospital Anxiety and Depression Scale (HADS) (>7 points for the depression subscale; >7 points for the anxiety subscale),38 the Fatigue Severity Scale (FSS) (≥4 points),39 and the Apathy Evaluation Scale–Self-report (AES-S) (≥34 points).40
Details about the testing of the cognitive and the emotional domains have previously been published.23
## Statistics
The Statistical Package for Social Sciences versions 27.0 and 28.0 were used for all statistical analyses.a Descriptive statistics and table analyses were performed, and data are presented with means and SDs for continuous variables and with proportions and percentages for categorical variables. The rating scale scores are represented with median and range where the data distribution was skewed. To compare the baseline and the follow-up characteristics of the participants who returned to full-time work and those who did not, we performed chi-square (χ2) test for the categorical and independent samples t test for continuous variables.
Logistic regression analyses were performed separately to identify possible factors associated with the dependent variable of not returning to full-time work 12 months after a mild stroke. First, unadjusted regression analyses were performed using baseline patients’ characteristics, stroke characteristics such as topography, TOAST classification, functional level expressed as NIHSS, Barthel Index score at discharge from hospital, treatment with intravenous thrombolysis and thrombectomy, mRS score at discharge and follow-up, presence of cognitive impairment, and the scores on the anxiety, depression, fatigue, and apathy scales at 12-month follow-up as independent variables in unadjusted regression analyses.
Age, sex, and variables associated with the outcome with a P value <.1 in unadjusted analyses were then entered into multivariate hierarchical logistic regression models. In model 1 potential explanatory variables were baseline characteristics, that is, age, sex, years of education, diabetes mellitus, and NIHSS score at admission and discharge. The mRS score at follow-up and the scores of the anxiety, depression, and fatigue scales at 12-month follow-up were entered in model 2. The factors that remained significant from the hierarchical logistic models were considered to be the independent predictors of the outcome. Regarding the use of the anxiety, depression, fatigue, and apathy scales, we decided not to dichotomize these variables but to use the raw scores in order not to lose information and statistical power.41 Results are presented as odds ratio with $95\%$ confidence interval. A P value <.05 was considered as the limit for statistical significance.
## Ethics
The Regional Committee South East for Medical and Health research ethics (register no. $\frac{2014}{1268}$) and the Oslo University Hospital's Data Protection Authority approved the study. Before inclusion, written informed consent was given by all patients.
## Results
In the main study, we included 127 patients, with follow-up data on 117 of them. This substudy comprises 84 patients (17 female [$20\%$]), with mean age 52.4±10.8 years who worked full-time at stroke onset. A total of 22 patients ($25\%$) were treated with intravenous thrombolysis. Four patients ($5\%$) were treated with mechanical thrombectomy, all combined with intravenous thrombolysis. Demographic and clinical baseline characteristics are presented in Table 1.Table 1Patient characteristics at baselineTable 1VariablesTotal Sample ($$n = 84$$)Male ($$n = 67$$)Female ($$n = 17$$)Demographics Male, n (%)67 [80] Age (y), mean ± SD52.4±10.854.0±10.446.0±10.5 Age (y), median (range)55 [30-70]56 [32-70]44 [30-67] Age group 30-45 y, n (%)26 [31]16 [24]10 [59] Age group >45 y, n (%)58 [69]51 [76]7 [41] Education in years, mean ± SD15.6±3.415.6±3.415.7±3.9 Living alone, n (%)22 [26]18 [27]4 [24]Risk factors, n (%) Hypertension39 [46]36 [54]3 [18] Hyperlipidemia61 [73]54 [81]7 [41] APOE-ε4 allele24 [30]20 [31]4 [24] Coronary disease8 [10]8 [12]0 [0] Atrial fibrillation6 [7]5 [8]1 [6] Cigarette smoking23 [27]19 [28]4 [24] Diabetes mellitus11 [13]10 [15]1 [6] BMI, mean ± SD26.9±4.227.3±4.225.5±3.8Given thrombolysis, n (%)22 [26]12 [18]10 [59]TOAST classification, n (%) Large vessel disease16 [19]13 [19]3 [18] Cardiac embolic disease25 [30]19 [28]6 [35] Small vessel disease20 [24]18 [27]2 [12] Stroke of unknown/other etiology23 [27]17 [25]6 [35]Topography, MRI/CT findings, n (%) Right hemisphere30 [36]26 [39]4 [24] Left hemisphere21 [25]15 [22]6 [35] Cerebellum/brainstem15 [18]11 [16]4 [24] Multiple brain regions14 [17]12 [18]2 [12] No acute infarcts4 [5]3 [5]1 [6]Assessments NIHSS at admission, median (range)1 [0-22]1 [0-22]2 [0-8] NIHSS at discharge, median (range)0 [0-3]0 [0-3]0 [0-3] mRS score at discharge, median (range)1 [0-4]1 [0-4]1 [0-3] mRS score 0-1 at discharge, n (%)67 [80]54 [81]13 [77] BI at discharge, mean ± SD19.8±1.119.7±1.319.9±0.3 BI at discharge, median (range)20 [11-20]20 [11-20]20 [19-20]Cognitive assessments IQCODE, mean ± SD3.0±0.13.2±0.13.0±0.0NOTE. Cigarette smoking, (≥1 cigarette/d); coronary artery disease: myocardial infarction, angina pectoris; hyperlipidemia: cholesterol >5 mmol/L or low-density lipoprotein cholesterol >3 mmol/L.Abbreviations: APOE-ε4, apolipoprotein E-epsilon 4; BI, Barthel Index; BMI, body mass index (calculated as weight in kilograms divided by height in meters squared); CT, computed tomography; IQCODE, Informant Questionnaire on Cognitive Decline in the Elderly; MRI, magnetic resonance imaging; TOAST, Trial of Org 10172 in Acute Stroke Treatment.
At 12-month follow-up, 78 patients ($93\%$) attended. During the 1-year poststroke period, 1 patient died, 3 had a recurrent stroke, and 2 withdrew from the follow-up. A total of 63 patients ($81\%$) had returned to work, 47 (60 %) of them to full-time work. Table 2 shows the distribution of the employment status as well as cognitive and emotional impairments at follow-up for all patients included in the study. Tables 3 and 4 show the baseline and follow-up characteristics of patients who returned to full-time work 1 year after stroke and those who did not. Table 2Patient characteristics at 12-month follow-upTable 2VariableTotal Sample $$n = 78$$Male $$n = 62$$Female $$n = 16$$Demographics Male, n (%)62 [79] Employment status, full time, n (%)47 [60]42 [68]5 [31] Employment status, part time, n (%)16 [21]10 [16]6 [38] Employment status, full AND part time, n (%)63 [81]52 [84]11 [69] Full-time work after thrombolysis ($$n = 21$$), n (%)11 [52]9 [75]2 [22]Assessments mRS score, median (range)1 [0-3]1 [0-3]1 [0-3] mRS score 0-1, n (%)59 [76]50 [81]9 [56] Barthel Index, median (range)20 [20-20]20 [20-20]20 [20-20]Impairments Cognitive impairments, n (%)49 [63]41 [66]8 [50]Emotional assessments HADS-Anxiety Scale, median (range)4 [0-15]4 [0-15]5 [0-11] HADS-Depression Scale, median (range)2 [0-19]1.50 [0-19]2 [0-7] FSS, median (range)2.66 (1.00-6.77)2.44 (1.00-6.77)4.44 (1.00-6.22) AES-S, median (range)27 [18-52]27 [18-52]26 [18-38]NOTE. Cognitive impairment, a score outside the reference range on at least 1 of the used cognitive tests.20Abbreviations: AES-S, Apathy Evaluation Scale–Self-report; FSS, Fatigue Severity Scale; HADS, Hospital Anxiety and Depression Scale. Table 3Baseline characteristics of patients who returned to FTW 1 year after stroke and those who did notTable 3VariableFTW $$n = 47$$NFTW $$n = 31$$P Value*Demographics Male, n (%)42 [89]20 [65] Female, n (%)5 [11]17 [35].01† Age (y), mean ± SD52.1±11.151.8±10.3.68 Age (y), median (range)53 [31-70]55 [30-67] Age group 30-45 y14 [30]11 [35] Age group >45 y33 [70]20 [65].46 Education in years, mean ± SD16.4±3.415.0±3.4.04† Living alone, n (%)11 [23]11 [35].38Risk factors, n (%) Hypertension22 [47]14 [45].96 Hyperlipidemia33 [70]24 [77].66 APOE-ε4 allele14 [31]8 [26].51 Coronary disease2 [4]2 [7]> 1.00 Atrial fibrillation3 [6]2 [7]>.99 Cigarette smoking10 [21]11 [35].45 Diabetes mellitus3 [6]7 [23].08 BMI, mean ± SD27.5±4.226.5±4.0.37Given thrombolysis11 [23]10 [32].42TOAST classification Large vessel disease9 [19]6 [19].98 Cardiac embolic disease13 [28]10 [32].66 Small vessel disease12 [26]7 [23].77 Stroke of unknown/other etiology13 [28]8 [26].86Topography, MRI/CT findings, n (%) Right hemisphere18 [38]9 [29].62 Left hemisphere10 [21]9 [29].62 Cerebellum/brainstem9 [19]6 [19].98 Multiple brain regions7 [15]6 [19].64 No acute infarcts3 [6]1 [3].65Assessments NIHSS at admission, median (range)0 [0-15]2 [0-22].04† NIHSS at discharge, median (range)0 [0-3]0 [0-3].01† mRS score at discharge, median (range)1 [0-4]1 [0-4].16 mRS score 0-1 at discharge, n (%)40 [85]23 [74].08 BI at discharge, mean ± SD19.9±0.419.5±1.8.11 BI at discharge, median (range)20 [18-20]20 [11-20]Cognitive assessments IQCODE, mean ± SD3.0±0.13.0±0.1.24NOTE. Cigarette smoking, (≥1 cigarette/d); coronary artery disease: myocardial infarction, angina pectoris; hyperlipidemia: cholesterol >5 mmol/L or low-density lipoprotein cholesterol >3 mmol/L.Abbreviations: APOE-ε4, apolipoprotein E-epsilon 4; BI, Barthel Index; BMI, body mass index (calculated as weight in kilograms divided by height in meters squared); CT, computed tomography; FTW, returned to full-time work; IQCODE, Informant Questionnaire on Cognitive Decline in the Elderly; MRI, magnetic resonance imaging; NFTW, did not return to full-time work; TOAST, Trial of Org 10172 in Acute Stroke Treatment.⁎P value: χ2 test is performed for categorical variables; independent samples t test is performed for continuous variables.†Statistical differences ($P \leq .05$).Table 4Follow-up characteristics of patients who returned to FTW 1 year after stroke and those who did notTable 4VariableFTW $$n = 47$$NFTW $$n = 31$$P Value*Assessments mRS score, median (range)1 [0-2]1.5 [0-3]<.01† mRS score 0-1, n (%)44 [94]15 [48]<.01† Barthel Index, median (range)20 [20-20]20 [20-20]Impairments Cognitive impairments, n (%)28 [60]21 [68].25Emotional assessments HADS-Anxiety Scale, median (range)3 [0-10]5.50 [0-15]<.01† HADS-Depression Scale, median (range)1 [0-10]3 [0-19]<.01† FSS, median (range)2.00 (1.00-5.77)4.11 (1.00-6.77)<.01† AES-S, median (range)27 [18-52]30.50 [18-50].22NOTE. Cognitive impairment: a score outside the reference range on at least 1 of the used cognitive tests.20Abbreviations: AES-S, Apathy Evaluation Scale–Self-report; FSS, Fatigue Severity Scale; FTW, returned to full-time work; HADS, Hospital Anxiety and Depression Scale; NFTW, did not return to full-time work.⁎P value: χ2 test is performed for categorical variables; independent samples t test is performed for continuous variables.†Statistical differences ($P \leq .05$).
## Factors determining not returning to full-time work 12 months after a mild ischemic stroke
In the adjusted regression models, a low functional level (mRS score >1) at follow-up was significantly associated with not returning to full-time work, and diabetes mellitus showed a borderline ($$P \leq .052$$) significant association. The relationship between not returning to full-time work and female sex, NIHSS score at discharge, and a higher score on the anxiety, depression, and fatigue scales were deflated after adjusting for relevant covariates (Table 5).Table 5Predictors of not returning to full-time work 12 months after a mild ischemic strokeTable 5UnadjustedAdjusted/Model 1Adjusted/Model 2VariableOR ($95\%$ CI)P ValueOR ($95\%$ CI)P ValueOR ($95\%$ CI)P ValueAge at baseline1.00 (0.96-1.04).911.10 (0.96-1.07).761.00 (0.94-1.07).95Female sex4.62 (1.41-15.09).01*7.13 (1.71-29.71).0073.43 (0.55-21.22).19Living alone1.80 (0.66-4.89).25Years of education0.88 (0.76-1.00).080.84 (0.70-1.01).060.86 (0.69-1.07).19BMI at admission0.94 (0.83-1.06).28Hyperlipidemia1.46 (0.51-4.15).48Diabetes mellitus4.28 (1.01-18.07).056.33 (1.19-33.63).0310.56 (0.98-113.47).052Hypertension0.94 (0.38-2.33).89Myocardial infarction1.55 (0.21-11.64).67Atrial fibrillation1.01 (0.16-6.43).99Presence of any APOE-ε4 allele0.77 (0.28-2.14).62Treatment with thrombolysis1.56 (0.57-4.29).39Treatment with thrombectomy4.93 (0.49-49.72).18Stroke location (no infarcts at MRI/CT)0.49 (0.05-4.93).54Stroke location (right hemisphere)0.66 (0.25-1.74).40Stroke location (left hemisphere)1.51 (0.53-4.30).44Stroke location (cerebellum/brainstem)1.01 (0.32-3.20).98Stroke location (multiple infarctions)1.37 (0.41-4.55).61TOAST (large vessel disease)1.01 (0.32-3.20).98TOAST (cardiac embolic)1.25 (0.46-3.34).66TOAST (small vessel disease)0.85 (0.29-2.47).77TOAST (underdetermined/other etiology)0.91 (0.33-2.54).86NIHSS at admission1.13 (0.99-1.30.071.10 (0.95-1.27).201.16 (0.96-1.40).13NIHSS at discharge1.59 (1.00-2.55).051.37 (0.75-2.50).311.11 (0.53-2.33).78Barthel Index score at discharge0.66 (0.33-1.32).24mRS score >1 at discharge1.99 (0.64-6.19).24HADS-Anxiety Scale score1.24 (1.07-1.45)<.01*1.20 (0.88-1.63).26HADS-Depression Scale score1.27 (1.07-1.49)<.01*0.98 (0.72-1.35).91FSS score2.01 (1.41-2.85)<.01*1.43 (0.77-2.63).26AES-S score1.05 (0.99-1.11).11Cognitive impairment1.43 (0.55-3.69).47mRS score >1 at follow-up15.64 (3.99-61.28)<.01*12.44 (2.37-65.43).003*NOTE. Cognitive impairment: a score outside the reference range on at least 1 of the used cognitive tests.20Abbreviations: AES-S, Apathy Evaluation Scale–Self-report; APOE-ε4, apolipoprotein E-epsilon 4; BMI, body mass index (calculated as weight in kilograms divided by height in meters squared); CI, confidence interval; CT, computed tomography; FSS, Fatigue Severity Scale; HADS, Hospital Anxiety and Depression Scale; MRI, magnetic resonance imaging; OR, odds ratio; TOAST, Trial of Org 10172 in Acute Stroke Treatment.⁎Statistical differences ($P \leq .05$).
## Discussion
We included patients who worked full time at onset of a first-ever mild ischemic stroke. At 12-month follow-up, 4 in 5 patients were back to any kind of employment, either full time or part time, of whom 3 in 4 worked full time. Reduced functional status (mRS score >1) at follow-up and having diabetes mellitus were associated with not returning to full-time work after 12 months. Female sex, NIHSS score at discharge, and a high score on anxiety, depression, and fatigue scales were also associated with not returning to full-time work, although these associations were only seen in the unadjusted regression models.
It has previously been reported that approximately $50\%$ of patients return to work 1 year after stroke, and approximately $60\%$ return 2 years after stroke.4,19,42 A greater percentage of our patients returned to paid work. This may not fully be explained by the young age in our sample because other studies4 also report on an average age between 37 and 55 years. Tables 1 and 3 show that the percentage of patients 45 years or younger ($\frac{14}{26}$ patients [$54\%$]) returning to full-time work is nearly the same as the percentage of patients older than 45 years ($\frac{33}{58}$ [$57\%$]). Generally, young working patients wish to return to their job and daily life as soon as possible after stroke.19 Older employees may be unable to work because of age-related changes or other health issues or prefer not to work as much as they did before stroke onset.43 On the other hand, it is possible that employers as well as the rehabilitation system focus more on younger than older working patients.
We found that low functional status at 12-month follow-up, operationalized as an mRS score >1, was an independent associative factor of not returning to full-time work. This may be because high mRS score expresses a higher dependence in daily life, which hinders patients from returning to work. An Indian study44 showed that functional disability after stroke and the type of job can determine return to work rather than psychosocial factors such as depression and anxiety. In addition, functional level at discharge19, 1 month45, and 3 months46 post stroke can affect work participation.
In patients with stroke, diabetes mellitus has been shown to be both a risk factor for cognitive impairment47 and reduced ability to return to work.6,8,19,43 We found that diabetes is borderline significantly associated with not returning to full-time work, but we did not find any significant relationship between cognitive impairment and unemployment 12 months after stroke. Anxiety, depressive symptoms, and fatigue were related to not returning to full-time work but only in the unadjusted analyses. Such invisible emotional impairments have previously been described as factors affecting both quality of life48, work, and social participation10,42,46,49 after stroke.
High education (especially at university level) is related to high socioeconomical status, private insurance, and high income, and these factors are associated with a higher probability of returning to work up to 4 years after stroke onset19,45,50. In our study, we did not show that educational level was associated with returning to full-time work. However, other socioeconomic factors, such as patients’ income and profession were not recorded, and so the importance of these factors on returning to full-time work remains unclear. According to data from several studies19,43,45,50, men seem to return to work sooner than women. One explanation might be that women are less likely to have a favorable outcome51 and a good health-related quality of life48 after stroke than men. In our study there were fewer women ($56\%$) with a favorable functional outcome (mRS score 0-1) than men ($79\%$).
Some studies have identified a relationship between living alone and not returning to any kind of paid work.44,52 Possible explanations for these findings could be that people who live alone after stroke may have more anxiety and less support and encouragement from a partner in their daily life52. We did not, however, find that living alone at stroke onset predicts not returning to full-time work.
The effect of reperfusion therapy on the outcome after stroke is widely investigated during the last years12, but there are only a few studies15,16 that describe that treatment of acute ischemic stroke with intravenous thrombolysis may be a positive predictor of returning to full-time work after stroke. Approximately $50\%$ of our patients treated with intravenous thrombolysis returned to full-time work 1 year later, but the association between thrombolysis and return to work was not statistically significant. This may be because of low power related to few patients or the study sample with patients who had a high functional level expressed as mRS score ≤1 at discharge ($74\%$).
## Study limitations and strengths
One limitation was that only 1 in 5 participants were female. The large number of male patients younger than 70 years, however, is in line with stroke epidemiology.53 One of our inclusion criteria was that patients were previously cognitively intact. To reduce the possibility of underreported prestroke cognitive decline in some patients, the information about cognitive health was obtained through interviews with patients, their next of kin, and their family physicians and by reviewing patients’ medical records. The dependents of all patients filled in the Informant Questionnaire on Cognitive Decline in the Elderly.25 Strengths of the study are the clear diagnosis and classification of a first-ever mild stroke. We used validated and widely used cognitive and emotional instruments to evaluate several cognitive and emotional domains. The tests were performed by either a physician, occupational therapist, or nurse. We did not perform any intrarater reliability tests of the tools used in our study, and this may represent a limitation.
Almost all patients aged 18-70 years admitted to the 2 stroke units during the inclusion period and who had been evaluated as eligible for inclusion, wished to participate. Only 3 of the included patients withdrew their consent before follow-up. Consequently, attrition bias was low.
## Conclusions
Younger patients can have difficulties with returning to paid work even after a mild stroke without visible neurologic deficits. Low functional level that persists 12 months after stroke may be related to not returning to full-time work. Patients with diabetes mellitus, female patients, and patients with a higher score on fatigue, anxiety, and depression scales may also be at risk of not returning to full-time job post stroke. We suggest that in future studies and clinical practice, working patients should be offered a poststroke follow-up with particular focus on identifying factors that can affect participation in work and social life.
## Supplier
a.SPSS versions 27.0 and 28.0; IBM.
## Supplementary materials
Image, application 1
## References
1. Brydsten A, Gustafsson PE, Hammarström A, San Sebastian M. **Does contextual unemployment matter for health status across the life course? A longitudinal multilevel study exploring the link between neighbourhood unemployment and functional somatic symptoms**. *Health Place* (2017) **43** 113-120. PMID: 27997864
2. Bacikova-Sleskova M, van Dijk JP, Geckova AM. **The impact of unemployment on school leavers' perception of health. Mediating effect of financial situation and social contacts?**. *Int J Public Health* (2007) **52** 180-187. PMID: 17958285
3. Alaszewski A, Alaszewski H, Potter J, Penhale B.. **Working after a stroke: survivors' experiences and perceptions of barriers to and facilitators of the return to paid employment**. *Disabil Rehabil* (2007) **29** 1858-1869. PMID: 17852252
4. Edwards JD, Kapoor A, Linkewich E, Swartz RH.. **Return to work after young stroke: a systematic review**. *Int J Stroke* (2018) **13** 243-256. PMID: 29189108
5. Carcel C, Farnbach S, Essue BM. **Returning to unpaid work after stroke: the psychosocial outcomes in stroke cohort study**. *Cerebrovasc Dis* (2019) **47** 1-7. PMID: 30654379
6. van der Kemp J, Kruithof WJ, Nijboer TCW, van Bennekom CAM, van Heugten C, Visser-Meily JMA.. **Return to work after mild-to-moderate stroke: work satisfaction and predictive factors**. *Neuropsychol Rehabil* (2019) **29** 638-653. PMID: 28441897
7. Westerlind E, Persson HC, Sunnerhagen KS.. **Return to work after a stroke in working age persons; a six-year follow up**. *PLoS One* (2017) **12**
8. Glozier N, Hackett ML, Parag V, Anderson CS.. **The influence of psychiatric morbidity on return to paid work after stroke in younger adults: the Auckland Regional Community Stroke (ARCOS) Study, 2002 to 2003**. *Stroke* (2008) **39** 1526-1532. PMID: 18369172
9. Radman N, Staub F, Aboulafia-Brakha T, Berney A, Bogousslavsky J, Annoni JM.. **Poststroke fatigue following minor infarcts: a prospective study**. *Neurology* (2012) **79** 1422-1427. PMID: 22955128
10. Pihlaja R, Uimonen J, Mustanoja S, Tatlisumak T, Poutiainen E.. **Post-stroke fatigue is associated with impaired processing speed and memory functions in first-ever stroke patients**. *J Psychosom Res* (2014) **77** 380-384. PMID: 25218164
11. Wardlaw JM, Murray V, Berge E, del Zoppo GJ.. **Thrombolysis for acute ischaemic stroke**. *Cochrane Database Syst Rev* (2014) **7**
12. Lindekleiv H, Berge E, Bruins Slot KM, Wardlaw JM. **Percutaneous vascular interventions versus intravenous thrombolytic treatment for acute ischaemic stroke**. *Cochrane Database Syst Rev* (2018) **10**
13. Manning NW, Warne CD, Meyers PM.. **Reperfusion and clinical outcomes in acute ischemic stroke: systematic review and meta-analysis of the stent-retriever-based, early window endovascular stroke trials**. *Front Neurol* (2018) **9** 301. PMID: 29867718
14. Cerasuolo JO, Mandzia J, Cipriano LE. **Intravenous thrombolysis after first-ever ischemic stroke and reduced incident dementia rate**. *Stroke* (2022) **53** 1170-1177. PMID: 34965738
15. Stefanovic Budimkic M, Pekmezovic T, Beslac-Bumbasirevic L. **Return to paid work after ischemic stroke in patients treated with intravenous thrombolysis**. *Neuroepidemiology* (2016) **46** 114-117. PMID: 26784442
16. Butt JH, Kragholm K, Kruuse C. **Workforce attachment after ischemic stroke - the importance of time to thrombolytic therapy**. *J Stroke Cerebrovasc Dis* (2021) **30**
17. Goldstein LB, Bertels C, Davis JN.. **Interrater reliability of the NIH stroke scale**. *Arch Neurol* (1989) **46** 660-662. PMID: 2730378
18. Röding J, Lindström B, Malm J, Ohman A.. **Frustrated and invisible–younger stroke patients' experiences of the rehabilitation process**. *Disabil Rehabil* (2003) **25** 867-874. PMID: 12851097
19. Morris R.. **The psychology of stroke in young adults: the roles of service provision and return to work**. *Stroke Res Treat* (2011) **2011**
20. Ashley KD, Lee LT, Heaton K.. **Return to work among stroke survivors**. *Workplace Health Saf* (2019) **67** 87-94. PMID: 30616463
21. Aarnio K, Rodríguez-Pardo J, Siegerink B. **Return to work after ischemic stroke in young adults: a registry-based follow-up study**. *Neurology* (2018) **91** e1909-e1917. PMID: 30315074
22. Vlachos G, Ihle-Hansen H, Wyller TB. **Predictors of cognitive and emotional symptoms 12 months after first-ever mild stroke**. *Neuropsychol Rehabil* (2022 Feb 24)
23. Vlachos G, Ihle-Hansen H, Bruun Wyller T. **Cognitive and emotional symptoms in patients with first-ever mild stroke: the syndrome of hidden impairments**. *J Rehabil Med* (2021) **53** jrm00135. PMID: 33145603
24. Fischer U, Baumgartner A, Arnold M. **What is a minor stroke?**. *Stroke* (2010) **41** 661-666. PMID: 20185781
25. Jorm AF.. **A short form of the Informant Questionnaire on Cognitive Decline in the Elderly (IQCODE): development and cross-validation**. *Psychol Med* (1994) **24** 145-153. PMID: 8208879
26. Aho K, Harmsen P, Hatano S, Marquardsen J, Smirnov VE, Strasser T.. **Cerebrovascular disease in the community: results of a WHO collaborative study**. *Bull World Health Organ* (1980) **58** 113-130. PMID: 6966542
27. Adams HP, Bendixen BH, Kappelle LJ. **Classification of subtype of acute ischemic stroke. Definitions for use in a multicenter clinical trial. TOAST. Trial of Org 10172 in Acute Stroke Treatment**. *Stroke* (1993) **24** 35-41. PMID: 7678184
28. Sulter G, Steen C, De Keyser J.. **Use of the Barthel Index and modified Rankin scale in acute stroke trials**. *Stroke* (1999) **30** 1538-1541. PMID: 10436097
29. Rankin J.. **Cerebral vascular accidents in patients over the age of 60. II. Prognosis**. *Scott Med J* (1957) **2** 200-215. PMID: 13432835
30. Strobel C, Engedal K.. (2008)
31. Shulman KI, Pushkar Gold D, Cohen CA, Zucchero CA. **Clock-drawing and dementia in the community: a longitudinal study**. *Int J Geriatr Psychiatry* (1993) **8** 487-496
32. Tombaugh TN.. **Trail Making Test A and B: normative data stratified by age and education**. *Arch Clin Neuropsychol* (2004) **19** 203-214. PMID: 15010086
33. Egeland J, Landro NI, Tjemsland E, Walbaekken K.. **Norwegian norms and factor-structure of phonemic and semantic word list generation**. *Clin Neuropsychol* (2006) **20** 716-728. PMID: 16980257
34. Fastenau PS, Denburg NL, Hufford BJ.. **Adult norms for the Rey-Osterrieth Complex Figure Test and for supplemental recognition and matching trials from the Extended Complex Figure Test**. *Clin Neuropsychol* (1999) **13** 30-47. PMID: 10937646
35. Delis DC, Kramer JH, Kaplan E. (1987)
36. Crum RM, Anthony JC, Bassett SS, Folstein MF.. **Population-based norms for the Mini-Mental State Examination by age and educational level**. *JAMA* (1993) **269** 2386-2391. PMID: 8479064
37. Sachdev P, Kalaria R, O'Brien J. **Diagnostic criteria for vascular cognitive disorders: a VASCOG statement**. *Alzheimer Dis Assoc Disord* (2014) **28** 206-218. PMID: 24632990
38. Aben I, Verhey F, Lousberg R, Lodder J, Honig A.. **Validity of the Beck Depression Inventory, Hospital Anxiety and Depression Scale, SCL-90, and Hamilton Depression Rating Scale as screening instruments for depression in stroke patients**. *Psychosomatics* (2002) **43** 386-393. PMID: 12297607
39. Cumming TB, Packer M, Kramer SF, English C.. **The prevalence of fatigue after stroke: a systematic review and meta-analysis**. *Int J Stroke* (2016) **11** 968-977. PMID: 27703065
40. Andersson S, Krogstad JM, Finset A.. **Apathy and depressed mood in acquired brain damage: relationship to lesion localization and psychophysiological reactivity**. *Psychol Med* (1999) **29** 447-456. PMID: 10218936
41. Lydersen S.. **Statistical review: frequently given comments**. *Ann Rheum Dis* (2015) **74** 323-325. PMID: 25261576
42. Andersen G, Christensen D, Kirkevold M, Johnsen SP.. **Post-stroke fatigue and return to work: a 2-year follow-up**. *Acta Neurol Scand* (2012) **125** 248-253. PMID: 21692753
43. Busch MA, Coshall C, Heuschmann PU, McKevitt C, Wolfe CD.. **Sociodemographic differences in return to work after stroke: the South London Stroke Register (SLSR)**. *J Neurol Neurosurg Psychiatry* (2009) **80** 888-893. PMID: 19276102
44. Bonner B, Pillai R, Sarma PS, Lipska KJ, Pandian J, Sylaja PN.. **Factors predictive of return to work after stroke in patients with mild-moderate disability in India**. *Eur J Neurol* (2016) **23** 548-553. PMID: 26518615
45. Hackett ML, Glozier N, Jan S, Lindley R. **Returning to paid employment after stroke: the Psychosocial Outcomes in Stroke (POISE) cohort study**. *PLoS One* (2012) **7** e41795. PMID: 22848610
46. Chun HY, Whiteley WN, Dennis MS, Mead GE, Carson AJ.. **Anxiety after stroke: the importance of subtyping**. *Stroke* (2018) **49** 556-564. PMID: 29437982
47. Pendlebury ST, Rothwell PM.. **Prevalence, incidence, and factors associated with pre-stroke and post-stroke dementia: a systematic review and meta-analysis**. *Lancet Neurol* (2009) **8** 1006-1018. PMID: 19782001
48. Sturm JW, Donnan GA, Dewey HM. **Quality of life after stroke: the North East Melbourne Stroke Incidence Study (NEMESIS)**. *Stroke* (2004 Oct) **35** 2340-2345. PMID: 15331799
49. Harris C.. **Return to work after stroke: a nursing state of the science**. *Stroke* (2014) **45** e174-e176. PMID: 25013019
50. Trygged S, Ahacic K, Kåreholt I.. **Income and education as predictors of return to working life among younger stroke patients**. *BMC Public Health* (2011) **11** 742. PMID: 21957999
51. Martínez-Sánchez P, Fuentes B, Fernández-Domínguez J. **Young women have poorer outcomes than men after stroke**. *Cerebrovasc Dis* (2011) **31** 455-463. PMID: 21346351
52. Doucet T, Muller F, Verdun-Esquer C, Debelleix X, Brochard P.. **Returning to work after a stroke: a retrospective study at the Physical and Rehabilitation Medicine Center La Tour de Gassies**. *Ann Phys Rehabil Med* (2012) **55** 112-127. PMID: 22386687
53. Thrift AG, Thayabaranathan T, Howard G. **Global stroke statistics**. *Int J Stroke* (2017) **12** 13-32. PMID: 27794138
|
---
title: Exacerbation of financial burden of insulin and overall glucose‐lowing medications
among uninsured population with diabetes
authors:
- Yilu Lin
- Hui Shao
- Vivian Fonseca
- Lizheng Shi
journal: Journal of Diabetes
year: 2023
pmcid: PMC10036254
doi: 10.1111/1753-0407.13360
license: CC BY 4.0
---
# Exacerbation of financial burden of insulin and overall glucose‐lowing medications among uninsured population with diabetes
## Abstract
Highlights Approximately 7.4 million Americans with diabetes use one or more formulations of insulin and the price of insulin tripled from 2002 to 2013. There are limited studies to explore the impact of insulin price, particularly on the uninsured population. For insured people, there was little impact on the out‐of‐pocket (OOP) payment for patients with diabetes using insulin treatment during the insulin price rise. The burden was borne mainly by insurance. When the high insulin price issue came into the uninsured population, the financial burden became urgent because the consequences of rationing insulin are deadly. After the Affordable Care Act was enacted, the uninsured population had $403.96 and $143.64 more OOP payments than people with public and private insurance, respectively.
### Background
Approximately 7.4 million Americans with diabetes used insulin. This study aimed to document the 10‐year trend of insulin and other glucose‐lowering medications expenditure among insured and uninsured populations and to examine the impact of insulin out‐of‐pocket (OOP) payment and insurance status on glucose‐lowering medication OOP expenditure.
### Methods
We extracted data from the Medical Expenditure Panel Survey (2009–2018) to document trends in the expenditure of insulin among people with diabetes. Total expenditures and OOP spending per person were documented on insulin and noninsulin glucose‐lowering medications among insured and uninsured populations. Multivariable regression was applied to assess the association of insulin OOP payment and insurance status on glucose‐lowering medication OOP expenditure.
### Results
Although insulin usage was stable over the decades, total insulin expenditure almost doubled per person per year after the Affordable Care Act (ACA) regardless of the insurance status. The OOP cost of insulin by the uninsured population increased from $1678 per person per year in the pre‐ACA period to $2800 per person per year in the post‐ACA period. After the ACA was enacted, the uninsured population had $403.96 and $143.64 more on OOP costs than the people with public and private insurance, respectively.
### Conclusion
For insured people, the rising financial burden of insulin was borne mainly by insurance. The uninsured population is bearing a heavy burden due to the high price of insulin. Policymakers should take action to reduce the insulin price and improve the transparency of the insulin pricing process.
## INTRODUCTION
Approximately 34 million Americans were estimated to have been diagnosed with diabetes in 2018. 1 Among them, 7.4 million people with diabetes used at least one formulation of insulin. 1 The population with type 1 diabetes mellitus (T1DM) would have to take insulin indefinitely because their bodies no longer make this hormone. 2 Individuals with type 2 diabetes mellitus (T2DM) could manage the disease with a healthier lifestyle. Insulin therapy is recommended for a person with advanced T2DM when other medications have failed to maintain glycemic control. 3 The rising prices of insulin products are anything but justifiable. One insulin Humalog (Lispro, 10 ml vial) was $21 in 1999, but cost $332 in 2019, reflecting a price increase of more than $1000\%$. 4 The high price being attributed to the “high cost of development” does not apply to insulin because even the latest and most commonly used analog insulin products have been on the market for over 20 years or longer. This soaring price contributed to the total economic burden of diagnosed diabetes, which was estimated to be $327 billion in 2017. 5, 6 One in three Medicare beneficiaries has diabetes, and 3.1 million Part D enrollees require insulin. 7 Around $25\%$ of the diabetes population covered by Medicare reported a reduction in the use of insulin owing to the rising cost. 8 Beyond the economic burden, the rising insulin price also had clinical implications, especially for the most vulnerable subgroups. To lower the burden caused by insulin prices, people with diabetes rationed their insulin. They either skipped insulin injections or did not take enough to prolong each dose. There were people with diabetes who suffered severe complications (eg, diabetic ketoacidosis and end‐stage renal disease) and died owing to insulin rationing with the poor affordability of insulin. 9 People even went to Canada to buy insulin because they could no longer afford insulin in the United States. 10 The medication expenditure could be reimbursed partially by the insurance. Thus, the out‐of‐pocket (OOP) cost matters to every insulin user through the rising insulin price issue. OOP cost includes deductibles, coinsurance, and copayments for covered services plus all costs for services that are not covered. 11 For uninsured and underinsured people, a large proportion of their medical expenses are OOP payments. Therefore, the vulnerable uninsured and underinsured population are more likely to live with restricted insulin access or even higher mortality risk, owing to the high cost of the insulin OOP payments.
The Affordable Care Act (ACA), which was enacted on March 23, 2010, addressed health insurance coverage, health care costs, and preventive care. In 2009, $17\%$ of all adults with diabetes under age 65 were uninsured. After the ACA took effect, that number declined to $5\%$. Among low‐income adults with diabetes, $33\%$ were uninsured before the ACA and $6\%$ were uninsured after. In all, an additional 1.9 million people with diabetes—more than half of whom were low income—gained insurance coverage after the ACA. 12 However, the act did not provide details about whether the expenses of prescription drugs including insulin syringes, insulin pumps, and infusion sets are covered and if these expenses would be applied before or after deductibles are met for those insured. 13 Moreover, those who remained uninsured and underinsured after ACA, as the most vulnerable population, were still suffering from the high price of insulin.
Therefore, there is an urgent need to examine insulin expenditure and its impact on individuals, especially on the uninsured population. Prior studies looked into the insulin spending on either T1DM or T2DM in only cohort or elderly populations. Lipska et al discussed the OOP cost of insulin among the T2DM population from 2000 to 2010 using the Optum Labs Data Warehouse, private insurance claim data. 14 *There is* another research group focused on insulin spending by the T1DM population. 15 Researchers also investigated Medicare spending on insulin. 7 Hua's team used the Medical Expenditure Panel Survey (MEPS) data from 2002 to 2013 and examined the price and expenditures of antihyperglycemic medication. 6 After the ACA, the uninsured population is gaining more attention on the rising insulin price. This study aimed to document the 10‐year trend in insulin spending and other glucose‐lowering medications spending among the insured population and the uninsured population and to examine the impact of OOP expenditure for insulin and insurance status on the overall OOP cost for glucose‐lowering medication from 2009 to 2018.
## METHODS
Ten‐year individual and prescription data from the MEPS 2009 to 2018 were extracted to describe and compare trends in the expenditure of insulin and insulin analogs among people with diabetes. The MEPS is a nationally representative household survey supplemented with data collected from pharmacies and other providers. Its Household Component (HC) collects demographic characteristics, health conditions, health status, use of medical services, charges and source of payments, access to care, satisfaction with care, health insurance coverage, income, and employment of each person interviewed. The overall response rate from MEPS‐HC in the study period ranges from $42.7\%$ to $57.2\%$. 16 To derive national estimates, MEPS data were weighted by the proportion of the population they represent. The longitudinal files derived from the respondents to the MEPS Panel and Full Year Consolidated Data File, Medical Conditions File, and Prescribed Medicines File were used for analysis.
Diabetes diagnosis was determined by the MEPS diabetes condition variable. If people were considered as treated for diabetes with medication, they must take glucose‐lowering medication and/or treat diabetes with insulin injections. Multum Lexicon Codes were used to identify glucose‐lowering medications. The non‐nsulin medications included metformin, thiazolidinediones, sulfonylureas, alpha‐glucosidase inhibitors, sodium‐glucose cotransporter‐2 (SGLT‐2) inhibitors, meglitinides, dipeptidyl peptidase‐4 (DPP‐4) inhibitors, amylin analogs, incretin mimetics, and antidiabetic combinations. OOP payment was indicated by “self/family” as a source of payment variable. Overall spending was aggregated by different sources, including Medicaid/Medicare, other public insurances (Veterans/CHAMPVA, Tricare, state and local government, other federal and other public), commercial insurances (private insurance, workers company insurance, and other private insurance), and OOP payments. Insurance status was grouped as insured (any private insurance or public insurance)) and uninsured. All expenditures were adjusted as the 2018 dollars using Consumer Price Index (CPI) for prescription drugs (CPI‐PMED). 17 All the information from MEPS data was self‐reported and validated by the MEPS pharmacy sector.
The study documented the trend of the gross per‐person spending on insulin versus noninsulin medications and total glucose‐lowering medication expenditure on insured and uninsured populations. Descriptive statistics were used to describe individuals' characteristics including demographic characteristics, insurance status, and glucose‐lowering drug use in each year through the decade. Glucose‐lowering medication spending on the use of insulin vs noninsulin glucose‐lowering medications by OOP payment and total payment of insured and uninsured in pre‐ and post‐ACA period was presented. Annual median spending was also presented. Because of the disadvantage of being affected by any single value being too high or too low compared to the rest of the sample, we used median as a representative midpoint measurement instead of mean. Moreover, multivariate regression was conducted to explore the impact of insulin OOP payment and insurance status on overall glucose‐lowering medication OOP expenditure, controlling for demographic characteristics, treated diabetes, insurance types, quantity of insulin used among participants who used insulin, and ACA. ACA has an impact on insurance coverage and in turn, influences medication use. With a concern of collinearity with the ACA dummy variable (year 2013 and after), the calendar year variable was not included in the regression. The regression was clustered by year and variance estimation primary sampling unit provided by the MEPS data file to avoid serial correlation. The interaction term of insurance status and ACA was also included in the regressions. SAS “proc survey” was used for analysis as recommended by MEPS. 18
## RESULTS
A total of 13 696 participants were extracted from the MEPS data from 2009 to 2018. The medication information was recorded annually. The prevalence of insulin purchase has a small growth from $25\%$ to $30\%$. Alpha‐glucosidase inhibitors, amylin analogs, and meglitinides were no longer recorded in the MEPS database since 2013, whereas SGLT‐2 started to take its place after its first agent in the class was approved in 2013. There was a reduction in the proportion of antidiabetic combinations, sulfonylureas, and thiazolidinediones use. The proportions of other medications including DPP‐4 inhibitors, incretin mimetics, and metformin increased over the years (Table 1).
**TABLE 1**
| Period | Measurement | Insulin | Alpha‐glucosidase inhibitors | Amylin analogs | Antidiabetic combinations | Dipeptidyl peptidase 4 inhibitors | Incretin mimetics | Meglitinides | Metformin | SGLT‐2 inhibitors | Sulfonylureas | Thiazolidinediones |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| 2009 | N | 585 | 7 | 3 | 204 | 132 | 36 | 27 | 1097 | | 728 | 326 |
| 2009 | Weighted N | 10 154 279 | 59 346 | 59 584 | 2 360 173 | 1 697 110 | 617 471 | 296 588 | 13 007 507 | | 8 261 593 | 3 371 217 |
| 2009 | % | 25.46 | 0.15 | 0.15 | 5.92 | 4.26 | 1.55 | 0.74 | 32.61 | | 20.71 | 8.45 |
| 2010 | N | 587 | 9 | 4 | 164 | 132 | 39 | 24 | 1053 | | 668 | 252 |
| 2010 | Weighted N | 11 317 965 | 104 835 | 58 048 | 1 931 802 | 1 472 766 | 598 533 | 319 106 | 14 059 910 | | 8 500 617 | 2 931 008 |
| 2010 | % | 27.41 | 0.25 | 0.14 | 4.68 | 3.57 | 1.45 | 0.77 | 34.05 | | 20.59 | 7.10 |
| 2011 | N | 706 | 8 | 2 | 159 | 159 | 46 | 18 | 1250 | | 714 | 207 |
| 2011 | Weighted N | 13 405 112 | 58 872 | 9349 | 1 853 624 | 1 848 915 | 714 732 | 164 260 | 15 194 909 | | 8 670 611 | 2 307 894 |
| 2011 | % | 30.31 | 0.13 | 0.02 | 4.19 | 4.18 | 1.62 | 0.37 | 34.36 | | 19.60 | 5.22 |
| 2012 | N | 730 | 12 | 3 | 178 | 196 | 43 | 21 | 1310 | | 750 | 136 |
| 2012 | Weighted N | 12 813 380 | 106 348 | 41 911 | 1 753 823 | 2 127 144 | 648 374 | 217 240 | 14 475 587 | | 8 396 786 | 1 594 361 |
| 2012 | % | 30.38 | 0.25 | 0.10 | 4.16 | 5.04 | 1.54 | 0.52 | 34.32 | | 19.91 | 3.78 |
| 2013 | N | 758 | | | 167 | 196 | 65 | | 1366 | | 708 | 91 |
| 2013 | Weighted N | 12 348 071 | | | 1 394 241 | 2 028 807 | 994 253 | | 15 441 175 | | 8 316 737 | 1 076 205 |
| 2013 | % | 29.68 | | | 3.35 | 4.88 | 2.39 | | 37.12 | | 19.99 | 2.59 |
| 2014 | N | 773 | | | 157 | 198 | 70 | | 1377 | 44 | 654 | 106 |
| 2014 | Weighted N | 16 115 118 | | | 1 973 929 | 1 949 723 | 1 149 108 | | 17 624 692 | 488 226 | 8 763 048 | 1 381 002 |
| 2014 | % | 32.59 | | | 3.99 | 3.94 | 2.32 | | 35.65 | 0.99 | 17.72 | 2.79 |
| 2015 | N | 777 | | | 180 | 217 | 87 | | 1405 | 88 | 706 | 100 |
| 2015 | Weighted N | 13 551 525 | | | 1 939 384 | 2 522 495 | 1 387 565 | | 16 756 480 | 1 202 470 | 8 475 283 | 1 062 595 |
| 2015 | % | 28.90 | | | 4.14 | 5.38 | 2.96 | | 35.73 | 2.56 | 18.07 | 2.27 |
| 2016 | N | 797 | | | 175 | 253 | 108 | | 1451 | 101 | 675 | 101 |
| 2016 | Weighted N | 14 222 474 | | | 1 592 458 | 3 014 126 | 1 340 502 | | 18 039 777 | 1 409 845 | 8 387 497 | 1 221 826 |
| 2016 | % | 28.89 | | | 3.23 | 6.12 | 2.72 | | 36.65 | 2.86 | 17.04 | 2.48 |
| 2017 | N | 790 | | | 136 | 259 | 153 | | 1379 | 119 | 633 | 90 |
| 2017 | Weighted N | 14 433 186 | | | 1 724 238 | 2 743 806 | 1 818 651 | | 17 616 713 | 1 682 094 | 8 199 396 | 1 084 780 |
| 2017 | % | 29.27 | | | 3.50 | 5.57 | 3.69 | | 35.73 | 3.41 | 16.63 | 2.20 |
| 2018 | N | 769 | | | 143 | 263 | 192 | | 1489 | 134 | 599 | 102 |
| 2018 | Weighted N | 15 301 094 | | | 1 793 750 | 3 042 702 | 2 706 130 | | 18 793 944 | 1 720 856 | 7 993 954 | 1 287 081 |
| 2018 | % | 29.07 | | | 3.41 | 5.78 | 5.14 | | 35.70 | 3.27 | 15.19 | 2.45 |
By comparing the median OOP expenditure and overall medication expenditure between the insured population and uninsured population, the insulin OOP cost of individuals with insurance even decreased by approximately $300, and their overall insulin expenditure doubled from $11 213 to $20 652 because of the implementation of ACA and Medicaid expansion. In contrast, for the uninsured population, the insulin OOP payment had an overall increasing trend from $1678 to $2800. The noninsulin OOP payments and the overall OOP payment, however, decreased after the ACA for uninsured population. These results suggested that the increased insulin expenditure carried a larger impact on the uninsured population because the main payment method for these people was OOP. When looking into the noninsulin medication costs, noninsulin OOP expenditure and noninsulin total expenditure decreased regardless of insurance status (Table 2). The medication OOP overall (noninsulin OOP payment and insulin OOP payment together) decreased by approximately $33\%$ (from $818 to $546) and $47\%$ (from $804 to $429) for the insured population and uninsured population, respectively (Table A1). It indicated that the decrease in glucose‐lowering medication OOP costs was contributed by noninsulin OOP payment because there was a large drop in noninsulin OOP. Moreover, though there was little variation in insulin OOP payment for the individuals with insurance coverage, the insulin OOP payment had a high increase from the uninsured population. Jonckheere–Terpstra test was conducted and indicated the trend was statistically significant.
**TABLE 2**
| Expenditure type | Period | Insured sample | Insured sample.1 | Insured sample.2 | Uninsured sample | Uninsured sample.1 | Uninsured sample.2 |
| --- | --- | --- | --- | --- | --- | --- | --- |
| Expenditure type | Period | Insulin | Non‐insulin | Total | Insulin | Noninsulin | Total |
| OOP | Pre‐ACA b | 1595 | 348 | 680 | 1678 | 331 | 678 |
| OOP | Post‐ACA c | 1226 | 205 | 489 | 2800 | 289 | 543 |
| Total expenditure a | Pre‐ACA b | 11 213 | 983 | 4134 | 4680 | 439 | 1220 |
| Total expenditure a | Post‐ACA c | 20 652 | 672 | 6417 | 7122 | 396 | 1077 |
In the regression analysis, age was transformed into a categorical variable by a cutoff point of 65 years old based on the Medicare coverage age limit. The regression explored the impact of insulin OOP payment and insurance status on glucose‐lowering medication OOP expenditure. Compared with White people, the Asian population spent $246.37 more on glucose‐lowering medication, whereas the Black and Hispanic populations spent $190.95 and $238.78 less on glucose‐lowering drugs OOP cost, respectively. The implementation of ACA increased the overall glucose‐lowering medications expenditure by $266.30. After the ACA, people with public and private insurance had $403.96 and $143.64 lower OOP payments than the uninsured population, respectively. These statistics indicated that the ACA did alleviate the OOP cost for the insured population; however, for the uninsured population, the impact of the ACA was exacerbated (Table 3).
**TABLE 3**
| Parameter | Estimate | Standard error | p Value |
| --- | --- | --- | --- |
| Age ≥ 65 years | ref | | |
| Age < 65 years | 60.59 | 79.77 | .4476 |
| Male | ref | | |
| Female | −75.92 | 69.05 | .2716 |
| Race White | ref | | |
| Race Asian | 246.37 | 463.93 | .5954 |
| Race Black | −190.95 | 76.76 | .0129 |
| Race Hispanic | −238.78 | 77.98 | .0022 |
| Race Other | −179.28 | 129.24 | .1655 |
| Treated diabetes with diet | ref | | |
| Treated diabetes with medication | 81.37 | 92.85 | .3809 |
| Insulin quantity | −1.33 | 0.66 | .0433 |
| Uninsured | ref | | |
| Any private insurance | 91.33 | 151.99 | .5480 |
| Public insurance only | 119.45 | 176.47 | .4985 |
| ACA | 266.30 | 245.45 | .2781 |
| Uninsured*ACA | ref | | |
| Any private insurance*ACA | −143.64 | 258.45 | .5791 |
| Public insurance only*ACA | −403.96 | 273.3 | .1395 |
| Insulin OOP | 1.01 | 0.0 | <.0001 |
## DISCUSSION
In this study, we observed an increasing trend in insulin expenditure. For people with insurance, insulin OOP cost was not greatly affected because health insurance payments covered the gap generated by the rapidly increasing insulin expenditure. From the uninsured population side, however, the OOP payment of this essential medication increased 1.7 times along with the increased insulin price. The ACA and its provisions like Medicaid Expansion and Marketplace subsidies did remove a large proportion of the population from uninsured status. The population that remained uninsured, however, was the most vulnerable group that was sensitive to the disease's financial burden. 19 The magnitude of estimated coefficients in the regression analysis indicated that health insurance, public or private, played an important role in the overall glucose‐lowering medication OOP expenditure. But for the uninsured population, high OOP expenditure caused by the high price of insulin treatment posed a large barrier that may decrease their adherence and threaten their lives.
Our findings are consistent with a study by the Centers for Disease Control and Prevention that used the private insurance MarketScan Claims database showing the increase in insulin spending was covered by payers. 20 For example, the policy on closing the Medicare Part D coverage gap traded off the impact of higher insulin prices, which kept the OOP cost in check. 21 Nevertheless, insurance companies could use their buyer power to bargain and receive discounts and rebates from the pharmacy companies on insulins in return for their formulary status. 22 Our study suggested that the rising insulin price became a big problem for uninsured people who would have to pay full price for insulin prescriptions, whereas people with insurance could be better shielded by the insurance coverage plan. It was always the uninsured population who were more vulnerable. One study confirmed our conclusion for the insured people that used claims data of privately insured enrollees, which indicated high reimbursement proportion. 14 A recent study found that $14.1\%$ of insulin users spent $40\%$ of their postsubsistence family income on insulin alone over 1 year, representing almost 1.2 million people. Approximately two thirds of people who experienced this high spending on insulin were Medicare beneficiaries. 23 Insulin expenditure is a great burden even for the insured population with government assistance. How this high OOP payment affects the uninsured population is imaginable. It was appalling to see a six‐times increase in the insulin OOP cost for the uninsured in our study. Moreover, the national health reform reduced the burden for people with preexisting conditions and increased insurance coverage; however, those unable to pay the insurance premium still existed and these people are the most vulnerable population. Therefore, previous studies and our study raised the concern about the uninsured/underinsured population because they had the high OOP cost of insulin without a third‐party payer. Considering the great impact on uninsured low‐income, minority populations who need insulin treatment and the unreasonably high price for this life‐saving medication, the disparities and inequities in insulin access need to be addressed through multifaceted national actions.
Both government and industry are making efforts to satisfy the need for insulin and to decrease the price of insulin in many ways. Congress and state governments are discussing action to relieve the burden of rising insulin prices. 24, 25, 26, 27 The Centers for Medicare & Medicaid *Services is* offering lower insulin costs through enhanced Part D prescription drug plans will have access to a broad set of insulins at a maximum $35 copay for a month's supply, from the beginning of the year through the Part D coverage gap. 28 Pharmaceutical companies offer support programs to defray the costs for those who are experiencing high OOP costs, including coupons, free samples, and patient assistance programs (PAPs). 29, 30, 31, 32 However, for the uninsured population, the accessibility of insulin was still limited because of the high price. Finding information about these PAPs can be difficult, and PAPs often have complicated income, insurance, and prescription requirements even with the information collected.
Our study had several limitations. We used patient‐reported data from a national survey with a response rate ranging from $42.7\%$ to $57.2\%$. The survey participants have the possibility of faulty recall or other ascertainment bias and volunteer bias. We did observe certain outliers in Table 2, although these outliers did not affect the overall increasing/decreasing trend. One explanation of this might be the different sample: MEPS data were not able to follow up with households for a very long period, which may change the characteristics largely in a certain year and caused the outliers. 33 Furthermore, the physical condition and diabetes duration were not taken into consideration as control variables in our study because self‐reports may not perfectly conform to diagnoses made by physicians. Insulin use adherence and appropriateness are important factors; however, the information is not contained in the MEPS data. Our study controlled only for insulin quantity, which may not get a full picture of the research question. Usage of free medication samples, insurance premiums, and the amount of tax paid were not included in our analyses because of the data availability. However, the payment amount was validated with pharmacy records for prescription drugs., which ensured our data quality and the credibility of the results.
## CONCLUSION
Our study found that the price of insulin and its analogs increased. For the insured cohort, the financial burden of this rapid price change was covered by insurance. The insulin OOP payment among insured individuals with diabetes was stable. When the high insulin price issue came to the uninsured population, the financial burden became urgent because the consequences of rationing insulin are significant. The irrational increase in insulin price, however, remained an unresolved issue after ACA. It is imperative to slow down the increasing expenditure trend by reducing insulin costs. Additionally, more supportive policies should be implemented for the uninsured diabetes population to get enough necessary insulin usage.
## ACKNOWLEGEMENTS
None.
## FUNDING STATEMENT
No funding.
## DISCLOSURE
No conflict of interest.
## ETHICS APPROVAL STATEMENT
Not applicable.
## PATIENT CONSENT STATEMENT
Not applicable.
## PERMISSION TO REPRODUCE MATERIAL FROM OTHER SOURCES
Not applicable.
## CLINICAL TRIAL REGISTRATION
Not applicable.
## DATA AVAILABILITY STATEMENT
MEPS is a publicly available survey data source.
## References
1. **United States Diabetes Surveillance System. Accessed January 10, 2021**
2. **What type of diabetes do I have? Accessed January 8, 2021**
3. Petznick A. **Insulin management of type 2 diabetes mellitus**. *Am Fam Physician* (2011.0) **84** 183-190. PMID: 21766768
4. **Accessed January 12, 2021**
5. **Economic costs of diabetes in the US in 2017**. *Diabetes Care* (2018.0) **41** 917-928. PMID: 29567642
6. Hua X, Carvalho N, Tew M, Huang ES, Herman WH, Clarke P. **Expenditures and prices of antihyperglycemic medications in the United States: 2002‐2013**. *JAMA* (2016.0) **315** 1400-1402. DOI: 10.1001/jama.2016.0126
7. Cubanski J, Neuman T, True S, Damico A. *How Much Does Medicare Spend on Insulin* (2019.0)
8. Herkert D, Vijayakumar P, Luo J. **Cost‐related insulin underuse among patients with diabetes**. *JAMA Intern Med* (2019.0) **179** 112-114. PMID: 30508012
9. **Diabetes Voice. Accessed January 8, 2021**
10. **RSS. Accessed January 8, 2021**
11. **Out‐of‐Pocket Costs. Accessed April 22, 2022**
12. Myerson R, Romley J, Chiou T, Peters AL, Goldman D. **The affordable care act and health insurance coverage among people with diagnosed and undiagnosed diabetes: data from the National Health and Nutrition Examination Survey**. *Diabetes Care* (2019.0) **42** e179-e180. PMID: 31548249
13. Burge MR, Schade DS. **Diabetes and the Affordable Care Act**. *Diabetes Technol Ther* (2014.0) **16** 399-413. DOI: 10.1089/dia.2014.0171
14. Lipska KJ, Ross JS, Van Houten HK, Beran D, Yudkin JS, Shah ND. **Use and out‐of‐pocket costs of insulin for type 2 diabetes mellitus from 2000 through 2010**. *JAMA* (2014.0) **311** 2331-2333. PMID: 24915266
15. Biniek JF, Johnson W. *Spending on Individuals with Type 1 Diabetes and the Role of Rapidly Increasing Insulin Prices* (2019.0)
16. **Medical Expenditure Panel Survey Household Component Response Rates. Accessed January 12, 2021**
17. **Using Appropriate Price Indices for Analyses of Health Care Expenditures or Income across Multiple Years. Accessed January 12, 2021**
18. **Using Statistical Software Packages to Produce Estimates from MEPS Data Files. Accessed December 7, 2022**
19. Rachel Garfield KO. **The Uninsured and the ACA: A Primer ‐ Key Facts about Health Insurance and the Uninsured amidst Changes to the Affordable Care Act. Accessed June 13, 2022**
20. Shao H, Laxy M, Benoit SR, Cheng YJ, Gregg E, Zhang P. **1239‐P: trend in total payment and out‐of‐pocket payment on a yearly supply of oral antidiabetic drug types among US adults with Private Health Insurance from 2003 to 2016**. *Diabetes* (2019.0)
21. Tseng CW, Masuda C, Chen R, Hartung DM. **Impact of higher insulin prices on out‐of‐pocket costs in Medicare part D**. *Diabetes Care* (2020.0) **43** e50-e51. DOI: 10.2337/dc19-1294
22. **Commonwealth Fund. Accessed January 12, 2021**
23. Bakkila BF, Basu S, Lipska KJ. **Catastrophic spending on insulin in the United States, 2017‐18**. *Health Aff (Millwood)* (2022.0) **41** 1053-1060. DOI: 10.1377/hlthaff.2021.01788
24. **The New York Times. Accessed January 8, 2021**
25. **Congressional Diabetes Caucus. Accessed January 8, 2021**
26. **Governor Walz Signs Alec Smith Insulin Affordability Act. Accessed January 10, 2021**
27. **HB21‐1307 Prescription Insulin Pricing And Access. Accessed May 6, 2022**
28. **CMS. Accessed January 10, 2021**
29. **AACE Prescription Affordability Resource Center. Accessed May 5, 2022**
30. **USA Today. Accessed January 10, 2021**
31. **Sanofi Provides Unprecedented Access to its Insulins for One Set Monthly Price. Accessed January 10, 2021**
32. **Lilly again Reduces List Price of Insulin Lispro Injection as Latest Change to Affordability Options. Accessed May 5, 2022**
33. **Agency for Healthcare Research and Quality R, Md. MEPS‐HC Panel Design and Data Collection Process. December 7, 2022**
|
---
title: The Association of Retinal age gap with metabolic syndrome and inflammation
authors:
- Zhuoting Zhu
- Dan Liu
- Ruiye Chen
- Wenyi Hu
- Huan Liao
- Katerina Kiburg
- Jason Ha
- Shuang He
- Xianwen Shang
- Yu Huang
- Wei Wang
- Honghua Yu
- Xiaohong Yang
- Mingguang He
journal: Journal of Diabetes
year: 2023
pmcid: PMC10036256
doi: 10.1111/1753-0407.13364
license: CC BY 4.0
---
# The Association of Retinal age gap with metabolic syndrome and inflammation
## Abstract
Highlights Retinal age gap is a non‐invasive and cost‐effective aging biomarker. Retinal age gap was significantly associated with metabolic syndrome and inflammation. Retinal age gap could be potentially used as a screening tool for metabolic syndrome in large populations.
### Background
Metabolic syndrome (MetS) is a clustering of cardiometabolic components, posing tremendous burdens in the aging society. Retinal age gap has been proposed as a robust biomarker associated with mortality and Parkinson's disease. Although MetS and chronic inflammation could accelerate the aging process and increase the risk of mortality, the association of the retinal age gap with MetS and inflammation has not been examined yet.
### Methods
Retinal age gap (retina‐predicted age minus chronological age) was calculated using a deep learning model. MetS was defined as the presence of three or more of the following: central obesity, hypertension, dyslipidemia, hypertriglyceridemia, and hyperglycemia. Inflammation index was defined as a high‐sensitivity C‐reactive protein level above 3.0 mg/L. Logistic regression models were used to examine the associations of retinal age gaps with MetS and inflammation.
### Results
We found that retinal age gap was significantly associated with MetS and inflammation. Specifically, compared to participants with retinal age gaps in the lowest quartile, the risk of MetS was significantly increased by $10\%$ and $14\%$ for participants with retinal age gaps in the third and fourth quartile (odds ratio [OR]:1.10; $95\%$ confidence interval [CI], 1.01,1.21;, $$p \leq .030$$; OR: 1.14, $95\%$ CI, 1.03,1.26; $$p \leq .012$$, respectively). Similar trends were identified for the risk of inflammation and combined MetS and inflammation.
### Conclusion
We found that retinal age gaps were significantly associated with MetS as well as inflammation. Given the noninvasive and cost‐effective nature and the efficacy of the retinal age gap, it has great potential to be used as a screening tool for MetS in large populations.
## INTRODUCTION
Metabolic syndrome (MetS) is a clustering of cardiometabolic components, including abdominal obesity, hypertension, elevated glucose levels, as well as dyslipidemia, which are highly prognostic of type 2 diabetes mellitus and cardiovascular diseases (CVD), causing morbidity and mortality eventually. 1, 2 Notably, inflammation has been reported to play a crucial role in the pathology of MetS. 3, 4 The incidence of MetS rises exponentially as the population ages, posing a tremendous burden to individuals and families in the aging society. 5 Early detection and risk stratification of MetS are needed to improve prevention and early intervention strategies of the diabetes and CVDs. A growing number of studies have investigated several potential screening tools for the MetS including anthropometric measurements and blood tests. 6, 7 Among all the tools proposed, noninvasive measurements such as body mass index, waist circumference, and blood pressure stand out. 8, 9 However, measurement errors and ethnicity heterogeneity have limited their further application for screening in large populations. 10, 11, 12 Therefore, a novel, reliable, and noninvasive method is urgently warranted to accurately screening for MetS.
Retinal age has been recently proposed as a novel reliable aging biomarker. Our research group leveraged deep learning (DL) to accurately predict chronological age based on retinal fundus images in healthy populations. 13 Retinal age gap was defined as the difference between retina‐predicted age and chronological age, showing the deviations from normal aging. Our group has verified that the retinal age gap could independently predict future death events and age‐related diseases events including cardiovascular diseases and Parkinson's disease. 13, 14, 15 Taken together, retinal age could reflect the overall health and thus has been considered as a reliable indicator of aging. As age is an independent risk factor for MetS, we hypothesized that retinal age gaps may be associated with MetS.
Herein, we aimed to investigate the associations of retinal age gap with MetS in a large population based on UK Biobank study. In addition, the association between retinal age gap and inflammation as a pathogenic component of MetS was also investigated in the present study.
## Study population
The UK *Biobank is* a population‐based cohort study with over 500 000 participants recruited in the United Kingdom between 2006 and 2010. A total of 22 assessment centers were set across England, Wales, and Scotland. Approximately 9.2 million residents aged 40–69 years within 25 miles of a nearest assessment center were invited to participant in this study, among whom $5.5\%$ were enrolled in the baseline assessment. All participants completed comprehensive health care questionnaires, detailed physical measurements, and biological sample collections. The overall study protocols and data are available elsewhere. 16 This study was restricted to a subset of participants with metabolic syndrome and/or inflammation data available at the initial assessment (March 2006 to December 2010).
## Ethical approval
The UK Biobank Study received approval from the National Information Governance Board for Health and Social Care and the National Health Service North West Multicentre Research Ethics Committee (11/NW/0382). Written informed consent was obtained from all participants in accordance with the Declaration of Helsinki.
## Fundus photography
Ophthalmic examinations were introduced from 2010. LogMAR visual acuity, autorefraction, and keratometry (Tomey RC5000, Tomey GmbH, Nuremberg, Germany); intraocular pressure (Ocular Response Analyzer, Reichert, New York, USA); and paired retinal fundus and optical coherence tomography imaging (OCT, Topcon 3D OCT 1000 Mk2, Topcon Corp, Tokyo, Japan) examinations were performed. A 45‐degree nonmydriatic and nonstereo fundus image centered on the macula with the optic disc included was taken for each eye. A total of 131 238 images from 66 500 participants were obtained from the UK Biobank study. A total of 80 170 images from 46 970 participants who passed the image quality check were used for the following analyses.
## Estimation of retina age and retinal age gap
Retinal age was predicted according to the methods described by Zhu et al. 13 Briefly, only color fundus images were fed into a DL model using an Xception architecture to predict biological age. DL is the advanced subset of machine learning with multiple neural networks and could self‐learn complex representations without requiring human engineering and domain expertise to design feature extractors or calculation methods.
The difference between the retinal age predicted by the DL model and chronological age was defined as the retinal age gap. A positive retinal age gap indicated an “older” appearing retina, whereas a negative one indicated a “younger” appearing retina.
## Metabolic syndrome and inflammation index
MetS was defined as the presence of three or more of the following: central obesity, hypertension, dyslipidemia, hypertriglyceridemia, and hyperglycemia. 1, 17 Waist circumference was measured at the smallest part of the trunk using a 200‐cm tape measure (SECA). Central obesity was defined as a waist circumference of 88 cm or above for women and 102 cm or above for men. Systolic and diastolic blood pressure were measured twice using an IntelliSense blood pressure monitor model HEM‐907XL (Omron) after participants rested for at least 5 min and the averages were taken as the final results. 18 Hypertension was defined as a systolic blood pressure of ≥130 mm Hg and/or a diastolic blood pressure of ≥80 mm Hg or taking antihypertensive drugs. Dyslipidemia was defined as levels of high‐density lipoprotein (HDL) cholesterol <50 mg/dL in women and <40 mg/dL in men. A triglyceride level ≥150 mg/dL was defined as hypertriglyceridemia. Hyperglycemia was defined as fasting blood glucose levels >110 mg/dL or taking hypoglycemic medication or using insulin. Inflammation index was defined as a high‐sensitivity C‐reactive protein (CRP) level >3.0 mg/L.
## Demographic and health variables
Demographic factors and health variables included baseline age, sex, ethnicity (recorded as white and nonwhite), Townsend deprivation indices (an area‐based proxy measure for socioeconomic status), education attainment (recorded as college or university degree, and others), smoking status (recorded as current/previous and never), drinking status (recorded as current/previous and never), physical activity level (recorded as above moderate/vigorous/walking recommendation and not), history of heart disease (angina or heart attack), history of stroke, and general health status (recorded as excellent/good and fair/poor).
## Statistical analyses
Data are presented as mean (SD) or median (interquartile range, IQR) for continuous variables and numbers (percentages) for categorical variables. The differences of the baseline characteristics of the study population among retinal age gap quartiles were compared using analysis of variance and chi‐square test. The differences of the baseline characteristics between the MetS group and non‐MetS group were compared using chi‐square test and/or t test. Logistic regression models were applied to investigate the association between retinal age gap (independent variable) and MetS and inflammation (outcome), respectively. We then investigated associations between retinal age gaps at different quartiles with metabolic syndrome and inflammation. The associations between retina age gap and abdominal obesity, hypertension, elevated serum HDL, elevated serum triglycerides, and hyperglycemia were also examined respectively by logistic regression models. All logistic regression models were adjusted for baseline age, sex, ethnicity, Townsend deprivation indices, educational level, smoking status, drinking status, physical activity level, general health status, history of heart disease, and history of stroke. Odds ratios (OR) with their $95\%$ confidence intervals (CIs) were reported. Variance inflation factors (VIF) procedure was used to test collinearity for all variables and all covariables' VIF were <2. For all the analyses, complete data were used. A two‐sided p value of <.05 indicated statistical significance. Analyses were performed using Stata (version 13, StataCorp, Texas, USA).
## Study sample
Of 35 918 included participants, the mean (SD) age was 56.6 (8.04) years, and 20 002 ($55.7\%$) were women. Table 1 shows baseline characteristics of the study participants overall and stratified by quartiles of retinal age gaps. As shown, participants with a higher quartile of retinal age gaps were more likely to be younger, of female gender, of nonwhite ethnicity, nonsmokers, physical inactivity, less healthy in general status, more deprived, better educated, and without a history of chronic heart diseases and stroke ($p \leq .001$).
**TABLE 1**
| Baseline characteristics | Total | Retinal age gap | Retinal age gap.1 | Retinal age gap.2 | Retinal age gap.3 | Test results | p value |
| --- | --- | --- | --- | --- | --- | --- | --- |
| Baseline characteristics | Total | Q1 | Q2 | Q3 | Q4 | Test results | p value |
| N | 35 918 | 8980 | 8979 | 8980 | 8979 | | ‐ |
| Age, years, mean (SD) | 56.6 (8.04) | 63.1 (4.80) | 59.3 (6.43) | 54.7 (7.34) | 49.9 (6.43) | 7424.45 a | <.001 |
| Sex, n (%) | Sex, n (%) | Sex, n (%) | Sex, n (%) | Sex, n (%) | Sex, n (%) | Sex, n (%) | Sex, n (%) |
| Men | 15 916 (44.3) | 4409 (49.1) | 3970 (44.2) | 3812 (42.4) | 3725 (41.5) | 125.08 b | <.001 |
| Women | 20 002 (55.7) | 4571 (50.9) | 5009 (55.8) | 5168 (57.6) | 5254 (58.5) | 125.08 b | <.001 |
| Ethnicity, n (%) | Ethnicity, n (%) | Ethnicity, n (%) | Ethnicity, n (%) | Ethnicity, n (%) | Ethnicity, n (%) | Ethnicity, n (%) | Ethnicity, n (%) |
| White | 33 480 (93.2) | 8475 (94.4) | 8431 (93.9) | 8318 (92.6) | 8256 (92.0) | 53.42 b | <.001 |
| Nonwhite | 2438 (6.79) | 505 (5.62) | 548 (6.10) | 662 (7.37) | 723 (8.05) | 53.42 b | <.001 |
| Townsend index, mean (SD) | −1.09 (2.96) | −1.45 (2.79) | −1.22 (2.88) | −0.99 (3.02) | −0.69 (3.08) | 108.07 a | <.001 |
| Attainable education, n (%) | Attainable education, n (%) | Attainable education, n (%) | Attainable education, n (%) | Attainable education, n (%) | Attainable education, n (%) | Attainable education, n (%) | Attainable education, n (%) |
| Above college/university | 12 462 (34.7) | 2723 (30.3) | 2976 (33.1) | 3186 (35.5) | 3577 (39.8) | 192.52 b | <.001 |
| Below college/university | 23 456 (65.3) | 6257 (69.7) | 6003 (66.9) | 5794 (64.5) | 5402 (60.2) | 192.52 b | <.001 |
| Smoking status, n (%) | Smoking status, n (%) | Smoking status, n (%) | Smoking status, n (%) | Smoking status, n (%) | Smoking status, n (%) | Smoking status, n (%) | Smoking status, n (%) |
| Never | 19 793 (55.4) | 4869 (54.5) | 4857 (54.3) | 4883 (54.6) | 5184 (58.0) | 33.80 b | <.001 |
| Former/current | 15 945 (44.6) | 4058 (45.5) | 4081 (45.7) | 4056 (45.4) | 3750 (42.0) | 33.80 b | <.001 |
| Drinking status, n (%) | Drinking status, n (%) | Drinking status, n (%) | Drinking status, n (%) | Drinking status, n (%) | Drinking status, n (%) | Drinking status, n (%) | Drinking status, n (%) |
| Never | 1586 (4.43) | 443 (4.94) | 357 (3.98) | 396 (4.42) | 390 (4.36) | 9.94 b | .019 |
| Former/current | 34 223 (95.6) | 8516 (95.1) | 8607 (96.0) | 8554 (95.6) | 8546 (95.6) | 9.94 b | .019 |
| Meeting physical education recommendation, n (%) | Meeting physical education recommendation, n (%) | Meeting physical education recommendation, n (%) | Meeting physical education recommendation, n (%) | Meeting physical education recommendation, n (%) | Meeting physical education recommendation, n (%) | Meeting physical education recommendation, n (%) | Meeting physical education recommendation, n (%) |
| No | 5307 (18.1) | 1132 (15.7) | 1274 (17.4) | 1383 (18.8) | 1518 (20.2) | 56.87 b | <.001 |
| Yes | 24 084 (81.9) | 6096 (84.3) | 6039 (82.6) | 5962 (81.2) | 5987 (79.8) | 56.87 b | <.001 |
| Health status, n (%) | Health status, n (%) | Health status, n (%) | Health status, n (%) | Health status, n (%) | Health status, n (%) | Health status, n (%) | Health status, n (%) |
| Excellent/good | 24 823 (69.5) | 6466 (72.3) | 6304 (70.5) | 6152 (69.0) | 5901 (66.3) | 81.97 b | <.001 |
| Fair/poor | 10 895 (30.5) | 2477 (27.7) | 2644 (29.5) | 2771 (31.0) | 3003 (33.7) | 81.97 b | <.001 |
| History of chronic heart diseases, n (%) | History of chronic heart diseases, n (%) | History of chronic heart diseases, n (%) | History of chronic heart diseases, n (%) | History of chronic heart diseases, n (%) | History of chronic heart diseases, n (%) | History of chronic heart diseases, n (%) | History of chronic heart diseases, n (%) |
| No | 34 383 (95.7) | 8399 (93.5) | 8547 (95.2) | 8667 (96.5) | 8770 (97.7) | 208.98 b | <.001 |
| Yes | 1535 (4.27) | 581 (6.47) | 432 (4.81) | 313 (3.49) | 209 (2.33) | 208.98 b | <.001 |
| History of stroke, n (%) | History of stroke, n (%) | History of stroke, n (%) | History of stroke, n (%) | History of stroke, n (%) | History of stroke, n (%) | History of stroke, n (%) | History of stroke, n (%) |
| No | 35 375 (98.5) | 8793 (97.9) | 8834 (98.4) | 8865 (98.7) | 8883 (98.9) | 35.32 b | <.001 |
| Yes | 543 (1.51) | 187 (2.08) | 145 (1.61) | 115 (1.28) | 96 (1.07) | 35.32 b | <.001 |
The distribution of retinal age gap of the included participants is shown in Figure 1. The mean (SD) and median (IQR) of the retinal age gap were −1.31 (4.82) and −1.18 (−4.18 to 1.79). The retinal age gap was divided into four groups of equal size, with Q1 as the lowest $25\%$ retinal age gap, Q2 as the lowest $25\%$–$50\%$, Q3 as $50\%$–$75\%$, and Q4 as the highest $25\%$ of retinal age gaps. The retinal age gap ranges for the four quartiles are Q1 (−27.9 to −4.18), Q2 (−4.18 to −1.18), Q3 (−1.18 to 1.79), and Q4 (1.79 to 19.0). We also added the fundus photographs of Q1‐Q4 stages in the Figure S1. Compared to the fundus images in Q1, the fundus images in Q4 tended to have more retinal vascular changes including arteriolar narrowing and blood vessel tortuosity.
**FIGURE 1:** *The distribution of retinal age gap of the included participants.*
## Metabolic syndrome and inflammation
Table 2 presents baseline characteristics of the study participants stratified by MetS. Compared to the nonmetabolic syndrome group, participants with MetS tend to be older, of male gender, former/current smokers, nondrinkers, physical inactivity and with greater deprivation, poor education, and poor general health status, and with a history of chronic heart diseases and stroke ($p \leq 0.001$).
**TABLE 2**
| Baseline characteristics | Metabolic syndrome group | Nonmetabolic syndrome group | Test result | p value |
| --- | --- | --- | --- | --- |
| N | 7921 | 27 997 | | ‐ |
| Age, years, mean (SD) | 57.9 (0.09) | 56.4 (0.05) | −14.25 a | <.001 |
| Sex, n (%) | Sex, n (%) | Sex, n (%) | Sex, n (%) | Sex, n (%) |
| Men | 3837 (48.4) | 12 079 (43.1) | 70.20 b | <.001 |
| Women | 4084 (51.6) | 15 918 (56.9) | 70.20 b | <.001 |
| Ethnicity, n (%) | Ethnicity, n (%) | Ethnicity, n (%) | Ethnicity, n (%) | Ethnicity, n (%) |
| White | 7371 (93.1) | 26 109 (93.3) | 0.39 b | .532 |
| Nonwhite | 550 (6.94) | 1888 (6.74) | 0.39 b | .532 |
| Townsend index, mean (SD) | −0.88 (0.03) | −1.15 (0.02) | −7.19 a | <.001 |
| Attainable education, n (%) | Attainable education, n (%) | Attainable education, n (%) | Attainable education, n (%) | Attainable education, n (%) |
| Above college/university | 2127 (26.9) | 10 335 (36.9) | 275.89 b | <.001 |
| Below college/university | 5794 (73.2) | 17 662 (63.1) | 275.89 b | <.001 |
| Smoking status, n (%) | Smoking status, n (%) | Smoking status, n (%) | Smoking status, n (%) | Smoking status, n (%) |
| Never | 4120 (52.4) | 15 673 (56.2) | 36.22 b | <.001 |
| Former/current | 3742 (47.6) | 12 203 (43.8) | 36.22 b | <.001 |
| Drinking status, n (%) | Drinking status, n (%) | Drinking status, n (%) | Drinking status, n (%) | Drinking status, n (%) |
| Never | 506 (6.42) | 1080 (3.87) | 94.49 b | <.001 |
| Former/current | 7378 (93.6) | 26 845 (96.1) | 94.49 b | <.001 |
| Meeting physical education recommendation, n (%) | Meeting physical education recommendation, n (%) | Meeting physical education recommendation, n (%) | Meeting physical education recommendation, n (%) | Meeting physical education recommendation, n (%) |
| No | 1475 (23.5) | 3832 (16.6) | 162.58 b | <.001 |
| Yes | 4787 (76.5) | 19 297 (83.4) | 162.58 b | <.001 |
| Health status, n (%) | Health status, n (%) | Health status, n (%) | Health status, n (%) | Health status, n (%) |
| Excellent/good | 4340 (55.2) | 20 483 (73.5) | 976.22 b | <.001 |
| Fair/poor | 3256 (44.8) | 7369 (26.5) | 976.22 b | <.001 |
| History of chronic heart diseases, n (%) | History of chronic heart diseases, n (%) | History of chronic heart diseases, n (%) | History of chronic heart diseases, n (%) | History of chronic heart diseases, n (%) |
| No | 7377 (93.1) | 27 006 (96.5) | 167.17 b | <.001 |
| Yes | 544 (6.87) | 991 (3.54) | 167.17 b | <.001 |
| History of stroke, n (%) | History of stroke, n (%) | History of stroke, n (%) | History of stroke, n (%) | History of stroke, n (%) |
| No | 7762 (98.0) | 27 613 (98.6) | 16.76 b | <.001 |
| Yes | 159 (2.01) | 384 (1.37) | 16.76 b | <.001 |
## Retinal age gap and metabolic syndrome and/or inflammation
Associations of retinal age gap with metabolic syndrome and/or inflammation are reported in Table 3. Each year increase in the retinal age gap was associated with a $1\%$ risk increase of MetS (OR: 1.01; $95\%$ CI, 1.00,1.02; $$p \leq .016$$), a $1\%$ risk increase of inflammation (OR: 1.01; $95\%$ CI, 1.00,1.02; $$p \leq .021$$), and a $1\%$ risk increase of MetS and inflammation combined (OR: 1.01; $95\%$ CI, 1.00,1.02; $$p \leq .011$$) in the fully adjusted model. Compared to participants with the lowest quartile of retinal age gaps, the risk of MetS significantly increased by $10\%$ and $14\%$ respectively for participants with retinal age gap in the third and fourth quartiles (OR: 1.10; $95\%$ CI, 1.01,1.21; $$p \leq .030$$; OR: 1.14; $95\%$ CI, 1.03,1.26; $$p \leq .012$$, respectively). A similar trend was identified for the risk of inflammation that participants with retinal age gaps in the third and fourth quartiles had $10\%$ and $25\%$ increased risks (OR: 1.10; $95\%$ CI, 1.01,1.21; $$p \leq .048$$; OR: 1.25; $95\%$ CI, 1.12,1.38; $p \leq .001$) compared to those in the first quartile, respectively. A significant association existed between retinal age gaps and a combined condition of inflammation and MetS. Table 4 shows the association between retinal age gaps and the five specific subcomponents of MetS. Per year increase in retinal age gaps was associated with a $2\%$ risk increase in abdominal obesity (OR: 1.02, $95\%$ CI, 1.01,1.02; $p \leq .001$), a $1\%$ risk increase in hypertension (OR: 1.01; $95\%$ CI, 1.00,1.02; $$p \leq .002$$), a $6\%$ risk increase in hyperglycemia (OR: 1.06, $95\%$ CI, 1.04,1.07; $p \leq .001$) but not associated with elevated serum HDL and elevated serum triglycerides. Compared with the participants with the lowest retinal age gap quartile, participants with higher quartiles tend to have a $10\%$ to $27\%$ risk increase in abdominal obesity and $14\%$ to $19\%$ risk increase in hypertension. Moreover, participants with the higher quartiles of retinal age gaps were associated with a $25\%$ to $104\%$ risk increase in hyperglycemia (OR: 1.25; $95\%$ CI, 1.10,1.42; $p \leq .001$; OR: 1.45; $95\%$ CI, 1.27,1.66;, $p \leq .001$; OR: 2.04; $95\%$ CI, 1.74,2.38; $p \leq .001$, respectively).
## DISCUSSION
In a large population of middle‐aged and older adults, we found that the retinal age gap was significantly associated with MetS and inflammation. Specifically, compared to participants with retinal age gaps in the lowest quartile, the risk of MetS was significantly increased by $10\%$ and $14\%$ respectively for participants with retinal age gaps in the third and fourth quartile. Similar trends were identified for the risk of inflammation and combined MetS and inflammation.
Previous studies have provided valuable clues for our investigations about the associations of retinal age gaps with MetS and inflammation. First, structural defects of the retina were found in patients with MetS and inflammation. 19 For example, the patients with MetS and high levels of CRP had thinner inner retinal layers and photoreceptor layer in OCT segmentation analysis. 20, 21 Second, retinal microvascular signs were identified in the fundus images from MetS patients. A population‐based study suggested that MetS was associated with retinal microvascular signs (microaneurysms, retinal hemorrhages, arteriovenous nicking, and focal arteriolar narrowing) based on retinal photographs. 22 Moreover, MetS and its components, as well as CRP, a marker of inflammation, have been associated with retinal diseases including age‐related macular degeneration, 23, 24 retinopathy, 25, 26 and glaucoma, 27 resulting in significant increases in morbidity and mortality. 2, 23 Ours findings might be explained by several underlying mechanisms. MetS and chronic inflammation may contribute to the retinal age gap between biological age and chorological age, as both of the conditions accelerate aging process and increase the risk of age‐related diseases. 23 Moreover, early signs of MetS could be presented in the retina at an early stage. The microvascular and macrovascular dysfunctions in MetS are the fundamental pathology in the development of MetS. 28 The retina, as a highly vascular organ, could serve as an instant window to the systemic vasculature, having the potential to catch the early vascular changes of MetS. Of note, the attention maps of our DL model for the prediction of retina age exactly highlighted areas near the retinal vessels.
The retinal age gap has provided a novel and reliable screening method for MetS and inflammation. Compared with previous screening tools based on blood tests or anthropometric measurements, the retinal age gap is measured by a deep learning model to integrate all features from fundus images automatically, so that it could minimize the manual assessments error and avoid invasive testing. Therefore, it has great potentials to be used as a diagnosis biomarker, characterized by noninvasiveness, reliability, and objectiveness. This guarantees its further application in screening for MetS as well as inflammation in large populations. Moreover, an early detection and risk stratification of MetS and inflammation could help to promote the prevention and intervention strategies for chronic diseases such as diabetes and CVDs, which may effectively alleviate the economic burden of the whole society.
Although the large population, standardized protocol, high quality of fundus image data, and objective measures based on a deep learning model have enhanced the robustness of our findings, the current study harbors several limitations. First, causality could not be drawn from this cross‐sectional study. Second, our results might underestimate the effects of retinal age gap on MetS, as individuals with poor health would be less likely to participate in this study. Longitudinal data are needed to investigate the association between retinal age gaps and incident MetS.
## CONCLUSION
We found that retinal age gaps were significantly associated with MetS as well as inflammation. Based on a DL model, the retinal age gap has great potentials to be used as a noninvasive, reliable, and objective tool for the screening of MetS in large populations. Longitudinal data are further needed to confirm the association between retinal age gaps and incident MetS.
## AUTHOR CONTRIBUTIONS
Study concept and design: Zhuoting Zhu, Dan Liu, Ruiye Chen, Mingguang He. Acquisition, analyses, or interpretation: all authors. Drafting of the manuscript: Ruiye Chen, Zhuoting Zhu, Dan Liu. Critical revision of the manuscript for important intellectual content: Wenyi Hu, Katerina Kiburg, Jason Ha, Wei Wang, Honghua Yu, Xiaohong Yang, Mingguang He. Statistical analyses: Ruiye Chen, Zhuoting Zhu, Dan Liu, Xianwen Shang. Obtained funding: Zhuoting Zhu, Wei Wang, Honghua Yu, Xiaohong Yang, Mingguang He. Administrative, technical, or material support: Wei Wang, Honghua Yu, Xiaohong Yang, Mingguang He. Study supervision: Wei Wang, Honghua Yu, Xiaohong Yang, Mingguang He.
## CONFLICT OF INTEREST
We declare no competing interests.
## References
1. Alberti KG, Eckel RH, Grundy SM. **Harmonizing the metabolic syndrome: a joint interim statement of the international diabetes federation task force on epidemiology and prevention; National Heart, Lung, and Blood Institute; American Heart Association; world heart federation; international atherosclerosis society; and International Association for the Study of obesity**. *Circulation* (2009) **120** 1640-1645. PMID: 19805654
2. Saklayen MG. **The global epidemic of the metabolic syndrome**. *Curr Hypertens Rep* (2018) **20** 12. PMID: 29480368
3. Dandona P, Aljada A, Chaudhuri A, Mohanty P, Garg R. **Metabolic syndrome: a comprehensive perspective based on interactions between obesity, diabetes, and inflammation**. *Circulation* (2005) **111** 1448-1454. PMID: 15781756
4. Mahajan A, Jaiswal A, Tabassum R. **Elevated levels of C‐reactive protein as a risk factor for metabolic syndrome in Indians**. *Atherosclerosis* (2012) **220** 275-281. PMID: 22104116
5. Eckel RH, Grundy SM, Zimmet PZ. **The metabolic syndrome**. *Lancet* (2005) **365** 1415-1428. PMID: 15836891
6. Meng W, Zhang C, Zhang Q. **Association between leukocyte and metabolic syndrome in urban Han Chinese: a longitudinal cohort study**. *PLoS One* (2012) **7**. PMID: 23209610
7. Lee MJ, Jung CH, Kang YM. **Serum bilirubin as a predictor of incident metabolic syndrome: a 4‐year retrospective longitudinal study of 6205 initially healthy Korean men**. *Diabetes Metab* (2014) **40** 305-309. PMID: 24951082
8. Romero‐Saldana M, Fuentes‐Jimenez FJ, Vaquero‐Abellan M, Alvarez‐Fernandez C, Molina‐Recio G, Lopez‐Miranda J. **New non‐invasive method for early detection of metabolic syndrome in the working population**. *Eur J Cardiovasc Nurs* (2016) **15** 549-558. PMID: 26743264
9. Bener A, Yousafzai MT, Darwish S, Al‐Hamaq AO, Nasralla EA, Abdul‐Ghani M. **Obesity index that better predict metabolic syndrome: body mass index, waist circumference, waist hip ratio, or waist height ratio**. *J Obes* (2013) **2013**. PMID: 24000310
10. Tillin T, Sattar N, Godsland IF, Hughes AD, Chaturvedi N, Forouhi NG. **Ethnicity‐specific obesity cut‐points in the development of type 2 diabetes ‐ a prospective study including three ethnic groups in the United Kingdom**. *Diabet Med* (2015) **32** 226-234. PMID: 25186015
11. Wang Z, Ma J, Si D. **Optimal cut‐off values and population means of waist circumference in different populations**. *Nutr Res Rev* (2010) **23** 191-199. PMID: 20642876
12. Romero‐Saldana M, Tauler P, Vaquero‐Abellan M. **Validation of a non‐invasive method for the early detection of metabolic syndrome: a diagnostic accuracy test in a working population**. *BMJ Open* (2018) **8**
13. Zhu Z, Shi D, Guankai P. **Retinal age gap as a predictive biomarker for mortality risk**. *Br J Ophthalmol* (2022). DOI: 10.1136/bjophthalmol-2021-319807
14. Hu W, Wang W, Wang Y. **Retinal age gap as a predictive biomarker of future risk of Parkinson's disease**. *Age Ageing* (2022) **51**. PMID: 35352798
15. Zhu Z, Chen Y, Wang W. **Association of retinal age gap with arterial stiffness and incident cardiovascular disease**. *Stroke* (2022) **53** 3320-3328. DOI: 10.1161/STROKEAHA.122.038809
16. Sudlow C, Gallacher J, Allen N. **UK biobank: an open access resource for identifying the causes of a wide range of complex diseases of middle and old age**. *PLoS Med* (2015) **12**. PMID: 25826379
17. White WB, Anwar YA. **Evaluation of the overall efficacy of the Omron office digital blood pressure HEM‐907 monitor in adults**. *Blood Press Monit* (2001) **6** 107-110. PMID: 11433132
18. van der Berg JD, Stehouwer CD, Bosma H. **Associations of total amount and patterns of sedentary behaviour with type 2 diabetes and the metabolic syndrome: the Maastricht study**. *Diabetologia* (2016) **59** 709-718. PMID: 26831300
19. Lopes de Faria JM, Russ H, Costa VP. **Retinal nerve fibre layer loss in patients with type 1 diabetes mellitus without retinopathy**. *Br J Ophthalmol* (2002) **86** 725-728. PMID: 12084737
20. Qiu F, Ma X, Shin YH. **Pathogenic role of human C‐reactive protein in diabetic retinopathy**. *Clin Sci (Lond)* (2020) **134** 1613-1629. PMID: 32602547
21. Zarei R, Anvari P, Eslami Y. **Retinal nerve fibre layer thickness is reduced in metabolic syndrome**. *Diabet Med* (2017) **34** 1061-1066. PMID: 28430372
22. Wang SB, Mitchell P, Plant AJ. **Metabolic syndrome and retinal microvascular calibre in a high cardiovascular disease risk cohort**. *Br J Ophthalmol* (2016) **100** 1041-1046. PMID: 26531050
23. Ghaem Maralani H, Tai BC, Wong TY. **Metabolic syndrome and risk of age‐related macular degeneration**. *Retina* (2015) **35** 459-466. PMID: 25207946
24. Seddon JM, Gensler G, Milton RC, Klein ML, Rifai N. **Association between C‐reactive protein and age‐related macular degeneration**. *JAMA* (2004) **291** 704-710. PMID: 14871913
25. Mbata O, Abo El‐Magd NF, El‐Remessy AB. **Obesity, metabolic syndrome and diabetic retinopathy: beyond hyperglycemia**. *World J Diabetes* (2017) **8** 317-329. PMID: 28751954
26. Yang XF, Deng Y, Gu H. **C‐reactive protein and diabetic retinopathy in Chinese patients with type 2 diabetes mellitus**. *Int J Ophthalmol* (2016) **9** 111-118. PMID: 26949620
27. Newman‐Casey PA, Talwar N, Nan B, Musch DC, Stein JD. **The relationship between components of metabolic syndrome and open‐angle glaucoma**. *Ophthalmology* (2011) **118** 1318-1326. PMID: 21481477
28. Czernichow S, Greenfield JR, Galan P. **Macrovascular and microvascular dysfunction in the metabolic syndrome**. *Hypertens Res* (2010) **33** 293-297. PMID: 20075933
|
---
title: 'Sex differences in risk factors for end‐stage kidney disease and death in
type 2 diabetes: A retrospective cohort study'
authors:
- Megumi Oshima
- Yasunori Iwata
- Tadashi Toyama
- Shinji Kitajima
- Akinori Hara
- Norihiko Sakai
- Miho Shimizu
- Kengo Furuichi
- Masakazu Haneda
- Tetsuya Babazono
- Hiroki Yokoyama
- Kunitoshi Iseki
- Shinichi Araki
- Toshiharu Ninomiya
- Shigeko Hara
- Yoshiki Suzuki
- Masayuki Iwano
- Eiji Kusano
- Tatsumi Moriya
- Hiroaki Satoh
- Hiroyuki Nakamura
- Hirofumi Makino
- Takashi Wada
journal: Journal of Diabetes
year: 2023
pmcid: PMC10036257
doi: 10.1111/1753-0407.13367
license: CC BY 4.0
---
# Sex differences in risk factors for end‐stage kidney disease and death in type 2 diabetes: A retrospective cohort study
## Abstract
Highlights Sex differences are observed in the progression of chronic kidney disease; however, it is uncertain whether sex is associated with the risk of kidney failure in type 2 diabetes. In this multicenter cohort study, men had a higher risk of end‐stage kidney disease (ESKD) than women, and moderately increased albuminuria was strongly associated with sex difference in developing ESKD in patients with type 2 diabetes. Sex differences may affect the risk of kidney disease progression because of type 2 diabetes.
### Background
This study investigated the sex differences in the risk of end‐stage kidney disease (ESKD) and mortality, as well as the effect modification of sex on associated factors in patients with type 2 diabetes.
### Methods
This multicenter observational cohort study included 4328 patients with type 2 diabetes. Hazard ratios (HRs) with $95\%$ confidence intervals (CIs) of sex for ESKD and death were estimated using Cox proportional regression with adjustment for baseline covariates. For assessing risk modification, HRs and incidence rates for ESKD and death were compared between sexes across patient characteristics using Cox proportional and Poisson regression models.
### Results
During a median follow‐up of 7 years, 276 patients ($70\%$ men) developed ESKD, and 241 patients ($68\%$ men) died. Men had higher risks of ESKD (HR 1.34; $95\%$ CI 1.02–1.75; $$p \leq .034$$) and death (HR 1.64; $95\%$ CI 1.24–2.16; $$p \leq .001$$) versus women after adjusting for multiple covariates. Among patients with microalbuminuria, men had a substantially higher risk of ESKD versus women, compared to those with normo‐ and macroalbuminuria (p for interaction.04). Incidence rates were also increased in men versus women with albuminuria of around 300 mg/g. No differences were detected in the association of sex and death across baseline patient subgroups.
### Conclusions
In type 2 diabetes, men had an increased risk of ESKD and death versus women. Moderately increased albuminuria was strongly associated with sex difference in developing ESKD.
## INTRODUCTION
Diabetic kidney disease (DKD) develops in ~$40\%$ of people with type 2 diabetes, and it remains a leading cause of end‐stage kidney disease (ESKD) and early mortality worldwide. 1, 2 Various explanatory factors have been associated with a greater risk of the onset and progression of kidney disease in patients with type 2 diabetes. 3 These factors can be used for early risk stratification and targeted intervention to prevent adverse kidney outcomes.
The cumulative incidence and risk of kidney replacement therapy have been reported to be higher in men than women with chronic kidney disease (CKD). 4 A study of patients with type 2 diabetes also demonstrated that men had a faster decline in kidney function versus women, 5 suggesting that sex differences may exist in the progression of CKD. However, it is uncertain whether sex is associated with the risk of ESKD and which factors affect the associations between sex and kidney outcomes in patients with type 2 diabetes. To address this knowledge gap, this study investigated the sex differences in kidney prognosis and the effect modification of sex on its associated factors in patients with type 2 diabetes using data from a multicenter cohort study with long‐term follow‐up.
## Patients and study design
The observational cohort used in this study included 4328 patients with type 2 diabetes treated across 10 centers in Japan between 1985 and 2011. Patients were diagnosed with type 2 diabetes according to the criteria of the Japan Diabetes Society (JDS). 6 The exclusion criteria were as follows: age <18 years, type 1 or secondary diabetes, kidney transplantation, maintenance dialysis, missing values of baseline covariates (including urine albumin‐to‐creatinine ratio [UACR], estimated glomerular filtration rate [eGFR], glycated hemoglobin [HbA1c], and systolic blood pressure [BP]), and refusal to provide informed consent. A detailed description of the study design was previously reported. 7 Other background therapies for glycemic management and control of cardiovascular risk factors were used according to the recommendation of the local guidelines. Patients were followed up until the onset of the first study outcome or the end of follow‐up in October 2011. All participants provided written informed consent and study protocols were approved by the local institutional ethics committees at each site.
## Baseline and follow‐up variables
A history of cardiovascular disease (CVD) was defined as having at least one of following; coronary heart disease, stroke, cerebral hemorrhage, heart failure, or arteriosclerosis obliterans. BP was measured in the sitting position. HbA1c was measured according to the standards of the JDS, using nonfasting blood samples, and converted into HbA1c via the National Glycohemoglobin Standardization Program (JDS + 0.4) method for the analysis. Serum creatinine level was measured using an enzymatic method, and eGFR was calculated using the equation proposed by the Japanese Society of Nephrology. 8 Urine albumin and creatinine levels were measured using a turbidimetric immunoassay and an enzymatic method applied on spot urine samples.
## Study outcome
The primary outcome was the occurrence of ESKD, defined as the need for renal replacement therapy or having an eGFR of <15 mL min−1 1.73 m−2. The secondary outcome was death.
## Statistical analyses
Baseline patient characteristics were summarized and stratified according to sex. Continuous variables are presented as mean with SD for variables with approximately symmetrical distributions. The results for variables with skewed distributions are shown as the median and interquartile range (IQR); these were transformed into natural logarithms before analysis. Categorical variables are reported as percentages.
Cox proportional regression was performed to estimate hazard ratios (HRs) with $95\%$ confidence intervals (CIs) of sex for ESKD. The models were adjusted for baseline covariates including age, history of CVD, systolic and diastolic BP, HbA1c, eGFR, and log‐transformed albuminuria and stratified by institutions. Additionally, Kaplan–Meier curves were determined among men and women. In the sensitivity analysis, we repeated the analysis after including death as a competing risk.
For assessing potential effect modification by sex, we evaluated the association between sex and ESKD according to baseline covariates, including age (< 65 or ≥ 65 years), sex, history of CVD, systolic BP (<140 or ≥140 mm Hg), HbA1c (<7.0 or ≥$7.0\%$ [<53 or ≥53 mmol/mol]), eGFR (< 60 or ≥60 mL min−1 1.73 m−2), and albuminuria (<30, 30–300, or >300 mg/g). The interaction was assessed by adding sex with subgroup interaction terms to the models. Further analyses were performed for the variables indicating effect modification by sex. Incidence rates (per 100 patient‐years of follow‐up) of ESKD were estimated according to the levels of variables using restricted cubic splines via Poisson regression models.
To address the association of sex with eGFR and UACR trajectories, least‐squares mean changes from baseline in eGFR and UACR over time were analyzed by sex using linear mixed‐effects models with restricted maximum likelihood‐based repeated measures. The models included the fixed effects of measurement timepoint and continuous fixed covariates of baseline value and baseline value‐by‐time point interaction. An unstructured covariance structure was used to model within‐patient errors. We also evaluated the association of sex with three phenotypes of clinical course of DKD using Cox proportional or logistic regression as appropriate: (a) progression of albuminuria defined as the development of macroalbuminuria (UACR ≥300 mg/g) from normo‐ or microalbuminuria (<300 mg/g); (b) regression of albuminuria defined as a transition from macroalbuminuria to normo‐ or microalbuminuria or from microalbuminuria (≥ 30 mg/g) to normoalbuminuria (<30 mg/g); and (c) rapid eGFR decline of ≥5 mL min−1 1.73 m−2 year−1 during follow‐up. These models were adjusted for baseline covariates described previously. Statistical significance was set at $p \leq .05$, and all analyses were performed using Stata version 16.
## Patient characteristics
Among 4328 patients, $61\%$ (2635 patients) were men. The study cohort had a mean age of 60 years (SD 12), mean HbA1c of $7.6\%$ (SD 1.7) (60 mmol/mol [19]), mean eGFR of 76 mL min−1 1.73 m−2 (SD 24), and median UACR of 25 mg/g (IQR 9–66) at baseline (Table 1). Men were more likely to be younger and have history of CVD, lower levels of systolic BP and HbA1c, and higher levels of UACR than women. Men and women had a similar mean eGFR at baseline. Among patients whom the year of registration was recorded, similar distributions were observed between men and women, irrespective of the period of registration (Table S1).
**TABLE 1**
| Unnamed: 0 | Total (n = 4328) | Men (n = 2635) | Women (n = 1693) | P value |
| --- | --- | --- | --- | --- |
| Age, years | 60 (12) | 59 (11) | 62 (12) | <0.001 |
| History of CVD, n (%) | 309 (7) | 215 (8) | 94 (6) | 0.001 |
| Systolic BP, mm Hg | 131 (19) | 130 (18) | 132 (19) | 0.005 |
| Diastolic BP, mm Hg | 74 (18) | 75 (12) | 74 (25) | 0.10 |
| HbA1c, % | 7.6 (1.7) | 7.5 (1.7) | 7.8 (1.7) | <0.001 |
| HbA1c, mmol/ml | 60 (19) | 58 (19) | 62 (19) | <0.001 |
| eGFR, ml min−1 1.73 m−2 | 76 (24) | 76 (24) | 76 (25) | 0.90 |
| eGFR, n (%) | | | | 0.24 |
| ≥90 | 1132 (26) | 675 (26) | 457 (27) | |
| 60 ≤ 90 | 2124 (49) | 1329 (50) | 795 (47) | |
| 45 ≤ 60 | 682 (16) | 406 (15) | 276 (16) | |
| 30 ≤ 45 | 267 (6) | 154 (6) | 113 (7) | |
| <30 | 123 (3) | 71 (3) | 52 (3) | |
| UACR, mg/g | 25 (9, 66) | 19 (8, 74) | 18 (9, 58) | 0.006 |
| UACR, n (%) | | | | 0.001 |
| <30 mg/g | 2679 (62) | 1581 (60) | 1098 (65) | |
| 30 ≤ 300 mg/g | 115 (26) | 693 (26) | 422 (25) | |
| ≥300 mg/g | 534 (12) | 361 (14) | 173 (10) | |
## Association between sex and ESKD
Over a median follow‐up of 7 years (IQR 4–8), 276 ($6.4\%$) patients ($70\%$ [$$n = 194$$] men) had ESKD and 241 ($5.6\%$) patients ($68\%$ [$$n = 164$$] men) died. More men reached ESKD than women (11.7 vs. 7.4 patients per 1000 patient‐years). Men were associated with a higher risk for ESKD (HR 1.34; $95\%$ CI 1.02–1.75; $$p \leq .03$$) compared to women after adjusting for multiple covariates (Figure 1). Older age and higher eGFR levels were also associated with a lower risk for ESKD, whereas higher levels of HbA1c and UACR were associated with a higher risk for ESKD. A similar relationship between sex and ESKD was observed in the Kaplan–Meier curves (Figure 2) and after including death as a competing risk (Table S2). Additional analysis revealed that men were also associated with an increased risk for death compared to women (HR 1.64; $95\%$ CI 1.24–2.16; $$p \leq .001$$) (Figures S1 and S2).
**FIGURE 1:** *Hazard ratios for the association between baseline covariates and ESKD. Adjusted for age, history of CVD, systolic BP, diastolic BP, HbA1c, eGFR, and log‐transformed UACR and stratified by institutions. BP, blood pressure; CI, confidence interval; CVD, cardiovascular disease; eGFR, estimated glomerular filtration ratio; ESKD, end‐stage kidney disease; HbA1c, glycated hemoglobin; UACR, urine albumin‐to‐creatinine ratio.* **FIGURE 2:** *Kaplan–Meier curves for ESKD in women and men. ESKD, end‐stage kidney disease.*
## Sex differences in risk factors for ESKD
No differences were observed in the association of sex and ESKD across baseline participant subgroups (p for interaction >.19) except for UACR (Figure 3). Among patients with a UACR of 30–300 mg/g, men had a higher risk of ESKD than women (HR 2.37; $95\%$ CI 1.12–5.02), compared to patient groups with UACR of <30 and ≥300 mg/g (p for interaction.04). A similar finding for UACR was detected after including death as a competing risk (Table S3). Incidence rates per 100 patient‐years were higher in men than women with a UACR of approximately 300 mg/g (Figure 4). No difference was observed in the incidence rates of ESKD between sexes across age at baseline (Figure S3). There were no differences in the relationship between sex and death across patient characteristics including UACR (p for interaction >.08) (Table S4).
**FIGURE 3:** *Association between sex (men versus women) and ESKD by baseline participant characteristics. Adjusted for age, history of cardiovascular disease, systolic BP, diastolic BP, HbA1c, eGFR, and log‐transformed UACR and stratified by institutions. BP, blood pressure; CI, confidence interval; CVD, cardiovascular disease; eGFR, estimated glomerular filtration ratio; HbA1c, glycated hemoglobin; HR, hazard ratio; UACR, urine albumin‐to‐creatinine ratio.* **FIGURE 4:** *Incidence rates of ESKD according to baseline UACR by sex. Adjusted for age, history of CVD, systolic BP, diastolic BP, HbA1c, and eGFR. BP, blood pressure; CVD, cardiovascular disease; eGFR, estimated glomerular filtration ratio; ESKD, end‐stage kidney disease; HbA1c, glycated hemoglobin; UACR, urine albumin‐to‐creatinine ratio.*
## eGFR and UACR trajectories by sex
The mean eGFR gradually decreased during follow‐up in both men and women, and no differences in mean eGFR were observed between sex (0.09 mL min−1 1.73 m−2 year−1 higher in men versus women, $95\%$ CI −0.09 to 0.28; $$p \leq .33$$) (Figure S4A). Geometric mean UACR slowly increased over time among both men and women and was $17\%$ higher in men versus women ($95\%$ CI $6.3\%$–$24\%$; $$p \leq .002$$ in men compared with women) (Figure S4B).
For the association of sex with clinical phenotypes of DKD, men were more likely to have progression of albuminuria (HR 1.18; $95\%$ CI 1.02–1.37; $$p \leq .03$$) and a lower likelihood of regression of albuminuria (HR 0.69; $95\%$ CI 0.52–0.93; $$p \leq .01$$) versus women (Table S5). In contrast, no associations were observed between sex and a rapid eGFR decline of >5 mL min−1 1.73 m−2 year−1 (HR 0.93; $95\%$ CI 0.65–1.35; $$p \leq .72$$).
## DISCUSSION
Among patients with type 2 diabetes, men had significantly increased risks of ESKD and death versus women, even after adjustment for previously known risk factors such as HbA1c, eGFR, and albuminuria. Additionally, the sex difference in ESKD risk was prominent among patients with moderately increased albuminuria. In particular, men were associated with a higher risk of progression of albuminuria and a lower likelihood of regression of albuminuria compared with women. On the other hand, no difference was found between sex and eGFR trajectories. These findings may have important implications for improving risk stratification and preventive strategies for kidney disease progression in patients with type 2 diabetes.
We found that men had a higher risk of DKD progression compared to women with type 2 diabetes, which is mostly consistent with previous reports. 4, 5, 9 In prospective studies of CKD population, men had a higher risk of ESKD and death both with and without adjustment for the presence of diabetes. 4, 9 A prospective study of type 2 diabetes also indicated that male sex was an independent risk factor for a steep eGFR decline of >3.5 mL min−1 1.73 m−2 year−1. 5 Additionally, several cross‐sectional studies of type 2 diabetes found that men had a higher prevalence of albuminuria than women. 10, 11 These findings were consistent with data from the Japanese Society of Dialysis Therapy registry, demonstrating substantially higher incidence rates of renal replacement therapy due to diabetic nephropathy in men than in women. 12 In contrast, a retrospective study of type 2 diabetes in Japan has reported conflicting findings that females had a greater eGFR decline versus males (−$3.5\%$ ± $2.7\%$ vs. −$2.0\%$ ± $2.2\%$ per year). 13 The association of sex with the risk of kidney outcome has been limited to type 2 diabetes only, but the current study was notable for evaluating ESKD in a large population with type 2 diabetes while adjusting for risk factors of kidney disease progression.
This study found a substantial increase in ESKD risk in men versus women with microalbuminuria at baseline. This may be partly explained by our findings that men were more likely to demonstrate an increase in albuminuria and were less likely to have a reduction in albuminuria compared with women during follow‐up. A previous prospective study of type 2 diabetes has similarly reported that men had an increased incidence of microalbuminuria compared to women. 14 Changes in albuminuria have been widely recognized as reliable surrogate endpoints for kidney prognosis beyond one‐point albuminuria measurement, 15, 16 which may have led to a worse kidney outcome in men than women in the current cohort. Another cohort study also suggested that the higher rates of obesity and smoking in men versus women can promote the development of proteinuria. 17 Moreover, because albuminuria is indicative of microvascular dysfunction, the sex‐albuminuria interaction may reflect the presence of sex differences in the pathophysiology of DKD. 18, 19 A prospective cohort study of adolescents with type 2 diabetes reported that females had a threefold greater risk of developing hyperfiltration over 5 years compared to males. 20 In short, our findings suggest that sex differences may exist in the impact of albuminuria on the progression of DKD, although further evaluation is needed to elucidate its underlying mechanism.
In addition to the sex‐albuminuria interaction, there are several possible explanations for the association of sex and kidney prognosis in type 2 diabetes. First, in a meta‐analysis of the general public and CKD patients, males had a higher risk of acute kidney injury compared to females. 21 Acute kidney injury is also frequently seen in patients with diabetes and is a well‐known risk factor for the development of CKD and kidney failure, 22 which may lead to an increased incidence of ESKD in men compared to women. Second, in a pooled analysis of six cohorts, hypertension was a stronger risk factor for CKD progression and ESKD in men than women. 23 However, in this study, risk modification was not observed according to BP levels at baseline. Lastly, gender‐related behavior, specifically regarding disease management and therapeutic strategies, may be involved in the sex‐based differences in kidney prognosis. 24 We observed no sex differences in eGFR trajectories over time, although the annual eGFR slope is a useful surrogate end point for kidney prognosis in recent clinical trials. 16 This observation can be explained by various reasons. Diabetes is a risk factor for acute kidney injury requiring dialysis 25; however, this study collected eGFR data on a yearly basis, which might have resulted in missed cases of acute kidney injury, leading to dialysis. Diabetes is also associated with the early initiation of dialysis in patients with high eGFR levels, 26 which may have partly resulted in similar eGFR values between the sexes. In addition, the timing of dialysis initiation is not based on eGFR alone, and other clinical manifestations, including electrolyte acid–base metabolism disorders, volume overload, and heart failure, may possibly require dialysis, regardless of eGFR. 27 Sex hormones can directly and indirectly mediate hemodynamics and inflammation in the kidney. The influence of sex hormones on the renin‐angiotensin system (RAS) may be related to response to RAS blockade, leading to the sex‐based differences in DKD progression. 28 Testosterone levels may also affect the development of hypertension and hypertension‐induced kidney injury in diabetes. 29 For the detailed mechanisms underlying the association between sex hormones and DKD progression, recent studies have demonstrated that both sex and sex hormones affect the expression of transforming growth factor‐β1, which can induce kidney injury. 30 Furthermore, some studies have reported that kidney complication manifests more than 10 years later in women than in men, 31 suggesting the presence of protective effects of estrogen on the kidney during premenopausal period. 32, 33 However, the current study did not find any difference in the association between sex and kidney outcomes across age.
The current study also demonstrated the sex differences in the association with mortality in type 2 diabetes, which was inconsistent with previous studies. Previous meta‐analyses have reported that women with diabetes showed a greater risk of all‐cause and cardiovascular mortality compared to men, although there was significant heterogeneity between studies. 34, 35 This may be because of the variations in the prevalence of cardiovascular risk factors across sex in type 2 diabetes.
Our study's strengths include real‐world clinical data with a large sample size and long follow‐up duration. However, there are several limitations. Although we adjusted for multiple risk factors, we cannot exclude the effects of residual confounding factors such as body mass index, smoking status, and medications (eg, hyperglycemic and hypertensive agents). Moreover, sex‐specific factors such as pregnancy, menopause, and hormone replacement were not recorded in this cohort, and thus we cannot consider the impact of those factors on sex differences in DKD progression. Underlying mechanisms for the sex differences require to be elucidated. In addition, variations between centers regarding the requirement and processes for obtaining informed patient consent and enrolling patients can also lead to selection bias.
In conclusion, men have a higher risk of ESKD compared to women with type 2 diabetes, suggesting that sex differences may affect the risk of kidney disease progression due to type 2 diabetes. Early detection and management of modifiable risk factors and other comorbidities may slow the progression of DKD and prevent kidney failure in patients with type 2 diabetes. Further studies are needed to elucidate the mechanisms behind these sex differences.
## FUNDING INFORMATION
This research did not receive any specific grant from funding agencies in the public, commercial, or not‐for‐profit sectors.
## CONFLICT OF INTEREST STATEMENT
There are no conflicts of interest to declare.
## References
1. 1
System USRD
. USRDS Annual Data Report: Epidemiology of Kidney Disease in the United States. National Institutes of Health, National Institute of Diabetes and Digestive and Kidney Diseases; 2020.. *USRDS Annual Data Report: Epidemiology of Kidney Disease in the United States* (2020)
2. International Diabetes Federation. *IDF Diabetes Atlas* (2021)
3. Oshima M, Shimizu M, Yamanouchi M. **Trajectories of kidney function in diabetes: a clinicopathological update**. *Nat Rev Nephrol* (2021) **17** 740-750. PMID: 34363037
4. Hecking M, Tu C, Zee J. **Sex‐specific differences in mortality and incident dialysis in the chronic kidney disease outcomes and practice patterns study**. *Kidney Int Rep* (2022) **7** 410-423. PMID: 35257054
5. de Hauteclocque A, Ragot S, Slaoui Y. **The influence of sex on renal function decline in people with type 2 diabetes**. *Diabet Med* (2014) **31** 1121-1128. PMID: 24773061
6. Seino Y, Nanjo K. **Report of the committee on the classification and diagnostic criteria of diabetes mellitus**. *J Diabetes Investig* (2010) **1** 212-228
7. Wada T, Haneda M, Furuichi K. **Clinical impact of albuminuria and glomerular filtration rate on renal and cardiovascular events, and all‐cause mortality in Japanese patients with type 2 diabetes**. *Clin Exp Nephrol* (2014) **18** 613-620. PMID: 24132561
8. Matsuo S, Imai E, Horio M. **Collaborators developing the Japanese equation for estimated GFR. Revised equations for estimated GFR from serum creatinine in Japan**. *Am J Kidney Dis* (2009) **53** 982-992. PMID: 19339088
9. Ricardo AC, Yang W, Sha D. **Sex‐related disparities in CKD progression**. *J Am Soc Nephrol* (2019) **30** 137-146. PMID: 30510134
10. Coll‐de‐Tuero G, Mata‐Cases M, Rodriguez‐Poncelas A. **Chronic kidney disease in the type 2 diabetic patients: prevalence and associated variables in a random sample of 2642 patients of a Mediterranean area**. *BMC Nephrol* (2012) **13** 87. PMID: 22905926
11. Al‐Rubeaan K, Youssef AM, Subhani SN. **Diabetic nephropathy and its risk factors in a society with a type 2 diabetes epidemic: a Saudi National Diabetes Registry‐based study**. *PLoS One* (2014) **9**. PMID: 24586457
12. Wakasugi M, Narita I. **Trends in the incidence of renal replacement therapy by type of primary kidney disease in Japan, 2006–2020**. *Nephrology* (2022) **28** 119-129. PMID: 36461735
13. Kajiwara A, Kita A, Saruwatari J. **Sex differences in the renal function decline of patients with type 2 diabetes**. *J Diabetes Res* (2016) **2016** 4626382-4626388. PMID: 27247948
14. Singh SS, Roeters‐van Lennep JE, Lemmers RFH. **Sex difference in the incidence of microvascular complications in patients with type 2 diabetes mellitus: a prospective cohort study**. *Acta Diabetol* (2020) **57** 725-732. PMID: 32025877
15. Heerspink HJL, Greene T, Tighiouart H. **Change in albuminuria as a surrogate endpoint for progression of kidney disease: a meta‐analysis of treatment effects in randomised clinical trials**. *Lancet Diabetes Endocrinol* (2019) **7** 128-139. PMID: 30635226
16. Levey AS, Gansevoort RT, Coresh J. **Change in albuminuria and GFR as end points for clinical trials in early stages of CKD: a scientific workshop sponsored by the National Kidney Foundation in collaboration with the US Food and Drug Administration and European Medicines Agency**. *Am J Kidney Dis* (2020) **75** 84-104. PMID: 31473020
17. Tozawa M, Iseki K, Iseki C, Oshiro S, Ikemiya Y, Takishita S. **Influence of smoking and obesity on the development of proteinuria**. *Kidney Int* (2002) **62** 956-962. PMID: 12164878
18. Bjornstad P, Cherney DZ. **Renal hyperfiltration in adolescents with type 2 diabetes: physiology, sex differences, and implications for diabetic kidney disease**. *Curr Diab Rep* (2018) **18** 22. PMID: 29557063
19. Furuichi K, Shimizu M, Yamanouchi M. **Clinicopathological features of fast eGFR decliners among patients with diabetic nephropathy**. *BMJ Open Diabetes Res Care* (2020) **8**
20. Bjornstad P, Nehus E, El Ghormli L. **Insulin sensitivity and diabetic kidney disease in children and adolescents with type 2 diabetes: an observational analysis of data from the TODAY clinical trial**. *Am J Kidney Dis* (2018) **71** 65-74. PMID: 29157731
21. Grams ME, Sang Y, Ballew SH. **A meta‐analysis of the Association of Estimated GFR, albuminuria, age, race, and sex with acute kidney injury**. *Am J Kidney Dis* (2015) **66** 591-601. PMID: 25943717
22. Advani A. **Acute kidney injury: a Bona fide complication of diabetes**. *Diabetes* (2020) **69** 2229-2237. PMID: 33082271
23. Weldegiorgis M, Woodward M. **The impact of hypertension on chronic kidney disease and end‐stage renal disease is greater in men than women: a systematic review and meta‐analysis**. *BMC Nephrol* (2020) **21** 506. PMID: 33238919
24. Tong A, Evangelidis N, Kurnikowski A. **Nephrologists' perspectives on gender disparities in CKD and dialysis**. *Kidney Int Rep* (2022) **7** 424-435. PMID: 35257055
25. Chen Z, McCulloch CE, Powe NR. **Exploring reasons for state‐level variation in incidence of dialysis‐requiring acute kidney injury (AKI‐D) in the United States**. *BMC Nephrol* (2020) **21** 336. PMID: 32778062
26. Escoli R, Luz I, Santos P, Vila LA. **Glomerular filtration rate and initiation of dialysis**. *Ther Apher Dial* (2017) **21** 606-610. PMID: 29205888
27. **Chinese clinical practice guideline for the management of “CKD‐PeriDialysis”‐the periods prior to and in the early‐stage of initial dialysis**. *Kidney Int Rep* (2022) **7** S531-S558. PMID: 36567827
28. Clotet S, Riera M, Pascual J, Soler MJ. **RAS and sex differences in diabetic nephropathy**. *Am J Physiol Renal Physiol* (2016) **310** F945-F957. PMID: 26962103
29. Regensteiner JG, Golden S, Huebschmann AG. **Sex differences in the cardiovascular consequences of diabetes mellitus: a scientific statement from the American Heart Association**. *Circulation* (2015) **132** 2424-2447. PMID: 26644329
30. Ziller N, Kotolloshi R, Esmaeili M. **Sex differences in diabetes‐ and TGF‐beta1‐induced renal damage**. *Cell* (2020) **9**
31. Cobo G, Hecking M, Port FK. **Sex and gender differences in chronic kidney disease: progression to end‐stage renal disease and haemodialysis**. *Clin Sci (Lond)* (2016) **130** 1147-1163. PMID: 27252402
32. Maric‐Bilkan C. **Sex differences in diabetic kidney disease**. *Mayo Clin Proc* (2020) **95** 587-599. PMID: 32138885
33. Shepard BD. **Sex differences in diabetes and kidney disease: mechanisms and consequences**. *Am J Physiol Renal Physiol* (2019) **317** F456-F462. PMID: 31241989
34. Wang Y, O'Neil A, Jiao Y. **Sex differences in the association between diabetes and risk of cardiovascular disease, cancer, and all‐cause and cause‐specific mortality: a systematic review and meta‐analysis of 5,162,654 participants**. *BMC Med* (2019) **17** 136. PMID: 31296205
35. Wang H, Ba Y, Cai RC, Xing Q. **Association between diabetes mellitus and the risk for major cardiovascular outcomes and all‐cause mortality in women compared with men: a meta‐analysis of prospective cohort studies**. *BMJ Open* (2019) **9**
|
---
title: 'Type 1 diabetes management: Room for improvement'
authors:
- Rita D. M. Varkevisser
- Erwin Birnie
- Dick Mul
- Peter R. van Dijk
- Henk‐Jan Aanstoot
- Bruce H. R. Wolffenbuttel
- Melanie M. van der Klauw
journal: Journal of Diabetes
year: 2023
pmcid: PMC10036258
doi: 10.1111/1753-0407.13368
license: CC BY 4.0
---
# Type 1 diabetes management: Room for improvement
## Abstract
Highlights Achievement of glycemic, lipid, and blood pressure targets are suboptimal. Individuals with type 1 diabetes and cardiovascular disease (CVD) have more difficulty reaching treatment targets than individuals with diabetes without CVD.More consideration may be required for individuals with a previous cardiovascular event and type 1 diabetes.
### Aims/Hypothesis
Optimal diabetes care and risk factor management are important to delay micro‐ and macrovascular complications in individuals with type 1 diabetes (T1D). Ongoing improvement of management strategies requires the evaluation of target achievement and identification of risk factors in individuals who do (or do not) achieve these targets.
### Methods
Cross‐sectional data were collected from adults with T1D visiting six diabetes centers in the Netherlands in 2018. Targets were defined as glycated hemoglobin (HbA1c) <53 mmol/mol, low‐density lipoprotein‐cholesterol (LDL‐c) <2.6 mmoL/L (no cardiovascular disease [CVD] present) or <1.8 mmoL/L (CVD present), or blood pressure (BP) <$\frac{140}{90}$ mm Hg. Target achievement was compared for individuals with and without CVD.
### Results
Data from 1737 individuals were included. Mean HbA1c was 63 mmol/mol ($7.9\%$), LDL‐c was 2.67 mmoL/L, and BP $\frac{131}{76}$ mm Hg. In individuals with CVD, $24\%$, $33\%$, and $46\%$ achieved HbA1c, LDL‐c, and BP targets respectively. In individuals without CVD these percentages were $29\%$, $54\%$, and $77\%$, respectively. Individuals with CVD did not have any significant risk factors for HbA1c, LDL‐c, and BP target achievement. In comparison, individuals without CVD were more likely to achieve glycemic targets if they were men and insulin pump users. Smoking, microvascular complications, and the prescription of lipid‐lowering and antihypertensive medication were negatively associated with glycemic target achievement. No characteristics were associated with LDL‐c target achievement. Microvascular complications and antihypertensive medication prescription were negatively associated with BP target attainment.
### Conclusion
Opportunities for improvement of diabetes management exist for the achievement of glycemic, lipid, and BP targets but may differ between individuals with and without CVD.
## INTRODUCTION
Individuals with type 1 diabetes (T1D) are at risk for morbidity and mortality as a result of diabetes‐related complications. 1 Although there has been an overall improvement in life expectancy and quality of life over the years, there is still a discrepancy of 11–13 years in life expectancy in individuals with T1D in comparison to controls without diabetes. 2 This reduction in life expectancy remains largely attributed to cardiovascular disease (CVD). 2 The Diabetes Control and Complications Trial/Epidemiology of Diabetes Interventions and Complications trial (DCCT/EDIC) unequivocally demonstrated the importance of strict glycemic control to prevent macrovascular complications. 3 Although glycemic control is the cornerstone for T1D management, studies have shown that both lipid and blood pressure levels also affect the development of micro‐ and macrovascular complications. 1, 4 Moreover, dyslipidemia and hypertension have been hypothesized to have synergistic effects in cardiovascular risk. 5 Lowering blood pressure, and to a lesser extent low‐density lipoprotein‐cholesterol (LDL‐c), has been shown to have protective effects against CVD in individuals with T1D. 6, 7 Assessing whether individuals are reaching treatment goals and identifying subgroup differences in reaching these targets is necessary to improve patient care and self‐management. Despite the large burden of CVD in individuals with T1D, reports on the achievement of LDL‐c and blood pressure targets are limited. In addition, percentages of individuals with T1D reaching treatment targets 6, 8 in glycated hemoglobin (HbA1c), lipid, and blood pressure achievement vary from $10\%$–$39\%$, $24\%$–$73\%$, to $62\%$–$84\%$, respectively. 6, 9 Importantly, assessing the subgroup of individuals with T1D who already experienced a cardiovascular event is particularly relevant, as these individuals are still at (high) risk for recurrent CVD. 10 Data on target achievement in this group are extremely limited.
In this study, we assess the percentage of adults with T1D who achieve HbA1c, LDL‐c, and blood pressure targets, with and without known CVD across six diabetes centers in the Netherlands. Furthermore, differences between those who achieve or do not achieve targets are described to find potential subgroups that require more, or different, attention for cardiovascular risk management.
## Study design and population
This is a cross‐sectional registry‐based study. Electronic patient data were collected from two centers: Diabeter, a specialized T1D treatment and research center with five locations throughout the Netherlands, and the University Medical Center Groningen (UMCG), the Netherlands. Individuals visiting these clinics between 1 January 2018 and 31 December 2018 were included if they were over the age of 18, were diagnosed with T1D, and had used insulin for at least 1 year. T1D diagnosis was determined by the presence of American Diabetes Association (ADA) criteria for diabetes mellitus and a clinical presentation typical for T1D or the presence of autoantibodies. 11 Individuals were excluded if no measurements were present for LDL‐c, blood pressure, or HbA1c levels in 2018.
The Medical Ethical Review Board of the UMCG, Groningen, the Netherlands, declared that this study was not subject to the Dutch Medical Research Involving Human Subjects Act (WMO) and a waiver was granted. The institutional review board approved the study protocol [202000883].
Data extraction for this study is described in detail elsewhere. 12 In summary, demographic, anthropometric, laboratory, and medication data were extracted from electronic medical records (EMRs) from the subjects' annual diabetes complication screening visits in 2018.
## Variable definitions
Demographic data included age and diabetes duration, sex, and ethnicity or parental country of birth. Ethnicity was classified as either western European or non‐western European, as no meaningful subgroups could be formed in the latter. When ethnicity was not available, parental country of birth was used to determine ethnicity. If at least one parent was born outside of western Europe, the individual was considered non‐western European.
CVD was defined as the presence of a positive medical history for coronary artery disease, cerebrovascular disease or transient ischemic attack, peripheral arterial disease, or the prescription of platelet aggregation inhibitors. An individual was considered to have microvascular complications if they had a positive medical history for retinopathy, neuropathy, or nephropathy.
Anthropometric data on blood pressure, height, and weight were also extracted. Body mass index (BMI) was calculated by dividing weight in kilograms by the height in meters squared.
Laboratory measurements included HbA1c, serum creatinine, and LDL‐c. Estimated glomerular filtration rate (eGFR) was calculated using the Chronic Kidney Disease Epidemiology Collaboration formula. 13 Medication prescribed was extracted from the EMRs and relevant medication was coded based on their mechanism of action: antihypertensive, lipid lowering, and platelet aggregation inhibitors. For antihypertensive medication, the number of antihypertensive medications prescribed was calculated and classified as either none, one, two, or three or more.
## Achievement of targets
The achievement of targets was assessed for the outcomes glycemic control, LDL‐c, and blood pressure. The ADA guidelines recommend the HbA1c target to be <53 mmol/mol ($7.0\%$), unless stricter targets can be achieved without risk of hypoglycemia (HbA1c <48 mmol/mol [$6.5\%$]) or when an individual's life expectancy is limited (HbA1c <64 mmol/mol [$8.0\%$]), or the harms outweigh the benefits. 14 The Dutch guidelines similarly recommend striving for a HbA1c <53 mmol/mol, unless the individual is over the age of 70 and has had T1D for longer than 10 years, in which case a higher limit is accepted (HbA1c <64 mmol/mol). 15 Because the distinction as to when to choose a goal of <48 or <64 mmol/mol is difficult to make with the available data, the HbA1c target of <53 mmol/mol was used, which is comparable to other studies. 9 LDL‐c targets were <2.6 mmoL/L for those without CVD and <1.8 mmoL/L for those with CVD. 16 The Dutch guidelines recommend a LDL‐c <2.6 mmoL/L in individuals with high risk of CVD morbidity and mortality. 16 Although the Dutch guidelines do not recommend lipid‐lowering medication in individuals <40 years with a low systematic coronary risk evaluation (SCORE), 16 treatment may be considered. 17 Blood pressure targets were achieved if the measured blood pressure in the outpatient clinic setting was <$\frac{140}{90}$ mm Hg. 16 Although individuals with nephropathy would be considered for stricter targets such as <$\frac{130}{80}$ mm Hg, for the purpose of macrovascular complication risk reduction we used the target of <$\frac{140}{90}$ mm Hg.
The achievement of HbA1c, LDL‐c, and blood pressure targets is shown in Figure 1. Overall, less than a third of the study population achieved an HbA1c target below 53 mmoL/L ($7.0\%$). In the CVD‐ group, the target was achieved slightly more frequently ($$p \leq .17$$). The LDL‐c target was achieved by about half of the study population. Despite lower overall LDL‐c in the CVD+ group, the achievement of this target was significantly lower in the CVD+ group in comparison to the CVD‐ group ($35\%$ vs $54\%$, $$p \leq .001$$). Blood pressure targets were achieved by about three quarters of the study population and were achieved more often by individuals in the CVD‐ group ($77\%$ vs $45\%$, $p \leq .001$).
**FIGURE 1:** *Distribution of targets and percentage of targets achieved for glycated hemoglobin (HbA1c), LDL‐cholesterol (LDL‐c), and systolic blood pressure. In blue the no cardiovascular disease (CVD) group, in pink the CVD group, and in black the total group are shown. (A) percentage of individuals achieving the target HbA1c of <53 mmol/mol (7.0%), (B) percentage of individuals achieving target LDL‐c of <1.8 mmoL/L (CVD) and <2.6 mmoL/L (no CVD), (f) percentage of individuals achieving blood pressure target <140/90 mm Hg. *** = p value <0.001.*
## Statistical analysis
All statistical analyses were conducted using R Statistical software, 18 R Studio Software, 19 and R packages. 20, 21, 22 The study population is described and presented as unadjusted means with SDs, median with interquartile range, or counts with percentages.
Differences in the characteristics between those with and without CVD were evaluated with unpaired t tests, Wilcoxon rank‐sum, chi‐square, or Fisher exact tests where appropriate.
The achievement of targets was further analyzed for those without CVD and those with CVD. Percentages of individuals achieving the target were calculated for each target separately for those with and without CVD.
For the CVD positive (CVD+) and CVD negative (CVD−) group, age‐adjusted odds ratios (aORs) were calculated using multiple logistic regression analysis to determine which risk factors may have an impact on target achievement. Adjustment was made for age as age is a confounder for the achievement of HbA1c, LDL‐c, and blood pressure. 23 aORs were calculated for the following risk factors: sex, continuous subcutaneous insulin infusion (CSII), BMI, microvascular complications, smoking, eGFR, lipid‐lowering medication (LLM), and antihypertensive medication (AHM). Diabetes duration was excluded as diabetes duration and age were highly correlated.
## RESULTS
A total of 2293 individuals visited the six diabetes centers for annual diabetes screening. After excluding individuals with missing data for HbA1c, LDL‐c, and blood pressure, 1737 were included in this study. Table 1 shows the characteristics of the study population by CVD status.
**TABLE 1**
| Characteristics | Whole population (n = 1737) | CVD‐ (n = 1650) | CVD+ (n = 87) |
| --- | --- | --- | --- |
| Age, years | 27 (22, 43) | 26 (22, 39) | 61 (53, 67) |
| Sex, n (%) women | 876 (50) | 840 (51) | 36 (41) |
| Ethnicity, n (%) Western European | 1639 (94) | 1553 (94) | 86 (99) |
| Diabetes duration, years | 16 (10, 24) | 15 (10, 22) | 40 (33, 50) |
| CSII, n (%) yes | 917 (53) | 892 (55) | 25 (29) |
| BMI, kg/m2 | 25.6 ± 4.4 | 25.5 ± 4.4 | 27.1 ± 5.0 |
| Systolic blood pressure, mm Hg | 131 ± 13 | 130 ± 13 | 141 ± 18 |
| Smoking, n (%) yes | Smoking, n (%) yes | Smoking, n (%) yes | Smoking, n (%) yes |
| Current smoker count, n (%) | 234 (14) | 216 (14) | 18 (21) |
| Former smoker count, n (%) | 61 (3.7) | 54 (3.5) | 7 (8.1) |
| Never smoker count, n (%) | 1355 (82) | 1294 (83) | 61 (71) |
| HbA1c, mmol/mol | 63 ± 16 | 63 ± 16 | 63 ± 13 |
| HbA1c, % | 7.9 ± 1.5 | 7.9 ± 1.5 | 7.9 ± 1.2 |
| eGFR, ml min¯1 1.73¯2 | 98 (82, 117) | 100 (84, 118) | 63 (53, 80) |
| Albumin creatinine ratio, mg/mmol | 0.9 (0.50, 1.91) | 0.83 (0.50, 1.83) | 1.70 (0.78, 7.55) |
| LDL‐cholesterol, mmol/L | 2.67 ± 0.79 | 2.69 ± 0.78 | 2.33 ± 0.90 |
| Microvascular complications, n (%) yes | 369 (21) | 310 (19) | 59 (68) |
| Lipid‐lowering medication, n (%) yes | 357 (21) | 281 (17) | 76 (87) |
| Antihypertensive medication, n (%) yes | 303 (17) | 224 (14) | 79 (91) |
| Platelet aggregation inhibitors, n (%) yes | 60 (35) | 12 (7.3) | 48 (55) |
| Coronary heart disease, n (%) yes | 57 (3.3) | ‐ | 57 (65) |
| Cerebral vascular accident or transient ischemic attack, n (%) yes | 13 (0.8) | ‐ | 13 (15) |
| Peripheral arterial disease, n (%) yes | 32 (1.8) | ‐ | 32 (37) |
Participants with CVD were significantly older, had a longer diabetes duration, a greater BMI and higher systolic blood pressure, and were more often smokers or former smokers in comparison to the CVD‐ group. The percentage of individuals with microvascular complications, and with any medication prescription ‐‐ for all working mechanisms ‐‐ was significantly greater in individuals with CVD, whereas CSII use was significantly lower among those with CVD.
## Characteristics of target achievement in CVD+
Characteristics of individuals with CVD achieving glycemic, LDL‐c, and blood pressure targets were heterogenous, and only a few characteristics differed significantly between those who did or did not achieve targets (Table S1). Individuals achieving glycemic targets were significantly less often smokers and were prescribed less LLM. Those achieving LDL‐c targets had significantly longer diabetes duration, poorer renal function and were prescribed more LLM and AHM. No significant differences were found between those who did or did not achieve blood pressure targets.
Figures 2, 3, 4 show the aORs for target achievement for each of the risk factors. Despite some significant differences in characteristics for glycemic and LDL‐c target achievement in individuals with CVD, the aORs showed no significant associations between characteristics and target achievement in this group.
**FIGURE 2:** *Odds ratios of HbA1c target achievement for those with and without cardiovascular disease, adjusted for age. AHM, antihypertensive medication; BMI, body mass index; CSII, continuous subcutaneous insulin infusion; CVD, cardiovascular disease; eGFR, estimated glomerular filtration rate; HbA1c, glycated hemoglobin; LLM, lipid lowering medication; OR, odds ratio.* **FIGURE 3:** *Odds ratios of LDL‐cholesterol (LDL‐c) target achievement for those with and without cardiovascular disease, adjusted for age. AHM, antihypertensive medication; BMI, body mass index; CSII, continuous subcutaneous insulin infusion; CVD, cardiovascular disease; eGFR, estimated glomerular filtration rate; LLM, lipid lowering medication; OR, odds ratio.* **FIGURE 4:** *Odds ratios of blood pressure target achievement for those with and without cardiovascular disease, adjusted for age. AHM, antihypertensive medication; BMI, body mass index; CSII, continuous subcutaneous insulin infusion; CVD, cardiovascular disease; eGFR, estimated glomerular filtration rate; LLM, lipid lowering medication; OR, odds ratio.*
## Characteristics of target achievement in CVD−
In contrast, in the CVD− group several factors were found to differ significantly between those who did or did not achieve targets (Table S2). In individuals achieving glycemic targets, individuals were significantly older, more often men, of western European decent, CSII users, more often nonsmokers, with poorer renal function, lower LDL‐c, and less LLM prescribed. BMI and HbA1c were significantly lower in individuals who did or did not achieve LDL‐c targets. Finally, individuals without CVD achieving blood pressure targets were younger, were more often women, had a shorter diabetes duration, were more often CSII users, had lower BMI, less microvascular complications, were more often nonsmokers, had better renal function, and used less LLM and AHM.
When adjusted for age, the association for achieving glycemic targets was significantly greater in men (aOR: 1.31, $$p \leq .013$$) and CSII users (aOR: 1.39, $$p \leq .003$$) (Figure 2). Those smoking (aOR: 0.38, $p \leq .001$), with microvascular complications (aOR: 0.73, $$p \leq .044$$), and prescribed LLM (aOR: 0.51, p = <.001) and AHM (aOR: 0.58, $$p \leq .004$$) were significantly less likely to achieve glycemic targets (Figure 2). Other characteristics such as age, diabetes duration, BMI, and renal function were found not to increase chances of glycemic target achievement.
Associations between characteristics and LDL‐c achievement are illustrated in Figure 3. Only diabetes duration and BMI were significantly associated; however, both showed no or very weak associations with LDL‐c achievement.
Finally, men (aOR: 0.57, $p \leq .001$), individuals with microvascular complications (aOR: 0.61, $$p \leq .002$$), and those prescribed AHM (aOR: 0.40, $p \leq .001$) were significantly less likely to achieve blood pressure targets (Figure 4). Diabetes duration, BMI, and renal function did not show greater odds of blood pressure target achievement.
## DISCUSSION
In this study, we described the achievement of glycemic, LDL‐c, and blood pressure targets in individuals with T1D with and without CVD. Furthermore, we reported the differences between those able to achieve these targets and any associations between characteristics and target achievement.
We found that the majority of our population did not achieve glycemic and LDL‐c targets. In contrast, blood pressure targets were achieved by a great majority. Although suboptimal achievement of glycemic, LDL‐c, and blood pressure targets has been described before in individuals with T1D without CVD, 9, 24, 25, 26 our study is one of the first to describe target achievement, specifically in individuals with T1D who have established CVD. Individuals with CVD were significantly less likely to achieve LDL‐c targets and blood pressure targets in comparison to those without CVD. As treatment of CVD improves and the life expectancy of individuals with T1D continues to rise, it becomes increasingly important to study ways to improve CVD prevention and target achievement in these individuals. 10 In individuals with CVD, no measured characteristics were found to be associated with the likelihood to achieve targets. This may suggest that there are other factors, not measured in this study, which are worth investigating. Diet, physical activity, and psychological factors such as stress and anxiety are among the risk factors that could potentially influence target achievement. 27 However, the small sample size of individuals with CVD may have contributed to the lack of significant associations found between those that did or did not achieve targets. Larger studies in this subgroup of individuals with T1D may be beneficial. As individuals with CVD are significantly less likely to achieve targets, there is great importance in improving management strategies in this group.
Within the subgroup of individuals without CVD, certain characteristics were demonstrated to affect the likelihood of target achievement, which may help to identify individuals who require more attention. Smoking and the prescription of LLM and AHM were negatively associated with glycemic target achievement, suggesting that individuals with known CVD risk factors such as smoking, dyslipidemia, and hypertension are less likely to achieve glycemic targets. Considering the importance of glycemic management for CVD risk reduction, these individuals may be at greater risk for future CVD events. 3, 28 Furthermore, in our study blood pressure targets were less likely to be achieved if an individual was a man, had microvascular complications, and was prescribed AHM.
Individuals who have developed microvascular complications have an even greater risk for developing CVD, particularly individuals with diabetic nephropathy. 28 As AHM prescription indicates either the presence of hypertension or microalbuminuria, our study further suggests that individuals who fail to achieve targets are those comorbid for established risk factors. 28 Although these findings illustrate an association between comorbid CVD risk factors and target achievement, further longitudinal studies could shed light on potential causal relationships.
Despite an overall lower LDL‐c in the group with CVD, fewer individuals were able to reach the target of 1.8 mmol/L. Recently, the European Society of Cardiology/European Atherosclerosis Society guidelines have further lowered the LDL‐c target for individuals with CVD to <1.4 mmoL/L. 29 Whether these targets can be met is ultimately based on the prescription practices of health care providers. 29 *Therapeutic inertia* as a result of insufficient training and lack of knowledge of treatment options are some of the barriers for optimal LDL‐c management among health care providers. 30 Addressing these issues will be necessary in order to improve LDL‐c management.
Lastly, it is important to acknowledge that for many individuals with T1D HbA1c targets are difficult to attain and can lead to a higher frequency of hypoglycemic episodes and, importantly, can become a source of frustration and feelings of failure. 31 Factors such as fear of hypoglycemia, lack of access to technological devices, and self‐efficacy are just a few areas that can influence glycemic management. 31 The role of the health care provider is to recognize these potential pitfalls and to help guide those in their care.
## Strengths and limitations
Strengths of this study include the large sample size, as well as the use of real‐world data from six diabetes centers in the Netherlands.
This study has three limitations. First, the cross‐sectional study design hampers any conclusion on the causality of the associations found. The presence of CVD risk factors was found to be negatively associated with target achievement. However, it is unclear whether the risk factors themselves were barriers to target achievement or a consequence of it. Nonetheless, this study provides interesting insights and provides some directions for further research. Second, data quality may be limited by the completeness of medical records. In particular, albumin creatinine ratios were unavailable for over >$50\%$ of individuals included in this study. However, the missing data rate was comparable to other registry‐based studies. Third, no data were available on the reasons for medication discontinuation. Particularly for LLM, intolerances, patient preferences, or use of alternative supplements such as red rice yeast were not recorded, which could have provided more insights as to why LLM was not prescribed for some individuals despite recommendations.
## Recommendations
Further research should be conducted on the temporal relationship between these characteristics and target achievement. In particular, more attention and research are required for the management of individuals with T1D and CVD, as these individuals are the least likely to achieve treatment targets.
## CONCLUSIONS
In conclusion, this study emphasizes that individuals with T1D with established CVD are at a greater risk for not achieving lipid and blood pressure targets. Opportunities for the improvement of glycemic, lipid, and blood pressure management exist but may differ between individuals with and without CVD.
## AUTHOR CONTRIBUTIONS
Rita D. M. Varkevisser contributed to the design, data analysis, and interpretation of the results and authored the paper. Erwin Birnie contributed to the design and interpretation of the results and supervised the work. Dick Mul and Peter R. van Dijk contributed to the design and interpretation of the results. Henk‐Jan Aanstoot, Bruce HR Wolffenbuttel, and Melanie M. van der Klauw supervised the work and, in that role, contributed to the design, data interpretation, and writing of the paper.
## CONFLICT OF INTEREST STATEMENT
Erwin Birnie, Dick Mul, and Henk‐Jan Aanstoot are employed at Diabeter Netherlands, an independent clinic (owned by Medtronic), with brand‐agnostic prescription under EU/Dutch health care laws. The research presented here was independently performed and there are no conflicts of interest.
## References
1. Chiesa ST, Marcovecchio ML. **Preventing cardiovascular complications in type 1 diabetes: the need for a lifetime approach**. *Front Pediatr* (2021.0) **9**. PMID: 34178905
2. Verges B. **Cardiovascular disease in type 1 diabetes: a review of epidemiological data and underlying mechanisms**. *Diabetes Metab* (2020.0) **46** 442-449. PMID: 32998054
3. Nathan DM, Cleary PA, Backlund J‐YC. **Intensive diabetes treatment and cardiovascular disease in patients with type 1 diabetes**. *N Engl J Med* (2005.0) **353** 2643-2653. PMID: 16371630
4. Tell S, Nadeau KJ, Eckel RH. **Lipid management for cardiovascular risk reduction in type 1 diabetes**. *Curr Opin Endocrinol Diabetes Obes* (2020.0) **27** 207-214. PMID: 32618632
5. Cleland SJ. **Cardiovascular risk in double diabetes mellitus—when two worlds collide**. *Nat Rev Endocrinol* (2012.0) **8** 476-485. PMID: 22488644
6. Salna I, Salna E, Pahirko L. **Achievement of treatment targets predicts progression of vascular complications in type 1 diabetes**. *J Diabetes Complications* (2021.0) **35**. PMID: 34635403
7. Kearney PM, Blackwell L, Collins R. **Efficacy of cholesterol‐lowering therapy in 18,686 people with diabetes in 14 randomised trials of statins: a meta‐analysis**. *Lancet* (2008.0) **371** 117-125. PMID: 18191683
8. Prigge R, McKnight JA, Wild SH. **Scottish diabetes research network epidemiology G. international comparison of glycemic control in people with type 1 diabetes: an update and extension**. *Diabet Med* (2021.0) **39**. PMID: 34890078
9. Shah VN, Grimsmann JM, Foster NC. **Undertreatment of cardiovascular risk factors in the type 1 diabetes exchange clinic network (United States) and the prospective diabetes follow‐up (Germany/Austria) registries**. *Diabetes Obes Metab* (2020.0) **22** 1577-1585. PMID: 32329127
10. Bebu I, Schade D, Braffett B. **Risk factors for first and subsequent CVD events in type 1 diabetes: the DCCT/EDIC study**. *Diabetes Care* (2020.0) **43** 867-874. PMID: 32001614
11. **2. Classification and diagnosis of diabetes: standards of medical Care in Diabetes—2021**. *Diabetes Care* (2021.0) **44** S15-S33. PMID: 33298413
12. Varkevisser RDM, Birnie E, Vollenbrock CE. **Cardiovascular risk management in people with type 1 diabetes: performance using three guidelines**. *BMJ Open Diabetes Res Care* (2022.0) **10**
13. Levey AS, Stevens LA, Schmid CH. **A new equation to estimate glomerular filtration rate**. *Ann Intern Med* (2009.0) **150** 604-612. PMID: 19414839
14. **6. Glycemic targets: standards of medical Care in Diabetes—2020**. *Diabetes Care* (2020.0) **43** S66-S76. PMID: 31862749
15. **Diabetes mellitus type 2 (M01)**. (2018.0)
16. **Cardiovascular Risk Management**. (2019.0)
17. **9. Cardiovascular disease and risk management: standards of medical Care in Diabetes—2018**. *Diabetes Care* (2018.0) **41** S86-S104. PMID: 29222380
18. 18
R Core Team
. R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing; 2021.. *R: A Language and Environment for Statistical Computing* (2021.0)
19. 19
RStudio Team
. RStudio: Integrated Development for R, PBC. RStudio; 2020.. *RStudio: Integrated Development for R, PBC* (2020.0)
20. Wickham Hadley, François Romain, Henry Lionel, Müller K. **A grammar of data manipulation**
21. Wickham H.. **tidyr: Tidy Messy Data. R package version 1.1.3 ed2021**
22. Hadley W. *ggplot2: Elegant Graphics for Data Analysis* (2016.0)
23. Singh GM, Danaei G, Pelizzari PM. **The age associations of blood pressure, cholesterol, and glucose: analysis of health examination surveys from international populations**. *Circulation* (2012.0) **125** 2204-2211. PMID: 22492580
24. Syed A, Hussain S, Nightingale P. **Cardiovascular risk factors and their management in 1282 adult people with type 1 diabetes**. *Curr Med Res Opin* (2007.0) **23** 2921-2927. PMID: 17931464
25. Lithovius R, Harjutsalo V, Forsblom C, Saraheimo M, Groop PH. **The consequences of failure to achieve targets of guidelines for prevention and treatment of diabetic complications in patients with type 1 diabetes**. *Acta Diabetol* (2015.0) **52** 31-38. PMID: 24849006
26. Sonmez A, Haymana C, Bayram F. **Turkish nationwide survey of glycemic and other metabolic parameters of patients with diabetes mellitus (TEMD study)**. *Diabetes Res Clin Pract* (2018.0) **146** 138-147. PMID: 30244051
27. Ruissen MM, Regeer H, Landstra CP. **Increased stress, weight gain and less exercise in relation to glycemic control in people with type 1 and type 2 diabetes during the COVID‐19 pandemic**. *BMJ Open Diabetes Res Care* (2021.0) **9**
28. Schofield J, Ho J, Soran H. **Cardiovascular risk in type 1 diabetes mellitus**. *Diabetes Ther* (2019.0) **10** 773-789. PMID: 31004282
29. Packard C, Chapman MJ, Sibartie M, Laufs U, Masana L. **Intensive low‐density lipoprotein cholesterol lowering in cardiovascular disease prevention: opportunities and challenges**. *Heart* (2021.0) **107** 1369-1375. PMID: 33795379
30. Dixon DL, Sharma G, Sandesara PB. **Therapeutic inertia in cardiovascular disease prevention: time to move the bar**. *JACC* (2019.0) **74** 1728-1731. PMID: 31558257
31. Adu MA‐O, Malabu UH, Malau‐Aduli AA‐O, Malau‐Aduli BS. **Enablers and barriers to effective diabetes self‐management: a multi‐national investigation**. *PLoS One* (2019.0) **14** 1932-6203
|
---
title: 'Remnant cholesterol is independently associated with diabetes, even if the
traditional lipid is at the appropriate level: A report from the REACTION study'
authors:
- Binqi Li
- Xin Zhou
- Weiqing Wang
- Zhengnan Gao
- Li Yan
- Guijun Qin
- Xulei Tang
- Qin Wan
- Lulu Chen
- Zuojie Luo
- Guang Ning
- Yiming Mu
journal: Journal of Diabetes
year: 2023
pmcid: PMC10036259
doi: 10.1111/1753-0407.13362
license: CC BY 4.0
---
# Remnant cholesterol is independently associated with diabetes, even if the traditional lipid is at the appropriate level: A report from the REACTION study
## Abstract
Highlights The first multicenter, large‐sample study of the relationship between remnant cholesterol (RC) and diabetes. RC is independently associated with diabetes. RC represents residual diabetes risk beyond traditional lipids.
### Background
The association between remnant cholesterol (RC) and diabetes remains unclear because of limited study and data. This study attempted to explore the association between RC and diabetes in a large sample, multicenter general population.
### Methods
The current study included 36 684 participants from eight provinces across China. Subjects were quartered according to the RC quartile. Logistic regression analysis was used to evaluate the association between RC and diabetes.
### Results
After adjusting for potential confounding factors, RC was still significantly associated with diabetes (Q4: odds ratio [OR]:1.147, $95\%$ confidence interval [CI]: 1.049–1.254, $$p \leq .003$$). In addition, RC and diabetes were still significantly associated when triglycerides (TG) were <1.7 mmol/L (Q4: OR: 1.155, $95\%$ CI: 1.005–1.327, $$p \leq .042$$), low‐density lipoprotein cholesterol (LDL‐C) <3.4 mmol/L (Q4: OR: 1.130, $95\%$ CI: 1.011–1.264, $$p \leq .032$$), or HDL‐C (high‐density lipoprotein cholesterol) ≥1.0 mmol/L (Q4: OR: 1.116, $95\%$ CI: 1.007–1.237, $$p \leq .037$$). In the stratification analysis, elevated RC was significantly associated with diabetes in subjects with systolic blood pressure (SBP) <140 mm Hg and diastolic blood pressure (DBP) <90 mm Hg, 60 ≤ estimated glomerular filtration rate (eGFR) ≤90 ml/min per 1.73 m2, younger than 55 years old and female.
### Conclusion
In the Chinese community, RC is significantly correlated with diabetes, even when TG, LDL‐C, or HDL‐C were controlled within the appropriate range recommended by the guidelines.
## INTRODUCTION
Diabetes is one of the most serious epidemic diseases, resulting in a 2 ‐ to 3‐fold increasing danger of cardiovascular disease (CVD), 1 a 2 ‐ to 3‐fold increasing danger of all‐cause mortality, and a 20‐year reduction in life expectancy. 2, 3 As of 2021, 536.6 million people aged 21 to 79 worldwide were reported to have diabetes. 4 The worrying thing is that if the prevalence of diabetes continues to rise dramatically, it is expected that by 2045, approximately $10.9\%$ of the global population (about 700 million people) would have diabetes. 5 However, current measures for preventing hyperglycemia are poorly beneficial, so there is an urgent need to find novel risk factors associated with diabetes.
In recent years, researchers have concentrated on the influence of dyslipidemia on diabetes. 6 Previous studies have found that people who take lipid‐lowering drugs have a lower risk of diabetes. 7, 8 Besides, a cohort study indicated that patients with dyslipidemia had 1.7 times the risk of diabetes than those with normal lipids. 9 However, few studies have focused on the relationship between diabetes and remnant cholesterol (RC), which is a hot topic of research in the cardiovascular field in recent years and is considered an important cause of the residual risk of CVD. 10 RC is the metabolic residues of triglyceride‐rich lipoproteins (TGRL). To be specific, it refers to the metabolic residues of very‐low‐density lipoprotein (VLDL), intermediate‐density lipoprotein (IDL), and chylomicron when not fasting and the metabolic residues of VLDL and IDL when fasting. 11 It has been reported that the higher the level of RC, the greater the likelihood of developing CVD, 10 especially in people with diabetes. 12 Recent studies have also revealed a positive association between RC and diabetic complications. 13, 14 However, large sample, multicenter studies on the relationship between RC and diabetes have not yet been reported. Hence, the purpose of the current study is to investigate the relationship between RC and diabetes in a large sample, multicenter Chinese community population.
## Study population
The data of the current work was from eight centers of the REACTION (Risk Evaluation of cAncers in Chinese diabeTic Individuals) study. 15 A total of 53 639 participants aged over 40 or older participated in the study between March and December 2012. A total of 3042 participants with preexisting kidney disease and other serious illnesses, 2231 participants using lipid‐lowering drugs, 11 280 participants with missing important data, and 402 participants with calculated RC values ≤0 were excluded. Finally, 36 684 participants were enrolled.
## Data collection
Trained investigators helped participants fill out detailed questionnaires including demographic data, past medical history, family history of diabetes, current medication, current smoking and drinking habits, current occupation, and physical activity. Information on physical activity was gathered using the short form of the International Physical Activity Questionnaire: high‐level physical activity was classified as vigorous physical activity that lasted longer than 10 min each within the past 7 days, which would make people breath with difficulty, for example, playing basketball, swimming, and running; moderate‐level physical activity was defined as a minimum of 10 min of physical activity in each of the past 7 days, which required people to breathe a little harder than normally, for instance jogging, playing table tennis, golf, or tai chi but no vigorous physical activity; low physical activity levels were determined as no physical activity or only some physical activity like walking, but could not meet the criteria for high‐ and medium‐level physical activity.
Systolic blood pressure (SBP) and diastolic blood pressure (DBP) were measured three times, each 5 min apart. The average of the three results was taken for statistical analysis. The height, weight, waist circumference (WC), and hip circumference were measured and recorded after the participants removed jackets and shoes.
The participants fasted for 8–10 h before the assay. Serum TG, low‐density lipoprotein cholesterol (LDL‐C), high‐density lipoprotein cholesterol (HDL‐C), total cholesterol (TC), aspartate transferase (AST), alanine transferase (ALT), glutamine transferase (GGT), serum creatinine (SCr), and other biochemical indexes were measured on an autoanalyzer; hemoglobin A1c (HbA1c) was measured by high‐performance liquid chromatography method; Fasting insulin was measured with chemiluminescent immunoassay. After fasting blood glucose extraction, participants without or with diabetes were tested for 75 g oral glucose tolerance or 100 g steamed‐bread meal, respectively, post‐load blood glucose (PBG) was extracted 2 h after the first blood sample was drawn.
## Calculations
RC = TC−HDL‐C−LDL‐C. 16 Body mass index (BMI) (kg/m2) = weight (Kg)/height2 (m). Waist‐to‐hip ratio (WHR) = WC/HC. eGFR (mL/min per 1.73 m2) = 175 × (SCr in mg/dL) −1.154 × age−0.203 × (0.742 for women) × (1.212 if African American).
## Definitions
RC was divided into four groups based on quartiles: <$25\%$ group (the control group), $25\%$–$50\%$ group, $50\%$–$75\%$ group, and ≥$75\%$ group. Diabetes was defined according to the American Diabetes Association's criteria 17: FPG ≥7.0 mmol/L or PBG ≥11.1 mmol/L or HbA1c ≥$6.5\%$ or self‐reported diabetes.
## Statistical analysis
Statistical analysis was performed using SPSS 24.0 (IBM, Chicago, IL). Continuous variables were expressed as mean ± SD and categorical variables were presented numerically (proportionally). One‐way analysis of variance (ANOVA) was used to test the difference between continuous variables and the chi‐square test was used to analyze the categorical variables. The association between RC and diabetes was delineated by logistic regression analyses. Adjusted variables were examined using collinearity diagnosis according to the following criteria: [1] variance inflation factor >5; [2] condition index >30; and [3] variance proportions >$50\%$. The adjusted variables had the following criteria: [1] in the current study, there was a significant difference between the diabetic group and the nondiabetic group, or it was significant for the occurrence and development of diabetes in clinical practice; and [2] there was no collinearity. Model 0 was unadjusted. Model 1 was adjusted for age, sex, and center. Model 2: Model 1 + myocardial infarction (MI), stroke, coronary heart disease (CHD), occupation, smoking habit, drinking habit, physical activity level, and family history of diabetes. Model 3 = Model 2 + ALT, AST, GGT, and eGFR. Model 4 = Model 3 + SBP, DBP, BMI, WHR, and insulin. Model 5 = Model 4 + HDL‐C and LDL‐C. Model 6 = Model 5 + TG. Moreover, according to the appropriate levels of LDL‐C, HDL‐C, and TG recommended by the 2016 edition of the guidelines for the management of dyslipidemia in Chinese adults, 18 the association between RC and diabetes was also tested when LDL‐C <3.4 mmol/L, HDL‐C ≥1.0 mmol/L, and TG <1.7 mmol/L, respectively. In addition, the current study explored the relationship between RC and diabetes in BMI, BP, age, and eGFR subgroups. All statistical tests were two‐sided and $p \leq .05$ was considered statistically significant.
## Clinical characteristics of the study population
The present study included 36 684 participants (Table 1), with 11 321 ($30.9\%$) males and 25 363 ($69.1\%$) females, and their median age (Q1–Q3) was 57 [52, 64]. Compared with participants without diabetes, participants with diabetes had older age, higher RC, LDL‐C, TC, TG, BMI, WHR, ALT, AST, GGT, SBP, and DBP and lower HDL‐C and eGFR. Meanwhile, males, frequent smoking, MI, stroke, CHD, manual labor, retirement or unemployment, and a family history of diabetes were more common in participants with diabetes than in participants without diabetes.
**TABLE 1**
| Variable | Total | No diabetes | Diabetes | p value |
| --- | --- | --- | --- | --- |
| N | 36 684 | 27 268 | 9416 | |
| RC, mmol/L | 0.69 (0.49, 0.95) | 0.47 (0.66， 0.91) | 0.55 (0.77, 1.08) | <.001 |
| Age, years | 57 (52, 64) | 56 (51, 62) | 60 (55, 68) | <.001 |
| Sex, (%) | | | | <.001 |
| Male | 11 321 (30.9) | 7817 (28.7) | 3504 (37.2) | |
| Female | 25 363 (69.1) | 19 451 (71.3) | 5912 (62.8) | |
| LDL‐C, mmol/L | 2.93 (2.36, 3.55) | 2.93 (2.36, 3.53) | 2.96 (2.36, 3.58) | .013 |
| HDL‐C, mmol/L | 1.29 (1.09, 1.52) | 1.32 (1.11, 1.55) | 1.21 (1.03, 1.42) | <.001 |
| TC, mmol/L | 5.05 (4.32, 5.80) | 5.04 (4.32, 5.76) | 5.10 (4.33, 5.89) | <.001 |
| TG, mmol/L | 1.37 (0.98, 1.97) | 1.29 (0.93, 1.83) | 1.65 (1.15, 2.36) | <.001 |
| BMI, Kg/m2 | 24.27 (22.13, 26.57) | 21.84 (23.92, 26.21) | 25.20 (23.12, 27.53) | <.001 |
| WHR | 0.89 (0.84, 0.93) | 0.88 (0.83, 0.92) | 0.90 (0.86, 0.95) | <.001 |
| ALT, U/L | 15 (11, 21) | 14 (10, 20) | 16 (12, 24) | <.001 |
| AST, U/L | 20 (17, 25) | 20 (17, 24) | 21 (17, 26) | <.001 |
| GGT, U/L | 21 (15, 32) | 19 (14, 29) | 25 (17, 39) | <.001 |
| SBP, mm Hg | 129 (117, 144) | 127 (115, 141) | 136 (123, 150) | <.001 |
| DBP, mm Hg | 77 (70, 84) | 76 (70, 83) | 78 (71, 85) | <.001 |
| eGFR, mL/(min·1.73 m2) | 89.78 (80.09, 100.68) | 90.43 (81.09, 00.94) | 87.80 (77.27, 99.73) | <.001 |
| Drinking, (%) | | | | .664 |
| Never | 27 424 (74.8) | 20 283 (74.4) | 7141 (75.8) | |
| Occasional | 6791 (18.5) | 5230 (19.2) | 1561 (16.6) | |
| Frequently | 2467 (6.7) | 1755 (6.4) | 714 (7.6) | |
| Smoking, (%) | | | | <.001 |
| Never | 31 477 (85.8) | 23 520 (86.3) | 7957 (84.5) | |
| Occasional | 864 (2.4) | 617 (2.3) | 247 (2.6) | |
| Frequently | 4343 (11.8) | 3131 (11.5) | 1212 (12.9) | |
| Physical activity, (%) | | | | .09 |
| Low | 29 283 (79.9) | 21 590 (79.2) | 7693 (81.7) | |
| Moderate | 5165 (14.1) | 3971 (14.6) | 1194 (12.7) | |
| High | 2236 (6.1) | 1707 (6.3) | 529 (5.6) | |
| Occupation (%) | | | | .011 |
| Manual labor: worker/peasant/soldier | 4399 (12.0) | 3283 (12.0) | 1116 (11.9) | |
| Administrative staff/doctor/teacher/scientist | 1124 (3.1) | 840 (3.1) | 284 (3.0) | |
| Individual household/merchant | 2728 (7.4) | 2250 (8.3) | 478 (5.1) | |
| Retire/unemployed | 28 433 (77.5) | 20 895 (76.7) | 7538 (80.1) | |
| Myocardial infarction, (%) | | | | <.001 |
| Yes | 123 (0.3) | 62 (0.2) | 61 (0.6) | |
| No | 36 561 (99.7) | 27 206 (99.8) | 9355 (99.4) | |
| Stroke, (%) | | | | <.001 |
| Yes | 425 (1.2) | 265 (1.0) | 160 (1.7) | |
| No | 36 259 (98.8) | 27 003 (99.0) | 9256 (98.3) | |
| Coronary heart disease, (%) | | | | <.001 |
| Yes | 1270 (3.5) | 720 (2.6) | 550 (5.8) | |
| No | 35 414 (96.5) | 26 548 (97.4) | 8866 (94.2) | |
| Family history of diabetes, (%) | | | | <.001 |
| Yes | 6428 (17.5) | 4203 (15.4) | 2225 (23.6) | |
| No | 30 256 (82.5) | 23 065 (84.6) | 7191 (76.4) | |
## Association between RC and diabetes
Table 2 shows the association between RC and diabetes after potential confounders were adjusted and separated by sex. In the total participants, RC exhibited a significant adjusted odds ratio (OR) in models 0–6. After all potential confounders were adjusted in model 6, the third and fourth quartile of RC were still significantly associated with diabetes (Q3: OR: 1.132, $95\%$ CI: 1.047–1.223, $$p \leq .002$$; Q4: OR:1.147, $95\%$ CI: 1.049–1.254, $$p \leq .003$$). In the female participants, RC remained significantly associated with diabetes though all potential confounders were adjusted in model 6 (Q3: OR: 1.168, $95\%$ CI: 1.057–1.291, $$p \leq .002$$; Q4: OR: 1.228, $95\%$ CI: 1.097–1.975, $p \leq .001$). In the male participants, RC and diabetes were significantly associated in model 5 (Q3: OR: 1.133, $95\%$ CI: 1.002–1.282, $$p \leq .046$$; Q4: OR: 1.144, $95\%$ CI: 1.010–1.295, $$p \leq .034$$), however, after TG was adjusted in Model 6, the association between RC and diabetes no longer existed (Q4: OR: 0.985, $95\%$ CI: 0.848–1.145, $$p \leq .846$$).
**TABLE 2**
| RC four categories | Model 0 | Model 0.1 | Model 1 | Model 1.1 | Model 2 | Model 2.1 | Model 3 | Model 3.1 | Model 4 | Model 4.1 | Model 5 | Model 5.1 | Model 6 | Model 6.1 |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| RC four categories | OR (95% CI) | p value | OR (95% CI) | p value | OR (95% CI) | p value | OR (95% CI) | p value | OR (95% CI) | p value | OR (95% CI) | p value | OR (95% CI) | p value |
| Association between RC and diabetes in the total subjects | Association between RC and diabetes in the total subjects | Association between RC and diabetes in the total subjects | Association between RC and diabetes in the total subjects | Association between RC and diabetes in the total subjects | Association between RC and diabetes in the total subjects | Association between RC and diabetes in the total subjects | Association between RC and diabetes in the total subjects | Association between RC and diabetes in the total subjects | Association between RC and diabetes in the total subjects | Association between RC and diabetes in the total subjects | Association between RC and diabetes in the total subjects | Association between RC and diabetes in the total subjects | Association between RC and diabetes in the total subjects | Association between RC and diabetes in the total subjects |
| Group 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 |
| Group 2 | 1.191 (1.110, 1.279) | <.001 | 1.116 (1.037, 1.201) | .003 | 1.105 (1.026, 1.189) | .008 | 1.093 (1.014, 1.179) | .02 | 1.091 (1.011, 1.178) | .026 | 1.088 (1.008, 1.175) | .031 | 1.063 (0.984, 1.148) | .119 |
| Group 3 | 1.485 (1.385, 1.591) | <.001 | 1.355 (1.260, 1.456) | <.001 | 1.337 (1.243, 1.438) | <.001 | 1.266 (1.175, 1.364) | <.001 | 1.228 (1.138, 1.325) | <.001 | 1.196 (1.108, 1.291) | <.001 | 1.132 (1.047, 1.223) | .002 |
| Group 4 | 2.145 (2.006, 2.294) | <.001 | 1.928 (1.798, 2.067) | <.001 | 1.893 (1.765, 2.031) | <.001 | 1.648 (1.532, 1.772) | <.001 | 1.477 (1.371, 1.592) | <.001 | 1.369 (1.269, 1.477) | <.001 | 1.147 (1.049, 1.254) | .003 |
| Association between RC and diabetes in female subjects | Association between RC and diabetes in female subjects | Association between RC and diabetes in female subjects | Association between RC and diabetes in female subjects | Association between RC and diabetes in female subjects | Association between RC and diabetes in female subjects | Association between RC and diabetes in female subjects | Association between RC and diabetes in female subjects | Association between RC and diabetes in female subjects | Association between RC and diabetes in female subjects | Association between RC and diabetes in female subjects | Association between RC and diabetes in female subjects | Association between RC and diabetes in female subjects | Association between RC and diabetes in female subjects | Association between RC and diabetes in female subjects |
| Group 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 |
| Group 2 | 1.253 (1.143, 1.374) | <.001 | 1.130 (1.028, 1.244) | .012 | 1.123 (1.020, 1.237) | .018 | 1.124 (1.019, 1.240) | .019 | 1.120 (1.013, 1.238) | .027 | 1.116 (1.010, 1.234) | .032 | 1.090 (0.986, 1.205) | .093 |
| Group 3 | 1.655 (1.514, 1.809) | <.001 | 1.373 (1.252, 1.507) | <.001 | 1.358 (1.237, 1.491) | <.001 | 1.316 (1.196, 1.448) | <.001 | 1.279 (1.159, 1.411) | <.001 | 1.238 (1.122, 1.367) | <.001 | 1.168 (1.057, 1.291) | .002 |
| Group 4 | 2.628 (2.413, 2.863) | <.001 | 2.050 (1.875, 2.242) | <.001 | 2.010 (1.836, 2.200) | <.001 | 1.827 (1.665, 2.005) | <.001 | 1.622 (1.474, 1.785) | <.001 | 1.479 (1.342, 1.631) | <.001 | 1.228 (1.097, 1.375) | <.001 |
| Association between RC and diabetes in male subjects | Association between RC and diabetes in male subjects | Association between RC and diabetes in male subjects | Association between RC and diabetes in male subjects | Association between RC and diabetes in male subjects | Association between RC and diabetes in male subjects | Association between RC and diabetes in male subjects | Association between RC and diabetes in male subjects | Association between RC and diabetes in male subjects | Association between RC and diabetes in male subjects | Association between RC and diabetes in male subjects | Association between RC and diabetes in male subjects | Association between RC and diabetes in male subjects | Association between RC and diabetes in male subjects | Association between RC and diabetes in male subjects |
| Group 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 |
| Group 2 | 1.161 (1.037, 1.300) | <.001 | 1.097 (0.978, 1.231) | .116 | 1.076 (0.957, 1.210) | .22 | 1.055 (0.937, 1.188) | .379 | 1.053 (0.934, 1.188) | .397 | 1.052 (0.933, 1.187) | .408 | 1.033 (0.915, 1.165) | .603 |
| Group 3 | 1.347 (1.201, 1.511) | <.001 | 1.297 (1.154, 1.457) | <.001 | 1.272 (1.129, 1.431) | <.001 | 1.177 (1.043, 1.328) | .008 | 1.144 (1.012, 1.294) | .031 | 1.133 (1.002, 1.282) | .046 | 1.084 (0.956, 1.229) | .209 |
| Group 4 | 1.577 (1.411, 1.763) | <.001 | 1.592 (1.421, 1.784) | <.001 | 1.558 (1.388, 1.749) | <.001 | 1.291 (1.144, 1.456) | <.001 | 1.196 (1.058, 1.352) | .004 | 1.144 (1.010, 1.295) | .034 | 0.985 (0.848, 1.145) | .846 |
## Association between RC and diabetes in participants with TG <1.7 mmol/L, LDL‐C < 3.4 mmol/L, or HDL‐C ≥1.0 mmol/L
In Table 3, RC and diabetes were still significantly associated when TG <1.7 mmol/L (Q4: OR: 1.155, $95\%$ CI: 1.005–1.327, $$p \leq .042$$), LDL‐C <3.4 mmol/L (Q2: OR: 1.103, $95\%$ CI: 1.008–1.208, $$p \leq .034$$; Q3: OR: 1.151, $95\%$ CI: 1.047–1.264, $$p \leq .003$$; Q4: OR: 1.130, $95\%$ CI: 1.011–1.264, $$p \leq .032$$), or HDL‐C ≥1.0 mmol/L (Q3: OR: 1.140, $95\%$ CI: 1.045–1.243, $$p \leq .003$$; Q4: OR: 1.116, $95\%$ CI: 1.007–1.237, $$p \leq .037$$).
**TABLE 3**
| RC four categories | TG <1.7 mmol/L | TG <1.7 mmol/L.1 | LDL‐C <3.4 mmol/L | LDL‐C <3.4 mmol/L.1 | HDL‐C ≥1.0mml/L | HDL‐C ≥1.0mml/L.1 |
| --- | --- | --- | --- | --- | --- | --- |
| RC four categories | OR (95% CI) | p value | OR (95% CI) | p value | OR (95% CI) | p value |
| Group 1 | 1 | 1 | 1 | 1 | 1 | 1 |
| Group 2 | 1.086 (1.000, 1.179) | .051 | 1.103 (1.008, 1.208) | .034 | 1.080 (0.991, 1.176) | .078 |
| Group 3 | 1.063 (0.969, 1.166) | .198 | 1.151 (1.047, 1.264) | .003 | 1.140 (1.045, 1.243) | .003 |
| Group 4 | 1.155 (1.005, 1.327) | .042 | 1.130 (1.011, 1.264) | .032 | 1.116 (1.007, 1.237) | .037 |
## Stratification analysis
The results of the stratification analysis were presented in Table 4. RC was significantly associated with diabetes regardless of whether participants were overweight or not (BMI <24 Kg/m2: Q3: OR: 1.148, $95\%$ CI: 1.015–1.297, $$p \leq .028$$; Q4: OR: 1.175, $95\%$ CI: 1.013–1.364, $$p \leq .033$$; BMI ≥24 Kg/m2: Q3: OR: 1.124, $95\%$ CI: 1.016–1.243, $$p \leq .024$$; Q4: OR: 1.125, $95\%$ CI: 1.006–1.259, $$p \leq .033$$). RC and diabetes were significantly correlated when SBP <140 mm Hg and DBP <90 mm Hg (Q3: OR: 1.126, $95\%$ CI: 1.010–1.255, $$p \leq .032$$; Q4: OR: 1.179, $95\%$ CI: 1.039–1.338, $$p \leq .011$$), but there was no correlation when SBP ≥140 mm Hg or DBP ≥90 mm Hg. For participants <55 years of age, RC was significantly associated with diabetes (Q3: OR: 1.238, $95\%$ CI: 1.047–1.427, $$p \leq .003$$; Q4: OR: 1.322, $95\%$ CI: 1.124–1.555, $$p \leq .001$$). However, there was no association between RC and diabetes among participants aged 55–65 years or ≥65 years. When 60 ≤ eGFR ≤90 ml/min per 1.73 m2, RC was significantly related to diabetes (Q3: OR: 1.170, $95\%$ CI: 1.044–1.311, $$p \leq .007$$; Q4: OR: 1.211, $95\%$ CI: 1.067–1.374, $$p \leq .003$$), whereas RC and diabetes were not associated when eGFR ≥90 ml/min per 1.73 m2 or eGFR <60 ml/min per 1.73 m2.
**TABLE 4**
| Variable | Group 1 | Group 2 | Group 2.1 | Group 3 | Group 3.1 | Group 4 | Group 4.1 |
| --- | --- | --- | --- | --- | --- | --- | --- |
| Variable | Reference | OR (95% CI) | p value | OR (95% CI) | p value | OR (95% CI) | p value |
| BMI, Kg/m2 | BMI, Kg/m2 | BMI, Kg/m2 | BMI, Kg/m2 | BMI, Kg/m2 | BMI, Kg/m2 | BMI, Kg/m2 | BMI, Kg/m2 |
| BMI < 24 | 1 | 1.088 (0.968, 1.224) | .158 | 1.148 (1.015, 1.297) | .028 | 1.175 (1.013, 1.364) | .033 |
| BMI ≥ 24 | 1 | 1.047 (0.945, 1.159) | .38 | 1.124 (1.016, 1.243) | .024 | 1.125 (1.006, 1.259) | .039 |
| Blood pressure, mm Hg | Blood pressure, mm Hg | Blood pressure, mm Hg | Blood pressure, mm Hg | Blood pressure, mm Hg | Blood pressure, mm Hg | Blood pressure, mm Hg | Blood pressure, mm Hg |
| SBP < 140 and DBP < 90 | 1 | 1.072 (0.964, 1.192) | .202 | 1.126 (1.010, 1.255) | .032 | 1.179 (1.039, 1.338) | .011 |
| SBP ≥ 140 or DBP ≥ 90 | 1 | 1.035 (0.925, 1.157) | .551 | 1.100 (0.984, 1.230) | .092 | 1.085 (0.957, 1.230) | .205 |
| Age, years old | Age, years old | Age, years old | Age, years old | Age, years old | Age, years old | Age, years old | Age, years old |
| age < 55 | 1 | 1.095 (0.952, 1.258) | .204 | 1.238 (1.047, 1.427) | .003 | 1.322 (1.124, 1.555) | .001 |
| 55 ≤ age < 65 | 1 | 1.028 (0.912, 1.159) | .648 | 1.114 (0.989, 1.255) | .075 | 1.108 (0.967, 1.270) | .138 |
| age ≥ 65 | 1 | 1.072 (0.924, 1.243) | .361 | 1.035 (0.890, 1.203) | .658 | 1.040 (0.871, 1.241) | .667 |
| eGFR, ml/min per 1.73 m2 | eGFR, ml/min per 1.73 m2 | eGFR, ml/min per 1.73 m2 | eGFR, ml/min per 1.73 m2 | eGFR, ml/min per 1.73 m2 | eGFR, ml/min per 1.73 m2 | eGFR, ml/min per 1.73 m2 | eGFR, ml/min per 1.73 m2 |
| eGFR ≥ 90 | 1 | 1.033 (0.929, 1.148) | .549 | 1.107 (0.991, 1.236) | .073 | 1.053 (0.922, 1.203) | .443 |
| 60 ≤ eGFR ≤ 90 | 1 | 1.118 (0.995, 1.255) | .06 | 1.170 (1.044, 1.311) | .007 | 1.211 (1.067, 1.374) | .003 |
| eGFR <60 | 1 | 0.883 (0.541, 1.441) | .618 | 0.968 (0.609, 1.539) | .89 | 1.508 (0.911, 2.495) | .11 |
Previous studies demonstrated that the causality between RC and low‐level inflammation still existed even in participants who were neither overweight or obese, 27 suggesting that the relationship between elevated RC and low‐grade inflammation is not attributable to weight gain. The present study also found that RC and diabetes remained strongly associated regardless of whether the participant was overweight or not. Zhou et al 21 found a positive correlation between RC and diabetes in 13 721 hypertensive patients; however, RC and diabetes were associated only in persons with normal blood pressure in the current investigation. We speculate that this is because the number of hypertension patients in the current research was modest in comparison to Zhou's team's investigation. Further validation of the relationship between RC and diabetes in a larger sample of hypertensive patients is needed in the future. We found the correlation was just significant in participants younger than 55 years old, not in older participants. We hypothesize that older adults tend to maintain healthy lifestyle habits and possess better compliance, which might help to a prevention of dyslipidemia and diabetes. Prior literature has demonstrated a U‐type relation between eGFR and all‐cause mortality, 42 highlighting the significance of both high and low eGFR. The current study found a correlation between RC and diabetes only in participants with 60 ≤ eGFR ≤90 ml/min per 1.73 m2. We consider that the damage caused by low or high eGFR might cover up the pathogenicity of elevated RC.
## Main findings
The present study's findings suggested a significant association between RC and diabetes in the general Chinese population. In addition, RC was still associated with diabetes even when TG, LDL‐C, or HDL‐C was at the appropriate level recommended by guidelines. Moreover, people with elevated RC were more likely to develop diabetes, especially females, those with normal blood pressure, those with an eGFR between 60 and 90 ml/min per 1.73 m2, and those younger than 55 years of age. Hence, people with these characteristics should pay more attention to monitoring RC values and timely correction of risk factors to prevent the occurrence of diabetes. To the best of our knowledge, this is the first large sample, multicenter study to investigate the relationship between RC and diabetes in the general population.
## Previous studies
Studies on the relationship between RC and diabetes are scarce and have some minor limitations. A single‐center study of 15 464 Japanese found that RC could predict diabetes. 19 However, their study did not measure PBG, so their definition of diabetes did not include PBG≥11.1, the definition of diabetes in the current multicenter study strictly followed the American Diabetes Association's criteria, making it more accurate and scientific, and the current study was multicenter and fully adjusted for confounders. Szili‐Torok et al 20 analyzed 480 patients who had undergone kidney transplantation, and they found a significant association between RC and new‐onset diabetes after kidney transplantation. Zhou et al 21 found a positive correlation between RC and diabetes in 13 721 Chinese hypertensive patients. Nevertheless, the current study population was the large sample general Chinese population, including subjects with various chronic diseases and healthy subjects; hence, the findings of the current study were more generally applicable.
## Potential mechanisms
The reason for the association between RC and diabetes remains unclear. Atherosclerosis might be a major contributor. Numerous investigations have shown a strong link between atherosclerosis and hyperglycemia. 22, 23, 24 Elevated RC levels aided in the atherosclerosis progression, 10, 25, 26, 27 which causes hyperglycemia. RC could be penetrable to the arterial wall, be phagocytosed by macrophages, and result in the formation of foam cells. Furthermore, RC speeds up the creation of foam cells via upregulating scavenger receptor expressions. 26 Atherosclerosis lowers blood circulation to the pancreas and impairs pancreatic function, which reduces insulin secretion levels and causes hyperglycemia. 28 Atherosclerosis also causes liver dysfunction and reduces glycogen production in the liver, thus leading to hyperglycemia. 29 Insulin resistance (IR) could potentially account for the link between RC and diabetes. In a study of 86 nonobese Japanese diabetics, researchers found that the IR group had higher levels of RC than the insulin‐sensitive group. 30 Funada et al discovered that RC could forecast IR independently. 31 A recent study found that empagliflozin reduced RC in diabetic patients, which was highly related to a reduction in IR. 32 Besides, RC could directly disrupt pancreas islet β‐cells' function, resulting in decreased insulin secretion. 33
## The association between RC and diabetes when TG, LDL‐C, or HDL‐C is within the appropriate range recommended by the guidelines
“Atherosclerotic lipid triad,” characterized by elevated levels of TG and LDL‐C and reduced levels of HDL‐C, is thought to predict diabetes risk. 19 Nevertheless, in the current study, when TG was <1.7 mmol/L, LDL‐C <3.4 mmol/L, or HDL‐C ≥1.0 mmol/L, the association between RC and diabetes remained significant. Because many human cells could degrade TG, but none could degrade cholesterol, the content of cholesterol in the residue (RC) is the primary cause of atherosclerosis. 27 An Icelandic study has shown that the detrimental effects of a TG‐elevating gene variant are mediated by atherogenic effects of the cholesterol component of RC. 34 On the other hand, increased RC concentrations have been reported to lead to atherosclerosis risk even in individuals with normal TG levels. 35 As our results show, even if TG levels are appropriate, elevated RC and the corresponding risk of diabetes could not be ignored.
Varbo et al found that elevated LDL‐C could generate atherosclerosis, but not inflammation, and the inflammatory component of atherosclerosis was driven by elevated RC. 27 Despite lowering LDL‐C to suggested levels, there is a significant adverse residual risk of atherosclerosis, 36 which could be explained by the link between RC and inflammation. In addition, because RC is ingested in an uncontrolled way through scavenger receptors, RC causes greater damage to β cells than LDL‐C. 20 It has been demonstrated that there is no causality between low HDL‐C and atherosclerotic incidents, and low HDL‐C is merely a strong symbol of elevated RC. 37 After getting into the intima, HDL particles could penetrate the media and leave the arterial wall, 13, 38 but RC might be trapped in the intima and generate the foam cells, causing regional injury and inflammation. 13, 38 Moreover, TGRL promotes the remodeling of HDL‐C into smaller, cholesterol‐poor particles that might lack atherosclerotic protection regardless of HDL‐C levels. 39 Combined with the results of the current study, we recommend that people whose TG, LDL‐C, or HDL‐C levels is at the appropriate level recommended by the guidelines should also be vigilant against the increased risk of atherosclerosis and diabetes associated with elevated RC.
## Gender difference
The gender difference in older populations might be due in part to the gender‐specific hormones. Our participants' median age (Q1–Q3) was 57 [52, 64], indicating that the women in the current study were either perimenopausal or postmenopausal. The atherogenic effect of RC is more pronounced in women at this stage. 40 Ossewaarde et al observed a $15\%$ reduction in RC levels in a group of healthy Dutch postmenopausal women given estrogen + progestin combination therapy for 3 months. 41 Hence, perimenopausal and postmenopausal women are more prone to develop RC abnormalities owing to hormonal changes, increasing their risk of atherosclerosis and diabetes.
## Limitations
The current study benefited from a vast collection of multiple community‐based samples, broadly representative of most Chinese people. Meanwhile, the current study also comprehensively adjusted for major risk factors and conducted a detailed stratification analysis. However, some limitations still remained. First, VLDL residues are a major component of RC during fasting, so the role of chylomicron residues may be underestimated. Second, the current study was cross‐sectional, so we could we could make only correlational inferences, not causal ones.
## CONCLUSIONS
The present study observed that elevated RC is significantly associated with diabetes in the *Chinese* general population, even when TG, LDL‐C, or HDL‐C was at the appropriate level recommended by guidelines. Population with elevated RC are at higher danger of diabetes, especially in subjects with normal blood pressure, 60 ≤ eGFR ≤ 90 ml/min per 1.73 m2, younger than 55 years old, and female. The current study's findings expand RC's application and provide new approaches to preventing diabetes. Calculating RC is an affordable method. It is necessary to measure RC along with monitoring other lipid markers in general people, even if traditional lipid components are at appropriate levels.
## AUTHOR CONTRIBUTIONS
All authors have read and approved the final manuscript. Binqi Li contributed to the conception and design of the study. Binqi Li, Xin Zhou, Weiqing Wang, Zhengnan Gao, Li Yan, Guijun Qin, Xulei Tang, Qin Wan, Lulu Chen, Zuojie Luo, Guang Ning, and Yiming Mu recruited the subjects and supervised the study. Binqi Li analyzed the data and wrote the initial draft of the paper. Yiming Mu and Binqi Li contributed to the manuscript's writing, reviewing, and revising.
## FUNDING INFORMATION
The study is supported by Beijing Municipal Science and Technology Commission Project (Z201100005520014), the Chinese Society of Endocrinology, the Key Laboratory for Endocrine and Metabolic Diseases of Ministry of Health (1994DP131044), the National Key New Drug Creation and Manufacturing Program of Ministry of Science and Technology (2012ZX09303006‐001), the National High Technology Research and Development Program of China (863 Program, 2011AA020107), National Science Foundation of China [81300717], National Science and Technology Major Project 288 (2011ZX09307‐001‐08).
## CONFLICT OF INTEREST
The authors declare no competing interests.
## DATA AVAILABILITY STATEMENT
The data sets are not freely available due to protection of participants' privacy.
## References
1. Goldberg RB, Orchard TJ, Orchard JP. **Effects of long‐term metformin and lifestyle interventions on cardiovascular events in the diabetes prevention program and its outcome study**. *Circulation* (2022) **145** 1632-1641. PMID: 35603600
2. Bragg F, Holmes MV, Iona A. **Association between diabetes and cause‐specific mortality in rural and urban areas of China**. *JAMA* (2017) **317** 280-289. PMID: 28114552
3. Yu M, Zhan X, Yang Z, Huang Y. **Measuring the global, regional, and national burden of type 2 diabetes and the attributable risk factors in all 194 countries**. *J Diabetes* (2021) **13** 613-639. PMID: 33486878
4. Sun H, Saeedi P, Karuranga S. **IDF diabetes atlas: global, regional and country‐level diabetes prevalence estimates for 2021 and projections for 2045**. *Diabetes Res Clin Pract* (2022) **183**. PMID: 34879977
5. Saeedi P, Petersohn I, Salpea P. **Global and regional diabetes prevalence estimates for 2019 and projections for 2030 and 2045: results from the international diabetes federation diabetes atlas, 9(th) edition**. *Diabetes Res Clin Pract* (2021) **157**
6. Ginsberg HN, Elam MB. **Effects of combination lipid therapy in type 2 diabetes mellitus**. *N Engl J Med* (2010) **17** 1563-1574
7. Tenenbaum AMMFE. **Effect of bezafibrate on incidence of type 2 diabetes mellitus in obese patients**. *Eur Heart J* (2005) **26** 2032-2038. PMID: 15872029
8. Tenenbaum AMMFE. **Peroxisome proliferator–activated receptor ligand Bezafibrate for prevention of type 2 diabetes mellitus in patients with coronary artery disease**. *Circulation* (2004) **109** 2197-2202. PMID: 15123532
9. Peng J, Zhao F, Yang X. **Association between dyslipidemia and risk of type 2 diabetes mellitus in middle‐aged and older Chinese adults: a secondary analysis of a nationwide cohort**. *BMJ Open* (2021) **11**
10. Varbo A, Benn M, Tybjaerg‐Hansen A. **Remnant cholesterol as a causal risk factor for ischemic heart disease**. *J Am Coll Cardiol* (2013) **61** 427-436. PMID: 23265341
11. Sandesara PB, Virani SS, Fazio S, Shapiro MD. **The forgotten lipids: triglycerides, remnant cholesterol, and atherosclerotic cardiovascular disease risk**. *Endocr Rev* (2019) **40** 537-557. PMID: 30312399
12. Yu D, Wang Z, Zhang X. **Remnant cholesterol and cardiovascular mortality in patients with type 2 diabetes and incident diabetic nephropathy**. *J Clin Endocrinol Metab* (2021) **106** 3546-3554. PMID: 34291804
13. Li B, Wang A, Wang Y. **A study on the correlation between remnant cholesterol and urinary albumin to creatinine ratio in Chinese community adults: a report from the REACTION study**. *J Diabetes* (2020) **12** 870-880. PMID: 32500969
14. Shan Y, Wang Q, Zhang Y. **High remnant cholesterol level is relevant to diabetic retinopathy in type 2 diabetes mellitus**. *Lipids Health Dis* (2022) **21** 12. PMID: 35057797
15. Ning G. **Risk evaluation of cAncers in Chinese diabeTic individuals: a longitudinal (REACTION) study**. *J Diabetes* (2012) **4** 172-173. PMID: 22221801
16. Friedewald WT, Levy RI, Fredrickson DS. **Estimation of the concentration of low‐density lipoprotein cholesterol in plasma, without use of the preparative ultracentrifuge**. *Clin Chem* (1972) **18** 499-502. PMID: 4337382
17. **Standards of medical care in diabetes—2011**. *Diabetes Care* (2011) **34** S11-S61. PMID: 21193625
18. **2016 Chinese guidelines for the management of dyslipidemia in adults**. *J Geriatr Cardiol* (2018) **15** 1-29. PMID: 29434622
19. Xie G, Zhong Y, Yang S, Zou Y. **Remnant cholesterol is an independent predictor of new‐onset diabetes: a single‐center cohort study**. *Diabetes Metab Syndr Obes* (2021) **14** 4735-4745. PMID: 34887671
20. Szili‐Torok T, Sokooti S, Oste MCJ. **Remnant lipoprotein cholesterol is associated with incident new onset diabetes after transplantation (NODAT) in renal transplant recipients: results of the TransplantLines biobank and cohort studies**. *Cardiovasc Diabetol* (2022) **21** 41. PMID: 35296331
21. Zhou W, Yu Y, Zhu L. **Positive association between triglyceride‐rich lipoprotein cholesterol and diabetes mellitus in hypertensive patients**. *Int J Endocrinol* (2021) **2021** 7722269-7722268. PMID: 34899903
22. Shin JY, Lee HR, Lee DC. **Increased arterial stiffness in healthy subjects with high‐normal glucose levels and in subjects with pre‐diabetes**. *Cardiovasc Diabetol* (2011) **10** 30. PMID: 21492487
23. Henry RMA, Kostense PJ, Spijkerman AMW, Dekker JM, Nijpels G, Heine RJ. **Arterial stiffness increases with deteriorating glucose tolerance status: the Hoorn study**. *Circulation* (2003) **107** 2089-2095. PMID: 12695300
24. Zheng M, Zhang X, Chen S. **Arterial stiffness preceding diabetes: a longitudinal study**. *Circ Res* (2020) **127** 1491-1498. PMID: 32985370
25. Jorgensen AB, Frikke‐Schmidt R, West AS. **Genetically elevated non‐fasting triglycerides and calculated remnant cholesterol as causal risk factors for myocardial infarction**. *Eur Heart J* (2013) **34** 1826-1833. PMID: 23248205
26. Mearns BM. **Dyslipidaemia: role of remnant cholesterol in IHD**. *Nat Rev Cardiol* (2013) **10** 553. PMID: 23979217
27. Varbo A, Benn M, Tybjaerg‐Hansen A. **Elevated remnant cholesterol causes both low‐grade inflammation and ischemic heart disease, whereas elevated low‐density lipoprotein cholesterol causes ischemic heart disease without inflammation**. *Circulation* (2013) **128** 1298-1309. PMID: 23926208
28. Levy BI, Schiffrin EL, Mourad J. **Impaired tissue perfusion: a pathology common to hypertension, obesity, and diabetes mellitus**. *Circulation* (2008) **118** 968-976. PMID: 18725503
29. Chirinos JA, Segers P, Hughes T, Townsend R. **Large‐artery stiffness in health and disease: JACC state‐of‐the‐art review**. *J Am Coll Cardiol* (2019) **74** 1237-1263. PMID: 31466622
30. Taniguchi A, Fukushima M, Sakai M. **Remnant‐like particle cholesterol, triglycerides, and insulin resistance in nonobese Japanese type 2 diabetic patients**. *Diabetes Care* (2000) **23** 1766-1769. PMID: 11128349
31. Funada J, Sekiya M, Otani T, Watanabe K, Sato M, Akutsu H. **The close relationship between postprandial remnant metabolism and insulin resistance**. *Atherosclerosis* (2004) **172** 151-154. PMID: 14709369
32. Hattori S. **Empagliflozin decreases remnant‐like particle cholesterol in type 2 diabetes patients with insulin resistance**. *J Diabetes Investig* (2018) **9** 870-874
33. Sokooti S, Flores‐Guerrero JL, Heerspink HJL, Connelly MA, Bakker SJL, Dullaart RPF. **Triglyceride‐rich lipoprotein and LDL particle subfractions and their association with incident type 2 diabetes: the PREVEND study**. *Cardiovasc Diabetol* (2021) **20** 156. PMID: 34321006
34. Bjornsson E, Thorleifsson G, Helgadottir A. **Association of genetically predicted lipid levels with the extent of coronary atherosclerosis in Icelandic adults**. *JAMA Cardiol* (2020) **5** 13-20
35. Chait A, Ginsberg HN, Vaisar T, Heinecke JW, Goldberg IJ, Bornfeldt KE. **Remnants of the triglyceride‐rich lipoproteins, diabetes, and cardiovascular disease**. *Diabetes* (2020) **69** 508-516. PMID: 32198194
36. Mihaylova B, Emberson J. **The effects of lowering LDL cholesterol with statin therapy in people at low risk of vascular disease: meta‐analysis of individual data from 27 randomised trials**. *Lancet* (2012) **380** 5811670e-1484607363e
37. Crosby J, Peloso GM, Auer PL. **Loss‐of‐function mutations in APOC3, triglycerides, and coronary disease**. *N Engl J Med* (2014) **371** 22-31. PMID: 24941081
38. Rutledge JC, Mullick AE, Gardner G, Goldberg IJ. **Direct visualization of lipid deposition and reverse lipid transport in a perfused artery: roles of VLDL and HDL**. *Circ Res* (2000) **86** 768-773. PMID: 10764410
39. Prenner SB, Mulvey CK, Ferguson JF, Rickels MR, Bhatt AB, Reilly MP. **Very low density lipoprotein cholesterol associates with coronary artery calcification in type 2 diabetes beyond circulating levels of triglycerides**. *Atherosclerosis* (2014) **236** 244-250. PMID: 25105581
40. Lamon‐Fava S, Herrington DM, Reboussin DM. **Changes in remnant and high‐density lipoproteins associated with hormone therapy and progression of coronary artery disease in postmenopausal women**. *Atherosclerosis* (2009) **205** 325-330. PMID: 19155011
41. Ossewaarde ME, Dallinga‐Thie GM, Bots ML. **Treatment with hormone replacement therapy lowers remnant lipoprotein particles in healthy postmenopausal women: results from a randomized trial**. *Eur J Clin Invest* (2003) **33** 376-382. PMID: 12713450
42. Matsushita KVDVM. **Association of estimated glomerular filtration rate and albuminuria with all‐cause and cardiovascular mortality in general population cohorts: a collaborative meta‐analysis**. *Lancet* (2010) **9731** 2073-2081
|
---
title: Transparent machine learning suggests a key driver in the decision to start insulin
therapy in individuals with type 2 diabetes
authors:
- Nicoletta Musacchio
- Rita Zilich
- Paola Ponzani
- Giacomo Guaita
- Carlo Giorda
- Rebeca Heidbreder
- Pierluigi Santin
- Graziano Di Cianni
journal: Journal of Diabetes
year: 2023
pmcid: PMC10036260
doi: 10.1111/1753-0407.13361
license: CC BY 4.0
---
# Transparent machine learning suggests a key driver in the decision to start insulin therapy in individuals with type 2 diabetes
## Abstract
Highlights ML suggests that when the HbA1c gap from a previous visit is >11 mmol/mol ($1.0\%$) there is a greater probability that insulin therapy will be initiated, but when HbA1c gap is <6.6 mmol/mol ($0.6\%$), a timely initiation of insulin therapy is less probable. Furthermore, for individuals initiated on insulin in a timely manner, the HbA1c gap is systematically higher than for those patients who have experienced clinical inertia. This key driver correlated with insulin therapy initiation could help combat clinical inertia.
### Aims
The objective of this study is to establish a predictive model using transparent machine learning (ML) to identify any drivers that characterize therapeutic inertia.
### Methods
Data in the form of both descriptive and dynamic variables collected from electronic records of 1.5 million patients seen at clinics within the Italian Association of Medical Diabetologists between 2005–2019 were analyzed using logic learning machine (LLM), a “clear box” ML technique. Data were subjected to a first stage of modeling to allow ML to automatically select the most relevant factors related to inertia, and then four further modeling steps individuated key variables that discriminated the presence or absence of inertia.
### Results
The LLM model revealed a key role for average glycated hemoglobin (HbA1c) threshold values correlated with the presence or absence of insulin therapeutic inertia with an accuracy of 0.79. The model indicated that a patient's dynamic rather than static glycemic profile has a greater effect on therapeutic inertia. Specifically, the difference in HbA1c between two consecutive visits, what we call the HbA1c gap, plays a crucial role. Namely, insulin therapeutic inertia is correlated with an HbA1c gap of <6.6 mmol/mol ($0.6\%$), but not with an HbA1c gap of >11 mmol/mol ($1.0\%$).
### Conclusions
The results reveal, for the first time, the interrelationship between a patient's glycemic trend defined by sequential HbA1c measurements and timely or delayed initiation of insulin therapy. The results further demonstrate that LLM can provide insight in support of evidence‐based medicine using real world data.
## INTRODUCTION
In spite of abundant evidence demonstrating that in type 2 diabetes early glycemic control is correlated with a reduction in long‐term complications, 1 data from many health systems indicate that delay in initiation and/or intensification of insulin therapy remains systemic. 2, 3 The end result is that glycemic control in type 2 diabetes is globally inadequate, and individuals live years with poor glycemic control. 4, 5 Several studies have provided explanations for the motivation behind the delay in glycemic control, 6, 7, 8 but none, to our knowledge, have examined real world data with artificial intelligence (AI) techniques to identify which factors are more likely to be associated with a provider's behavior (presence or absence of inertia) with regard to initiation of insulin therapy.
In 2005, the Italian Association of Medical Diabetologists (AMD) initiated the Annals project that led to the creation of a network of diabetes clinics, representing half of the total number of diabetes clinics in Italy, with the aim of monitoring, standardizing, and sharing the main parameters used for the evaluation of the quality of care given to patients. In this way, the AMD Annals were able to collect from patient electronic records up to 180 parameters, including clinical, pharmacological, organization related, and provider related, from each patient 9, 10, 11 to create “big data” sets (approximately 1.5 million patients and about 9 million visits), which permitted a thorough analysis using AI techniques.
Previously published studies have utilized AI methods to investigate other aspects of diabetes and have yielded promising results. 12 The present study applied AI techniques, specifically a transparent machine learning (ML) methodology, to overcome the problems associated with “black box” AI algorithms, which can deliver performant models but do not furnish explanations as to how the results were obtained. 13, 14 The transparent ML technique used in this study is based on a proprietary algorithm of “explainable artificial intelligence” known as logic learning machine (LLM), which yields performance that is on par to the best ML algorithms while at the same time allowing full control over the algorithmic logic and permitting the correlation between predictive factors and outcome. 14 LLM has already been used to great effect in the analysis of biomedical datasets included in the Statlog benchmark. 15, 16 For the purposes of this study, LLM was used to generate predictive and explanatory models that identify the combination of factors (clinical, personal, organizational) correlated with provider inertia in situations which would otherwise require the initiation of insulin therapy. This approach has already been previously used to good effect to analyze other types of clinical data. 17, 18
## Participants
Data from people with type 2 diabetes were obtained from electronic medical records located on the AMD Annals database. The medical records are from patients who, prior to being referred to one of the clinics within the AMD, showed a high risk for the development of type 2 diabetes or had lab values that were indisputably consistent with type 2 diabetes. No patient referred to one of the diabetes clinics was initiated on insulin prior to their first visit at the clinic. Every patient had visited at least one of the Italian diabetes clinics between 2005 and the first half of 2019. 5 A total of 2.3 billion data points corresponding to information on 1 186 247 people with a confirmed diagnosis of type 2 diabetes (as indicated in the diagnosis field of the electronic medical record found on the database) and 9 954 976 visits were selected. These individuals were followed over time, and a total of 91 variables, including glycated hemoglobin (HbA1c), were checked periodically (on average every 0.6 years). The data preparation can be summarized as follows (for more in‐depth information please see 17):Time interval between two HbA1c measurements ≥2 months. For each HbA1c measurement, “clinical factors” (eg, blood pressure, lipid panel, albuminuria, etc.) were tracked over time with an interval of maximum 4 months before and after the date of each measurement. For each HbA1c measurement, irreversible comorbidities (eg, acute myocardial infarction, stroke, etc.) were tracked starting from the date of first detection.
Table 1 provides the inclusion criteria for this study. Following the application of the inclusion criteria, measurements from 129 373 individuals were included.
**TABLE 1**
| Inertia‐yes | Inertia‐no |
| --- | --- |
| Patient not currently pregnant AND | Patient not currently pregnant AND |
| 2Patient currently on dual or triple therapy AND | 2Patient currently on dual or triple therapy AND |
| 31–2 consecutive measurements of above threshold HbA1c >7.5% (58.5 mmol/mol) if patient is ≤75 years old; >8% (63.9 mmol/mol) if patient is >75 years old AND | 31–2 consecutive measurements of above threshold HbA1c >7.5% (58.5 mmol/mol) if patient is ≤75 years old; >8% (63.9 mmol/mol) if patient is >75 years old AND |
| 4 NO prescribed insulin therapy (basal, basal‐bolus or rapid) after second above threshold measurement | 4 YES prescribed insulin therapy (basal, basal‐bolus or rapid) after either first or second above threshold measurement |
| The application of these criteria led to a total of 96 621 patients. Of this total 43 375 were started on insulin. The remaining 53 246 never received insulin | The application of these criteria led to a total of 32 752 patients |
Data related to drug therapies and comorbidities were grouped as described in our previous study. 17 Prescribed medications were grouped into eight main diabetes therapies to simplify the number of drug combinations, yielding 18 combinations. To ensure a robust estimate of comorbidities, we grouped information from across different fields in the electronic medical record.
Figure 1 provides a flow chart with participant characteristics.
**FIGURE 1:** *Flow chart illustrating the number of participants being included/excluded at each stage. HbA1C, glycated hemoglobin*
Table 2 provides the means, median, SDs, and interquartile for the characteristics of the cohorts.
**TABLE 2**
| Cohort characteristics | INERT | NON‐INERT |
| --- | --- | --- |
| BMI mean | 30.5 kg/m2 | 29.9 kg/m2 |
| BMI SD | 5.6 kg/m2 | 5.9 kg/m2 |
| BMI median | 29.8 kg/m2 | 29.1 kg/m2 |
| BMI (25 percentile) | 26.7 kg/m2 | 25.8 kg/m2 |
| BMI (75 percentile) | 33.5 kg/m2 | 33.2 kg/m2 |
| HbA1c mean | 69 mmom/mol (8.5%) | 79 mmom/mol (9.4%) |
| HbA1c SD | 8.8 mmom/mol (0.8%) | 14.3 mmom/mol (1.3%) |
| HbA1c median | 67 mmom/mol (8.3%) | 79 mmom/mol (9.1%) |
| HbA1c (25 percentile) | 63 mmom/mol (7.9%) | 68 mmom/mol (8.4%) |
| HbA1c (75 percentile) | 73 mmom/mol (8.8%) | 87 mmom/mol (10.1%) |
| Serum glutamic oxaloacetic transaminase (GOT) mean | 23.1 mU/mL | 25.1 mU/mL |
| GOT SD | 13.4 mU/mL | 27.8 mU/mL |
| GOT median | 20.0 mU/mL | 19.0 mU/mL |
| GOT (25 percentile) | 16.0 mU/mL | 15.0 mU/mL |
| GOT (75 percentile) | 26.0 mU/mL | 26.0 mU/mL |
| Serum glutamic pyruvic transaminase (GPT) mean | 29.1 mU/mL | 30.6 mU/mL |
| GPT SD | 20.3 mU/mL | 31.8 mU/mL |
| GPT median | 24.0 mU/mL | 23.0 mU/mL |
| GPT (25 percentile) | 17.0 mU/mL | 16.0 mU/mL |
| GPT (75 percentile) | 34.0 mU/mL | 34.0 mU/mL |
| High density lipoproteins (HDL) mean | 47.7 mg/dL | 46.6 mg/dL |
| HDL SD | 13.1 mg/dL | 14.0 mg/dL |
| HDL median | 46.0 mg/dL | 44.0 mg/dL |
| HDL (25 percentile) | 39.0 mg/dL | 37.0 mg/dL |
| HDL (75 percentile) | 55.0 mg/dL | 54.0 mg/dL |
| Low density lipoproteins (LDL) mean | 99.5 mg/dL | 99.4 mg/dL |
| LDL SD | 34.8 mg/dL | 37.5 mg/dL |
| LDL median | 96.8 mg/dL | 96.0 mg/dL |
| LDL (25 percentile) | 76.0 mg/dL | 73.8 mg/dL |
| LDL (75 percentile) | 120.0 mg/dL | 121.0 mg/dL |
| Systolic blood pressure (BP) mean | 138.5 mmHg | 137.7 mmHg |
| Systolic BP SD | 18.3 mmHg | 19.4 mmHg |
| Systolic BP median | 140.0 mmHg | 140.0 mmHg |
| Systolic BP (25 percentile) | 128.0 mmHg | 121.0 mmHg |
| Systolic BP (75 percentile) | 150.0 mmHg | 150.0 mmHg |
| Diastolic BP mean | 79.2 mmHg | 78.6 mmHg |
| Diastolic BP SD | 9.6 mmHg | 10.1 mmHg |
| Diastolic BP median | 80.0 mmHg | 80.0 mmHg |
| Diastolic (25 percentile) | 70.0 mmHg | 70.0 mmHg |
| Diastolic (75 percentile) | 83.0 mmHg | 82.0 mmHg |
| Triglyceride mean | 158.2 mg/dL | 177.5 mg/dL |
| Triglyceride SD | 104.1 mg/dL | 146.6 mg/dL |
| Triglyceride median | 135.0 mg/dL | 144.0 mg/dL |
| Triglyceride (25 percentile) | 98.0 mg/dL | 102.0 mg/dL |
| Triglyceride (75 percentile) | 189.0 mg/dL | 209.0 mg/dL |
| Fasting glycemia mean | 171.5 mg/dL | 207.6 mg/dL |
| Fasting glycemia SD | 42.9 mg/dL | 65.7 mg/dL |
| Fasting glycemia median | 167.0 mg/dL | 198.0 mg/dL |
| Fasting glycemia (25 percentile) | 144.0 mg/dL | 163.0 mg/dL |
| Fasting glycemia (75 percentile) | 193.0 mg/dL | 243.0 mg/dL |
| Glomerular filtration rate (GFR) mean | 80.6 mL/min/1.73 m2 | 74.3 mL/min/1.73 m2 |
| GFR SD | 19.3 mL/min/1.73 m2 | 23.9 mL/min/1.73 m2 |
| GFR median | 83.9 mL/min/1.73 m2 | 77.6 mL/min/1.73 m2 |
| GFR (25 percentile) | 67.6 mL/min/1.73 m2 | 56.1 mL/min/1.73 m2 |
| GFR (75 percentile) | 94.5 mL/min/1.73 m2 | 93.1 mL/min/1.73 m2 |
| Age mean | 65.4 years | 67.2 years |
| Age SD | 9.8 years | 11.3 years |
| Age median | 66.0 years | 68.0 years |
| Age (25 percentile) | 59.0 years | 59.0 years |
| Age (75 percentile) | 72.0 years | 76.0 years |
## LLM characteristics and ML modeling
ML has the ability to both analyze data without making any a priori assumptions and predict new output values from the data. The ML technique, “rule generation methods,” builds models described by a set of intelligible rules that permit the extraction of knowledge about variables in the analysis as well as their relationships with a target attribute. Two different paradigms for rule generation have been proposed in the literature. Decision trees 19 adopt a divide‐and‐ conquer approach for generating the final model. Methods based on Boolean function reconstruction follow an aggregative procedure for building the set of rules. 20, 21 LLM is a proprietary algorithm that implements the switching neural network model, 22 which allows for solving classification problems and produces sets of intelligible rules expressed in the form: “if premise …, then consequence …,” where “premise” includes one or more conditions on the input variables, and “consequence” contains the output value or information about the target function in terms of “yes or no.” Thus, the LLM rule generation technique produces a subset of relevant variables associated with a specific outcome and informs on explicit intelligible conditions related to a particular outcome as well as relevant thresholds for each input variable. Furthermore, the “clear box” approach used by LLM yields “explainable AI,” which provides comprehensible and trustworthy results and output created by the ML algorithms. 23 In the present study, data were subjected to a first stage of modeling to allow ML to select automatically the most relevant factors related to inertia. The model incorporated both descriptive variables (clinical and demographic) and dynamic variables (HbA1c gap and drop speed, mean, SD, and trend for several clinical measurements) collected from each individual (Table S1).
After the preliminary phase, four further modeling steps were completed to individuate the key variables that discriminate the presence or absence of inertia. The role and relevance of the different variables that influenced YES/NO inertia were taken through several modeling steps (learning set = $70\%$ and test set = $30\%$) as outlined in Table 3.
**TABLE 3**
| Step description | Input parameters from the AMD Annals database | Accuracy & AUC (area under the ROC curve) | Comment |
| --- | --- | --- | --- |
| Step 1 (first model iteration) All variables measured at the time of the patient visit (“static” clinical measurements from the patient and organizational measurements from the medical center), except for preceding HbA1c that was added later following LLM results and suggestions | Age, sex, body mass index (BMI), systolic blood pressure (BP), diastolic BP, HbA1c at current visit, fasting glucose, hypertension, dyslipidemia, triglycerides, High‐density lipoproteins (HDL), low‐density lipoproteins (LDL), creatinine, estimated glomerular filtration rate (eGFR), micro–macro/albuminuria, serum uric acid, nephropathy, atrial fibrillation, heart failure, stroke, cardiac complications, vasculopathy, lower limb complications, neuropathy, foot complications, eye complications, hepatopathy, serum glutamic oxaloacetic transaminase (GOT), serum glutamic pyruvic transaminase (GPT), drug therapy (double or triple), “years of clinical observation” (considered a proxy of duration of diabetes), Q‐score (quality of care summary score calculated for each year of observation, developed and validated in two previous studies) | Accuracy: 0.70 AUC: 0.760 | This step in the model revealed a modest predictive precision (0.70), which indicated that the “static” variables were insufficient for the clear discrimination/forecast of the provider's decision (YES/NO inertia) |
| Step 2 (second model iteration) All variables measured at the time of the patient visit (“static” clinical measurements from the patient and organizational measurements from the medical center) as well as derived dynamic variables served as input for ML. In addition, HbA1c at previous visit was added based on Step 1 results and suggestions | Same as step1 + HbA1c at previous visit, HbA1c gap (HbA1c actual – HbA1c previous visit), HbA1c drop speed (speed of HbA1c yearly reduction) + mean, SD, and trend for HbA1c, fasting glycemia, systolic BP, diastolic BP, creatinine, eGFR, triglycerides, HDL, LDL, GOT, GPT, BMI | Accuracy: 0.79 AUC: 0.876 | This model had very good predictive precision (0.79) and was >10% more precise than the first iteration. This second step clearly indicates that derived dynamic variables have a determining influential role on the provider's decision‐making process (YES/NO inertia) |
| Step 3 (third model iteration) For this iteration, only the variables linked to glycemia (static or dynamic) served as input for ML | HbA1c at current visit, HbA1c at previous visit + HbA1c drop speed, HbA1c gap, mean, SD, and trend for HbA1c and fasting glucose | Accuracy: 0.78 AUC: 0.845 | This third iteration had decidedly good predictive precision (0.78), which is only slightly inferior to the second step in the model despite the fact that variables other than those related to glycemia were not included as input. This result indicates that glycemia and HbA1c play a dominant role over any other variable with respect to a provider's decision‐making process (YES/NO inertia) |
| Step 4 (Fourth model iteration) For this iteration all static and dynamic variables, excluding static and glycemic variables linked to glycemia (static or dynamic) (HbA1c and fasting glucose) served as input for ML | Same as first iteration (excluding HbA1c at current visit, fasting glucose + mean, SD, and trend for systolic BP, diastolic BP, creatinine, triglycerides, HDL, LDL, GOT, GPT, BMI, and eGFR) | Accuracy: 0.64 AUC: 0.676 | This model confirms the dominant role of HbA1c and glycemia. This model, which excludes glycemic variables, has low predictive validity (0.64) relative to the first two. The variables used for input were not able to discriminate the outcome despite also being dynamic variables |
LLM affords the advantages of ML, which permits the analysis of very large number of variables, along with ability to have access to (transparency) of the ranking of the most relevant variables that can help guide the analysis.
As such, Step 1 began by incorporating all of the descriptive variables (Table S1) into the model, which resulted in accuracy = 0.70 and area under the curve (AUC) = 0.76 (also reported in Table S1). This was followed by Step 2 to verify if the addition of dynamically derived‐variables could improve the performance of the model. The hypothesis was that a medical practitioner's inertia could be influenced by factors related to the patient's progress across time rather than only by static parameters related to a single visit. This hypothesis was confirmed by the results from Step 2 in that the performance of the model significantly improved accuracy = 0.79 and AUC = 0.87. Furthermore, the relevant variables revealed by the transparent ML highlighted the important role of those variables that are related to the patient's progress. For example, the HbA1c gap achieved the second position in the ranking and in third position one finds the average HbA1c across 4 years (Table 3).
Given that the first three positions in the ranking of the variables in Step 2 were all related to HbA1c, we were driven to carry out a third step to verify if the dominant role of glycemia and HbA1c was real. Therefore, for Step 3, input for ML consisted only of variables that were related to glycemic factors (static and dynamic values for fasting HbA1c and glycemia). Step 3 results had accuracy = 0.78 and AUC = 0.84, which confirm the dominant role of the glycemic factors as determinants of the medical practitioner's decision to initiate insulin therapy in a patient.
As a counterproof to Step 3, we carried out a final step in modeling that included the use of all variables, both dynamic and descriptive, EXCEPT for those related to fasting HbA1c and glycemia, as input for ML. The results from Step 4 of modeling resulted in accuracy = 0.64 and AUC = 0.67. These results were modest compared to the previous three modeling steps and serve to demonstrate the small, but not absent, influence that other variables have on the medical practitioner's decision to initiate insulin therapy.
## RESULTS
The total data pool was comprised of 129 373 individuals, 32 752 of whom were started on insulin therapy in accordance with the 2020 guidelines of the American Diabetes Association, 43 375 whose insulin therapy initiation was delayed, and 53 246 who never received insulin as reported in Table 1.
The results indicate that the best model was derived from the second modeling iteration. This best performing model, which includes all the dynamic and descriptive variable outlined in Table S1, underscores the relevance of the variables selected for the analysis and prediction of the phenomenon studied. The area under the receiver operating characteristic curve (ROC) of the model is = 0.87 (Figure 3), accuracy = 0.79, sensitivity = 0.76, specificity = 0.78, and precision = 0.91. The 10 main variables in terms of relevance highlight the dominant role of current HbA1c values as well as the immediately preceding values.
Table 4 provides a ranking of the most relevant factors in the ML model. The factors with the highest relevance, namely those having the strongest correlation with outcome inertia‐yes/inertia‐no, are the first three listed and are related to consecutive HbA1c measurements and the measured difference between them, or HbA1c gap. Average HbA1c across 4 years, fasting glucose, as well as yearly HbA1c reduction speed are also relevant, but to a lesser degree than the HbA1c gap.
**TABLE 4**
| Relevant factors | Threshold inertia‐no | Threshold inertia‐yes | Relevance |
| --- | --- | --- | --- |
| Factors related to glycemia | Factors related to glycemia | Factors related to glycemia | Factors related to glycemia |
| HbA1c at current visit | >73 mmol/mol (>8.8%) | <72 mmol/mol (<8.7%) | 0.741 |
| Change in HbA1c from previous visit (HbA1c gap) | >11 mmol/mol (>1.0%) | <6.6 mmol/mol (< 0.6%) | 0.396 |
| Mean HbA1c (4 previous years) | >69 mmol/mol (>8.5%) | < 66 mmol/mol (<8.2%) | 0.199 |
| Fasting glucose | >242 | <210 | 0.186 |
| HbA1c reduction speed (a negative drop in speed indicates an annual worsening of HbA1c relative to the indicated threshold) | <−11 mmol/mol (<−1.0%) | >−5.94 mmol/mol (>−0.54%) | 0.158 |
| Mean fasting glucose (4 previous years) | | <203 | 0.026 |
| Fasting glucose trend | | <1.33 | 0.025 |
| HbA1c trend | | <0.05 | 0.013 |
| Factors related to insulin resistance | Factors related to insulin resistance | Factors related to insulin resistance | Factors related to insulin resistance |
| Mean diastolic BP (4 previous years) | < 88 | | 0.029 |
| BMI | | >24 | 0.028 |
| Diastolic BP trend | | >−0.53 & <+0.15 | 0.020 |
| Triglyceride trend | | >−1.97 & <+2.34 | 0.017 |
| Average BMI (4 previous years) | | >24 | 0.010 |
| Comorbidity | Comorbidity | Comorbidity | Comorbidity |
| Mean eGFR (4 previous years) | < 59.99 | >69 | 0.044 |
| Hepatopathy (threshold) | | Not diagnosed | 0.040 |
| Nephropathy | | Not diagnosed (OR current dialysis) | 0.039 |
| Hyperuricemia | | Not diagnosed | 0.028 |
| Average creatinine (4 previous years) | >1.09 | <0.75 | 0.026 |
| eGFR trend | | >−0.99 & <+0.01 | 0.026 |
| Heart disease | Diagnosed | Not diagnosed | 0.019 |
| eGFR | <77 | | 0.016 |
| Personal characteristics | Personal characteristics | Personal characteristics | Personal characteristics |
| Age | >78 | <64 | 0.155 |
| History and care | History and care | History and care | History and care |
| Months of follow‐up | >101 | | 0.111 |
| Q‐score | | <21 | 0.048 |
The model indicates that the glycemic profile, which for the purposes of this study refers specifically to the change in either HbA1c or glycemia seen in a patient from one visit to the next, and a patient's glycemic trend, that is, the direction that change takes, has a greater effect on a provider's therapeutic inertia than any one individual datapoint in the patient's static profile. ML indicates that for threshold HbA1c values, on average, an HbA1c gap of <6.6 mmol/mol (<$0.6\%$) is correlated with inertia. On the other hand, an HbA1c gap of >11 mmol/mol (>$1.0\%$) is correlated with non‐inertia. Thus, the data suggest that the HbA1c gap between two consecutive visits appears to play a crucial role in the decision to start insulin therapy in a person with type 2 diabetes. Moreover, not only are the current HbA1c and the change in HbA1c between two consecutive measurements the key drivers with the strongest influence on the presence or absence of inertia, but the average HbA1c across 4 years is also relevant in the decision‐making process. Furthermore, in terms of relevance, following the HbA1c level at the time of the most recent visit, ML suggests that below a value of 72 mmol/mol ($8.7\%$), clinical inertia is most probable, whereas values above 73 mmol/mol ($8.8\%$) lead to a greater probability that insulin will be initiated.
There are other parameters not directly linked to HbA1c but which in all iterations of modeling also suggest some relevance. For example, estimated glomerular filtration rate (eGFR) (mean and trend) stood out in all three iterations of modeling as the most important comorbidity and, when the eGFR mean is <59.99, it is correlated with non‐inertia. A body mass index >24, either as a static variable or as the mean over 4 years, is correlated with inertia. Both stable triglycerides and the absence of complications in particular cardiac, hepatopathy, and hyperuricemia are all correlated with a greater probability of inertia.
The model was also able to confirm data previously reported in the literature. Namely, HbA1c values of less than 68 mmol/mol ($8.4\%$) are associated with insulin therapy inertia whereas values above 75 mmol/mol ($9.0\%$) are associated with an increased probability of insulin therapy initiation. The presence of a sudden increase in HbA1c as a driver of non‐inertia led to a more in‐depth analysis to verify the accuracy of the results provided by LLM.
Given that among the transparent ML rankings the HbA1c gap is an innovative finding in that it has not yet been described in the literature as a factor that plays a role in insulin inertia, we wanted to verify with traditional statistics the correlation between the variables revealed by transparent ML and insulin inertia. To confirm the results from the model, a statistical analysis was carried out on those individuals identified by ML who at some point in their clinical history had HbA1c values of >58 mmol/mol (>$7.5\%$) for one or two consecutive visits. HbA1c was compared between inertia‐YES and inertia‐NO conditions at "T0." In the inertia‐YES condition, T0 represents the point at which an individual presents with an HbA1c >58 mmol/mol (>$7.5\%$) for the second consecutive time and is the point at which insulin therapy initiation would have been appropriate. 24 In the inertia‐NO condition, T0 represents the point at which a patient who has had an HbA1c >58 mmol/mol (>$7.5\%$) for one or two consecutive visits is prescribed insulin. The total data pool comprised 129 373 individuals, 32 752 of whom were started on insulin therapy in accordance with the 2020 guidelines of the American Diabetes Association, 43 375 whose insulin therapy initiation was delayed, and 53 246 who never received insulin as reported in Table 1.
Figure 2 provides an illustration of the stratification of HbA1c levels across patients who experienced a delay in insulin initiation (inertia‐yes) and those who did not (inertia‐no).
**FIGURE 2:** *Stratification of HbA1c levels at "T0" across patients who experienced a delay in insulin initiation (Inertia‐yes) and those who did not (Inertia‐no)* **FIGURE 3:** *Area under the receiver operating characteristic (AUC ROC) curve of the best performing model* **FIGURE 4:** *Average HbA1c gap across HbA1c ranges at T0 for individuals who did and did not experience therapeutic inertia. Mann–Whitney (U) and number of subjects is given for each range. (*) indicates statistical significance* **FIGURE 5:** *Area under the receiver operating characteristic (AUC ROC) Curve of the HbA1c gap relative to YES/NO inertia* **FIGURE 6:** *HbA1c variation for subgroups of individuals who are initiated on insulin either at T0 + 1 or on subsequent visits (T0 + 2…T0 + n). Mann–Whitney (U) and number of subjects is given for each range. (*) indicates statistical significance*
The primary goal of the statistical analyses was to verify if the HbA1c gap was able to discriminate the presence or absence of inertia for any HbA1c level at a particular visit. Figure 4 shows the calculated HbA1c gap across different ranges of HbA1c at T0 for individuals who did and did not experience therapeutic inertia. What can be observed in Figure 4 is that the HbA1c gap is systematically higher across all ranges of HbA1c in those individuals who did not experience therapeutic inertia. To confirm the statistical validity of the results, we began by determining whether the data have a normal distribution by submitting the entire range of HbA1c values to a Jarque–Bera test, which is typically used for large data sets such as ours. The results of the test indicated a nonnormal distribution ($p \leq .05$). The data set were then submitted to a Mann–Whitney test for each of the pairwise comparisons starting with 58–64 mmol/mol ($7.5\%$–$8\%$) and ending with 119 mmol/mol ($13\%$). The analyses confirmed a statistically significant difference in the average HbA1c gap between the two groups (presence of absence of inertia) ($p \leq .01$, two tailed test) across all ranges.
A further verification was made by calculating the ROC curve for the HbA1c gap related to inertia. Figure 5 shows the AUC (0.776) that confirms the ability of the HbA1c gap to discriminate between situations of therapeutic inertia from those where no therapeutic inertia was present as suggested by ML. The Youden index for the HbA1c gap indicates a threshold value equal to 5.5 mmol/mol ($0.505\%$), which agrees with the threshold value obtained with ML. It should be noted that the threshold values obtained with the two techniques were not identical, given that ML takes into consideration the HbA1c gap along with all other variables input into the model, and the Youden index refers strictly to the HbA1c as a single variable.
Finally, in order to verify if the HbA1c gap plays a role in the discrimination between situations in which therapeutic inertia is present and those situations in which it is not even after T0, all patient visits after T0 (T0 + 1, T0 + 2…T0 + n) were analyzed and the HbA1c gap was calculated. Figure 6 illustrates the results of this analysis for individuals who received insulin and those who continued to experience therapeutic inertia. Figure 6 once again shows that even for patient visits subsequent to T0, the HbA1c gap is systematically elevated in those individuals who receive insulin relative to those who continue to experience therapeutic inertia. The Jarque–Bera test was also applied to this series of data and results indicated a nonnormal distribution ($p \leq .05$). A Mann–Whitney test revealed that for all periods following T0, average values between the two groups were statistically significant ($p \leq .01$, two‐tailed test). This finding further confirms the discriminating role of HbA1c between situations in which therapeutic inertia is present and those in which it is not.
## DISCUSSION
The current state of therapeutic inertia and delay in initiation of insulin therapy in individuals with type 2 diabetes is systemic and unsettling. 2, 3 This inadequate approach to glycemic control in individuals with type 2 diabetes is a global concern and one that is related to poor patient outcome as well as increased socioeconomic and clinical burden. 4, 5, 25, 26 Though several studies have provided explanations for the motivation behind the delay in insulin therapy, 27, 28 clearly, new approaches are needed to get at the root of the factors that are associated with therapeutic inertia, particularly the identification of key drivers that may break a health care provider's tendency to delay initiation of therapy.
Our objective for this study was to uncover as yet unrecognized factors that motivate a provider to move away from behavioral inertia and initiate insulin therapy using ML techniques and evidence‐based medicine. LLM, a “clear box” ML with “explainable AI,” is able to delineate the characteristics of those individuals who face therapy initiation inertia compared with those who undergo treatment with insulin in accordance with established guidelines. It was thus possible to establish a predictive model capable of identifying key drivers associated with the initiation of insulin therapy with high accuracy (0.79).
Specifically, our results revealed that a medical practitioner was more likely to initiate insulin not only as a consequence of excessively elevated HbA1c as one might expect, but also when, across two consecutive visits, a patient showed a difference in HbA1c of $1\%$ or greater. This difference, which we are calling the HbA1c gap, is correlated with a movement toward insulin initiation irrespective of the absolute HbA1c value and could play an important role in the decision to start insulin. That is to say, it was the HbA1c gap that prompted insulin initiation rather than only the absolute HbA1c at a current visit, which in our study varied between $7.5\%$ and $11\%$.
We believe that our results could provide the medical practitioner with an additional and new measure to monitor in existing or new patients. One could even envision the inclusion of an AI algorithm as part of a patient's electronic health record, which could alert the provider in real time as to risks related not only to inertia but to other health concerns that can be modeled in a manner similar to what has been described in this report. This type of analysis which has been deemed “augmented intelligence” is a way to use AI to improve the quality of decision‐making rather than substitute or automate human decision‐making. Rather than waiting for a critical point at which the patient begins to experience new and potentially dangerous comorbidities as the moment to begin insulin therapy, the medical practitioner could remain increasingly more aware of variabilities in a patient's condition in real time and intervene more promptly and appropriately.
The model's high level of precision confirms the adequacy and utility of the input variables to identify key drivers that influence a health care provider's tendency to exhibit or not exhibit inertia when faced with an individual with type 2 diabetes. The model clearly points to the role of dynamic variables related to glycemia as crucial for the determination of whether or not a health care provider will make a timely decision or remain inert with respect to the initiation of insulin therapy.
A weakness in the study is related to the need to validate findings in a pilot study first. Furthermore, a weakness of any ML model is its reliance on electronic medical data, which to be fully functional for use with ML, must somehow remain in a public domain and not be privately owned data. Otherwise, access to the data can be terminated when an attempt is made at integration with ML modeling from external software.
The data from this study provide new insight for health care providers as they face the challenges of understanding their patients and provide the individualized care that each needs. The information generated by these LLM analyses not only will allow health care providers to gain awareness into the factors that drive their behavior leading to therapeutic inertia but also clearly paves the way for in‐depth investigations of other unknown factors using ML techniques that could help identify further subgroups at greater risk for therapeutic inertia.
In sum, the presence of inertia in the absence of complications suggests how the importance of timely and decisive therapeutic action is still often underestimated even though current guidelines clearly indicate that timely therapeutic action prevents future complications. The present real‐life study has demonstrated how research on inertia can generate new points of view and novel approaches, which will provide healthcare providers the ability to create innovative, effective, and realistic training on this highly debated topic.
## DISCLOSURE
The authors declare that there is no conflict of interest. This study used existing data located on a database and as such did not involve the direct participation of humans or animals. Therefore, no ethics board approval was required. Furthermore, no data used in this study had any identifiable features that could be traced back to the individual from whom the data were obtained.
## References
1. Selvin E, Parrinello CM, Daya N, Bergenstal RM. **Trends in insulin use and diabetes control in the US: 1988–1994 and 1999–2012**. *Diabetes Care* (2016) **39** e33-e35. PMID: 26721815
2. Khunti K, Millar‐Jones D. **Clinical inertia to insulin initiation and intensification in the UK: a focused literature review**. *Prim Care Diabetes* (2017) **11** 3-12. PMID: 27727005
3. Harris S, Yale J‐F, Dempsey E, Gerstein H. **Can family physicians help patients initiate basal insulin therapy successfully?: randomized trial of patient‐titrated insulin glargine compared with standard oral therapy: lessons for family practice from the Canadian INSIGHT trial**. *Can Fam Physician* (2008) **54** 550-558. PMID: 18411384
4. Khunti K, Wolden ML, Thorsted BL, Andersen M, Davies MJ. **Clinical inertia in people with type 2 diabetes: a retrospective cohort study of more than 80,000 people**. *Diabetes Care* (2013) **36** 3411-3417. PMID: 23877982
5. Michelli A, Cimino E, Manicardi V. **AMD annals 2020 synopsis on type 2 diabetes evaluation of AMD quality of indicators of type 2 diabetes care in Italy**. *J AMD* (2021) **24** 19-29
6. Ross SA. **Breaking down patient and physician barriers to optimize glycemic control in type 2 diabetes**. *Am J Med* (2013) **126** S38-S48. PMID: 23953078
7. Russell‐Jones D, Pouwer F, Khunti K. **Identification of barriers to insulin therapy and approaches to overcoming them**. *Diabetes Obes Metab* (2018) **20** 488-496. PMID: 29053215
8. Khunti S, Davies MJ, Khunti K. **Clinical inertia in the management of type 2 diabetes mellitus: a focused literature review**. *Br J Diabetes* (2015) **15** 65-69
9. Rossi MC, Nicolucci A, Arcangeli A. **Baseline quality‐of‐care data from a quality‐improvement program implemented by a network of diabetes outpatient clinics**. *Diabetes Care* (2008) **31** 2166-2168. PMID: 18694979
10. Nicolucci A, Rossi M, Arcangeli A. **Four‐year impact of a continuous quality improvement effort implemented by a network of diabetes outpatient clinics: the AMD‐annals initiative**. *Diabet Med* (2010) **27** 1041-1048. PMID: 20722678
11. Rossi MC, Candido R, Ceriello A. **Trends over 8 years in quality of diabetes care: results of the AMD annals continuous quality improvement initiative**. *Acta Diabetol* (2015) **52** 557-571. PMID: 25528003
12. Abhari S, Kalhori SRN, Ebrahimi M, Hasannejadasl H, Garavand A. **Artificial intelligence applications in type 2 diabetes mellitus care: focus on machine learning methods**. *Healthcare Inf Res* (2019) **25** 248-261
13. Challen R, Denny J, Pitt M, Gompels L, Edwards T, Tsaneva‐Atanasova K. **Artificial intelligence, bias and clinical safety**. *BMJ Qual Safety* (2019) **28** 231-237
14. Verda D, Parodi S, Ferrari E, Muselli M. **Analyzing gene expression data for pediatric and adult cancer diagnosis using logic learning machine and standard supervised methods**. *BMC Bioinf* (2019) **20** 1-13
15. Muselli M. **Extracting knowledge from biomedical data through logic learning machines and Rulex**. *EMBnet J* (2012) **18(B)** 56-58
16. Michie D, Spiegelhalter DJ, Taylor CC. *Machine Learning, Neural and Statistical Classification* (1994)
17. Giorda CB, Pisani F, De Micheli A. **Determinants of good metabolic control without weight gain in type 2 diabetes management: a machine learning analysis**. *BMJ Open Diabetes Res Care* (2020) **8**
18. Masi D, Risi R, Biagi F. **Application of a machine learning Technology in the Definition of metabolically healthy and unhealthy status: a retrospective study of 2567 subjects suffering from obesity with or without metabolic syndrome**. *Nutrients* (2022) **14** 373. PMID: 35057554
19. Duda RO, Hart PE. *Pattern Classification and Scene Analysis* (1973)
20. Boros E, Hammer PL, Ibaraki T, Kogan A, Mayoraz E, Muchnik I. **An implementation of logical analysis of data**. *IEEE Trans Knowl Data Eng* (2000) **12** 292-306
21. Muselli M, Ferrari E. **Coupling logical analysis of data and shadow clustering for partially defined positive Boolean function reconstruction**. *IEEE Trans Knowl Data Eng* (2009) **23** 37-50
22. Muselli M. *Switching Neural Networks: A New Connectionist Model for Classification* (2005) 23-30
23. Linardatos P, Papastefanopoulos V, Kotsiantis S. **Explainable ai: a review of machine learning interpretability methods**. *Entropy* (2021) **23** 18
24. Doyle‐Delgado K, Chamberlain JJ, Shubrook JH, Skolnik N, Trujillo J. **Pharmacologic approaches to glycemic treatment of type 2 diabetes: synopsis of the 2020 American Diabetes Association's standards of medical care in diabetes clinical guideline**. *Ann Intern Med* (2020) **173** 813-821. PMID: 32866414
25. Bain SC, Bekker Hansen B, Hunt B, Chubb B, Valentine WJ. **Evaluating the burden of poor glycemic control associated with therapeutic inertia in patients with type 2 diabetes in the UK**. *J Med Econ* (2020) **23** 98-105. PMID: 31311364
26. Ali SN, Dang‐Tan T, Valentine WJ, Hansen BB. **Evaluation of the clinical and economic burden of poor glycemic control associated with therapeutic inertia in patients with type 2 diabetes in the United States**. *Adv Ther* (2020) **37** 869-882. PMID: 31925649
27. McCoy RG, O'Connor PJ. **Overcoming therapeutic inertia in type 2 diabetes care—timing, context, and appropriateness of treatment intensification**. *JAMA Netw Open* (2021) **4**. PMID: 34673968
28. Pantalone KM, Misra‐Hebert AD, Hobbs TM. **Clinical inertia in type 2 diabetes management: evidence from a large, real‐world data set**. *Diabetes Care* (2018) **41** e113-e114. PMID: 29678811
|
---
title: 'Assessment of subclinical left ventricular systolic dysfunction in patients
with type 2 diabetes: Relationship with HbA1c and microvascular complications'
authors:
- Yanyan Chen
- Ying Zhang
- Yi Wang
- Shengjun Ta
- Min Shi
- Yingni Zhou
- Mengying Li
- Jianfang Fu
- Li Wang
- Xiangyang Liu
- Zuowei Lu
- Liwen Liu
- Zeping Li
- Jie Zhou
- Xiaomiao Li
journal: Journal of Diabetes
year: 2023
pmcid: PMC10036261
doi: 10.1111/1753-0407.13369
license: CC BY 4.0
---
# Assessment of subclinical left ventricular systolic dysfunction in patients with type 2 diabetes: Relationship with HbA1c and microvascular complications
## Abstract
Highlights In asymptomatic type 2 diabetic patients with preserved left ventricular ejection fraction, a negative correlation of glycated hemoglobin with subclinical left ventricular systolic dysfunction was revealed. Furthermore, glycated hemoglobin may make prognostic significance for the progression of myocardial damage.
### Background
We aimed to examine the association between glycated hemoglobin (HbA1c), microvascular complications, and subclinical left ventricular (LV) systolic dysfunction, and to determine the strength of the correlation in asymptomatic patients with type 2 diabetes mellitus (T2DM).
### Methods
Global longitudinal strain (GLS) was employed to assess the subclinical LV function of 152 enrolled T2DM patients with preserved LV ejection fraction, with the cutoff for subclinical LV systolic dysfunction predefined as GLS < $18\%$.
### Results
According to univariate analysis, the reduced GLS exhibited association with the clinical features including HbA1c, triglyceride, systolic blood pressure, fasting glucose, heart rate, diabetic retinopathy, and urinary albumin creatinine ratio (UACR) (all $p \leq .05$). After the factors of gender, age, and related clinical covariables adjusted, multiple logistic regression analysis revealed the HbA1c (odds ratio [OR] 1.66; $95\%$ confidence interval [CI] 1.30–2.13; $p \leq .001$), UACR (OR 2.48; $95\%$ CI 1.12–5.47; $$p \leq .025$$) and triglyceride (OR 1.84; $95\%$ CI 1.12–3.03; $$p \leq .017$$) as the independent risk factors for the reduced GLS. Receiver operating characteristic curve showed a predictive value of the HbA1c for the subclinical LV systolic dysfunction (area under curve: 0.74; $p \leq .001$).
### Conclusions
In asymptomatic T2DM patients, subclinical LV systolic dysfunction was associated with HbA1c, diabetic complications, and triglyceride. More prominently, HbA1c may exert a prognostic significance for the progression of myocardial damage.
## INTRODUCTION
Diabetes mellitus and heart failure as the mutual risk factors exert an influence on each other, 1 in which disturbances of cardiac lipid and glucose metabolism are considered the early deterioration events of cardiac function in diabetes mellitus. 2 The chronic hyperglycemia resulting from insulin resistance or insulin deficiency has been demonstrated to be the starting point of the cascade that launches the diabetes‐related cardiomyopathy. 3 A linear relationship between glycemic levels and the long‐term mortality of heart failure emerges even before the occurrence of clinical symptoms of diabetes. 4 The clinical pathologic entity was first proposed for “diabetic cardiomyopathy” as early as 1972. 5 Most studies prefer support the conventional hypothesis that left ventricle (LV) diastolic dysfunction is the first transformation in cardiac function during the progression of diabetic cardiomyopathy. 6, 7 However, recently, it has been proposed that systole impairment may occur at a more previous stage of the diabetic cardiomyopathy process prior to the detectable transformation of ejection fraction (EF), and advised subclinical systolic dysfunction might as the first indicator of diabetic cardiomyopathy. 8, 9 *Under this* context, a series of novel diagnostic techniques have been proposed as the deepened research on the evaluation of myocardial function. For instance, the reduced global longitudinal strain (GLS) has been demonstrated to effectively assess global LV systolic function 10 and is considered a sensitive indicator in identifying subtle variations in LV myocardium. The concept of “common soil” theory indicates a great contribution of the diabetes‐related microvascular malfunction to the adverse myocardium alterations. 11 On the other hand, multiple studies have confirmed the relation of glycated hemoglobin (HbA1c) to incident cardiac events. 12 In the UK Prospective Diabetes Study clinical study, each $1\%$ reduction in HbA1c was correlated with a $16\%$ decrease in the risk of heart failure. 13 Furthermore, in a large cohort study HbA1c >$10.0\%$ was correlated with a 1.56‐fold elevated risk of cardiac failure in comparison to HbA1c <$7.0\%$. 14 However, several investigations have recently stated that the intended glycemic control could not achieve a reduced risk of heart failure and cardiovascular events among individuals with diabetes. 15 Whether the level of HbA1c and microvascular complications contribute to subclinical impairment of LV systolic function remains to be investigated. Herein, the current study was carried out to investigate the association between HbA1c, microvascular complications and GLS, so as to determine the strength of the correlation in asymptomatic patients with type 2 diabetes mellitus (T2DM) to estimate the independent effect of those on subclinical LV systolic function.
## Study population
It was initially designed as a 1‐year cross‐sectional survey, although terminated early because of the outbreak of the COVID‐19 pandemic, only half a year's data results were displayed. Patients defined as type 2 diabetes at the Department of Endocrinology of Xijing Hospital of Air Force Medical University from June 2021 to December 2021 were enrolled. The diagnosis of T2DM was based on the World Health Organization criteria. 16 Patients with [1] type 1 diabetes; [2] coronary artery disease or other heart disease history; [3] anemia; [4] severe valvular disorders; [5] LVEF <$50\%$; [6] atrial fibrillation; and [7] blood pressure >$\frac{180}{100}$ mm Hg were excluded. All patients received conventional echocardiography, tissue doppler echocardiography, two‐dimensional speckle tracking echocardiography (2D STE) and bilateral carotid ultrasound examination. The poor echocardiography images were excluded from analysis. Ultimately, 152 subjects were identified and signed informed consent. This study was approved by the local ethics committee of our institution (No. XJLL‐KY20222107).
## Patient information
Demographic information covering diabetes duration, gender, age, medication, history of hypertension, and systolic and diastolic pressures was obtained from the electronic medical record system. The fasting blood samples were taken the next morning using an automatic biochemical analyzer (Centrifugation was performed at 4000 g for 5 minutes) and employed to detect traditional lipid profiles, apolipoproteins A1 and apolipoproteins B, and uric acid. Fasting plasma glucose was measured by the glucose oxidase method. The enzyme was used to measure the triglyceride and total cholesterol. Low‐density lipoprotein‐cholesterol (LDL‐C), high‐density lipoprotein‐cholesterol (HDL‐C), and other biochemical indicators were determined by direct analysis method (Hitachi Automatic Biochemical Analyzer, 7170). HbA1c was checked by high‐performance liquid chromatography (Tocho Liquid Chromatograph, G8‐90SL). Height and weight were collected to calculate the body mass index (BMI) by weight/height 2 (kg/m 2). Urinary albumin‐to‐creatinine ratio (UACR) was examined using random urine and detected by immunoturbidimetry (COBAS INTEGRA 400 plus autoanalyzer, Germany). The clinical diagnosis of diabetic nephropathy was carried out according to an increase in UACR and, or decrease in estimated glomerular filtration rate (<60 mL/min/1.73 m 2) with other chronic kidney diseases excluded by an experienced clinician. UACR <30 mg/g in two of three consecutive measurements was defined as normoalbuminuric, and UACR ≥30 mg/g as increased urinary albumin excretion. The diagnosis of diabetic retinopathy was assessed using a fundoscopy performed by an ophthalmologist, with the bias controlled by the double‐blind method. Also, the carotid intima‐media thickness (cIMT) was recorded simultaneously using the high‐resolution B‐mode ultrasonography to obtain the left and right common carotid artery; the maximal intima‐media thickness on both sides of these was recorded and averaged.
## Conventional echocardiography
The measurements of all subjects were collected based on the American Society of Echocardiography guidelines, 17 and every individual underwent transthoracic echocardiography (Philips Healthcare, iE33 system, X5‐1 probe) synchronously connected to the electrocardiogram. LV fractional shortening, LVEF, heart rate, and stroke volume were subsequently analyzed. In addition, peak velocity in early diastole (E‐wave) and late diastole (A‐wave) were measured to calculate the E/A ratio. The early diastolic mitral annulus velocity (E') was measured by the pulsed wave tissue doppler imaging, to obtain the E/E′ ratio that was assessed as an index of LV filling pressures.
## Speck‐tracking echocardiography
The 2D STE analysis was performed on each subject, with the 2D gray scale dynamic diagrams of three consecutive cardiac cycles of LV apical four‐chamber view and apical two‐chamber view and apical three‐chamber view collected under calm breathing. The endocardial and epicardial borders were automatically traced using the software of QLAB 8.1 2D strain analysis, to display the 2D strain‐time curve and bull's‐eye plot of 17 segments of LV. The image would be reread if two or more components of the global average strain were missed until no more than one fragment was rejected. The untraceable images of spots resulting from atrial fibrillation were excluded from the analysis. The average value of the three peak strains in systole was employed to evaluate the LV systolic function by calculating the LV GLS. According to the current guidelines of the European Association of Cardiovascular Imaging, subclinical LV systolic dysfunction was defined as GLS < $18\%$. 17, 18
## Statistical analysis
The normal distribution was determined according to the P–P plot and Kolmogorov–Smirnov Test, and a nonparametric test was employed if the assumption of normality was not met. According to the distribution, categorical variables were displayed as percentages n (%) and continuous variables as means ± SD, or median and interquartile range. The comparison of quantitative variables was conducted using Student's t or Mann–Whitney U tests. The relationship between continuous variables was assessed by the Pearson correlation coefficients. Categorical variables were compared using Fisher's exact or the Pearson's hi‐square test. The independent risk factors of injured GLS were determined by multivariable logistic regression analysis. The age, gender, and the significant variables with p values <.05 in the univariate logistic regression were adjusted by backward stepwise selection, with the odds ratio (OR) and $95\%$ confidence interval (CI) provided. No multicollinearity was detected between variables by variance inflation factor check. The predictive capability of HbA1c was estimated using receiver‐operating characteristic curves (ROC), with the optimum cutoff value, sensitivity, and specificity given. Based on the distribution, UACR was converted into logarithmic form for each analysis. Patients with retinopathy and, or nephropathy were classified as the complications group. To prevent the interobserver and intraobserver variability of GLS, the measurement of GLS was performed by one professional physician. To avoid the ambiguity of negative size to a value, the GLS was displayed in the absolute value form. All statistical analysis was conducted by SPSS statistics version 26.0, with p value on two‐sided <.05 regarded as significant.
## Basic features of patients with GLS <18% and ≥18%
According to GLS levels, the basic T2DM patient characteristics were summarized in Table 1. A total of 152 subjects were included, among whom $46.7\%$ suffered GLS <$18\%$ ($$n = 71$$), with an increased possibility to suffer higher levels of HbA1c and poor blood glucose control, hypertriglyceridemia, and higher systolic pressure and heart rate. Moreover, they showed a higher prevalence of diabetic microvascular complications (neuropathy and retinopathy). No obvious difference was found in gender, age, diabetes duration, BMI, and prevalence of hypertension between patients with GLS <$18\%$ and GLS ≥$18\%$, accompanied with the similar levels of conventional echocardiographic parameters and cIMT (all $p \leq .05$).
**TABLE 1**
| Characteristic | Patients with GLS ≥18% (n = 81) | Patients with GLS <18% (n = 71) | p value |
| --- | --- | --- | --- |
| Male gender, n (%) | 52 (64) | 45 (63) | .92 |
| Age, years | 54.9 ± 13.6 | 51.1 ± 14.5 | .11 |
| BMI, kg/m2 | 23.5 ± 3.6 | 24. 6 ± 3.9 | .09 |
| Diabetes duration, years | 9.9 ± 7.2 | 10.3 ± 7.3 | .80 |
| Heart rate, bpm | 74.1 ± 11.5 | 79.2 ± 13.3 | .012 |
| SBP, mm Hg | 130.2 ± 14.1 | 136.6 ± 21.2 | .028 |
| DBP, mm Hg | 77.1 ± 9.1 | 79.7 ± 12.7 | .15 |
| cIMT, mm | 0.41 ± 0.5 | 0.45 ± 0.3 | .64 |
| HbA1c, % | 8.0 ± 1.5 | 9.8 ± 2.3 | <.001 |
| FPG, mmol/L | 10.8 ± 4.6 | 12.7 ± 5.1 | .015 |
| Total cholesterol, mmol/L | 4.0 ± 1.0 | 4.1 ± 1.5 | .39 |
| HDL, mmol/L | 1.2 ± 0.5 | 1.0 ± 0.3 | .063 |
| LDL, mmol/L | 2.4 ± 1.2 | 2.4 ± 1.1 | .90 |
| Triglyceride, mmol/L | 1.4 ± 0.8 | 2.0 ± 1.7 | .01 |
| Apolipoproteins A1, g/L | 1.2 ± 0.2 | 1.1 ± 0.2 | .23 |
| Apolipoproteins B, g/L | 0.7 ± 0.4 | 0.7 ± 0.3 | .92 |
| Albuminuria, mg/L | 10.7 (8.3–14.3) | 14.5 (8.4–48.3) | .01 |
| UACR, mg/mmoL | 1.2 (0.8–2.4) | 2.2 (1.4–5.9) | <.001 |
| Uric acid, umol/L | 320.6 ± 74.1 | 321.9 ± 90.0 | .92 |
| Diabetic nephropathy, n (%) | 14 (17) | 21 (30) | .072 |
| Diabetic retinopathy, n (%) | 6 (7) | 15 (21) | .018 |
| Hypertension, n (%) | 32 (40) | 37 (52) | .119 |
| Medical treatment | Medical treatment | Medical treatment | Medical treatment |
| ACEI/ARB, n (%) | 14 (17) | 20 (28) | .108 |
| CCB, n (%) | 13 (16) | 18 (25) | .156 |
| Statin, n (%) | 15 (19) | 20 (28) | .159 |
| Insulin, n (%) | 40 (49) | 43 (61) | .167 |
| DPP‐4I, n (%) | 10 (12) | 7 (10) | .627 |
| SGLT‐2I, n (%) | 9 (11) | 5 (7) | .387 |
| GLP‐1RA, n (%) | 6 (7) | 9 (13) | .277 |
| Metformin, n (%) | 60 (74) | 45 (63) | .155 |
| αGI, n (%) | 28 (35) | 26 (37) | .792 |
| Echocardiographic indexes | Echocardiographic indexes | Echocardiographic indexes | Echocardiographic indexes |
| LV EF, % | 60.2 ± 4.3 | 59.9 ± 5.1 | .65 |
| LV FS, % | 32.0 ± 3.4 | 32.0 ± 4.3 | .98 |
| Stroke volume, mL | 46.2 ± 8.0 | 48.6 ± 9.8 | .12 |
| E/A ratio | 1.1 ± 1.9 | 0.9 ± 0.4 | .48 |
| E/E' ratio | 9.9 ± 3.8 | 10.2 ± 3.1 | .67 |
| GLS, % | 20.5 ± 2.1 | 15.1 ± 2.2 | <.001 |
## Association of clinical features with reduced GLS
The comprehensive evaluation of relationship between clinical and biochemical features and GLS was provided in Table 2. Univariate logistic regression analysis indicated the reduced GLS <$18\%$ was prominently associated with HbA1c, triglyceride, UACR, systolic blood pressure, fasting plasma glucose, heart rate, and diabetic retinopathy (all $p \leq .05$). Accordingly, after adjusting for gender, age, and relevant clinical covariables (defined as $p \leq .05$ in the univariate analysis), the most diabetes‐related clinical features difference was attenuated, whereas the independent association of higher HbA1c (OR: 1.66; $95\%$ CI = 1.30–2.13; $p \leq .001$), UACR (log‐transformed; OR 2.48; $95\%$ CI: 1.12–5.47; $$p \leq .025$$), and triglyceride (OR = 1.84; $95\%$ CI: 1.12–3.03; $$p \leq .017$$) with GLS < $18\%$ remained (Table 2). Furthermore, Pearson correlation analysis revealed a negative correlation of the HbA1c level (r = −0.48, $p \leq .001$) with GLS on plotting in the scatter diagram (Figure 1).
The diagnostic performance of HbA1c and the proposed cutoff value was presented in Figure 2 by ROC curves. Notably, the cutoff value of HbA1c (area under the curve = 0.74; $p \leq .001$) was $8.8\%$, revealing a sensitivity of $69.0\%$ and specificity of $72.5\%$ for predictive subclinical LV systolic dysfunction.
**FIGURE 2:** *Receiver‐operating characteristic curves for prediction of left ventricular systolic dysfunction in patients with type 2 diabetes using HbA1c. AUC, area under curve; HbA1c, glycated hemoglobin.*
Figure 3 provided the representative cases of GLS ≥$18\%$ and GLS < $18\%$ in a bull's‐eye plot of patients with type 2 diabetes.
**FIGURE 3:** *Representative cases of GLS ≥18% and GLS <18% in a bull's eye plot of patients with type 2 diabetes. GLS, global longitudinal stain; HbA1c, glycated hemoglobin; LVEF, left ventricular ejection fraction; T2DM, type 2 diabetes mellitus.*
## DISCUSSION
Our study demonstrated the main potential contributors of suboptimal glycemic control, hyperlipidemia, and microangiopathy to the impaired myocardial systolic function in asymptomatic T2DM patients. More significantly, HbA1c may exert a certain prognostic significance for the progression of myocardial damage.
## Clinical risk factors and LV longitudinal myocardial function
In this study, a total of $46.7\%$ of patients exhibited the decreased GLS, which reached an accordance with the prior reports of subclinical LV longitudinal systolic dysfunction in people with diabetes. In various studies, GLS was adopted as the preferential indicator to evaluate the global systolic function, for the longitudinal subendocardium fibers as the most vulnerable part are the first to suffer damage from metabolism disorder in the early stage of the clinical pathologic entity proposed for “diabetic cardiomyopathy.” 19 The more specific mechanism of this disorder remains unclear. However, metabolic characterizations indicate that chronic hyperglycemia and glucotoxicity exert direct damage on myocardial cells through the accumulated advanced glycosylation end products or lead to myocardial ischemia by eliciting the steady deterioration of endothelial function. 20 Lipotoxic injury from lipid oversupply is also considered to play a critical role in developing myocardial injury. 21 Conclusively, glucotoxicity, lipotoxicity, and coronary microcirculation dysfunction might jointly affect longitudinal cardiac dysfunction, manifested by the more advanced transformation in the early form of myocardial deformation.
In addition, research based on the Framingham Heart Study has revealed the significant conjoint associations of hypertension and diabetes with GLS, suggesting a synergistic effects on the reduced GLS. 22 Ballo et al demonstrated that diabetes, but not hypertension, exerted a negative effect on LV systolic function. 23 The current study revealed that T2DM patients with reduced GLS preferentially suffered higher systolic blood pressure, compared with those with GLS ≥$18\%$, further validating the adverse influence of hypertension on the subclinical myocardial injury. Moreover, the relationship between carotid atherosclerosis and GLS was also a focus. In a study covering 338 young individuals with concomitant obesity and type 2 diabetes, the increased cIMT was revealed to be independently associated with the reduced GLS, 24 which, however, was not observed in the present study. This may be related to the fact that thickened cIMT is generally taken as a risk marker for diabetic macrovascular disease, 25 whereas microangiopathy tends to exert a more prominent effect on the impaired global strain in young subjects. Moreover, another clinical trial by Yamauchi et al demonstrated a close association of heart rate with LV longitudinal myocardial function and highlighted a great contribution of high heart rate ≥70 bpm to early LV damage in asymptomatic T2DM patients. 9 The consistent result was also revealed by the univariable analysis of this study but weakened in multiple regression analysis, which might result from the higher proportion of patients with GLS ≤$18\%$ taking cardiovascular protective medications.
## Hyperglycemia and LV longitudinal myocardial function
In this study, it is notable that fasting plasma glucose was associated with reduced GLS, but this did not emerge in multivariate analysis. The underlying pathological mechanism could be unstable fasting glycemia that is susceptible to interference by other stress factors such as diet and medication. Moreover, other investigators have proposed that the cardiomyopathy induced by glucose toxicity could be reasonably considered the cause of cardiac dysfunction in diabetic patients. 26 For instance, epidemiological studies have demonstrated the association of each $1\%$ increase in HbA1c with the relative risk of cardiovascular disease increase by $18\%$ and with an $8\%$ increase in heart failure among T2DM patients. 27 The present study found a significant correlation of HbA1c with the reduced GLS, independent of conventional cardiovascular risk factors, which corresponded to the previous research, 28 demonstrating a close association of poor glycemic control with LV longitudinal dysfunction. And, most important, the results of ROC showed that HbA1c > $8.8\%$ could identify patients with high risk of early alterations of myocardial function. This finding is clinically relevant and supports that HbA1c serves as an independent warning and a screening predictor of early diagnosis of subclinical ventricular dysfunction.
## Microvascular complications and LV longitudinal myocardial function
Although the relationship between retinopathy and the heart has been established, the correlation with subclinical LV systolic function remains to be discovered. A cross‐sectional study including 82 T2DM patients demonstrated that GLS was free from the impacts of retinopathy. 29 Similarly, Pararajasingam et al also pointed out that none of the subgroups of retinopathies had a relation to GLS. 30 However, increasing studies demonstrated the prominent relation of the diabetic retinopathy to the impaired LV systolic function as evaluated by GLS <$18\%$. 31, 32 Accordingly, out of 71 patients with reduced GLS in present study, only 15 ($21.1\%$) retinopathy patients exhibited GLS <$18\%$. Nonetheless, there remained a significant difference in GLS between patients with or without retinopathy, which further supported the latter conclusion.
More recent evidence suggests that diabetic nephropathy, especially albuminuria, was not only a robust predictor of cardiovascular events in diabetes but more an early marker of widespread vascular injury with high sensitivity. 33 A study reported the close relation of the declined myocardial flow reserve to albuminuria based on positron emission tomography/computed tomography outcomes. 34 Additionally, according to the Adolescent Type 1 Diabetes Cardio‐Renal Intervention Trial, the determination of UACR has been suggested to exert more prognostic value for cardiovascular complications of diabetes. 35 Here, our result revealed the UACR as an independent and contributing factor for the reduced GLS, showing a significant inverse relationship with GLS. These data were consistent with Mochizuki et al's and Levelt et al‘s research. 32, 36 To some extent, the theory of “common soil” could partly interpret those findings, and it is thus more logical that microvascular complications, especially albuminuria, may serve as an early clinical warning of vascular damage and endothelial dysfunction, illustrating an increased heart burden and impairing cardiac performance.
## Serum lipids and LV longitudinal myocardial function
More recent evidence suggests that under hyperglycemia conditions, the increased fatty acid uptake by myocardium exceeds the oxidation capacity of nonadipose tissues to free fatty acids, which results in the excessive deposition within heart in the form of triglyceride known as lipotoxicity. 37 In terms of molecular mechanism, the transition of cardiac metabolism leading to lipid toxic injury greatly contributes to the pathogenesis of diabetic cardiomyopathy. Generally, triglyceride is relatively inert and thus does not directly mediate lipotoxicity but rather the intermediate product of triglyceride primarily responsible for cardiac dysfunction. 38 However, there exist debates on the role of hypertriglyceridemia as an independent cardiovascular risk factor. 39 More recent epidemiological studies tended to support the cascading effect of triglyceride that the excessive production of proinflammatory cytokines and the reduced endothelial nitric oxide biosynthesis jointly contribute to cardiac microcirculation dysfunction. 40 Depending on the results of magnetic resonance spectroscopy, research conducted by Ng et al confirmed an association of high myocardial triglyceride content with LV myocardial longitudinal strain. 41 *In a* study of type1 diabetes by Vinereanu et al, LDL rather than HbA1c was found the only independent predictor of the cutoff value of $18.6\%$ of abnormal LV GLS, 42 which was consistently revealed by other authors in T2DM, followed by the inverse relation of LV longitudinal function to LDL. 43 In contrast, another study covering 144 type 1 and type 2 diabetes has identified that hypertriglyceridemia but not low‐ and high‐density lipoprotein, resulted in further damage to GLS. 36 Similarly, a study carried out on a population of Chinese ethnicity in Taiwan has emphasized that triglycerides potentially played an adverse effect on the progress of cardiovascular, free from other lipid parameters. 44 Similar outcomes were also obtained in Chinese patients with T2DM and proved in the present study of ours, illustrating a possible involvement of triglyceride in the occurrence and development of LV systolic function impairment. Additionally, BMI was not an independent risk factor as identified here, which showed bias compared to the study of Tseng et al, where BMI was determined as a significant predictor for heart disease in T2DM patients in Taiwan. 45 This discrepancy could perhaps be explained by the selection bias, lack of adequate sample size, and the deficiency in accuracy of BMI measurements. Notably, the patients with reduced GLS were more prone to receiving lipid‐lowering drugs (especially statins) compared to those with normal GLS. As was reported, statins contributed a lot to the correlation between glycemic control, dyslipidemia, and cardiovascular disease. 46 Another nonnegligible observation was the increased exogenous insulin use but the decreased metformin use on subjects with reduced GLS, which has exhibited protective effects on the development of hypertension 47 and heart failure. 48 Despite no statistical effect of medication on GLS was observed, the results should be interpreted as the important confounding effects of the use of statins and antidiabetic drugs.
Expectedly, the present study has certain clinical implications. It demonstrated that glucotoxicity and lipotoxicity promoted by metabolic disturbances were the triggers to the hallmark pathological transformations in the diabetic heart. In addition, microvascular complications may provide incremental diagnostic benefit in identifying transformations in subclinical myocardial function, which in turn allows for an earlier identification of risk factors to prevent heart failure progression. However, the data on the correlation between subclinical LV dysfunction and microvascular complications in diabetic patients are currently still limited. The findings in this study further display a close relationship between microvascular complications and the reduced GLS. In addition, not only the reduced GLS concerning the uncontrolled glucose and serum lipid levels was further confirmed, but the newly clinical significance of HbA1c was also a focus. The results of the ROC curve indicated HbA1c >$8.8\%$ provided a reasonable specificity to identify high‐risk patients with diabetes. Conclusively, our research demonstrates the advantage of wide clinical applicability that evaluation of HbA1c, including blood lipids, is a convenient and quick screening tool with low cost. It not only could identify altered metabolic homeostasis in earlier individuals but could also serve as an initial warning for clinical monitoring of myocardial dysfunction, emphasizing the intensive control of glucose and blood lipids as a potentially promising strategy in patients with type 2 diabetes without the history of cardiovascular events.
The limitations of this study should also be considered. First, as a cross‐sectional study, whether strict glycemic control is associated with improvement in GLS remains unclear, and further observation by follow‐up is required. Second, invasive coronary angiography was not performed on patients without coronary artery disease. In addition, this retrospective study was carried out for hospitalized type 2 diabetes patients, with a single‐center population enrolled, so that selection bias cannot be fully ruled out, resulting in the limited extensibility of the research results. Finally, considering the epidemic prevention and the control measures that may affectt the lifestyle and therapeutic drugs of patients, stratified cluster sampling is recommended so that large samples will be adopted for prospective research to reduce bias in future studies.
In conclusion, the present study illustrated a robust correlation between diabetic complications with subclinical LV dysfunction. In addition, HbA1c and hypertriglyceridemia could serve as independent risk factors for the early stage of LV myocardial dysfunction in asymptomatic T2DM patients. More important, HbA1c may exert prognostic significance for the progression of myocardial damage.
## CONFLICT OF INTEREST
No potential conflict of interest relevant to this article was reported.
## References
1. Tseng CH. **Clinical features of heart failure hospitalization in younger and elderly patients in Taiwan**. *Eur J Clin Invest* (2011) **41** 597-604. PMID: 21158853
2. Bayeva M, Sawicki KT, Ardehali H. **Taking diabetes to heart—deregulation of myocardial lipid metabolism in diabetic cardiomyopathy**. *J Am Heart Assoc* (2013) **2**. PMID: 24275630
3. Jia G, Demarco VG, Sowers JR. **Insulin resistance and hyperinsulinaemia in diabetic cardiomyopathy**. *Nat Rev Endocrinol* (2016) **12** 144-153. PMID: 26678809
4. Jia G, Hill MA, Sowers JR. **Diabetic cardiomyopathy: an update of mechanisms contributing to this clinical entity**. *Circ Res* (2018) **122** 624-638. PMID: 29449364
5. Rubler S, Dlugash J, Yuceoglu YZ, Kumral T, Branwood AW, Grishman A. **New type of cardiomyopathy associated with diabetic glomerulosclerosis**. *Am J Cardiol* (1972) **30** 595-602. PMID: 4263660
6. Galderisi M. **Diastolic dysfunction and diabetic cardiomyopathy: evaluation by Doppler echocardiography**. *J Am Coll Cardiol* (2006) **48** 1548-1551. PMID: 17045886
7. From AM, Scott CG, Chen HH. **The development of heart failure in patients with diabetes mellitus and pre‐clinical diastolic dysfunction a population‐based study**. *J Am Coll Cardiol* (2010) **55** 300-305. PMID: 20117433
8. Ernande L, Bergerot C, Rietzschel ER. **Diastolic dysfunction in patients with type 2 diabetes mellitus: is it really the first marker of diabetic cardiomyopathy?**. *J Am Soc Echocardiogr* (2011) **24** 1268-1275.e1261. PMID: 21907542
9. Yamauchi Y, Tanaka H, Yokota S. **Effect of heart rate on left ventricular longitudinal myocardial function in type 2 diabetes mellitus**. *Cardiovasc Diabetol* (2021) **20** 87. PMID: 33894777
10. Minciună IA, Hilda Orășan O, Minciună I. **Assessment of subclinical diabetic cardiomyopathy by speckle‐tracking imaging**. *Eur J Clin Invest* (2021) **51**. PMID: 33326612
11. Adameova A, Dhalla NS. **Role of microangiopathy in diabetic cardiomyopathy**. *Heart Fail Rev* (2014) **19** 25-33. PMID: 23456446
12. Rubin J, Matsushita K, Ballantyne CM, Hoogeveen R, Coresh J, Selvin E. **Chronic hyperglycemia and subclinical myocardial injury**. *J Am Coll Cardiol* (2012) **59** 484-489. PMID: 22281251
13. Stratton IM, Adler AI, Neil HA. **Association of glycaemia with macrovascular and microvascular complications of type 2 diabetes (UKPDS 35): prospective observational study**. *BMJ* (2000) **321** 405-412. PMID: 10938048
14. Iribarren C, Karter AJ, Go AS. **Glycemic control and heart failure among adult patients with diabetes**. *Circulation* (2001) **103** 2668-2673. PMID: 11390335
15. Zhan J, Chen C, Wang DW, Li H. **Hyperglycemic memory in diabetic cardiomyopathy**. *Front Med* (2022) **16** 25-38. PMID: 34921674
16. Alberti KG, Zimmet PZ. **Definition, diagnosis and classification of diabetes mellitus and its complications. Part 1: diagnosis and classification of diabetes mellitus provisional report of a WHO consultation**. *Diabet Med* (1998) **15** 539-553. PMID: 9686693
17. Lang RM, Badano LP, Mor‐Avi V. **Recommendations for cardiac chamber quantification by echocardiography in adults: an update from the American Society of Echocardiography and the European Association of Cardiovascular Imaging**. *Eur Heart J Cardiovasc Imaging* (2015) **16** 233-270. PMID: 25712077
18. Lang RM, Badano LP, Mor‐Avi V. **Recommendations for cardiac chamber quantification by echocardiography in adults: an update from the American Society of Echocardiography and the European Association of Cardiovascular Imaging**. *J Am Soc Echocardiogr* (2015) **28** 1-39.e14. PMID: 25559473
19. Dillmann WH. **Diabetic cardiomyopathy**. *Circ Res* (2019) **124** 1160-1162. PMID: 30973809
20. Wang M, Li Y, Li S, Lv J. **Endothelial dysfunction and diabetic cardiomyopathy**. *Front Endocrinol (Lausanne)* (2022) **13**. PMID: 35464057
21. Zhang C, Zhang B, Zhang X, Wang M, Sun X, Sun G. **Panax notoginseng Saponin protects against diabetic cardiomyopathy through lipid metabolism modulation**. *J Am Heart Assoc* (2022) **11**. PMID: 35112884
22. Von Jeinsen B, Vasan RS, Mcmanus DD, Mitchell GF, Cheng S, Xanthakis V. **Joint influences of obesity, diabetes, and hypertension on indices of ventricular remodeling: findings from the community‐based Framingham heart study**. *PLoS One* (2020) **15**. PMID: 33301464
23. Ballo P, Cameli M, Mondillo S. **Impact of diabetes and hypertension on left ventricular longitudinal systolic function**. *Diabetes Res Clin Pract* (2010) **90** 209-215. PMID: 20822822
24. Mehta S, Khoury PR, Madsen NL, Dolan LM, Kimball TR, Urbina EM. **Arterial thickness and stiffness are independently associated with left ventricular strain**. *J Am Soc Echocardiogr* (2018) **31** 99-104. PMID: 29174337
25. Nakanishi K, Daimon M, Yoshida Y. **Carotid intima‐media thickness and subclinical left heart dysfunction in the general population**. *Atherosclerosis* (2020) **305** 42-49. PMID: 32629183
26. Yu W, Zha W, Ren J. **Exendin‐4 and Liraglutide attenuate glucose toxicity‐induced cardiac injury through mTOR/ULK1‐dependent autophagy**. *Oxid Med Cell Longev* (2018) **2018** 5396806-5396814. PMID: 29849901
27. Buse JB, Ginsberg HN, Bakris GL. **Primary prevention of cardiovascular diseases in people with diabetes mellitus: a scientific statement from the American Heart Association and the American Diabetes Association**. *Circulation* (2007) **115** 114-126. PMID: 17192512
28. Silverii GA, Toncelli L, Casatori L. **Assessment of left ventricular global longitudinal strain in patients with type 2 diabetes: relationship with microvascular damage and glycemic control**. *Nutr Metab Cardiovasc Dis* (2022) **32** 994-1000. PMID: 35168822
29. Karagöz A, Bezgin T, Kutlutürk I. **Subclinical left ventricular systolic dysfunction in diabetic patients and its association with retinopathy: a 2D speckle tracking echocardiography study**. *Herz* (2015) **40** 240-246
30. Pararajasingam G, Heinsen LJ, Larsson J. **Diabetic microvascular complications are associated with reduced global longitudinal strain independent of atherosclerotic coronary artery disease in asymptomatic patients with diabetes mellitus: a cross‐sectional study**. *BMC Cardiovasc Disord* (2021) **21** 269. PMID: 34078282
31. Zhen Z, Chen Y, Shih K. **Altered myocardial response in patients with diabetic retinopathy: an exercise echocardiography study**. *Cardiovasc Diabetol* (2015) **14** 123. PMID: 26382215
32. Roy S, Kant R, Kumar B, Khapre M, Bairwa M. **Systolic dysfunction in asymptomatic type 2 diabetic patients, a harbinger of microvascular complications: a cross‐sectional study from North India**. *Diab Vasc Dis Res* (2020) **17** 1479164120944134. PMID: 32713196
33. Scurt FG, Menne J, Brandt S. **Systemic inflammation precedes microalbuminuria in diabetes**. *Kidney Int Rep* (2019) **4** 1373-1386. PMID: 31701047
34. Potier L, Chequer R, Roussel R. **Relationship between cardiac microvascular dysfunction measured with 82Rubidium‐PET and albuminuria in patients with diabetes mellitus**. *Cardiovasc Diabetol* (2018) **17** 11. PMID: 29325551
35. Marcovecchio ML, Chiesa ST, Armitage J. **Renal and cardiovascular risk according to tertiles of urinary albumin‐to‐creatinine ratio: the adolescent type 1 diabetes cardio‐renal intervention trial (AdDIT)**. *Diabetes Care* (2018) **41** 1963-1969. PMID: 30026334
36. Mochizuki Y, Tanaka H, Matsumoto K. **Clinical features of subclinical left ventricular systolic dysfunction in patients with diabetes mellitus**. *Cardiovasc Diabetol* (2015) **14** 37. PMID: 25889250
37. Hu L, Zha YF, Wang L. **Quantitative evaluation of vertebral microvascular permeability and fat fraction in Alloxan‐induced diabetic rabbits**. *Radiology* (2018) **287** 128-136. PMID: 29156149
38. Leggat J, Bidault G, Vidal‐Puig A. **Lipotoxicity: a driver of heart failure with preserved ejection fraction?**. *Clin Sci* (2021) **135** 2265-2283
39. Nakamura M, Sadoshima J. **Cardiomyopathy in obesity, insulin resistance and diabetes**. *J Physiol* (2020) **598** 2977-2993. PMID: 30869158
40. Reiner Ž. **Hypertriglyceridaemia and risk of coronary artery disease**. *Nat Rev Cardiol* (2017) **14** 401-411. PMID: 28300080
41. Ng AC, Delgado V, Bertini M. **Myocardial steatosis and biventricular strain and strain rate imaging in patients with type 2 diabetes mellitus**. *Circulation* (2010) **122** 2538-2544. PMID: 21126971
42. Jędrzejewska I, Król W, Światowiec A. **Left and right ventricular systolic function impairment in type 1 diabetic young adults assessed by 2D speckle tracking echocardiography**. *Eur Heart J Cardiovasc Imaging* (2016) **17** 438-446. PMID: 26160403
43. Vinereanu D, Nicolaides E, Tweddel AC. **Subclinical left ventricular dysfunction in asymptomatic patients with type II diabetes mellitus, related to serum lipids and glycated haemoglobin**. *Clin Sci* (2003) **105** 591-599
44. Tseng CH, Tseng CP, Chong CK, Cheng JC, Tai TY. **Independent association between triglycerides and coronary artery disease in Taiwanese type 2 diabetic patients**. *Int J Cardiol* (2006) **111** 80-85. PMID: 16213608
45. Tseng CH. **Body mass index and waist circumference as determinants of coronary artery disease in Taiwanese adults with type 2 diabetes mellitus**. *Int J Obes (Lond)* (2006) **30** 816-821. PMID: 16418754
46. Chen SC, Tseng CH. **Dyslipidemia, kidney disease, and cardiovascular disease in diabetic patients**. *Rev Diabet Stud* (2013) **10** 88-100. PMID: 24380085
47. Tseng CH. **Exogenous insulin use and hypertension in adult patients with type 2 diabetes mellitus**. *Arch Intern Med* (2006) **166** 1184-1189. PMID: 16772245
48. Tseng CH. **Metformin use is associated with a lower risk of hospitalization for heart failure in patients with type 2 diabetes mellitus: a retrospective cohort analysis**. *J Am Heart Assoc* (2019) **8**. PMID: 31630591
|
---
title: 'Does the toxicity of endocrine therapy persist into long-term survivorship?:
Patient-reported outcome results from a follow-up study beyond a 10-year-survival'
authors:
- Albertini Carmen
- Oberguggenberger Anne
- Sztankay Monika
- Egle Daniel
- Giesinger Johannes
- Meraner Verena
- Hubalek Michael
- Brunner Christine
journal: Breast Cancer Research and Treatment
year: 2022
pmcid: PMC10036266
doi: 10.1007/s10549-022-06808-9
license: CC BY 4.0
---
# Does the toxicity of endocrine therapy persist into long-term survivorship?: Patient-reported outcome results from a follow-up study beyond a 10-year-survival
## Abstract
### Background
Endocrine treatment (ET) is a highly effective breast cancer treatment but can distinctly impair breast cancer patients’ quality of life (QOL). In a patient-reported outcome (PROs) study conducted by the authors in 2011, patients reported higher ET-induced symptom levels than known from the registration trials, and was underestimated. Based on these study results, we investigated the long-term sequelae of ET reported by breast cancer survivors (BCS) in a follow-up study conducted 5–10 years after an earlier assessment.
### Methods
BCS who had participated in the earlier study ($$n = 436$$) were approached for study participation either at one of their routine follow-up appointments or via mail; consenting patients were asked to completed the same PRO assessment used in the original study (FACT-B + ES). BCS with relapse/ progressive disease were excluded from the analysis. We compared long-term endocrine symptomatology and overall QOL outcome (i.e. FACT-G and -ES sum score).
### Results
A final sample of 268 BCS was included in the analysis. BCS reported a significant improvement of the overall endocrine symptomatology (baseline mean = 59 vs. follow-up mean = 62, $p \leq 0.001$), physical (baseline = 23.9 mean vs. follow-up mean = 24.8, $p \leq 0.01$) and functional well-being (baseline mean = 21.7 vs. follow-up mean = 22.7, $$p \leq 0.013$$) and overall QOL (mean baseline = 88.3 vs. mean follow-up = 90.9, $$p \leq 0.011$$). However, the prevalence of particular symptoms, well-known to be ET induced, did not change over time such as joint pain (baseline = $45.5\%$ vs. $44.2\%$, n.s. difference), lack of energy ($36.4\%$ vs $33.8\%$, n.s. difference), weight gain ($36.8\%$ vs. $33.9\%$, n.s. difference) or vaginal dryness ($30.2\%$ vs. $33\%$, n.s. difference) and the proportion reporting lack of interest in sex increased ($40.4\%$ vs. $48.7\%$, $p \leq 0.05$).
### Conclusion
Presented results indicate that BCS recover well in terms of overall endocrine symptomatology and quality of life but experience some clinically relevant and unfavorable ET-related long-term effects.
## Introduction
With constantly increasing survival rates over the last decade, the group of long-term breast cancer survivors (i.e. permanent survivorship according to ASCO, www.cancer.net) has been expanding. Personalized treatments such as endocrine therapy (ET) applied for multiple years after initial treatment make a distinct contribution to these increased survival rates. More than $75\%$ of women diagnosed with breast cancer would receive at least 5 years of ET as part of their treatment [1, 2]. Though increasing survival, women are undergoing these highly effective treatments at the cost of an (potential) enduring impairment of their quality of life (QOL) [3, 4]. Hot flashes, joint pain, sexual problems or emotional instability are among the most prevalent ET side effects challenging patients’ QOL [5, 6]. Some evidence claims these ET treatment side effects and QOL impairments to occur not only during treatment but to persist after treatment completion far into the survivorship stage [7–9]. Hence, survivorship issues such as QOL including not only physical but also psychosocial recovery in the long-term gain importance when it comes to comprehensive survivorship care [10–13]. An essential step in this regard is the systematic identification of ET long-term sequelae most detracting to breast cancer survivors (BCS) including the patient’s subjective experience. For this purpose, patient-reported outcomes (PROs) have been proven to give comprehensive insight into the patient’s physical and psychosocial health complementing provider-generated information [14]. In a study called PRO-BETh (PROs in Breast cancer patients undergoing ET), performed 2009–2011, the authors were able to demonstrate the value of PROs for the understanding of ET treatment toxicity [14, 15]. *Evidence* generated by this study suggested high rates of ET-induced toxicity for both, pre- and postmenopausal women. The prevalence of most side effects observed in this “real-life” study (i.e. a sample within routine after-care) significantly exceeded those reported by the original registration trials [16–18]. Joint pain, hot flashes, loss of interest in sex and lack of energy were the most prevalent symptoms reported by patients. In order to gain more insight into the long-term sequelae of ET, the authors conducted a follow-up study to the research project PRO-BETh.
The main aim of this follow-up study was the determination of patient-reported ET-associated toxicity and QOL outcomes in BCS 5–10 years after the initial assessment.
## PRO-BETh study description
The original PRO-BETh study [14, 15] was designed as a cross-sectional observation study targeting on the assessment of prevalence and severity of ET-induced side effects from a subjective patient perspective. For this purpose, BC patients undergoing up-front ET with either AIs or tamoxifen (with or without Zoladex) at the time of assessment completed a comprehensive PRO-battery on QOL including physical side effects and psychosocial burden. Reported symptom prevalence rates were compared to data derived from pivotal phase IV trials (ATAC 2005, BIG1-98 [16]. Overall, PROs resulted in significantly higher prevalence rates as compared to physician ratings for most symptoms published in pivotal clinical trials. The authors concluded that ET toxicity seems to be underestimated in clinical routine care. Please find further study details in the respective publications [11, 12].
## Sample
All BC patients who had participated in the original study were eligible and approached for participation in the follow-up assessment. Contact data were taken from the medical records of the Department of Gynecology and Obstetrics at the Medical University of Innsbruck. Inclusion criteria for this study were defined as followed:Participation in the initial PRO-BETh studyBreast cancer survivor having undergone endocrine treatment - defined as patient who had completed the primary treatment (maintenance treatment can be ongoing) by the EORTC Cancer Survivorship Task Force [19]No overt cognitive impairmentWritten informed consentFluency in German From the 436 patients on endocrine treatment originally surveyed in 2009–11, 27 patients ($6.2\%$) were deceased, this corresponds to an OS of $93\%$. A total of 290 breast cancer long-term survivors participated in the follow-up study. The remaining 119 patients did either not agree to fill out questionnaires because of personal reasons ($11.9\%$) or could not be contacted due to logistic reasons ($15.4\%$). Hence, a response rate of $70\%$ could be achieved. Among the patients in the final analysis, a total of $8\%$ reported a relapse ($3.4\%$ in the AI group and $4.5\%$ in the tamoxifen group). We excluded those patients from the further analysis to provide group homogeneity. Hence, we report data of a final sample of 268 BCS. Please find details in the flow chart below (Fig. 1).Fig. 1Flow diagram of the inclusion/exclusion of the study participants and response rate Patients participated after a median follow-up period of 8 years (range 6–9 years; mean = 8.02). At the time of the follow-up assessment, patients were aged 65 years on average and $90\%$ were postmenopausal. Patients who had received tamoxifen were significantly younger than patients with AI therapy ($p \leq 0.001$) as tamoxifen has been the first-line ET for premenopausal patients at the time the original study was performed (and AIs for postmenopausal patients). Details on clinical data are presented in Table 1.Table 1Clinical patient data ($$n = 268$$) at follow-upPatients with AI treatment $$n = 159$$Patients with tamoxifen treatment $$n = 109$$All BCS $$n = 268$$Frequency (%)Frequency (%)Frequency (%)Age at the follow-upMean (SD)72.8 (SD 7.2) years#54.5 (SD 5.3) years#65.3 (SD 11) yearsRange56–9436–68(36–94 years)Time since diagnosisMedian (SD)10.4 (SD 2) years9.2 (SD 1.3) years10 (SD 1.9) yearsMean10.9 years9.4 years10.3 (SD 1.9) yearsRange7–17 years7–13 years7–17.6 yearsTime between assessmentsMedian (SD)8.2 (SD 0.5) years7.9 (0.6) years8 (SD 0.6) yearsMean8.2 years7.8 years8.0 yearsRange6.9–9.0 years6.4–9.10 years6.4–9.1 yearsTime between ET termination and follow-upMedian (SD)4.9 (SD 2) years3.8 (SD 1.8) years4.5 (SD 2) yearsMean4.9 years3.9 years4.4 (SD 2) yearsRangeongoing -9 yearsongoing—8 yearsongoing—9 yearsMenopausal statePremenopausal–27 ($25\%$)27 ($10\%$)postmenopausal159 ($100\%$)81 ($75\%$)240 ($90\%$)Overall duration of adjuvant endocrine therapy (months)Mean (SD)5.8 (SD 1.9) years5.3 (SD 1.7) years5.5 (1.8) yearsRange2.7–15 years1.7–13 years1.7–15 yearsET intake (treatment duration groups) < 5 years10 ($6.4\%$)16 ($14.8\%$)26 ($10\%$)Regular 5 years Extended110 ($70.1\%$)76 ($70.4\%$)186 ($70\%$)37 ($23.5\%$)16 ($14.8\%$)53 ($20\%$)ET ongoing at follow-upyes7 ($4.4\%$)8 ($7.3\%$)15 ($5.6\%$)BMIMean25.625.625.7Range16.9–46.916.9–46.918.2–40.8#Significant difference between patients treated with tamoxifen vs. AIs regarding age ($p \leq 0.001$)
## Procedure
Following the recruitment procedure of the original project, the data assessment was conducted at the outpatient unit of the Department of Gynecology and Obstetrics at the Medical University of Innsbruck.
Breast cancer survivors (BCS) were approached for study participation either at one of their routine follow-up appointments (in Austria, BC patients have lifelong routine check-ups at the primary care center) or via mail after an introductory telephone call explaining the study purpose. Patients completed written informed consent. In case of consenting to study participation, BCS completed the same PRO assessment used in the original study (see below). Patients returned the questionnaires pseudo-anonymized (ID indicated by a number) in an envelope either via mail or personally at the outpatient unit (paper–pencil assessment). Clinical data for participants were derived from the medical records.
## PRO instruments
The original questionnaire battery included the Functional assessment of cancer therapy-breast (FACT-B) and Functional assessment of cancer therapy-endocrine subscale (Fact-ES). The Functional Assessment of Cancer Therapy-Breast and Endocrine Subscale (FACT-B + -ES) consists of 36 items assessing QOL in BC patients. The questionnaire uses a five-point Likert scale and relates to the FIM framework for the past seven days. The answer format ranges from 0 (not at all) to 4 (very much). The maximum scoring for general well-being ranges from 0 to 108, for emotional well-being from 0 to 24 and for physical and functional well-being from 0 to 28. High values indicate a good QOL. The FACT-B is supplemented by the endocrine subscale (FACT-ES), which measures symptoms and side effects related to ET for breast cancer such as hot flashes, joint pain and loss of libido [3]. The FACT-ES comprises 19 items. Further details have been published elsewhere [14, 15].
## Statistical analysis
Sample characteristics are described using absolute and relative percentages, means and standard deviations.
Primary analysis: In order to investigate long-term ET toxicity, we analyzed the FACT-B + -ES on single item level following the analysis of the original study. i.e. we compared FACT-B + -ES data of each patient from the first assessment to her data at the follow-up assessment. We present the prevalence of patient-reported physical and psychological symptoms related to ET (derived from the FACT-B + -ES) as percentages and $95\%$ confidence intervals for baseline and follow-up time points. Symptom frequency was calculated by summarizing percentages of patients selecting the categories 'somewhat', 'quite a bit' and 'very much' on single item level of the FACT-B + -ES. Confidence intervals were calculated using the modified Wald method [20]. The Sign Test was used to compare symptom frequencies between the two assessment-time points. We further aimed at the clarification of the impact of age on symptoms. For this purpose, age was considered a relevant covariate already at the first assessment with a continuous effect on the outcome. We hence were interested in the impact of age on symptom change over time rather than assessing its effect at the follow-up assessment only. For this purpose, we calculated the difference between the first- and follow-up assessment for the FACT-B + -ES items and compared age groups (< 50, 50–59, 60–69 and > 70 years) for this difference using the Kruskal–Wallis Test.
Secondary analysis: For the investigation of overall long-term QOL outcome (i.e. FACT-G and -ES sum score), we used a mixed linear model. In this analysis, the dependent variables were log-transformed to obtain normal distribution. Time point (first assessment, follow-up) was included as a fixed effect and post-hoc, we conducted pairwise comparisons between time points and tamoxifen- vs. aromatase inhibitor treatment (with Bonferroni-correction for multiplicity). To assess the association of age with change over time we included the two-way interaction age-by-time point in the model. To account for correlations between repeated measurements, we used a first order autoregressive covariance matrix. P-values below 0.05 were considered statistically significant. All analyses were done with SPSS 22.0.
We obtained ethical approval for this follow-up project from the Ethics Committee of the Medical University of Innsbruck (Innsbruck, 22.04.2017/Ah).
## Changes of ET-related toxicity
We observed a significant improvement of the overall endocrine symptomatology in the long-term (FACT-ES baseline mean = 59 vs. follow-up mean = 62, $p \leq 0.001$); this significant improvement was found for both, patients who had received tamoxifen (FACT-ES baseline mean = 58.4 vs. follow-up mean = 61.2) as well as those with previous AI treatment (FACT-ES baseline mean = 59.7 vs. follow-up mean = 62.7).
In detail, vasomotor symptoms including hot flashes and cold/night sweats decreased significantly with time in both groups. In contrast, gynecologic symptoms did not change over time except for vaginal discharge, which decreased significantly; loss of interest in sex even increased in long-term in percentages. Interesting to note, of the overall sample $9.3\%$ did not complete both questions on sexuality and $20.7\%$ answered only one of both questions (i.e. pain with intercourse and loss of interest in sex) at the second assessment.
No significant difference was observed for gastrointestinal symptomatology ($p \leq 0.05$ for all gastrointestinal symptoms); a total of $38.4\%$ and $30.4\%$ of patients reported weight problems in the tamoxifen and AI group, respectively, at follow-up. Finally, the typically ET-related symptoms joint pain, lack of energy and mood swings were highly prevalent at the follow-up assessment time point. Details are presented in Table 2, 3, 4.Table 2Prevalence physical and psychological symptoms in BCS (entire sample)All BCS $$n = 2681$$st assessmentFollow-upDifference in %Median time since diagnosis1.9 (SD 1.7) years10.0 (SD 1.9) years% (CI$95\%$)% (CI$95\%$)Vasomotor symptoms Hot flashes$68.7\%$ (62–74)$39.5\%$ (33–46)*− 29.2 Cold sweats$30.0\%$ (24–36)$16.3\%$ (12–21)*− 13.7 Night sweats$47.0\%$ (40–53)$24.3\%$ (19–30)*− 22.7 Sleeping difficulties$14.9\%$ (11–20)$9.5\%$ (6–14)#− 5.4Gynecologic symptoms Vaginal discharge$12.4\%$ (9–17)$4.6\%$ (2–8)*− 7.8 Bleeding or spotting$1.6\%$ (0.5–4)$1.4\%$ (0.3–4)− 0.2 Vaginal itching/irritation$9.8\%$ (6–14)$10.3\%$ (7–14) + 0.5 Vaginal dryness$30.2\%$ (25–36)$33.0\%$ (27–39) + 2.8 Breast sensitivity/tenderness$24.4\%$ (19–30)$19.1\%$ (14–24)− 5.3 Pain or discomfort with intercourse$16.5\%$ (12–22)$20.5\%$ (15–27) + 4 Lost interest in sex$40.4\%$ (34–46)$48.7\%$ (42–55)* + 8.3Gastrointestinal symptoms Weight gain$36.8\%$ (31–43)$33.9\%$ (28–40)− 2.9 Emesis$1.2\%$ (0–3)$0.9\%$ (0–2)− 0.3 Diarrhoea$5.2\%$ (3–9)$5.8\%$ (3–10) + 0.6 Feeling bloated$14.7\%$ (11–19)$9.5\%$ (6–14)− 5.2 Nausea$4.5\%$ (2–8)$3.6\%$ (2–7)− 0.9Pain Headaches$16.4\%$ (12–21)$14.7\%$ (10–20)− 1,7 Joint pain$45.5\%$ (40–51)$44.2\%$ (38–50)− 1.3Psychological symptoms Feeling lightheaded (dizziness)$13.9\%$ (10–19)$13.9\%$ (10–19)– Mood swings$36.8\%$ (31–43)$30.2\%$ (24–36)*− 6.6 Being irritable$30.6\%$ (25–36)$27.5\%$ (22–33)− 3.1 Lack of energy$36.4\%$ (30–42)$33.8\%$ (28–40)− 2.6*Significant difference on a $p \leq 0.01$ (based on the Sign Test)#Significant difference on a $p \leq 0.05$ (based on the Sign Test)Table 3Prevalence physical and psychological symptoms in BCS having received ET with tamoxifenPatients with TAM treatment $$n = 1591$$st assessmentFollow-upDifference in %Median time since diagnosis1.2 (SD 1.0) years1.2 (SD 1.0) years % (Cl$95\%$) % (Cl$95\%$)Vasomotor symptoms Hot flashes$83.8\%$ (75–89)$50.0\%$ (40–59)*− 33.8 Cold sweats$39.6\%$ (30–49)$21.6\%$ (14–30)*− 18.0 Night sweats$60.6\%$ (51–69)$29.9\%$ (21–39)*− 30.7 Sleeping difficulties$14.2\%$ (8–22)$5.9\%$ (2–12)*− 8.3Gynecologic symptoms Vaginal discharge$25.0\%$ (17–34)$5.2\%$ (2–12)*− 19.8 Bleeding or spotting$3.8\%$ (1–9)$2.1\%$ (0–7)− 1.7 Vaginal itching/ irritation$14.3\%$ (8–22)$13.3\%$ (7–21)− 1 Vaginal dryness$21.9\%$ (15–30)$35.1\%$ (26–45)# + 13.2 Breast sensitivity/ tenderness$21\%$ (14–29)$19.4\%$ (13–28)− 0.6 Pain or discomfort with intercourse$14.4\%$ (8–22)$21.5\%$ (14–31) + 7.1 Lost interest in sex$21.4\%$ (14–30)$34.7\%$ (26–44)# + 13.3Gastrointestinal symptoms Weight gain$37.5\%$ (29–47)$38.4\%$ (29–48) + $0.9\%$ Emesis$1.0\%$ (0–5)$0.0\%$ [0]− 1 Diarrhoea$2.9\%$ (0.6–8)$3.0\%$ (0.6–8)− 0.1 Feeling bloated$14.3\%$ (8–22)$10.2\%$ (5–18)− 4.1 Nausea$3.9\%$ (1–10)$2.1\%$ (0–7)− 1.8Pain Headaches$13.3\%$ (8–21)$12.1\%$ (7–20)− 1.2 Joint pain$30.5\%$ (22–40)$35.1\%$ (26–45) + 4.6Psychological symptoms Feeling lightheaded (dizziness)$8.6\%$ (4–15)$6.1\%$ (3–19)− 2.5 Mood swings$38.2\%$ (30–48)$34.0\%$ (25–44)− 4.2 Being irritable$34.6\%$ (26–44)$25.8\%$ (18–35)− 8.8 Lack of energy$32.7\%$ (24–42)$27.5\%$ (20–37)− 5.2*Significant difference on a $p \leq 0.01$ (based on the Sign Test)#Significant difference on a $p \leq 0.05$ (based on the Sign Test)Table 4Prevalence physical and psychological symptoms in BCS having received ET with aromatase inhibitorsPatients with AI treatment $$n = 109$$BaselineFollow-upDifference in %Median time since diagnosis2.2 (SD 1.9) years10.4 (SD 2) years% (CI$95\%$)% (CI$95\%$)Vasomotor symptoms Hot flashes$57.8\%$ (49–65)$31.2\%$ (23–39)*− 26.6 Cold sweats$23.2.\%$ (17–30)$12.1\%$ (7–19)*− 11.1 Night sweats$37.4\%$ (30–45)$19.8\%$ (13–27)*− 17.4 Sleeping difficulties$15.4\%$ (10–22)$12.4\%$ (7–19)− $3\%$Gynecologic symptoms Vaginal discharge$3.4\%$ (1–8)$4.1\%$ (1–9) + 0.7 Bleeding or spotting$0.0\%$ [0]$0.8\%$ (0–5) + 0.8 Vaginal itching/ irritation$7.5\%$ (4–13)$7.1\%$ (3–13)− 0.4 Vaginal dryness$36.1\%$ (28–44)$31.5\%$ (24–40)− 4.6 Breast sensitivity/ tenderness$26.9\%$ (20–34)$18.9\%$ (12–27)#− 8 Pain or discomfort with intercourse$18.3\%$ (12–26)$19.6\%$ (13–29) + 1.3 Lost interest in sex$55.3\%$ (46–63)$61.5\%$ (52–70) + 6.2Gastrointestinal symptoms Weight gain$36.3\%$ (29–44)$30.4\%$ (23–39)− 5.9 Emesis$1.4\%$ (0–5)$1.6\%$ (0–6) + 0.2 Diarrhoea$6.9\%$ (3–12)$8.0\%$ (4–14) + 1.1 Feeling bloated$15.1\%$ (10–21)$9.0\%$ (5–15)− 6.1 Nausea$4.9\%$ (2–10)$4.6\%$ (2–9)− 0.3Pain Headaches$18.6\%$ (13–25)$16.7\%$ (11–24)− 1.9 Joint pain$56\%$ (48–63)$51.2\%$ (42–60)− 4.8Psychological symptoms Feeling lightheaded (dizziness)$17.7\%$ (12–25)$20.0\%$ (14–28) + 2.3 Mood swings$35.8\%$ (28–43)$27.2\%$ (20–35)− 8.6 Being irritable$27.7\%$ (21–35)$28.8\%$ (21–37) + 1.1 Lack of energy$38.9\%$ (31–47)$38.5\%$ (30–47)− 0.4*Significant difference on a $p \leq 0.01$ (based on the Sign Test)#Significant difference on a $p \leq 0.05$ (based on the Sign Test) Regarding the effect of age—independent from the treatment received—on symptom change over time, we found no difference for most symptoms across age groups (results not shown) with the exception of vaginal discharge ($p \leq 0.001$), headaches ($$p \leq 0.023$$) and mood swings ($p \leq 0.001$) (details in Table 5).Table 5Mean symptom difference from baseline to follow-up across age groups (significant results $p \leq 0.05$)Symptoms age at follow-upvaginal discharge mean difference (SD)Headaches mean difference (SD)mood swings mean difference (SD) < 50 years− 1 (1.13)− 0.3 (0.9)− 0.9 (1.6)50–59 years− 0.4 (0.9)0.12 (0.9)− 0.014 (1.6)60–69 years− 0.23 (0.7)− 0.35 (0.9)0.54 (1.7) > = 70 years0.03 (0.5)− 0.04 (0.9)1 (1.5)
## QOL outcome
Overall, QOL according to the FACT-global score was significantly higher in long-term BCS compared to QOL in patients on ET treatment (mean baseline = 88.3 vs. mean follow-up = 90.9, $$p \leq 0.011$$). This was true for patients who had received tamoxifen (mean baseline = 89.3 vs. mean follow-up = 92.8) as well as those with previous AI treatment (mean baseline = 87.5 vs. mean follow-up = 89.5). However, in terms of clinical relevance the improvement seems to be minor [21].
BCS reported significantly higher levels of physical well-being (FACT-physical well-being baseline = 23.9 mean vs. follow-up mean = 24.8, $p \leq 0.01$) and functional well-being (FACT-functional well-being baseline mean = 21.7 vs. follow-up mean = 22.7, $$p \leq 0.013$$) than patients on ET treatment. For functional well-being, we observed a trend towards a higher increase in patients who had received tamoxifen, i.e. in the younger patient group (interaction effect $$p \leq 0.079$$) compared to patients in the AI-group.
No changes were observed for emotional well-being (FACT-emotional well-being baseline = 19.7 mean vs. follow-up mean = 20, $p \leq 0.05$) and social well-being (FACT-emotional well-being baseline mean = 22.5 vs. follow-up mean = 22.5, $p \leq 0.05$). This was true for both ET groups.
For all QOL scales, age had no significant impact on symptom change over time (results not shown).
## Discussion
While previous evidence suggests that ET-associated toxicity is high and distinctly impairs patient QOL during intake [14, 22] we lack evidence on the patients’ experience of these symptoms in the long run, in particular after ET termination. In this paper, we aimed at shedding light to the physical and psychosocial long-term outcome after ET in BCS from a patient perspective.
Overall, BCS experienced a decrease of the overall ET-related symptomatology in the long-term (as indicated by the overall score of the FACT-ES subscale). In particular, the vasomotor symptomatology—a major side effect of ET—seems to decrease over time significantly. In addition, patients reported a small but significant increase of the overall physical and functional well-being score as well as their general QOL over time. This observation is consistent with the results of Schmitt et al. demonstrating improvement of physical- and role-functioning in-between 5 years after the end of cancer treatment and even exceeding levels of an aged-matched healthy population [23, 24]. Others suggest levels of overall QOL in BCS to be comparable to those of a population without a previous cancer disease [7, 8]. We might hence conclude that there is some sort of stabilization of an overall symptomatology and QOL in BCS.
Consequently, we observed a lack of recovery when it comes to specific ET-related symptoms. Several symptoms seem to persist at high levels: Joint pain, loss of interest in sex or weight gain as well as lack of energy have been indicated as the most prevalent long-term follow-up problems for BCS in our study. Our results complement existing evidence illustrating specific cancer treatment-related symptoms to challenge patients in long-term. For instance, Haidinger [9] and others [25] observed high levels of joint pain in BCS after ET termination. Van Leuuwen [2018] identified joint pain among the chronic symptoms highly relevant and burdensome for cancer survivors when asking patients to quote QOL topics relevant for their cancer survivorship [19]. Evidence for the persistence of fatigue and lack of energy as among the most prevalent long-term sequels of a cancer disease is robust [7]. Weight gain is a well-known problem related to ET [26]. Particularly, patients receiving tamoxifen (i.e. younger patients) continue to struggle with their weight over years [26]. In the study presented herein, more than one third of both, AI-patients and TAM-patients, reported persistent problems with weight gain. Potentially resulting in obesity, weight gain is not only a problem for the subjective overall well-being, body image or feeling of attractiveness but mediates disease control and clinical outcome [27].
Moreover, this study again proves the impact of ET on sexual health in BCS. We observed not only a lack of recovery of interest in sexuality in long-term but even a tendency towards symptom deterioration in the “younger” (originally premenopausal) patient group. The same was true for vaginal dryness in premenopausal patients. This is in line with two recent meta-analyses highlighting a high prevalence of female sexual dysfunction in BCS [28, 29]. The authors reported recently that about $70\%$ of BCS complain clinically relevant sexual dysfunction [30]. In addition, $10\%$ of the participants had not answered the two questions about sexuality and $20\%$ only answered one of the two questions. This observation supports the notion that the topic of sexuality is still a taboo in clinical care patients are reluctant to talk about [31–33]. Sexuality is a complex issue influenced by numerous bio-psycho-social factors and underlies natural changes over the life span: For instance, age or menopausal status are well-known to affect interest in sexuality or libido [34, 35]. In this study, we were not able to clearly isolate an independent effect of factors potentially contributing to the patients’ sexual outcome as we lack a baseline assessment of QOL before the start of ET. However, in view of more than $50\%$ of BCS on- and off- treatment indicating sexual impairments, sexuality should be considered as a major, persistent care demand relevant for BC (survivorship) care.
With regard to the psychological domain, BCS reported no change over time. Established evidence supports the notion that psychological issues continue to be high far into the survivorship stage while patients recover physically [9]. Mood swings – though decreasing over time – can be a continuing problem at least for about a third of the younger patients as observed in this study. Other studies [7, 9] also describe comparable results with regard to emotional well-being. In particular, young BC patients need to meet the challenges of a cancer diagnosis in the middle of their work life, educational stage or during the phases of family planning; life plans need to be postponed or can finally not be achieved, thereby, requiring (psychological) adaptation to new requirements and are psychologically challenging. ( Irreversible) hormonal changes can—as induced by ET—might aggravate the emotional challenges; the latter was not investigated in this study but will be the topic of further studies.
In conclusion, a distinct proportion of BCS experiences a chronification of specific symptoms after treatment completion. These health impairments can significantly interfere with a management of daily life independently from others thereby having a profound impact on patient QOL. The persistence of some ET-related physical and emotional symptoms should not be underestimated—this is true also beyond the actual intake of ET.
## Limitations
A limiting factor for the interpretation of study results is the lack of knowledge of QOL outcome in the non-participant group. Though we have a very satisfactory response rate of $70\%$, a potential selection bias towards a worse OR better QOL cannot be excluded. Furthermore, a direct comparison of data with an age-matched population without a cancer disease would have enriched the interpretability of data on QOL outcome. Data from such a reference sample could help identify other factors modulating QOL over a span of almost a decade beside the cancer diagnosis and related treatments e.g. natural menopause and age are well-known to have an effect on interest in sex or mood swings independently from cancer. The investigation of the independent effect of age on QOL outcome was further limited in the presented analysis due to the following: The type of ET prescribed originally had been based on patients’ menopausal state (i.e. tamoxifen for premenopausal patients and aromatase inhibitors for postmenopausal patients), so that age is an immanent factor related to the type of treatment (i.e. a high inter-correlation of the covariates age and type of treatment). An independent effect of age is therefore difficult to obtain. Finally, the sample heterogeneity in terms of treatment duration and time since ET termination to the follow-up assessment limits the interpretation of results. Clearly, a longitudinal design with a baseline assessment before the start of ET and a more homogenous sample at the first and second assessment would have contributed to a more accurate picture of the true extent of long-term toxicity caused by adjuvant ET—this limitation from the original study persists to the follow-up assessment. However, presented results clearly indicate that BCS experience unfavorable long-term effects that need to be better understood and should be subject to further research.
## Clinical implication
Our results are of importance for clinical survivorship care: Women after ET seem to recover well overall when it comes to QOL issues. However, they still suffer from particular health impairments presenting a high potential for QOL limitations. Most persistent problems seem to be sexual health issues, psychological demands and joint pain. Survivorship care efforts should focus on these problems. This includes the provision of more information on long-term sequel of breast cancer and ET in patient education, a systematic assessment of the respective symptoms at after-care visits, and the integration of targeted, supportive treatment individually tailored to the BCSs’ demands in long-term care plans. This might also include the adjustment of ET treatment application towards individual demands. For instance, the SOLE study [36] proved an intermittent administration of Letrozole as a safe and advantageous option in terms of QOL. The option for a treatment interruption might help patients to stabilize their OOL and ultimately better adhere to the treatment regime. Beside the increase of survival, the prevention of long-term QOL problems should be an ultimate goal for BC survivorship care.
## Appendix
See Table Table 6Physical and psychological symptoms in BCS at the first and the follow-up assessmentAll BCS $$n = 268$$SymptomsSymptom persistencea (%)Symptom present at first but not at follow-up assessment (%)Symptom not present a first but at follow-up assessment (%)Total symptom prevalence (%)Vasomotor symptoms Hot flashes34.436.35.139.5 Cold sweats8.6237.716.3 Night sweats17.229.77.124.3 Sleeping difficulties5.89.849.5Gynecologic symptoms Vaginal discharge2.911.51.74.6 Bleeding or spotting1.41.401.4 Vaginal itching/irritation4.86.55.510.3 Vaginal dryness18.413.715.133.0 Breast sensitivity/tenderness16.18.510.419.1 Pain or discomfort with intercourse86.812.320.5 Lost interest in sex28.510.220.448.7Gastrointestinal symptoms Weight gain17.818.617.333.9 Emesis0.91.400.9 Diarrhoea0.54.25.15.8 Feeling bloated3.811.36.19.5 Nausea3.64.403.6Pain Headaches7.410.2714.7 Joint pain29.814.214.244.2Psychological symptoms Feeling lightheaded (dizziness)5.69.37.913.9 Mood swings19.917.59.530.2 Being irritable151511.727.5 Lack of energy18.81714.833.8aGroup of patients with the respective symptom at both assessment time points.bGroup of patients with the respective symptom at the first assessment only (not reported at the follow-up assessment).cGroup of patients with the respective symptom at the follow-up assessment only (not reported at the first assessment).6.
## References
1. Davies C, Godwin J, Gray R, Clarke M, Cutter D, Darby S, McGale P, Pan HC, Taylor C, Wang YC, Dowsett M, Ingle J, Peto R. **Relevance of breast cancer hormone receptors and other factors to the efficacy of adjuvant tamoxifen: patient-level meta-analysis of randomised trials**. *Lancet* (2011) **378** 771-784. DOI: 10.1016/S0140-6736(11)60993-8
2. **Effects of chemotherapy and hormonal therapy for early breast cancer on recurrence and 15-year survival: an overview of the randomised trials**. *Lancet* (2005) **365** 1687-1717. DOI: 10.1016/S0140-6736(05)66544-0
3. Fallowfield L, Cella D, Cuzick J, Francis S, Locker G, Howell A. **Quality of life of postmenopausal women in the Arimidex, Tamoxifen, Alone or in Combination (ATAC) Adjuvant Breast Cancer Trial**. *J Clin Oncol* (2004) **22** 4261-4271. DOI: 10.1200/JCO.2004.08.029
4. Knobf MT. **The influence of endocrine effects of adjuvant therapy on quality of life outcomes in younger breast cancer survivors**. *Oncologist* (2006) **11** 96-110. DOI: 10.1634/theoncologist.11-2-96
5. Winer EP, Hudis C, Burstein HJ, Wolff AC, Pritchard KI, Ingle JN, Chlebowski RT, Gelber R, Edge SB, Gralow J, Cobleigh MA, Mamounas EP, Goldstein LJ, Whelan TJ, Powles TJ, Bryant J, Perkins C, Perotti J, Braun S, Langer AS, Browman GP, Somerfield MR. **American Society of Clinical Oncology technology assessment on the use of aromatase inhibitors as adjuvant therapy for postmenopausal women with hormone receptor-positive breast cancer: status report 2004**. *J Clin Oncol* (2005) **23** 619-629. DOI: 10.1200/JCO.2005.09.121
6. Riggs BL, Hartmann LC. **Selective estrogen-receptor modulators – mechanisms of action and application to clinical practice**. *N Engl J Med* (2003) **348** 618-629. DOI: 10.1056/NEJMra022219
7. Doege D, Thong MS, Koch-Gallenkamp L, Bertram H, Eberle A, Holleczek B, Pritzkuleit R, Waldeyer-Sauerland M, Waldmann A, Zeissig SR, Jansen L, Brenner H, Arndt V. **Health-related quality of life in long-term disease-free breast cancer survivors versus female population controls in Germany**. *Breast Cancer Res Treat* (2019) **175** 499-510. DOI: 10.1007/s10549-019-05188-x
8. Koch L, Jansen L, Herrmann A, Stegmaier C, Holleczek B, Singer S, Brenner H, Arndt V. **Quality of life in long-term breast cancer survivors - a 10-year longitudinal population-based study**. *Acta Oncol* (2013) **52** 1119-1128. DOI: 10.3109/0284186X.2013.774461
9. Haidinger R, Bauerfeind I. **Long-term side effects of adjuvant therapy in primary breast cancer patients: results of a web-based survey**. *Breast Care (Basel)* (2019) **14** 111-116. DOI: 10.1159/000497233
10. Kantsiper M, McDonald EL, Geller G, Shockney L, Snyder C, Wolff AC. **Transitioning to breast cancer survivorship: perspectives of patients, cancer specialists, and primary care providers**. *J Gen Intern Med* (2009) **24** S459-466. DOI: 10.1007/s11606-009-1000-2
11. Rowland JH, Kent EE, Forsythe LP, Loge JH, Hjorth L, Glaser A, Mattioli V, Fossa SD. **Cancer survivorship research in Europe and the United States: where have we been, where are we going, and what can we learn from each other?**. *Cancer* (2013) **119** 2094-2108. DOI: 10.1002/cncr.28060
12. Bouskill K, Kramer M. **The impact of cancer and quality of life among long-term survivors of breast cancer in Austria**. *Support Care Cancer* (2016) **24** 4705-4712. DOI: 10.1007/s00520-016-3319-7
13. Mols F, Vingerhoets AJ, Coebergh JW, van de Poll-Franse LV. **Quality of life among long-term breast cancer survivors: a systematic review**. *Eur J Cancer* (2005) **41** 2613-2619. DOI: 10.1016/j.ejca.2005.05.017
14. Oberguggenberger A, Hubalek M, Sztankay M, Meraner V, Beer B, Oberacher H, Giesinger J, Kemmler G, Egle D, Gamper EM, Sperner-Unterweger B, Holzner B. **Is the toxicity of adjuvant aromatase inhibitor therapy underestimated? Complementary information from patient-reported outcomes (PROs)**. *Breast Cancer Res Treat* (2011) **128** 553-561. DOI: 10.1007/s10549-011-1378-5
15. Oberguggenberger A, Goebel G, Beer B, Oberacher H, Meraner V, Sztankay M, Sperner-Unterweger B, Zeimet AG, Marth C, Hubalek M, Holzner B. **Getting the whole picture: adding patient-reported outcomes to adjuvant endocrine treatment evaluation in premenopausal breast cancer patients**. *Breast J* (2014) **20** 555-557. DOI: 10.1111/tbj.12323
16. Thurlimann B, Keshaviah A, Coates AS, Mouridsen H, Mauriac L, Forbes JF, Paridaens R, Castiglione-Gertsch M, Gelber RD, Rabaglio M, Smith I, Wardley A, Price KN, Goldhirsch A. **A comparison of letrozole and tamoxifen in postmenopausal women with early breast cancer**. *N Engl J Med* (2005) **353** 2747-2757. DOI: 10.1056/NEJMoa052258
17. Howell A, Cuzick J, Baum M, Buzdar A, Dowsett M, Forbes JF, Hoctin-Boes G, Houghton J, Locker GY, Tobias JS. **Results of the ATAC (Arimidex, Tamoxifen, Alone or in Combination) trial after completion of 5 years' adjuvant treatment for breast cancer**. *Lancet* (2005) **365** 60-62. DOI: 10.1016/S0140-6736(04)17666-6
18. Cella D, Fallowfield L, Barker P, Cuzick J, Locker G, Howell A. **Quality of life of postmenopausal women in the ATAC ("Arimidex", tamoxifen, alone or in combination) trial after completion of 5 years' adjuvant treatment for early breast cancer**. *Breast Cancer Res Treat* (2006) **100** 273-284. DOI: 10.1007/s10549-006-9260-6
19. van Leeuwen M, Husson O, Alberti P, Arraras JI, Chinot OL, Costantini A, Darlington AS, Dirven L, Eichler M, Hammerlid EB, Holzner B, Johnson CD, Kontogianni M, Kjaer TK, Morag O, Nolte S, Nordin A, Pace A, Pinto M, Polz K, Ramage J, Reijneveld JC, Serpentini S, Tomaszewski KA, Vassiliou V, Verdonck-de Leeuw IM, Vistad I, Young TE, Aaronson NK, van de Poll-Franse LV, Eortc QLG. **Understanding the quality of life (QOL) issues in survivors of cancer: towards the development of an EORTC QOL cancer survivorship questionnaire**. *Health Qual Life Outcomes* (2018) **16** 114. DOI: 10.1186/s12955-018-0920-0
20. Vollset SE. **Confidence intervals for a binomial proportion**. *Stat Med* (1993) **12** 809-824. DOI: 10.1002/sim.4780120902
21. Anderson DJ, Seib C, McCarthy AL, Yates P, Porter-Steele J, McGuire A, Young L. **Facilitating lifestyle changes to manage menopausal symptoms in women with breast cancer: a randomized controlled pilot trial of the pink women's wellness program**. *Menopause* (2015) **22** 937-945. DOI: 10.1097/GME.0000000000000421
22. Ruhstaller T, von Moos R, Rufibach K, Ribi K, Glaus A, Spaeti B, Koeberle D, Mueller U, Hoefliger M, Hess D, Boehme C, Thuerlimann B. **Breast cancer patients on endocrine therapy reveal more symptoms when self-reporting than in pivotal trials: an outcome research study**. *Oncology* (2009) **76** 142-148. DOI: 10.1159/000195540
23. Schmidt ME, Wiskemann J, Steindorf K. **Quality of life, problems, and needs of disease-free breast cancer survivors 5 years after diagnosis**. *Qual Life Res* (2018) **27** 2077-2086. DOI: 10.1007/s11136-018-1866-8
24. Maurer T, Thone K, Obi N, Jung AY, Behrens S, Becher H, Chang-Claude J. **Health-related quality of life in a cohort of breast cancer survivors over more than 10 years post-diagnosis and in comparison to a control cohort**. *Cancers (Basel)* (2021). DOI: 10.3390/cancers13081854
25. Cella D. **Quality of life and clinical decisions in chemotherapy-induced anemia**. *Oncology (Williston Park)* (2006) **20** 25-28. PMID: 16925108
26. Raghavendra A, Sinha AK, Valle-Goffin J, Shen Y, Tripathy D, Barcenas CH. **Determinants of weight gain during adjuvant endocrine therapy and association of such weight gain with recurrence in long-term breast cancer survivors**. *Clin Breast Cancer* (2018) **18** e7-e13. DOI: 10.1016/j.clbc.2017.11.006
27. Ewertz M, Jensen MB, Gunnarsdottir KA, Hojris I, Jakobsen EH, Nielsen D, Stenbygaard LE, Tange UB, Cold S. **Effect of obesity on prognosis after early-stage breast cancer**. *J Clin Oncol* (2011) **29** 25-31. DOI: 10.1200/JCO.2010.29.7614
28. Jing L, Zhang C, Li W, Jin F, Wang A. **Incidence and severity of sexual dysfunction among women with breast cancer: a meta-analysis based on female sexual function index**. *Support Care Cancer* (2019) **27** 1171-1180. DOI: 10.1007/s00520-019-04667-7
29. Maiorino MI, Chiodini P, Bellastella G, Giugliano D, Esposito K. **Sexual dysfunction in women with cancer: a systematic review with meta-analysis of studies using the female sexual function index**. *Endocrine* (2016) **54** 329-341. DOI: 10.1007/s12020-015-0812-6
30. Oberguggenberger A, Martini C, Huber N, Fallowfield L, Hubalek M, Daniaux M, Sperner-Unterweger B, Holzner B, Sztankay M, Gamper E, Meraner V. **Self-reported sexual health: breast cancer survivors compared to women from the general population - an observational study**. *BMC Cancer* (2017) **17** 599. DOI: 10.1186/s12885-017-3580-2
31. Reese JB, Sorice KA, Zimmaro LA, Lepore SJ, Beach MC. **Communication about sexual health in breast cancer: what can we learn from patients' self-report and clinic dialogue?**. *Patient Educ Couns* (2020) **103** 1821-1829. DOI: 10.1016/j.pec.2020.03.029
32. Stabile C, Goldfarb S, Baser RE, Goldfrank DJ, Abu-Rustum NR, Barakat RR, Dickler MN, Carter J. **Sexual health needs and educational intervention preferences for women with cancer**. *Breast Cancer Res Treat* (2017) **165** 77-84. DOI: 10.1007/s10549-017-4305-6
33. Rasmusson EM, Plantin L, Elmerstig E. **'Did they think I would understand all that on my own?' A questionnaire study about sexuality with Swedish cancer patients**. *Eur J Cancer Care (Engl)* (2013) **22** 361-369. DOI: 10.1111/ecc.12039
34. Harder H, Starkings RML, Fallowfield LJ, Menon U, Jacobs IJ, Jenkins VA, trialists U. **Sexual functioning in 4,418 postmenopausal women participating in UKCTOCS: a qualitative free-text analysis**. *Menopause* (2019) **26** 1100-1009. DOI: 10.1097/GME.0000000000001377
35. da Silva Lara LA, Useche B, Rosa ESJC, Ferriani RA, Reis RM, de Sa MF, de Carvalho BR, Carvalho MA, de Sa RESAC. **Sexuality during the climacteric period**. *Maturitas* (2009) **62** 127-133. DOI: 10.1016/j.maturitas.2008.12.014
36. Jerusalem G, Farah S, Courtois A, Chirgwin J, Aebi S, Karlsson P, Neven P, Hitre E, Graas MP, Simoncini E, Abdi E, Kamby C, Thompson A, Loibl S, Gavila J, Kuroi K, Marth C, Muller B, O'Reilly S, Gombos A, Ruhstaller T, Burstein HJ, Rabaglio M, Ruepp B, Ribi K, Viale G, Gelber RD, Coates AS, Loi S, Goldhirsch A, Regan MM, Colleoni M, Investigators S. **Continuous versus intermittent extended adjuvant letrozole for breast cancer: final results of randomized phase III SOLE (Study of Letrozole Extension) and SOLE Estrogen Substudy**. *Ann Oncol* (2021) **32** 1256-1266. DOI: 10.1016/j.annonc.2021.07.017
|
---
title: Presence of immunogenic alternatively spliced insulin gene product in human
pancreatic delta cells
authors:
- René van Tienhoven
- Maria J. L. Kracht
- Arno R. van der Slik
- Sofia Thomaidou
- Anouk H. G. Wolters
- Ben N. G. Giepmans
- Juan Pablo Romero Riojas
- Michael S. Nelson
- Françoise Carlotti
- Eelco J. P. de Koning
- Rob C. Hoeben
- Arnaud Zaldumbide
- Bart O. Roep
journal: Diabetologia
year: 2023
pmcid: PMC10036285
doi: 10.1007/s00125-023-05882-y
license: CC BY 4.0
---
# Presence of immunogenic alternatively spliced insulin gene product in human pancreatic delta cells
## Abstract
### Aims/hypothesis
Transcriptome analyses revealed insulin-gene-derived transcripts in non-beta endocrine islet cells. We studied alternative splicing of human INS mRNA in pancreatic islets.
### Methods
Alternative splicing of insulin pre-mRNA was determined by PCR analysis performed on human islet RNA and single-cell RNA-seq analysis. Antisera were generated to detect insulin variants in human pancreatic tissue using immunohistochemistry, electron microscopy and single-cell western blot to confirm the expression of insulin variants. Cytotoxic T lymphocyte (CTL) activation was determined by MIP-1β release.
### Results
We identified an alternatively spliced INS product. This variant encodes the complete insulin signal peptide and B chain and an alternative C-terminus that largely overlaps with a previously identified defective ribosomal product of INS. Immunohistochemical analysis revealed that the translation product of this INS-derived splice transcript was detectable in somatostatin-producing delta cells but not in beta cells; this was confirmed by light and electron microscopy. Expression of this alternatively spliced INS product activated preproinsulin-specific CTLs in vitro. The exclusive presence of this alternatively spliced INS product in delta cells may be explained by its clearance from beta cells by insulin-degrading enzyme capturing its insulin B chain fragment and a lack of insulin-degrading enzyme expression in delta cells.
### Conclusions/interpretation
Our data demonstrate that delta cells can express an INS product derived from alternative splicing, containing both the diabetogenic insulin signal peptide and B chain, in their secretory granules. We propose that this alternative INS product may play a role in islet autoimmunity and pathology, as well as endocrine or paracrine function or islet development and endocrine destiny, and transdifferentiation between endocrine cells. INS promoter activity is not confined to beta cells and should be used with care when assigning beta cell identity and selectivity.
### Data availability
The full EM dataset is available via www.nanotomy.org (for review: http://www.nanotomy.org/OA/Tienhoven2021SUB/6126-368/). Single-cell RNA-seq data was made available by Segerstolpe et al [13] and can be found at https://sandberglab.se/pancreas. The RNA and protein sequence of INS-splice was uploaded to GenBank (BankIt2546444 INS-splice OM489474).
### Supplementary Information
The online version of this article (10.1007/s00125-023-05882-y) contains peer-reviewed but unedited supplementary material..
## Introduction
Polyhormonal endocrine cells have been shown to reside in human fetal pancreatic islets and in individuals with type 2 diabetes and chronic pancreatitis but fully differentiated endocrine cells are classically dedicated to produce a single hormone (i.e. glucagon by alpha cells, insulin by beta cells and somatostatin by delta cells) [2–5]. Under this definition, insulin gene (INS) expression is restricted to pancreatic beta cells. Yet, accumulating data indicate that even mature human beta cells are more plastic than previously assumed [6]. While the differentiated state of beta cells is maintained by reinforcement of specific gene regulatory networks and repression of other transcriptional programmes [7–10], specific circumstances such as metabolic and mechanical stress have been shown to cause spontaneous dedifferentiation and transdifferentiation of human beta cells [11, 12]. Conversion of beta cells into alpha and delta cell-like states observed in individuals with type 2 diabetes has been proposed to contribute to reduced functional beta cell mass and beta cell failure [3]. Furthermore, in vitro disruption of human islet integrity has been reported to cause the spontaneous conversion of some beta cells into glucagon-producing cells. This endocrine plasticity has been proposed to allow dysfunctional beta cells to escape apoptosis due to environmental stress as well as replenish beta cell mass [13]. In situ hybridisation and single-cell transcriptome analysis of human islet cells have confirmed the presence of INS mRNA in alpha and delta cells [1]. Approximately $46\%$ of islet cells were found to express more than one additional hormonal transcript per cell, with a considerable portion containing both insulin and somatostatin transcripts [14].
Alternative splicing increases proteome diversity by generating multiple mRNA transcripts from a single gene that differ in their assembly of exons and introns. Approximately $95\%$ of the human transcriptome is estimated to derive from alternatively spliced transcripts [15]. Tissue-specific splicing patterns allow expression of genes in different cell types to produce protein isoforms that differ in biological composition and activity [16]. Alternative splicing networks are implicated in a broad variety of biological processes, including; maintenance of pluripotency, directing cell differentiation, cell lineage commitment and tissue identity [17]. Splicing patterns are highly dynamic and therefore provide a mechanism to allow swift adaptation to changes in the local microenvironment [18]. Many tumours undergo alternative splicing, potentially generating neoantigens that are prominent targets in cancer immunotherapy [19]. Likewise, splice variants generated in beta cells may contribute to autoimmunity and type 1 diabetes [20, 21]. Type 1 diabetes autoimmunity is hallmarked by insulitis in which beta cells are specifically destroyed by autoreactive cytotoxic T lymphocytes (CTLs) that are highly reactive to preproinsulin (PPI) epitopes [22–24]. The beta cell transcriptome was shown to be highly impacted by inflammatory and metabolic insults [25]. Long RNA sequencing and ribosomal profiling revealed the extreme diversity of the beta cell transcriptome and proteome [26]. Experiments conducted in HEK293T cells overexpressing INS demonstrated the presence of cryptic splice sites in INS mRNA, as multiple PPI-coding insulin transcript variants were detected [27, 28].
We investigated alternative splicing of INS mRNA in human islets and determined the expression and immunogenicity of alternative insulin protein products in endocrine cells.
## Human islets
Pancreatic islets were obtained from human cadaveric donor pancreases with consent. The reported investigations were carried out in accordance with the declaration of Helsinki [2008]. Islets were isolated as previously described [29]. See electronic supplementary material (ESM) Methods for details. The checklist for reporting human islet preparations is presented in ESM Table 1.
## Cell culture
HEK293T cells (ATCC CRL-3216) were maintained in high-glucose DMEM (Gibco-BRL, Breda, the Netherlands) supplemented with $8\%$ FBS (Gibco-BRL), 100 units/ml penicillin and 100 μg/ml streptomycin (Gibco-BRL). Cells were mycoplasma negative.
## DNA constructs and transfection
Insulin-expressing vectors were cloned using human genomic DNA and verified by Sanger sequencing. HEK293T cells were transfected using polyethylenimine and harvested 48 h post transfection. See ESM Methods for details.
## Western blotting
Standard western blotting protocols were followed using HEK293T cell lysate and antibodies against insulin, C-peptide, actin, GFP, SPLICE81-95 and somatostatin. See ESM Methods for details.
## Flow cytometry
Human islets were dispersed into single cells, fixed, permeabilised, stained with SPLICE81-95 antiserum and analysed using FACS Aria II (BD Biosciences, USA). See ESM Methods for details.
## ELISA
Plates were coated with recombinant polypeptide and blocked with $2\%$ BSA, followed by incubation with antibodies against DRiP1-13, SPLICE81-95 or C-peptide. Antibodies were visualised with horseradish peroxidase (HRP)-conjugated secondary antibodies and HRP substrate. Absorbance was measured at 450 nm. See ESM Methods for details.
## Electron microscopy
Electron microscopy (EM) islet datasets were created from nPOD donors. Sections (80 nm) were placed on formvar-coated copper grids and contrasted with uranyl acetate. Sections were immunolabelled with gold or quantum dots, using SPLICE81-95 antiserum. Data were acquired by Supra 55 scanning EM (SEM; Zeiss, Oberkochen, Germany) using a scanning transmission EM detector at 28 kV with 2.5 nm pixel size and an external scan generator ATLAS 5 (Fibics, Ottawa, ON, Canada). See ESM Methods for details.
## Single-cell transcriptome analysis
Single-cell RNA-seq data from non-diabetic donors were acquired online [1]. Data from all delta and beta cells were merged into single BAM files per donor. Reads in the region of interest (chr11:2157102-2163862) were extracted and Sashimi plots were generated using the Integrative Genomics Browser.
## Generation of custom polyclonal antisera
Custom polyclonal antisera were generated by immunising rabbits with the synthetic peptides MLYQHLLPLPAGEC (DRiP1-13; cysteine served as anchor residue for the carrier) and LLHRERWNKALEPAK (SPLICE81-95) (Eurogentec, Belgium). The rabbits were repeatedly boosted for 28 days with synthetic peptide and bled before and after immunisation. Immune reactivity to the specific peptides was tested by ELISA performed by the manufacturer.
## Immunohistochemistry and microscopy
Paraffin-embedded human tissues were cut into 4 μm sections, deparaffinised in xylene and rehydrated. Antigen retrieval was performed prior to staining with antibodies against insulin, C-peptide, glucagon, somatostatin, DRiP1-13, SPLICE81-95, insulin B chain, insulin-degrading enzyme (IDE) and proinsulin. Immunofluorescence was detected with a Leica SP8 (Leica, Germany) or Zeiss LSM880 confocal microscope. Manders co-localisation coefficients (MCCs) were determined using QuPath 0.2.3 [30]; the analysis script is available in the ESM (ESM QuPath Colocalisation Script). See ESM Methods for details.
## Single-cell western blot
Human islets were dispersed into single cells using trypsin and filtration. Single islet cells were loaded onto scWest chips (Protein Simple, San Jose, CA, USA), then lysed, electrophoresed and UV-crosslinked according to the manufacturer’s protocol. Chips were probed with primary antibodies against somatostatin (1:30, A0566; Dako, Denmark), insulin B chain (1:10, M093-3; MBL, USA) and insulin C-peptide (1:15, ab14181; Abcam, UK). The appropriate Alexa-conjugated secondary antibodies were used. DNA was stained with Yoyo1 (Invitrogen, USA). Immunofluorescence was detected with the GenePix 4400A microarray scanner (Molecular Devices, USA) at 2.5 μm resolution.
## Recombinant polypeptides
Human recombinant polypeptides were synthesised as previously described [31]. Protein encoding cDNA was obtained from human pancreatic islets by PCR and cloned in pDest17 for protein production in *Escherichia coli* using gateway cloning technology (Invitrogen, Carlsbad, CA, USA). Recombinant proteins were purified by His6 affinity purification tag and freeze-dried. Purified polypeptides were dissolved in $0.05\%$ acetic acid in MQ/PBS to a stock concentration of 1 mg/ml.
## CTL activation assay
HEK293T cells expressing the alternatively spiced INS mRNA were cocultured with CTLs directed against the PPI signal peptide PPI15-24. The supernatant fraction was used for detection of MIP-1β production by the CTLs. See ESM methods for details.
## IDE cleavage assay
Recombinant alternatively spliced insulin product (INS-splice; 0.25 μg/μl) was incubated with recombinant human IDE (0.05 μg/μl, 2496-ZN; R&D Systems, USA) in cleavage buffer (50 mmol/l Tris, 1 mol/l NaCl, pH 7.5) at 37°C for 48 h, following the manufacturer’s recommendations. Samples were heat-inactivated at 70°C for 5 min, diluted in LDS sample buffer (Thermo Fisher Scientific, USA) and heated at 70°C for 10 min before loading onto a $12\%$ Bis-Tris gel (Thermo Fisher Scientific). Gel was stained with Coomassie blue (Bio-Rad, USA) according to the manufacturer’s protocol.
## Statistical analysis
All data points are presented as mean values (±SD). Statistical calculations were carried out using Graphpad Prism 9 (Graphpad software, San Diego, CA, USA). Statistical tests are indicated in the figure legends. A p value of <0.05 was considered significant.
## Evidence of alternative INS RNA splicing in human islets
Analyses performed on RNA isolated from human islets of three different donors identified two major INS RNA variants (Fig. 1). Nucleotide sequencing of these INS cDNA variants indicated that the larger, more-abundant INS RNA variant represents full-length PPI in which intron 1 and 2 have been fully spliced out (ESM Fig. 1). This full-length INS mRNA has been shown to generate an insulin defective ribosomal product (INS-DRiP), in particular under endoplasmic reticulum (ER) stress, which is a target of islet autoimmunity and associated with type 1 diabetes pathology [32]. The shorter, less-abundant cDNA variant resulted from a cryptic splicing site within exon 3 at position 1338 (ESM Fig. 1), predicted by in silico analysis (not shown). The open reading frame that is formed by this alternative splicing may lead to the translation of a polypeptide in which the signal peptide and B chain of the canonical PPI are intact but the C-terminal end of the molecule differs because of RNA translation into the +2 reading frame (referred to as INS-splice). Coincidently, this C-terminal region is identical to the C-terminus of INS-DRiP except for the first ten-amino-acid immunodominant N-terminus that is unique to INS-DRiP [32] (Fig. 1). Fig. 1Alternative INS RNA splicing in human islets. Analysis of INS splicing by PCR on RNA derived from human pancreatic islets of three different donors, visualised on DNA gel. A schematic overview of the human insulin pre-mRNA is shown with the exons annotated by numbers (1–3) and the intronic regions represented by a black solid line. Normal INS mRNA splicing and alternative INS mRNA splicing are indicated by black and red dashed lines, respectively (showing the start codon, AUG). The resulting mRNA products with translation initiation sites are depicted underneath. For each mRNA molecule the potential protein products are displayed. Regular protein translation of the regular spliced transcript produces preproinsulin (PPI) with the signal peptide (orange), B chain (green), C-peptide (blue) and insulin A chain (yellow). Alternative translation of this transcript produces INS-DRiP, with the previously identified CTL epitope (dark red). Translation of the alternatively spliced transcript produces a splice protein variant, referred to as INS-splice. Amino acid sequences are indicated with corresponding colours and letters indicate the presence of the complete chain. The non-stop characteristic of INS-DRiP and INS-splice proteins is visualised by decreasing gradient. A, insulin A chain; B, insulin B chain; C, C-peptide; E, CTL epitope; M, DNA marker; INS, insulin; SP, signal peptide
## Alternatively spliced INS mRNA is a template for translation in delta cells
To investigate these alternative INS-derived proteins, rabbits were immunised with a short polypeptide unique to INS-DRIP (DRiP1-13) and a short polypeptide of the C-terminus shared between INS-DRiP and the predicted polypeptide INS-splice (‘SPLICE81-95’). The peptides were selected from analysis of the UniProt human protein Knowledgebase using the basic local alignment search tool (BLAST) to avoid cross-reactivity to other known proteins (data not shown). Antiserum specificity was confirmed by ELISA using recombinant PPI, INS-DRiP and INS-splice (ESM Fig. 2). As expected, neither antisera cross-reacted with PPI. While the anti-DRiP1-13 antiserum specifically detected the INS-DRiP polypeptide recognised by cytolytic T cells in individuals with type 1 diabetes, the anti-SPLICE81-95 antiserum reacted with both recombinant INS-DRiP and INS-splice proteins (ESM Fig. 2).
To investigate whether the INS-derived polypeptides are generated in islets, human pancreatic sections were stained with the pre-immunisation or post-immunisation antisera. The localisation of the N-terminal INS-DRiP polypeptide within beta cells is consistent with our previous findings and supports beta cell destruction by CTLs directed against INS-DRiP [32] (Fig. 2a). Yet, the SPLICE81-95 antiserum raised to the C-terminus shared between INS-DRiP and INS-splice did not co-localise with insulin, implying that SPLICE81-95+ cells are not beta cells (Fig. 2b). To assess the identity of these SPLICE81-95+ cells, human pancreatic sections were co-stained with various endocrine cell markers (insulin, glucagon and somatostatin). Staining of the SPLICE81-95 epitope proved restricted to delta cells as indicated by its exclusive co-localisation with somatostatin (Fig. 2c). Fig. 2SPLICE81-95 antiserum stained delta cells and INS-splice protein is localised to somatostatin granules. ( a, b) Immunohistochemistry of human pancreas sections with pre-immunisation serum and post-immunisation serum (green) in combination with insulin (red). Serum derived from DRiP1-13 immunised rabbits (a) and SPLICE81-95 immunised rabbits (b) was used. Scale bar, 30 μm. ( c) Human pancreas sections stained for glucagon (white) and somatostatin (red), and SPLICE81-95 antiserum (green). Enlarged images of the grey enclosure are shown. Nuclei were visualised by DAPI staining (blue). Scale bar, 30 μm. ( d) EM images of human pancreas sections labelled for INS-splice (quantum dots, red arrows) and insulin (immunogold, green arrows) visible as black dots. Scale bar, 200 nm. The granules were identified by their morphology. The full dataset is available via www.nanotomy.org (for review, see http://www.nanotomy.org/OA/Tienhoven2021SUB/6126-368/). ( e, f) Quantification of the insulin-immunogold+ (e) and INS-splice-quantum dot+ (f) granules in beta and delta cells. Each granule is represented as a point. The graphs represent the means of 30 beta and 30 delta cell granules Since the alternatively spliced insulin isoform product shares an N-terminus with PPI, we tested whether the presence of the signal peptide contributes to post-translational processing and intracellular localisation of INS-splice. Detailed examination of pancreatic slices by high-resolution EM with quantum dot-labelled SPLICE81-95 antiserum demonstrated that INS-splice was localised to secretory granules of delta cells (Fig. 2d–f), which could be clearly distinguished from insulin secretory granules of beta cells by their unique ultrastructure [33]. This confirms that INS-splice is transported to delta cell granules and implies that it is secreted upon degranulation. Of note, staining of other endocrine tissues demonstrated that the presence of the INS-splice polypeptide is limited to pancreatic islets (ESM Fig. 3). Immunohistochemistry of mouse pancreas sections revealed the presence of INS-splice in delta cells, similar to humans (ESM Fig. 4).
## SPLICE81-95 antiserum does not cross-react with somatostatin
To validate the presence of INS-splice in delta cells and exclude cross-reactivity with somatostatin, HEK293T cells expressing INS were generated. Expression of INS in these cells led to expression of both full-length PPI mRNA and the alternatively spliced insulin mRNA variant, as observed in human islets (ESM Fig. 5a, b). Western blot analysis of lysates of these surrogate beta cells indicated that PPI is expressed, as well as an insulin isoform detected by SPLICE81-95 antiserum (ESM Fig. 5c). To confirm that INS-splice was detected with the SPLICE81-95 antiserum (and not INS-DRiP), both INS mRNA variants were isolated and their cDNAs cloned into separate expression plasmids. Western blot analysis of lysates of HEK293T cells transfected with the alternatively spliced insulin cDNA plasmid demonstrated that the SPLICE81-95 antiserum specifically detects INS-splice only, whereas cells transfected with the full-length PPI cDNA plasmid showed C-peptide expression only (ESM Fig. 5d–f).
SPLICE81-95 antiserum did not cross-react with recombinant somatostatin, as assessed by western blot (ESM Fig. 5g). In addition, antibody blocking assays using recombinant somatostatin did not affect detection of SPLICE81-95 antiserum to recombinant INS-splice, while antibody blocking with recombinant INS-splice markedly reduced INS-splice detection (ESM Fig. 5h). Similarly, blocking of SPLICE81-95 antiserum using the immunisation peptide reduced the mean fluorescence of the SPLICE81-95+ islet cell population compared with irrelevant peptide (ESM Fig. 6).
## Alternatively spliced INS RNA is expressed in beta and delta cells
To further validate the presence of spliced INS mRNA in delta cells, we used a publicly available single islet cell transcriptome dataset and adopted the validated cell type classification of Segerstolpe et al [1], characterised by discrete clusters of endocrine cell types (ESM Fig. 7a). All delta cells showed high expression levels of somatostatin and negligible levels of beta cell-specific transcription factor MafA (ESM Fig. 7b). We searched for supporting reads of the alternative splice junction in both beta and delta cells. Several insulin transcripts were present in beta cells and delta cells (Fig. 3). Among the alternative INS mRNA splice variants, the one using the cryptic splicing site within exon 3 was detected in a subset of delta cells and beta cells. Splicing of insulin transcripts was studied in beta cells and delta cells of five non-diabetic human donors. Of note, aside from the regular and alternatively spliced insulin transcripts coding for PPI and INS-splice, respectively, we report additional alternatively spliced insulin transcripts detectable in subsets of delta cells and beta cells with alternative splice acceptor sites in exon 2 and exon 3 of the INS RNA (Fig. 3). Furthermore, low levels of alternatively spliced INS mRNA were detected in the alpha, epsilon and gamma cell clusters, although mean transcripts per million (TPM) values were 2.8, 2.8 and 2.4 times lower, respectively, compared with delta cells, and 79, 77, 67 times lower, respectively, compared with beta cells (ESM Fig. 7b). Fig. 3Splicing events of insulin transcripts in human beta cells and human delta cells. ( a) Splicing pattern of PPI (T1, green) and an in silico predicted INS RNA splice variant (T2, red). Exons are shown as boxes and introns as lines. ( b, c) Sashimi plots show splicing events of INS RNA for beta cells (b, green) and delta cells (c, blue). Human pancreas donor identity numbers are indicated by HP. The numbers of splicing events are shown. Splicing of PPI mRNA is shown by alignment with the regular splicing pattern from (a) (T1, green boxes). Alternative splicing of insulin transcripts is defined as any sequence that does not align with the regular splicing pattern as displayed in T1
## Insulin B chain expressed in delta cells
Since an alternatively spliced insulin gene product containing the insulin B chain was detected in delta cells, we analysed the presence of the insulin B chain fragment in the human pancreas by immunofluorescence. Co-localisation of insulin B chain, insulin and somatostatin was determined (Fig. 4a–c). Although complete co-localisation between insulin and insulin B chain was expected, we only found $22\%$ co-localisation, indicating that the insulin B chain staining is a gross underestimation, likely due to a low sensitivity of the antibody against insulin B chain compared with the ‘gold standard’ insulin polyclonal antibody from Dako (Fig. 4a). Importantly, some locations showed co-localisation of somatostatin and insulin B chain in the absence of insulin staining (Fig. 4b,d, grey enclosure). 3D reconstruction showed that this insulin B chain staining was indeed inside delta cells (Fig. 4d and ESM video) ESM Video3D reconstruction of insulin B-chain expressing delta cell. Pancreas section was stained for insulin (green), insulin B-chain (white), somatostatin (red) and Hoechst (blue). 3D colocalization analysis was performed using Imaris 9.7.1 software and a video was created using Amira 2019.1 software. Insulin B-chain was shown inside the delta cell (MP4 11584 kb). Control staining with secondary antibody alone was negative (ESM Fig. 8). These data confirm insulin B chain expression in a subset of delta cells. Fig. 4Insulin B chain expression in delta cells. ( a–c) Pancreas section was stained for insulin (green), insulin B chain (white), somatostatin (red) and Hoechst. Z-stack images were made and used for 3D reconstruction and co-localisation analysis (Imaris). Co-localisation of insulin and insulin B chain (a), insulin B chain and somatostatin (b), and insulin and somatostatin (c) was determined and a co-localisation channel (blue) was created for double-positive voxels. The MCC was 0.22, 0.13 and 0.22, respectively. Scale bar, 10 μm. ( d) Enlarged images of the grey enclosure show insulin B chain expression in a delta cell, including a 3D reconstructed image (see ESM Video). Scale bar, 5 μm. ( e) Single-cell western blot was performed on dispersed pancreatic islet cells from two different donors and stained for DNA (green), insulin B chain (red), C-peptide (white) and somatostatin (blue). Doublets were excluded by measuring a composite of DNA intensity and hormone content of all endocrine cells. Single cells (black circles) were included for analysis on the basis of their single-cell DNA intensity and hormone content. Doublets (red circles) were excluded because of their high DNA intensity and/or high hormone content. Examples of doublets and single cells are shown. The single-cell example shows single delta cells that are positive for somatostatin and insulin B chain (δB chain+, black open shapes), of which some are negative for C-peptide (δB chain+, C-pep−, open squares). DNA intensity and examples of cells are shown from one out of six single-cell western blot chips. In total, 554 single delta cells were included for analysis; 54 single delta cells expressed insulin B chain besides somatostatin ($9.7\%$), of which seven were negative for C-peptide ($1.3\%$). Β, beta cell; δ, delta cell; INS, insulin; INS-B, insulin B chain; SST, somatostatin To validate and enumerate insulin B chain expression in delta cells, a single-cell western blot was performed on dispersed islet cells from two human donors (1000 islet equivalents each), using antibodies against somatostatin, C-peptide and insulin B chain. Only single delta cells ($$n = 554$$) were included for analysis, and doublets were excluded by measuring a composite of their DNA intensity and hormonal content (Fig. 4e). A subset of delta cells was found that expressed insulin B chain besides somatostatin ($$n = 54$$, $9.7\%$), of which some were negative for C-peptide ($$n = 7$$, $1.3\%$). Detection of insulin B chain without C-peptide indicates the presence of INS-splice and excludes the presence of PPI in this subset of delta cells. These results provide evidence of the presence of INS products, in particular the immunogenic B chain, in a subset of delta cells.
## INS-splice activates PPI-specific CTLs
INS-splice includes the complete signal peptide and B chain sequences of PPI that contain highly immunogenic epitopes targeted in individuals with type 1 diabetes [22, 24]. To investigate the immunogenicity of INS-splice, HEK293T cells were transfected with a vector encoding for INS-splice-IRES-GFP and GFP expression was validated by RT-PCR 24 h post transfection (Fig. 5a). The activation of PPI15-24-specific CTLs was determined by measuring their MIP-1β secretion into the supernatant fraction after co-culture with INS-splice-expressing HEK293T cells. In the absence of INS-splice expression, MIP-1β secretion was basal, while its secretion was highly upregulated in the presence of INS-splice in an effector dose-dependent manner (Fig. 5b). These data demonstrate that human cells can generate immunogenic epitopes from the INS-splice polypeptide that are processed and presented to patient-derived cytolytic T cells, suggesting that INS-splice-expressing delta cells may be targeted by autoreactive T cells in type 1 diabetes immunopathology. Fig. 5INS-splice activates PPI15-24-specific CTLs. ( a) GFP mRNA expression in HEK293T cells transfected with INS-splice-IRES-GFP for 24 h. Gene expression levels are corrected for GAPDH used as housekeeping gene and presented as the induction ratio (mock control set to 1) ($$n = 3$$). ( b) MIP-1β secretion of PPI15-24-specific CTLs after co-culture with INS-splice-expressing HEK293T (red) or mock control (blue) cells. Effector/target ratios were 1:1, 2:1 and 4:1. The dotted line shows the basal MIP-1β secretion of PPI15-24-specific CTLs in the absence of target cells. Data are shown as mean ± SD, $$n = 4$.$ E, effector; T, target
## IDE cleaves INS-splice and lack of IDE expression in delta cells correlates with presence of INS-splice
The alternatively spliced INS mRNA was detectable in both beta and delta cells, whereas the INS-splice protein was only detected in delta cells using SPLICE81-95 antiserum. To reconcile this conundrum, we hypothesised whether this INS-splice protein could be selectively degraded in some endocrine cell types but spared in others. IDE is known to capture the insulin B chain to degrade insulin [34]. Since INS-splice contains the full insulin B chain, we tested whether detection of IDE in human pancreas sections by immunohistochemistry could help explain the expression of INS-splice protein in delta cells vs beta cells. While IDE expression was confirmed ubiquitously in islets and surrounding exocrine tissue, including beta cells, it was undetectable in delta cells (Fig. 6a). Co-localisation was quantified by analysing 36 islets from six donors and the mean MCC of IDE with somatostatin (0.13) was significantly lower than that for IDE with insulin (0.82) (Fig. 6b). Next, we tested whether IDE could cleave INS-splice. Recombinant INS-splice was incubated with human IDE and visualised by Coomassie staining. INS-splice was indeed digested by IDE (Fig. 6c and ESM Fig. 9). This supports a role for IDE in the selective expression of INS-splice protein in delta cells, reconciling the discrepancy between INS-splice RNA and protein expression in beta and delta cells. Fig. 6IDE is not expressed in delta cells and cleaves INS-splice. ( a) Pancreatic IDE protein expression was determined by immunohistochemistry of human pancreas sections by staining for proinsulin (green), IDE (white) and somatostatin (red). IDE was expressed ubiquitously in the exocrine and endocrine pancreas except in delta cells. Scale bar, 10 μm. Nuclei were visualised by Hoechst staining (blue). ( b) Co-localisation of IDE with insulin and IDE with somatostatin was quantified using MCC (QuPath). Thirty-six islets from six pancreas donors were analysed. Data are shown as mean ± SD and statistical analysis was performed using a paired two-tailed Student’s t test (***$p \leq 0.001$). ( c) Coomassie staining of INS-splice after IDE cleavage assay. Absence or presence of IDE is indicated by − or +, respectively. Full-length recombinant INS-splice is 14 kDa. INS, insulin; M, protein marker; SST, somatostatin
## Discussion
While insulin has been widely studied for its role in glucose homeostasis and islet autoimmunity in type 1 diabetes, little is known about alternative INS-derived proteins. In this study, we investigated two alternative INS-derived polypeptides; INS-DRiP, a defective ribosomal INS product; and INS-splice, an insulin splice variant that shares sequence homologies with PPI and INS-DRiP. We have previously identified INS-DRiP in type 1 diabetes immunopathology as target of cytolytic CTLs in individuals with type 1 diabetes [32]. We now report that alternative splicing of INS pre-mRNA resulted in a translated polypeptide with an N-terminus overlapping with PPI and a C-terminus overlapping with INS-DRiP. Similar to full-length PPI mRNA, the alternatively spliced INS mRNA variant was detected in both beta and delta cells by analysis of online single-cell transcriptome databases of human pancreatic islets [1], while protein detection of PPI in beta cells and INS-splice in delta cells appeared mutually exclusive. This points to cell-specific translation or protein degradation machinery in adult endocrine cells. The N-terminus of INS-DRiP containing the CTL epitope is lacking in INS-splice and can be detected in pancreatic beta cells by immunohistochemistry, supporting our previous data showing beta cell selectivity. INS-DRiP-reactive CTLs that were isolated from individuals with type 1 diabetes would therefore not be able to cross-react with INS-splice [32]. T cells reactive to INS-DRiP were present in insulitic lesions of individuals with type 1 diabetes [24]. The absence of the SPLICE81-95 antiserum epitope shared between INS-DRiP and INS-splice protein in beta cells suggests that INS-DRiP is rapidly targeted for protein degradation during translation in beta cells, consistent with the classic degradation process of non-stop proteins [35]. Similar to INS-DRiP, the INS-splice open reading frame lacks a stop codon, but INS-splice was detected specifically in delta cell granules using SPLICE81-95 antiserum. The presence of the PPI signal peptide may target the co-translational translocation of INS-splice to the ER lumen and subsequently into secretory granules, which may in turn explain why this non-stop protein was not subject to the degradation process that removes INS-DRiP from beta cells. Non-stop proteins targeted to the ER block translocon channels, which should be cleared rapidly to allow normal protein influx into the ER. Clearance of the blocked translocon channel allows the release of non-stop proteins into the ER, preventing proteasomal degradation in the cytosol [36]. We propose that IDE may be involved in the degradation and removal of INS-splice protein from beta cells. Indeed, IDE can digest INS-splice in vitro and IDE expression is ubiquitous in the pancreas except delta cells, possibly explaining why INS-splice protein is only detected in delta cells but not in beta cells despite the presence of INS-splice mRNA in both islet cell types.
Both single-cell RNA-seq and single-cell western blot have limited sensitivity in detecting rare targets [37, 38]. The low expression rate of INS-derived products in delta cells (i.e. alternatively spliced insulin RNA and INS-splice protein) is conceivably close to the sensitivity limit of these single-cell-based methods; this could help explain why SPLICE81-95 antiserum detected INS-splice in all delta cells using immunohistochemistry, while INS-splice mRNA and the insulin B chain were only detected in a delta cell subset. Alternatively, the observation that delta cells stained with SPLICE81-95 antiserum only partially stain with anti-insulin B chain antibody suggests that additional INS-splice products may be expressed in delta cells, as indicated by our transcriptome analyses as well as by detection of C-peptide traces in some delta cells containing insulin B chain. Detection of these potential additional alternatively spliced INS products is hampered by variation in protein length (lacking stop codon), rapid degradation of non-conventional proteins, low expression levels and limited availability of specific detection methods. Our data suggests that there are subpopulations of delta cells, as previously demonstrated in beta cells. Delta cell subsets showed heterogeneity regarding RNA expression and INS-derived protein expression.
The presence of the INS-splice protein in human and mouse delta cells is intriguing and may have implications for studies using the insulin promotor as a supposedly specific reporter for beta cells [39–41]. Our results imply that use of the insulin promotor activity to specifically target beta cells could lead to off-target effects in delta cells. In addition, the presence of INS-splice protein implies the activity of the insulin signal peptide and B chain in delta cells as confirmed by single-cell western blots in a subpopulation of delta cells. These peptides are major targets for islet autoimmunity [22, 23, 42–44]. INS-splice-expressing cells activated PPI-specific CTLs, validating the immunogenicity of the INS-splice peptide. Hence, a subset of delta cells may produce and present diabetogenic epitopes in HLA, making them vulnerable to attack by diabetogenic T cells. Delta cells have not yet been thoroughly investigated for their involvement in islet autoimmunity in type 1 diabetes. While impaired delta cell function has been reported [45], data on delta cell destruction by autoreactive CTLs are still lacking. INS-splice shares sequence homology with a previously described INS-IGF2 protein [46] and a 74-amino-acid proinsulin protein [47], although their C-termini differ. Since all three isoforms retained the insulin signal peptide and B chain, the intracellular distribution and function may overlap.
The relevance of INS mRNA expression in non-beta endocrine cells remains unclear. Delta cells have an important role in beta cell development during organogenesis [48, 49]. Human beta cells have also been shown to change identity via de- and transdifferentiation [11]. Alternative splicing is involved in maintaining lineage differentiation and tissue identity as well as maintenance of cell pluripotency and is influenced by the microenvironment [17]. It remains unknown whether INS promoter activity in non-beta cell endocrine cells is a remnant of their common progenitor cell or contributes to maintaining endocrine cell plasticity in adolescence. While the function of INS-splice protein is still enigmatic, its presence in secretory granules implies that INS-splice is co-secreted with somatostatin during exocytosis and may have paracrine or endocrine function in the developmental destiny of human islet cells.
## Supplementary information
ESM(PDF 1.99 mb)ESM QuPath Colocalisation Script(DOCX 17.7 kb)ESM Video3D reconstruction of insulin B-chain expressing delta cell. Pancreas section was stained for insulin (green), insulin B-chain (white), somatostatin (red) and Hoechst (blue). 3D colocalization analysis was performed using Imaris 9.7.1 software and a video was created using Amira 2019.1 software. Insulin B-chain was shown inside the delta cell (MP4 11.3 MB)
## Authors’ relationships and activities
The authors declare that there are no relationships or activities that might bias, or be perceived to bias, their work.
## Contribution statement
BOR, AZ and ARvdS conceived and designed the work. RvT and MJLK drafted the manuscript. BOR and AZ interpreted the data, revised the manuscript and supervised the project together with RCH. RvT, MJLK, ARvdS, ST, AHGW, JPRR and MSN acquired and analysed the data. AHGW and BNGG performed EM and analysed the data. JPRR analysed the single-cell RNA-seq data. MSN wrote the QuPath script and analysed immunohistochemical data. FC and EJPK supervised the human islet isolation and analyses. All co-authors contributed to discussions about and revisions of the manuscript, assume personal accountability for their own contributions and approved the final version of this work. BOR and AZ are the guarantors of this work and take full responsibility for the manuscript.
## References
1. 1.Segerstolpe A, Palasantza A, Eliasson P et al (2016) Single-cell transcriptome profiling of human pancreatic islets in health and type 2 diabetes. Cell Metab 24(4):593–607. 10.1016/j.cmet.2016.08.020
2. De Krijger RR, Aanstoot HJ, Kranenburg G, Reinhard M, Visser WJ, Bruining GJ. **The midgestational human fetal pancreas contains cells coexpressing islet hormones**. *Dev Biol* (1992.0) **153** 368-375. DOI: 10.1016/0012-1606(92)90121-v
3. Cinti F, Bouchi R, Kim-Muller JY. **Evidence of beta-cell dedifferentiation in human type 2 diabetes**. *J Clin Endocrinol Metab* (2016.0) **101** 1044-1054. DOI: 10.1210/jc.2015-2860
4. Moin ASM, Dhawan S, Cory M, Butler PC, Rizza RA, Butler AE. **Increased frequency of hormone negative and polyhormonal endocrine cells in lean individuals with type 2 diabetes**. *J Clin Endocrinol Metab* (2016.0) **101** 3628-3636. DOI: 10.1210/jc.2016-2496
5. Moin ASM, Cory M, Choi J. **Increased chromogranin a-positive hormone-negative cells in chronic pancreatitis**. *J Clin Endocrinol Metab* (2018.0) **103** 2126-2135. DOI: 10.1210/jc.2017-01562
6. Szabat M, Lynn FC, Hoffman BG, Kieffer TJ, Allan DW, Johnson JD. **Maintenance of beta-cell maturity and plasticity in the adult pancreas: developmental biology concepts in adult physiology**. *Diabetes* (2012.0) **61** 1365-1371. DOI: 10.2337/db11-1361
7. Schaffer AE, Taylor BL, Benthuysen JR. **Nkx6.1 controls a gene regulatory network required for establishing and maintaining pancreatic Beta cell identity**. *PLoS Genet* (2013.0) **9** e1003274. DOI: 10.1371/journal.pgen.1003274
8. Swisa A, Avrahami D, Eden N. **PAX6 maintains beta cell identity by repressing genes of alternative islet cell types**. *J Clin Invest* (2017.0) **127** 230-243. DOI: 10.1172/JCI88015
9. Taylor BL, Liu FF, Sander M. **Nkx6.1 is essential for maintaining the functional state of pancreatic beta cells**. *Cell Rep* (2013.0) **4** 1262-1275. DOI: 10.1016/j.celrep.2013.08.010
10. Gutierrez GD, Bender AS, Cirulli V. **Pancreatic beta cell identity requires continual repression of non-beta cell programs**. *J Clin Invest* (2017.0) **127** 244-259. DOI: 10.1172/JCI88017
11. Swisa A, Glaser B, Dor Y. **Metabolic stress and compromised identity of pancreatic beta cells**. *Front Genet* (2017.0) **8** 21. DOI: 10.3389/fgene.2017.00021
12. Galli A, Algerta M, Marciani P. **Shaping pancreatic beta-cell differentiation and functioning: the influence of mechanotransduction**. *Cells* (2020.0) **9** 413. DOI: 10.3390/cells9020413
13. Talchai C, Xuan S, Lin HV, Sussel L, Accili D. **Pancreatic beta cell dedifferentiation as a mechanism of diabetic beta cell failure**. *Cell* (2012.0) **150** 1223-1234. DOI: 10.1016/j.cell.2012.07.029
14. Teo AKK, Lim CS, Cheow LF. **Single-cell analyses of human islet cells reveal de-differentiation signatures**. *Cell Death Discov* (2018.0) **4** 14. DOI: 10.1038/s41420-017-0014-5
15. Pan Q, Shai O, Lee LJ, Frey BJ, Blencowe BJ. **Deep surveying of alternative splicing complexity in the human transcriptome by high-throughput sequencing**. *Nat Genet* (2008.0) **40** 1413-1415. DOI: 10.1038/ng.259
16. Baralle FE, Giudice J. **Alternative splicing as a regulator of development and tissue identity**. *Nat Rev Mol Cell Biol* (2017.0) **18** 437-451. DOI: 10.1038/nrm.2017.27
17. Gabut M, Samavarchi-Tehrani P, Wang X. **An alternative splicing switch regulates embryonic stem cell pluripotency and reprogramming**. *Cell* (2011.0) **147** 132-146. DOI: 10.1016/j.cell.2011.08.023
18. Brosseau JP, Lucier JF, Nwilati H. **Tumor microenvironment-associated modifications of alternative splicing**. *RNA* (2014.0) **20** 189-201. DOI: 10.1261/rna.042168.113
19. 19.Kahles A, Lehmann KV, Toussaint NC et al (2018) Comprehensive analysis of alternative splicing across tumors from 8,705 patients. Cancer Cell 34(2):211–224 e216. 10.1016/j.ccell.2018.07.001
20. de Jong VM, Abreu JR, Verrijn Stuart AA. **Alternative splicing and differential expression of the islet autoantigen IGRP between pancreas and thymus contributes to immunogenicity of pancreatic islets but not diabetogenicity in humans**. *Diabetologia* (2013.0) **56** 2651-2658. DOI: 10.1007/s00125-013-3034-6
21. 21.Gonzalez-Duque S, Azoury ME, Colli ML et al (2018) Conventional and neo-antigenic peptides presented by beta cells are targeted by circulating naive CD8+ T cells in type 1 diabetic and healthy donors. Cell Metab 28(6):946–960 e946. 10.1016/j.cmet.2018.07.007
22. Skowera A, Ellis RJ, Varela-Calvino R. **CTLs are targeted to kill beta cells in patients with type 1 diabetes through recognition of a glucose-regulated preproinsulin epitope**. *J Clin Invest* (2008.0) **118** 3390-3402. DOI: 10.1172/JCI35449
23. Coppieters KT, Dotta F, Amirian N. **Demonstration of islet-autoreactive CD8 T cells in insulitic lesions from recent onset and long-term type 1 diabetes patients**. *J Exp Med* (2012.0) **209** 51-60. DOI: 10.1084/jem.20111187
24. Anderson AM, Landry LG, Alkanani AA. **Human islet T cells are highly reactive to preproinsulin in type 1 diabetes**. *Proc Natl Acad Sci U S A* (2021.0) **118** e2107208118. DOI: 10.1073/pnas.2107208118
25. Eizirik DL, Sammeth M, Bouckenooghe T. **The human pancreatic islet transcriptome: expression of candidate genes for type 1 diabetes and the impact of pro-inflammatory cytokines**. *PLoS Genet* (2012.0) **8** e1002552. DOI: 10.1371/journal.pgen.1002552
26. Thomaidou S, Slieker RC, van der Slik AR. **Long RNA sequencing and ribosome profiling of inflamed beta-cells reveal an extensive translatome landscape**. *Diabetes* (2021.0) **70** 2299-2312. DOI: 10.2337/db20-1122
27. Kralovicova J, Gaunt TR, Rodriguez S, Wood PJ, Day IN, Vorechovsky I. **Variants in the human insulin gene that affect pre-mRNA splicing: is -23HphI a functional single nucleotide polymorphism at IDDM2?**. *Diabetes* (2006.0) **55** 260-264. DOI: 10.2337/diabetes.55.01.06.db05-0773
28. Laub O, Rutter WJ. **Expression of the human insulin gene and cDNA in a heterologous mammalian system**. *J Biol Chem* (1983.0) **258** 6043-6050. DOI: 10.1016/S0021-9258(18)32370-6
29. Ricordi C, Lacy PE, Finke EH, Olack BJ, Scharp DW. **Automated method for isolation of human pancreatic islets**. *Diabetes* (1988.0) **37** 413-420. DOI: 10.2337/diab.37.4.413
30. Bankhead P, Loughrey MB, Fernandez JA. **QuPath: open source software for digital pathology image analysis**. *Sci Rep* (2017.0) **7** 16878. DOI: 10.1038/s41598-017-17204-5
31. Franken KL, Hiemstra HS, van Meijgaarden KE. **Purification of his-tagged proteins by immobilized chelate affinity chromatography: the benefits from the use of organic solvent**. *Protein Expr Purif* (2000.0) **18** 95-99. DOI: 10.1006/prep.1999.1162
32. Kracht MJ, van Lummel M, Nikolic T. **Autoimmunity against a defective ribosomal insulin gene product in type 1 diabetes**. *Nat Med* (2017.0) **23** 501-507. DOI: 10.1038/nm.4289
33. de Boer P, Pirozzi NM, Wolters AHG. **Large-scale electron microscopy database for human type 1 diabetes**. *Nat Commun* (2020.0) **11** 2475. DOI: 10.1038/s41467-020-16287-5
34. Shen Y, Joachimiak A, Rosner MR, Tang WJ. **Structures of human insulin-degrading enzyme reveal a new substrate recognition mechanism**. *Nature* (2006.0) **443** 870-874. DOI: 10.1038/nature05143
35. Bengtson MH, Joazeiro CA. **Role of a ribosome-associated E3 ubiquitin ligase in protein quality control**. *Nature* (2010.0) **467** 470-473. DOI: 10.1038/nature09371
36. Arakawa S, Yunoki K, Izawa T, Tamura Y, Nishikawa S, Endo T. **Quality control of nonstop membrane proteins at the ER membrane and in the cytosol**. *Sci Rep* (2016.0) **6** 30795. DOI: 10.1038/srep30795
37. Haque A, Engel J, Teichmann SA, Lonnberg T. **A practical guide to single-cell RNA-sequencing for biomedical research and clinical applications**. *Genome Med* (2017.0) **9** 75. DOI: 10.1186/s13073-017-0467-4
38. Kang CC, Yamauchi KA, Vlassakis J, Sinkala E, Duncombe TA, Herr AE. **Single cell-resolution western blotting**. *Nat Protoc* (2016.0) **11** 1508-1530. DOI: 10.1038/nprot.2016.089
39. Ray MK, Fagan SP, Moldovan S, DeMayo FJ, Brunicardi FC. **Development of a transgenic mouse model using rat insulin promoter to drive the expression of CRE recombinase in a tissue-specific manner**. *Int J Pancreatol* (1999.0) **25** 157-163. DOI: 10.1007/BF02925964
40. Chai R, Chen S, Ding J, Grayburn PA. **Efficient, glucose responsive and islet-specific transgene expression by a modified rat insulin promoter**. *Gene Ther* (2009.0) **16** 1202-1209. DOI: 10.1038/gt.2009.114
41. Fukazawa T, Matsuoka J, Naomoto Y. **Development of a novel beta-cell specific promoter system for the identification of insulin-producing cells in in vitro cell cultures**. *Exp Cell Res* (2006.0) **312** 3404-3412. DOI: 10.1016/j.yexcr.2006.07.015
42. Pinkse GG, Tysma OH, Bergen CA. **Autoreactive CD8 T cells associated with beta cell destruction in type 1 diabetes**. *Proc Natl Acad Sci U S A* (2005.0) **102** 18425-18430. DOI: 10.1073/pnas.0508621102
43. Tan S, Li Y, Xia J. **Type 1 diabetes induction in humanized mice**. *Proc Natl Acad Sci U S A* (2017.0) **114** 10954-10959. DOI: 10.1073/pnas.1710415114
44. Michels AW, Landry LG, McDaniel KA. **Islet-derived CD4 T cells targeting proinsulin in human autoimmune diabetes**. *Diabetes* (2017.0) **66** 722-734. DOI: 10.2337/db16-1025
45. Rorsman P, Huising MO. **The somatostatin-secreting pancreatic delta-cell in health and disease**. *Nat Rev Endocrinol* (2018.0) **14** 404-414. DOI: 10.1038/s41574-018-0020-6
46. Kanatsuna N, Taneera J, Vaziri-Sani F. **Autoimmunity against INS-IGF2 protein expressed in human pancreatic islets**. *J Biol Chem* (2013.0) **288** 29013-29023. DOI: 10.1074/jbc.M113.478222
47. Liu QR, Zhu M, Zhang P. **Novel human insulin isoforms and calpha-peptide product in islets of langerhans and choroid plexus**. *Diabetes* (2021.0) **70** 2947-2956. DOI: 10.2337/db21-0198
48. Chera S, Baronnier D, Ghila L. **Diabetes recovery by age-dependent conversion of pancreatic delta-cells into insulin producers**. *Nature* (2014.0) **514** 503-507. DOI: 10.1038/nature13633
49. Gregg BE, Moore PC, Demozay D. **Formation of a human beta-cell population within pancreatic islets is set early in life**. *J Clin Endocrinol Metab* (2012.0) **97** 3197-3206. DOI: 10.1210/jc.2012-1206
|
---
title: Tcf7l2 in hepatocytes regulates de novo lipogenesis in diet-induced non-alcoholic
fatty liver disease in mice
authors:
- Da Som Lee
- Tae Hyeon An
- Hyunmi Kim
- Eunsun Jung
- Gyeonghun Kim
- Seung Yeon Oh
- Jun Seok Kim
- Hye Jin Chun
- Jaeeun Jung
- Eun-Woo Lee
- Baek-Soo Han
- Dai Hoon Han
- Yong-Ho Lee
- Tae-Su Han
- Keun Hur
- Chul-Ho Lee
- Dae-Soo Kim
- Won Kon Kim
- Jun Won Park
- Seung-Hoi Koo
- Je Kyung Seong
- Sang Chul Lee
- Hail Kim
- Kwang-Hee Bae
- Kyoung-Jin Oh
journal: Diabetologia
year: 2023
pmcid: PMC10036287
doi: 10.1007/s00125-023-05878-8
license: CC BY 4.0
---
# Tcf7l2 in hepatocytes regulates de novo lipogenesis in diet-induced non-alcoholic fatty liver disease in mice
## Abstract
### Aims/hypothesis
Non-alcoholic fatty liver disease (NAFLD) associated with type 2 diabetes may more easily progress towards severe forms of non-alcoholic steatohepatitis (NASH) and cirrhosis. Although the Wnt effector transcription factor 7-like 2 (TCF7L2) is closely associated with type 2 diabetes risk, the role of TCF7L2 in NAFLD development remains unclear. Here, we investigated how changes in TCF7L2 expression in the liver affects hepatic lipid metabolism based on the major risk factors of NAFLD development.
### Methods
Tcf7l2 was selectively ablated in the liver of C57BL/6N mice by inducing the albumin (Alb) promoter to recombine Tcf7l2 alleles floxed at exon 5 (liver-specific Tcf7l2-knockout [KO] mice: Alb-Cre;Tcf7l2f/f). Alb-Cre;Tcf7l2f/f and their wild-type (Tcf7l2f/f) littermates were fed a high-fat diet (HFD) or a high-carbohydrate diet (HCD) for 22 weeks to reproduce NAFLD/NASH. Mice were refed a standard chow diet or an HCD to stimulate de novo lipogenesis (DNL) or fed an HFD to provide exogenous fatty acids. We analysed glucose and insulin sensitivity, metabolic respiration, mRNA expression profiles, hepatic triglyceride (TG), hepatic DNL, selected hepatic metabolites, selected plasma metabolites and liver histology.
### Results
Alb-Cre;Tcf7l2f/f essentially exhibited increased lipogenic genes, but there were no changes in hepatic lipid content in mice fed a normal chow diet. However, following 22 weeks of diet-induced NAFLD/NASH conditions, liver steatosis was exacerbated owing to preferential metabolism of carbohydrate over fat. Indeed, hepatic Tcf7l2 deficiency enhanced liver lipid content in a manner that was dependent on the duration and amount of exposure to carbohydrates, owing to cell-autonomous increases in hepatic DNL. Mechanistically, TCF7L2 regulated the transcriptional activity of Mlxipl (also known as ChREBP) by modulating O-GlcNAcylation and protein content of carbohydrate response element binding protein (ChREBP), and targeted Srebf1 (also called SREBP1) via miRNA (miR)-33-5p in hepatocytes. Eventually, restoring TCF7L2 expression at the physiological level in the liver of Alb-Cre;Tcf7l2f/f mice alleviated liver steatosis without altering body composition under both acute and chronic HCD conditions.
### Conclusions/interpretation
In mice, loss of hepatic Tcf7l2 contributes to liver steatosis by inducing preferential metabolism of carbohydrates via DNL activation. Therefore, TCF7L2 could be a promising regulator of the NAFLD associated with high-carbohydrate diets and diabetes since TCF7L2 deficiency may lead to development of NAFLD by promoting utilisation of excess glucose pools through activating DNL.
### Data availability
RNA-sequencing data have been deposited into the NCBI GEO under the accession number GSE162449 (www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE162449).
### Supplementary Information
The online version of this article (10.1007/s00125-023-05878-8) contains peer-reviewed but unedited supplementary material..
## Introduction
Transcription factor 7-like 2 (TCF7L2, also known as transcription factor 4 [TCF4]) belongs to the T cell-specific high-mobility group box-containing family of transcription factors that mediates the canonical Wnt signalling pathway [1, 2]. It is well known for its important role in cell development and regeneration [1, 2]. It has also been described as a critical transcription factor in maintaining glucose homeostasis [1–8]. In mice, reduced TCF7L2 expression is associated with impaired glucose metabolism and defective insulin signalling in the pancreatic beta cells [3] and adipose tissue [4, 5]. Loss of hepatic Tcf7l2 disturbs hepatic glucose metabolism by promoting hepatic gluconeogenesis, resulting in impaired insulin signalling pathways [6–8].
According to genome-wide association studies in humans, Tcf7l2 is one of the strongest risk factors associated with type 2 diabetes [9, 10]. However, it remains unclear whether the genetic alteration of Tcf7l2 affects its expression. Therefore, Tcf7l2-deficient in vitro and in vivo models are being used to track the association between SNPs in Tcf7l2 and Tcf7l2 expression, and to gain insights into diabetes-related molecular pathways [1, 11]. Many studies have shown that alterations in Tcf7l2 expression are metabolically associated with type 2 diabetes [1, 11].
Tcf7l2 has also been proposed as a genetic factor that can explain the association between type 2 diabetes and non-alcoholic fatty liver disease (NAFLD), as genetic alterations in Tcf7l2 can make individuals susceptible to NAFLD [12]. Recently, the coexistence of NAFLD and type 2 diabetes has emerged as an important issue because individuals with NAFLD plus type 2 diabetes have a poor metabolic profile and a higher risk of worsening with more advanced forms of disease, such as non-alcoholic steatohepatitis (NASH), cirrhosis and hepatocellular carcinoma (HCC) [13–15]. However, the association of TCF7L2 with NAFLD remains unclear and little is known about the underlying mechanisms and function of TCF7L2 in the pathogenesis of NAFLD.
NAFLD is characterised by excessive hepatic triglyceride (TG) accumulation, which is primarily caused by excessive exogenous fatty acid uptake and increased endogenous fatty acid synthesis (de novo lipogenesis [DNL]) in the liver [16–19]. In this study, to determine the role of hepatic Tcf7l2 deficiency in NAFLD, we generated liver-specific Tcf7l2 knockout (KO) mice (Alb-Cre;Tcf7l2f/f). Based on the major risk factors for NAFLD development [18], mice lacking hepatic Tcf7l2 were given a high-fat diet (HFD) to provide fatty acids, or a refeeding/high-carbohydrate diet (HCD) to stimulate DNL. We thoroughly explored the underlying mechanisms and metabolic phenotype of liver TG accumulation owing to hepatic Tcf7l2 deficiency in in vivo and in vitro systems.
## Methods
For detailed methods, please refer to the electronic supplementary material (ESM).
## Human liver tissue samples
Human liver tissue samples in ESM Fig. 1a were collected from 32 individuals who underwent hepatectomy or cholecystectomy at the university-affiliated Severance Hospital, Yonsei University College of Medicine, Republic of Korea, between September 2014 and October 2019, as previously described [20]. Liver histology was assessed by an experienced pathologist and, according to diagnostic criteria [21], the samples were put into the following groups: [1] normal ($$n = 13$$); [2] simple steatosis (NAFLD; $$n = 10$$); and [3] NASH ($$n = 9$$). All individuals gave informed consent and the study protocol was approved by the Institutional Review Board at Severance Hospital (IRB No 4–2014–0674).
## Animal experiments
Seven-week-old male C57BL/6N mice were purchased from ORIENT BIO (ORIENT BIO, Republic of Korea). Hepatocyte-selective Tcf7l2 ablation in C57BL/6N mice (liver-specific Tcf7l2 KO mice: Alb-Cre;Tcf7l2f/f) was obtained by crossing mice carrying Cre recombinase driven by the albumin (Alb) promoter with mice carrying Tcf7l2 alleles floxed at exon 5. Six-week-old male Alb-Cre;Tcf7l2f/f and their wild-type littermates (Tcf7l2f/f mice; developed in-house from Tcf7l2+/− mice [Tcf7l2tm2a(EUCOMM)Wtsi; the EUCOMM Consortium; www.mousephenotype.org/data/genes/MGI:1202879, accessed 8 September 2022]) were given either an HFD (catalogue no. D12492; Research Diet, USA) or HCD (catalogue no. TD.98090; Envigo, USA) for the indicated periods. For GTTs and ITTs, mice were fasted for 16 h or 6 h, respectively, and injected intraperitoneally with 1.5 g/kg of glucose or 1 U/kg of insulin. Blood glucose levels were measured in samples taken from the tail vein (LifeScan; One-Touch, USA). \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \dot{V}{\mathrm{O}}_2 $$\end{document}V˙O2, carbon dioxide production (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \dot{V}\mathrm{C}{\mathrm{O}}_2 $$\end{document}V˙CO2), respiratory exchange ratio (RER; \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \dot{V}\mathrm{C}{\mathrm{O}}_2 $$\end{document}V˙CO2/\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \dot{V}{\mathrm{O}}_2 $$\end{document}V˙O2) and energy expenditure were measured in mice at 10 weeks using the Oxylet Pro system (Panlab, Spain). All animal procedures were approved by the Institutional Animal Care and Use Committee of the Korea Research Institute of Bioscience & Biotechnology (KRIBB-AEC-21051) and were performed in accordance with the guidelines for the Care and Use of Laboratory Animals published by the US National Institutes of Health.
## Mouse primary hepatocyte culture
Primary hepatocytes were isolated from 8-week-old male C57BL/6N mice by collagenase perfusion [7]. The cells were maintained in Medium 199 (catalogue no. M4530; Sigma-Aldrich, UK) supplemented with $10\%$ (vol./vol.) FBS, $1\%$ (vol./vol.) antibiotics and 10 nmol/l dexamethasone, and were infected with adenovirus for 24 h. Subsequently, the cells were maintained in medium 199 without $10\%$ FBS and were treated with 25 mmol/l glucose (catalogue no. G7021; Sigma-Aldrich) and/or 100 nmol/l insulin (catalogue no. I6634; Sigma-Aldrich) for 48 h, 250 μmol/l palmitic acid (catalogue no. P9767; Sigma-Aldrich) for 24 h, 300 μmol/l AICAR (catalogue no. A611700; Toronto Research Chemicals, Canada) for 1 h, or 1 μmol/l T0901317 (catalogue no. T2320; Sigma-Aldrich) for 24 h.
## Quantitative PCR
Total RNA from human liver tissues, mouse tissues or mouse primary hepatocytes was extracted using the easy-spin Total RNA Extraction Kit (catalogue no. 17221; iNtRON Biotechnology, Republic of Korea). Quantitative PCR (qPCR) was performed using the SYBR green PCR kit in a C1000 Touch Thermal Cycler (Bio-Rad Laboratories, USA). All data were normalised to the expression of ribosomal L32. Primer sequences are listed in ESM Table 1. miRNA (miR) expression was analysed as previously described [22] and the miR primer sequences are listed in ESM Table 2.
## Western blotting
Western blot analysis of extracts from mouse tissues and whole cells was performed as previously described [23]. Primary and secondary antibodies were used according to manufacturer’s instructions and are listed in ESM Table 3.
## Metabolite measurements
Plasma insulin and IGF-1 levels were measured with a Mouse Insulin ELISA Kit (catalogue no. 80-INSMSE01; ALPCO, USA) and Mouse IGF1 ELISA Kit (catalogue no. EMIGF1; Thermo Fisher Scientific), respectively. Total lipids from mouse livers or mouse primary hepatocytes were extracted using the Folch method as described previously [7]. Hepatic TG, cellular TG, plasma TG and NEFA levels were measured with colorimetric assay kits (TG: catalogue no. 461-09092; NEFA: catalogue no. 438-91691; Wako, Japan). Hepatic glycogen was measured using an EnzyChrom Glycogen Assay Kit (catalogue no. E2GN-100; BioAssay Systems, USA). Hepatic β-hydroxybutyrate (β-OH) was measured with a fluorometric assay kit (catalogue no. 700740; Cayman Chemical, USA).
## Immunofluorescence staining
Mouse pancreas tissues in ESM Fig. 1e and ESM Fig. 2d were fixed with $4\%$ (wt/vol.) paraformaldehyde and were embedded in paraffin. Sliced specimens were immunostained using anti-insulin (catalogue no. 53-9769-82; Thermo Fisher Scientific) and anti-glucagon (catalogue no. ab92517; Abcam, USA) antibodies with DAPI (catalogue no. H-1800; Vector Laboratories, USA). The fluorescence images were taken using an OLYMPUS IX73 microscope (Olympus, Japan). Insulin-positive areas (beta cells), glucagon-positive areas (alpha cells) and pancreatic areas were quantified using Image J software (v 1.50i; https://imagej.nih.gov/ij/download.html). Antibody information is listed in ESM Table 4.
## Histological analysis
Liver tissue sections were stained with Oil Red O (catalogue no. O0625; Sigma-Aldrich) and H&E, as previously described [24]. Liver fibrosis was detected with Sirius red staining (catalogue no. ab150681; Abcam, USA) according to the manufacturer’s instructions. NAFLD Activity Score (NAS) and fibrosis stage were calculated as the sum of the scores for steatosis (0–3), lobular inflammation (0–3), hepatocyte ballooning (0–2), as assessed by H&E staining, and fibrosis (0–4), assessed by Sirius red staining [21, 25]. Histological samples were anonymised and assigned a number before the slides were analysed by an experienced pathologist.
## Fatty acid uptake assay
Mouse primary hepatocytes isolated from Tcf7l2f/f or C57BL/6N mice were transfected with the indicated plasmid DNA vectors using Lipofectamine 3000 transfection reagent (catalogue no. L3000015; Invitrogen, USA). Fatty acid uptake in mouse primary hepatocytes was measured with the Fatty Acid Uptake Assay Kit (catalogue no. K408-100; Biovision, USA) according to the manufacturer’s instructions. Fluorescent fatty acid probes (green) were detected by a fluorescence microscope (DM IL LED FLUO; Leica Microsystems, Germany) and fluorescence intensity was quantified using Image J software (v 1.50i).
## Glucose uptake assay
Mouse primary hepatocytes isolated from Tcf7l2f/f mice were infected with adenoviruses expressing gfp only and Cre and treated with 1 nmol/l insulin for 30 min. 2-Deoxyglucose (2-DG) levels in mouse primary hepatocytes were measured using the 2-DG Uptake Measurement Kit (catalogue no. CSR-OKP-PMG-K01; Cosmo Bio, Japan) according to the manufacturer’s instructions.
## Analysis of oxygen consumption rate via Seahorse assay
Fatty acid oxidation in mouse primary hepatocytes was assessed by analysing oxygen consumption rate (OCR) with an XF24 extracellular flux analyser (Seahorse Bioscience, USA). Mouse primary hepatocytes were seeded on collagen-coated XF24 cell culture microplates. Cells were treated with BSA or 250 μmol/l palmitate and/or 100 μmol/l etomoxir (catalogue no. E1905; Sigma-Aldrich), and then stimulated with 2.5 μmol/l oligomycin (catalogue no. O4876; Sigma-Aldrich), 10 μmol/l fluoro-carbonylcyanide phenylhydrazone (FCCP; catalogue no. C2920; Sigma-Aldrich), and 2 μmol/l rotenone (catalogue no. R8875; Sigma-Aldrich) plus 5 μmol/l antimycin A (catalogue no. A8674; Sigma-Aldrich), according to the manufacturer’s instructions. OCR levels were normalised to the amount of protein in each sample.
## DNL measurements
For in vitro DNL analysis, mouse primary hepatocytes were treated with 37 kBq of 14C-labelled glucose (catalogue no. NEC042; PerkinElmer, USA) and 10 nmol/l insulin for 48 h. Lipids were extracted from cells with chloroform and 14C radioactivity was measured using the Tri-Carb 2910 TR liquid scintillation analyser (PerkinElmer). For in vivo DNL analysis, mice were fasted for 24 h, then refed an HCD for 12 h. Subsequently, they were intraperitoneally injected with 14C-labelled sodium acetate (555 kBq/mouse; catalogue no. NEC553; PerkinElmer), or $3\%$ (vol./vol.) ethanol in saline (136.9 mmol/l NaCl) as a control, and euthanised 1 h post injection. Lipids were extracted using the Folch method [7] and incorporation rates of [1-14C]acetic acid into lipids were measured using the Tri-Carb 2910 TR liquid scintillation analyser.
## mRNA sequencing
Total RNA from mouse liver was extracted using the easy-spin Total RNA Extraction Kit (iNtRON Biotechnology). The library preparation, mRNA sequencing and analysis were performed by eBiogen (Republic of Korea). Differentially expressed genes were determined based on counts from unique and multiple alignments using coverage in Bedtools (v 2.25.0; https://bedtools.readthedocs.io/en/latest/index.html). The read count (RC) data were processed based on the quantile normalisation method using EdgeR (v 3.20.1; https://bioconductor.org/packages/release/bioc/html/edgeR.html) within R (v 3.4.4; www.r-project.org) using Bioconductor (v 3.6; https://bioconductor.org/install/). From the results obtained by setting the RC (log2) value to 5 or more, genes with a log2 fold change greater than 1.5 (log2 fold-change cut-off >1.5) in KO vs wild-type comparisons were selected and analysed using Gene Set Enrichment Analysis (GSEA; www.gsea-msigdb.org/gsea/msigdb/mouse/annotate.jsp, accessed 1 November 2021), based on canonical pathways gene sets derived from the Reactome pathway database (www.gsea-msigdb.org/gsea/msigdb/mouse/genesets.jsp?collection=CP:REACTOME, accessed 1 November 2021).
## Plasmids and recombinant adenoviruses
Expression vectors for mouse Tcf7l2, nuclear Srebf1c and Mlxipl(α) were amplified by RT-PCR using liver RNA derived from C57BL/6N mice and inserted into pcDNA3-Flag or pcDNA3-HA expression vectors. The pGL4-6X SRE-luc (containing six copies of SRE) and pGL4-4X ChoRE-luc (containing four copies of ChoRE) constructs were a kind gift from S.-H. Koo (Korea University, Korea). Mouse Fasn (−2000/+200), Pklr (−191/+200), Srebf1c (−1195/+77), miR-33-5p (−852/+16), miR-33-5p (−1412/+16), Mlxipl(α) (−2538/+124) and Mlxipl(β) (−536/+352) promoter sequences were amplified by PCR using mouse genomic DNA and inserted into the pGL4-luciferase reporter vector. miRNA mimics (mmu-miR-33-5p, mmu-miR-132-3p, mmu-miR-212-3p and negative control) were purchased from GenePharma (Shanghai, China). Primer sequences are listed in ESM Table 5. In total, 50 μmol/l miRNA mimics were transfected into cells using Lipofectamine 3000 transfection reagent (Invitrogen), according to the manufacturer’s instructions. Adenoviruses expressing gfp only, Tcf7l2 and Cre have been described previously [7]. For animal experiments, the viruses were purified on a CsCl gradient, dialysed against PBS buffer containing $10\%$ (vol./vol.) glycerol and stored at −80°C.
## Luciferase assay
Luciferase assays were performed to determine the effects of TCF7L2 on the sterol regulatory element binding protein 1 (SREBP1)c, ChREBP/max-like protein X (MLX) transcriptional activities and the miR-33-5p promoter activity in HEPG2 cells (catalogue no. HB-8065; ATCC, USA) and Srebf1 3′UTR promoter activity in HEK293T cells (catalogue no. CRL-3216, ATCC). HEPG2 and HEK293T cells were tested for mycoplasma using the BioMycoX Mycoplasma PCR Detection Kit (catalogue no. D-50; CellSafe, Republic of Korea) and were found to be mycoplasma negative. Promoter activity was measured using a luciferase reporter assay system (catalogue no. E1910; Promega, USA) and normalised to β-galactosidase activity.
## Protein stability assay
HEPG2 cells were transfected with HA-tagged TCF7L2 and Flag-tagged ChREBPα for 48 h and treated with 10 μmol/l MG132 (catalogue no. M1157; AG Scientific, USA) or DMSO (catalogue no. D2650; Sigma-Aldrich) for 3 h.
## Immunoprecipitation and wheat germ agglutinin purification
Total lysates from liver tissues were centrifuged at 16,000 × g for 10 min and the supernatant was extracted. For immunoprecipitation, 3 mg of protein lysates were incubated with carbohydrate response element binding protein (ChREBP) antibody and 30 μl of protein G plus A agarose beads (catalogue no. IP05; Millipore, USA) was added to each sample. For wheat germ agglutinin (WGA) precipitation, 3 mg of protein lysates were incubated with 30 μl of WGA agarose beads (catalogue no. AL-1023; Vector Laboratories), after which beads were eluted in 2× sample buffer without β-mercaptoethanol.
## Tcf7l2-KO in alpha-mouse-liver-12 cell lines
*To* generate Tcf7l2-KO alpha-mouse-liver-12 (AML12) cell lines (catalogue no. CRL-2254; ATCC), we developed a CRISPR/Cas9 system for gene editing using standard methods [26]. The single-guide RNA (sgRNA) sequence was cloned into pSpCas9(BB)-2A-Puro(PX459), which was a gift from F. Zhang (Broad Institute of MIT and Harvard, Cambridge, USA). The target guide RNA (gRNA) sequence was 5′-AGCAATGAACACTTCACCCC-3′ (Tcf7l2 exon 5; Genome Reference Consortium Mouse Reference 39 [GRCm39;]: 19:55,896,921–55,896,940; https://asia.ensembl.org/Mus_musculus/Info/Annotation). The vector expressing gRNA was transfected into AML12 cells and then cells were selected using puromycin. Cells tested negative for mycoplasma contamination. Genomic DNA was extracted from cells using the Exgene Tissue SV Kit (catalogue no. 104-152; GeneAll, Republic of Korea) Gene knockout was verified by PCR and western blot analysis. Cells were transfected with the pGL4-Pklr (−191/+200) promoter and then co-treated with 25 mmol/l glucose and 40 μmol/l OSMI-1 (catalogue no. ab235455; Abcam, UK) for 24 h.
## Chromatin immunoprecipitation
Cross-linking, nuclear isolation and chromatin immunoprecipitation (ChIP) assays were performed on AML12 cell line samples and mouse primary hepatocyte samples, using previously reported methods [27]. The precipitated DNA fragments were analysed by PCR.
## Statistical analysis
Statistical differences between two experimental groups were evaluated by the two-tailed unpaired Student’s t test. One-way ANOVA with Tukey’s multiple comparison test was performed using GraphPad Prism 8.0.1 (GraphPad Software, USA) when comparing three or more groups, as reported in the figure legends. Data are shown as mean±SD or mean±SEM. A p value <0.05 was considered statistically significant.
## Loss of hepatic TCF7L2 exerts lipogenic potential
We found that Tcf7l2 expression was significantly reduced in the liver of individuals with NAFLD and NASH (ESM Fig. 1a). Therefore, to gain new insights into the role of TCF7L2 in NAFLD development, we generated liver-specific Tcf7l2 KO mice (Alb-Cre;Tcf7l2f/f; Fig. 1a,b and ESM Fig. 1b). Essentially, Alb-Cre;Tcf7l2f/f mice exhibited increased fasting glucose levels without changes in body weight or plasma insulin and IGF-1 levels (Fig. 1c–e and ESM Fig. 1c). They also displayed impaired glucose and insulin tolerance (Fig. 1f,h and ESM Fig. 1d). However, there were no changes in insulin concentration patterns during GTT (Fig. 1g) or in islet glucagon-positive alpha cell and insulin-positive beta cell areas (ESM Fig. 1e,f). Additionally, hepatic Tcf7l2 deficiency did not lead to changes in hepatic insulin signalling, RER, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \dot{V}{\mathrm{O}}_2 $$\end{document}V˙O2, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \dot{V}\mathrm{C}{\mathrm{O}}_2 $$\end{document}V˙CO2 and energy expenditure, as compared with wild-type controls (Fig. 1i,j and ESM Fig. 1g). Although there were no significant changes in liver TG content (Fig. 1k), hepatic Tcf7l2 depletion significantly enhanced hepatic expression of lipogenic genes, without altering genes involved in β-oxidation and lipolysis (Fig. 1l). Additionally, there were no changes in Igf1 and Igf1R mRNA expression (ESM Fig. 1h). Fig. 1Metabolic phenotypes of Alb-Cre;Tcf7l2f/f mice on a normal chow diet (NCD). ( a) Strategy for generating Alb-Cre;Tcf7l2f/f (mice with Tcf7l2 liver-specific KO). ( b) Representative western blot showing TCF7L2 expression in the peripheral tissues of 10-week-old Tcf7l2f/f ($$n = 3$$) and Alb-Cre;Tcf7l2f/f ($$n = 3$$) mice. Quantification of TCF7L2 protein levels is also shown. eWAT, epididymal white adipose tissue; WT, wild-type. ( c–e) 10-week-old Tcf7l2f/f ($$n = 9$$) and Alb-Cre;Tcf7l2f/f ($$n = 6$$) mice fed an NCD were assessed for body weight (c), blood glucose levels (d) and plasma insulin levels (e). ( f) GTT (1.5 g/kg body weight) in 10-week-old Tcf7l2f/f ($$n = 9$$) and Alb-Cre;Tcf7l2f/f ($$n = 6$$) mice fed an NCD, with glucose AUC is presented. * $p \leq 0.05$, **$p \leq 0.01$, Alb-Cre;Tcf7l2f/f vs wild-type mice at the same time point, analysed by unpaired Student’s t test. ( g) Plasma insulin concentration during GTT in 10-week-old Tcf7l2f/f ($$n = 9$$) and Alb-Cre;Tcf7l2f/f ($$n = 6$$) mice fed an NCD, with insulin AUC presented. ( h) ITT (1 U/kg body weight) in 10-week-old Tcf7l2f/f ($$n = 9$$) and Alb-Cre;Tcf7l2f/f ($$n = 6$$) mice fed an NCD, with glucose AUC is presented. * $p \leq 0.05$, **$p \leq 0.01$, ***$p \leq 0.001$, Alb-Cre;Tcf7l2f/f vs wild-type mice at the same time point, analysed by unpaired Student’s t test. ( i) Ten-week-old Tcf7l2f/f ($$n = 12$$) and Alb-Cre;Tcf7l2f/f ($$n = 12$$) mice were fasted for 6 h and then injected intraperitoneally with insulin (10 U/kg body weight) or saline for 10 min. Representative western blot showing the effects of hepatic Tcf7l2 depletion on the hepatic insulin signalling pathway. The p-Akt/Akt and phosphorylated glycogen synthase kinase-3 β (p-GSK3β/GSK3β ratios are presented. ( j) RER in 10-week-old Tcf7l2f/f ($$n = 12$$) and Alb-Cre;Tcf7l2f/f ($$n = 11$$) mice fed an NCD. RER was measured hourly in each metabolic chamber using the Oxylet Pro system. Average RER values for light (09:00–21:00 hours) and dark (21:00–09:00 hours) phases in a 12 h/12 h light/dark cycle are presented. ( k, l) Ten-week-old Tcf7l2f/f ($$n = 6$$) and Alb-Cre;Tcf7l2f/f ($$n = 5$$) mice were fasted for 6 h, after which hepatic TG levels (k) and gene expression (l) were analysed. ( l) Quantitative PCR (qPCR) analysis showing mRNA expression of genes involved in lipid metabolism in the liver of mice. The key in (b) also applies to (c–i) (k) and (l). Data in (b), (i) and (l) are presented as mean±SD; data in (c–h) (j) and (k) are presented as mean±SEM. * $p \leq 0.05$, **$p \leq 0.01$, ***$p \leq 0.001$, analysed by t test
## Hepatic Tcf7l2 deficiency contributes to NAFLD development by inducing preferential metabolism of carbohydrates
In rodents, studies have shown that feeding with an HFD or HCD for 22 weeks leads to the development of NAFLD, including non-alcoholic fatty liver (NAFL; simple steatosis) and NASH [28–30]. As with individuals with NAFLD, Tcf7l2 expression was markedly decreased to below-normal levels in the liver of C57BL/6N mice fed an HFD or HCD for 22 weeks (Fig. 2a,b). To assess the role of TCF7L2 in the development of NAFLD, Alb-Cre;Tcf7l2f/f mice were fed an HFD or HCD for 22 weeks (Fig. 2c). Although hepatic TCF7L2 expression was decreased in the liver of wild-type Tcf7l2f/f mice by both HFD and HCD feeding for 22 weeks, hepatic TCF7L2 expression in wild-type Tcf7l2f/f mice was clearly higher compared with that in Alb-Cre;Tcf7l2f/f mice (ESM Fig. 2a). Under HCD conditions, hepatic Tcf7l2 deficiency promoted lipid droplet formation and TG accumulation in the liver, resulting in elevated alanine aminotransferase (ALT) levels (Fig. 2c–e). In contrast, these differences were not observed when mice were fed an HFD (Fig. 2c–e). Consistently, genes responsible for lipogenesis and inflammation in NAFLD/NASH were found to be significantly increased in the liver of Alb-Cre;Tcf7l2f/f mice, as compared with wild-type mice, following HCD feeding, whilst elevation of these genes was less consistently observed following HFD feeding (ESM Fig. 2b,c). However, there were no significant changes in islet size or islet glucagon-positive alpha cell and insulin-positive beta cell areas in mice fed an HFD or HCD (ESM Fig. 2d-g). Representative liver images of Alb-Cre;Tcf7l2f/f mice fed an HCD for 22 weeks are shown in Fig. 2f. In response to 22 weeks of HCD, hepatic Tcf7l2 deficiency caused impairment of glucose and insulin tolerance (Fig. 2g,h). This impeded hepatic insulin signalling, as confirmed by the reduced level of Akt phosphorylation on Ser473 in the liver of Alb-Cre;Tcf7l2f/f mice vs wild-type mice (Fig. 2i). However, it did not affect Akt phosphorylation levels in epididymal white adipose tissue and skeletal muscle (ESM Fig. 2h). Fig. 2Effects of hepatic Tcf7l2 depletion following 22 weeks of HCD and HFD feeding, used as models for diet-induced NAFLD/NASH. ( a) Representative western blot showing TCF7L2, ACC and FAS protein levels in the liver of mice fed an HCD or HFD for 22 weeks. ( b) Graph showing relative protein levels of TCF7L2, ACC and FAS in the liver of mice fed an HCD ($$n = 4$$) or HFD ($$n = 7$$) for 22 weeks compared with a normal chow diet (NCD; $$n = 3$$). * $p \leq 0.05$, **$p \leq 0.01$, 22 weeks vs 0 weeks, analysed by unpaired Student’s t test. ( c–e) Six-week-old Tcf7l2f/f and Alb-Cre;Tcf7l2f/f mice were fed an NCD for 22 weeks ($$n = 4$$ for both genotypes; control groups), or were given an HCD ($$n = 7$$ for both genotypes) or an HFD for 22 weeks (Tcf7l2f/f, $$n = 7$$; Alb-Cre;Tcf7l2f/f, $$n = 5$$) to induce NAFLD progression. Subsequently, frozen liver sections were stained using Oil Red O and H&E (representative images are shown; ×20 magnification; scale bars, 200 μm; c), and hepatic TG levels (d) and plasma alanine aminotransferase (ALT) and aspartate aminotransferase (AST) levels (e) were measured. ( f) Six-week-old Tcf7l2f/f ($$n = 9$$) and Alb-Cre;Tcf7l2f/f mice ($$n = 11$$) were fed an HCD for 22 weeks. Subsequently, intra-abdominal liver images were taken (representative images shown). ( g, h) Six-week-old Tcf7l2f/f ($$n = 9$$) and Alb-Cre;Tcf7l2f/f mice ($$n = 11$$) were fed an HCD for 22 weeks. Following this, a GTT (1.5 g/kg body weight; g) and ITT (1 U/kg body weight; h) were conducted. Glucose AUCs are also presented for both GTT and ITT. * $p \leq 0.05$, **$p \leq 0.01$, ***$p \leq 0.001$, Alb-Cre;Tcf7l2f/f vs wild-type mice at same time point, analysed by unpaired Student’s t test. ( i) Six-week-old Tcf7l2f/f ($$n = 8$$) and Alb-Cre;Tcf7l2f/f ($$n = 9$$) mice were fed an HCD for 22 weeks. Mice were fasted for 6 h and then injected intraperitoneally with insulin (10 U/kg body weight) or saline for 10 min. Presented is a representative western blot showing the effects of hepatic Tcf7l2 deficiency on the hepatic insulin signalling pathway. The p-Akt/Akt ratio is also shown. HSP90, heat shock protein 90. Key in (d) also applies to (e–h). Data in (b) and (i) are presented as mean±SD; data in (d) (e), (g) and (h) are presented as mean±SEM. * $p \leq 0.05$, **$p \leq 0.01$, ***$p \leq 0.001$, analysed by t test
## TCF7L2 does not affect HFD-induced intrahepatic TG accumulation
Compared with an HCD, an HFD did not have as marked an effect on NAFLD development in hepatic Tcf7l2-deficient mice. To investigate whether this could have been owing to saturation of liver lipid content under an HFD diet, we used Alb-Cre;Tcf7l2f/f mice fed an HFD for 4 or 12 weeks. Hepatic Tcf7l2 depletion exacerbated HFD-induced glucose intolerance and insulin intolerance without causing changes in body weight (ESM Fig. 3a–c). However, it did not alter hepatic TG content or plasma TG and NEFA levels (ESM Fig. 3d–g). To specifically address why hepatic Tcf7l2 did not affect HFD-induced TG content, we analysed the expression of genes involved in various lipid-related metabolic pathways (ESM Fig. 3h). Notably, hepatic Tcf7l2 deficiency increased the expression of several lipogenic genes (Srebf1c [encoded by the *Srebf1* gene], Fasn and Acaca), whereas it decreased the expression of several fatty acid transporter genes (Cd36, Slc27a1, Slc27a4 and Fabp3; ESM Fig. 3h). In contrast, adenovirus-mediated expression of hepatic Tcf7l2 inversely regulated many of these genes (Srebf1c, Fasn, Cd36, Slc27a4 and Fabp3) without altering HFD-induced hepatic TG content (ESM Fig. 3i,j). Further, in a cell-autonomous manner, Tcf7l2 KO via Cre deletion of floxed sequences in primary hepatocytes impeded insulin-induced fatty acid uptake, as confirmed by use of a fluorescent fatty acid probe (ESM Fig. 4a), whereas forced expression of Tcf7l2 in C57BL/6N primary mouse hepatocytes promoted it (ESM Fig. 4h). It is well known that HFD suppresses DNL [31]. Accordingly, fatty acids can inhibit insulin-stimulated glucose uptake [32, 33]. Therefore, we hypothesised that hepatic Tcf7l2 deficiency did not affect HFD-induced liver lipid content, despite there being a decrease in fatty acid uptake, because lipogenesis is balanced by a relative increase in glucose-sensing in response to reduced fatty acid uptake. Indeed, Tcf7l2 KO in primary hepatocytes impaired fatty acid-induced inhibition of glucose uptake (ESM Fig. 4b). Expression of L-type pyruvate kinase (L-PK; encoded by the pklr gene) and Pklr, a glucose-sensing marker, increased in the liver of Alb-Cre;Tcf7l2f/f mice vs wild-type mice, even under HFD conditions (ESM Fig. 4c,d).
Fatty acid β-oxidation, another lipid-associated metabolic pathway, was also altered by hepatic Tcf7l2 modulation under HFD conditions (ESM Fig. 3h,j). The AMP-activated protein kinase (AMPK)/acetyl-CoA carboxylase (ACC)/carnitine palmitoyltransferase 1α (CPT1α) axis is a key pathway for regulating fatty acid oxidation [34]. However, neither KO nor overexpression of Tcf7l2 functionally affected fatty acid β-oxidation, as determined by changes in OCR following treatment with fatty acid and/or etomoxir, an irreversible inhibitor of CPT1α (ESM Fig. 4e,i). Further, Tcf7l2 KO or overexpression did not affect AMPK phosphorylation levels following treatment with 5-aminoimidazole-4-carboxamide ribonucleotide (AICAR), a selective activator of AMPK, under HFD conditions (ESM Fig. 4f,g,j,k).
## Hepatic TCF7L2 regulates hepatic DNL in a cell-autonomous fashion
HFD provides excessive exogenous fatty acids, whereas HCD and refeeding mechanistically stimulate endogenous fatty acid synthesis from glucose by postprandial blood glucose and glucose-responsive insulin action, which are increased following carbohydrates consumption. Therefore, we hypothesised that the difference in effects of TCF7L2 on HCD- and HFD-induced fatty liver might originate from differential responses to exogenous fatty acid and glucose/insulin pools. Hepatic TCF7L2 expression decreased in primary hepatocytes treated with palmitic acid, the most common saturated fatty acid in the human body, and in the liver of HFD-fed mice (Fig. 3a). Conversely, hepatic TCF7L2 expression significantly increased following in vitro treatment with glucose/insulin and following in vivo refeeding (Fig. 3b and ESM Fig. 5a–c). To further investigate the relationship between TCF7L2 expression and carbohydrate sensing, we measured hepatic TCF7L2 expression at the same time as measuring the expression of lipogenic factors over varying HCD exposure time. Hepatic TCF7L2 expression increased after 2 and 4 weeks of HCD feeding vs baseline but was downregulated after 8 weeks of HCD feeding (Fig. 3c and ESM Fig. 5d,e). However, expression of the lipogenic factors ACC (encoded by Acaca) and fatty acid synthase (FAS; encoded by Fasn), and hepatic TG levels increased with HCD feeding (Fig. 3c,d and ESM Fig. 5d,e). Eventually, as shown in Fig. 2a, hepatic TCF7L2 expression decreased to below-normal levels after 22 weeks of HCD feeding, at which point excessive intrahepatic fatty acid was expected. Fig. 3Cell-autonomous role of hepatic TCF7L2 in regulating DNL. ( a) Representative western blots showing protein levels of TCF7L2, ACC and FAS in mouse primary hepatocytes treated with BSA or 250 μmol/l palmitic acid (PA) for 24 h (data are representative of $$n = 3$$ independent experiments) and TCF7L2 protein levels in the liver of C57BL/6N mice fed an HFD ($$n = 4$$) or a normal chow diet (NCD; $$n = 4$$) for 12 weeks. ( b) Representative western blots showing protein levels of TCF7L2, ACC and FAS in mouse primary hepatocytes treated with 25 mmol/l glucose (G) and/or 10 nmol/l insulin (I) or saline (control [C]) for 48 h (data are representative of $$n = 3$$ independent experiments), and TCF7L2, ACC, FAS, p-Akt and Akt levels in the liver of 10-week old C57BL/6N mice under ad libitum (ad lib; $$n = 4$$), 24 h fasted ($$n = 4$$) or 24 h refed ($$n = 4$$) conditions. Quantification data are presented in ESM Fig. 5a and ESM Fig. 5c. ( c, d) C57BL/6N mice were fed an HCD for 2, 4 and 8 weeks ($$n = 5$$ per group; for this experiment, at '0 weeks of HCD feeding', $$n = 5$$ mice that had been fed an NCD for 8 weeks were used). Representative western blot showing TCF7L2, ACC and FAS protein levels is shown alongside the quantified relative protein levels (TCF7L2/heat shock protein 90 [HSP90], ACC/HSP90 and FAS/HSP90 ratios) (c; further quantification data, including details of statistical significance, are presented in ESM Fig. 5d). Hepatic TG levels are also presented (d). * $p \leq 0.05$, analysed by one-way ANOVA with Tukey’s post hoc test. ( e) Schematic diagram showing the hypothesised function of hepatocyte TCF7L2 in hepatic lipid metabolism. FA, fatty acid. ( f) Primary hepatocytes were isolated from 10-week-old C57BL/6N mice. Cells were infected with adenovirus expressing Tcf7l2 (Ad-Tcf7l2) or Ad-gfp and treated with 25 mmol/l glucose and 10 nmol/l insulin for 48 h. Subsequently, quantitative PCR (qPCR) analysis of Tcf7l2 and glycolytic (Gck and Pklr) and lipogenic (Acaca and Fasn) genes was conducted ($$n = 3$$ per group). ( g) Primary hepatocytes were isolated from 10-week-old C57BL/6N mice. Cells were infected with Ad-gfp or Ad-Tcf7l2 and treated with 37 kBq of [14C]glucose and 10 nmol/l insulin or $3\%$ (vol./vol.) ethanol in saline (control [Con]) for 48 h. Incorporation of [14C]glucose into TG was measured to determine DNL rate. ( h) Primary hepatocytes were isolated from 10-week-old Tcf7l2f/f mice. Cells were infected with adenovirus expressing Cre (Ad-Cre) or Ad-gfp and treated with 25 mmol/l glucose and 10 nmol/l insulin for 48 h. Subsequently, qPCR analysis of Tcf7l2 and glycolytic and lipogenic gene expression was conducted ($$n = 3$$ per group). ( i) Primary hepatocytes were isolated from 10-week-old Tcf7l2f/f mice. Cells were infected with Ad-gfp or Ad-Cre and treated with 37 kBq of [14C]glucose and 10 nmol/l insulin or $3\%$ (vol./vol.) ethanol in saline (Con) for 48 h. Incorporation of [14C]glucose into TG was measured to determine DNL rate. ( j) Primary hepatocytes were isolated from 10-week-old C57BL/6N mice. Cells were infected with Ad-gfp, adenovirus expressing wild-type Tcf7l2 (Ad-Tcf7l2) or adenovirus expressing Tcf7l2 DN mutant (Ad-Tcf7l2DN) and treated with 25 mmol/l glucose and 10 nmol/l insulin for 48 h. Subsequently, qPCR analysis of Tcf7l2 and glycolytic and lipogenic gene expression was conducted ($$n = 3$$ per group). * $p \leq 0.05$, ***$p \leq 0.001$, analysed by one-way ANOVA with Tukey’s post hoc test. ( k) Primary hepatocytes were isolated from 10-week-old C57BL/6N mice. Cells were infected with Ad-gfp, Ad-Tcf7l2WT or Ad-Tcf7l2DN and treated with 37 kBq of [14C]glucose and 10 nmol/l insulin or $3\%$ (vol./vol.) ethanol in saline (Con) for 48 h. Incorporation of [14C]glucose into TG was measured to determine DNL rate. ** $p \leq 0.01$, ***$p \leq 0.001$, analysed by one-way ANOVA with Tukey’s post hoc test. Data in (c), (d), (g) and (i–k) are presented as mean±SEM; data in (f) and (h) are presented as mean±SD. * $p \leq 0.05$, **$p \leq 0.01$, ***$p \leq 0.001$, analysed by t test, unless stated otherwise Hence, we hypothesised that although excessive increases in intrahepatic fatty acid as a result of long-term HCD feeding can eventually reduce hepatic TCF7L2 expression and activity, hepatic TCF7L2 restrains endogenous fatty acid synthesis to maintain lipid homeostasis (Fig. 3e). Therefore, we investigated whether TCF7L2 could regulate hepatic DNL in a cell-autonomous fashion. Adenovirus-mediated Tcf7l2 expression suppressed the expression of glucose-/insulin-stimulated DNL genes involved in glycolysis (Gck and Pklr) and lipogenesis (Acaca and Fasn) and inhibited DNL in primary hepatocytes, as compared with control cells infected with Ad-gfp (Fig. 3f,g). Conversely, Tcf7l2f/f primary hepatocytes infected with adenovirus expressing Cre (Ad-Cre) exhibited increased DNL genes and promoted DNL, as compared with cells infected with Ad-gfp (Fig. 3h,i). Consistent with previous reports, Tcf7l2 deficiency did not significantly affect hepatic insulin signalling pathway in a cell-autonomous manner (ESM Fig. 5f). Further, to investigate Tcf7l2 functional knockdown whilst avoiding potential redundancy of other TCF protein members, we generated a TCF7L2 dominant-negative (DN) mutant (TCF7L2DN) that lacks the N-terminal β-catenin interaction domain. TCF7L2DN inhibited the suppressive effects of TCF7L2 on DNL-associated genes and DNL in primary hepatocytes (Fig. 3j,k).
## Hepatic Tcf7l2 depletion increases hepatic DNL in a time- and amount-dependent manner following carbohydrate loading
To further assess the physiological function of hepatic Tcf7l2 deficiency in hepatic DNL, Alb-Cre;Tcf7l2f/f mice were fasted and then refed. Compared with wild-type mice, this promoted the expression of DNL genes involved in glycolysis (Slc2a2, Gck and Pklr) and lipogenesis (Acly, Acaca and Fasn) following refeeding in a time-dependent manner (Fig. 4a). *These* genes are shown (in red) in Fig. 4b, which depicts the DNL process. However, there were no changes in the expression of genes involved in lipolysis and β-oxidation upon refeeding (ESM Fig. 6). Importantly, loss of hepatic Tcf7l2 led to increased liver TG content after 24 h of refeeding, but not levels of hepatic glycogen and β-OH, which are glucose products that are dispersed into and metabolised by other pathways during the DNL process (Fig. 4c–e). Additionally, there were no changes in plasma TG and NEFA levels (Fig. 4f,g). These data suggest that the activation of the glycolytic/lipogenic pathway for DNL is the primary factor that affects fat deposition in the liver of Alb-Cre;Tcf7l2f/f mice. Fig. 4Changes in intrahepatic DNL levels in Alb-Cre;Tcf7l2f/f mice following carbohydrate loading. ( a) Ten-week-old Tcf7l2f/f ($$n = 5$$) and Alb-Cre;Tcf7l2f/f ($$n = 7$$) mice were fed regular chow ad libitum (AL; Tcf7l2f/f, $$n = 5$$; Alb-Cre;Tcf7l2f/f, $$n = 7$$), were fasted for 24 h (F; Tcf7l2f/f, $$n = 6$$; Alb-Cre;Tcf7l2f/f, $$n = 7$$), or were fasted for 24 h and refed for either 6 h (6R; Tcf7l2f/f, $$n = 6$$; Alb-Cre;Tcf7l2f/f, $$n = 7$$) or 24 h (24R; Tcf7l2f/f, $$n = 6$$; Alb-Cre;Tcf7l2f/f, $$n = 7$$). ( a) The schematic diagram shows the experimental design. Also shown is quantitative PCR (qPCR) analysis of hepatic expression of glycolytic and lipogenic genes in mouse livers. * $p \leq 0.05$, **$p \leq 0.01$, ***$p \leq 0.001$ Alb-Cre;Tcf7l2f/f vs wild-type mice under the same conditions, analysed by unpaired Student’s t test. ( b) Schematic diagram of the DNL pathway. Red text, DNL-associated genes; blue text, metabolites associated with DNL. G6P, glucose 6-phosphate; TCA, tricarboxylic acid. ( c–g) Ten-week-old Tcf7l2f/f and Alb-Cre;Tcf7l2f/f mice were fed regular chow ad libitum (AL; Tcf7l2f/f, $$n = 5$$; Alb-Cre;Tcf7l2f/f, $$n = 7$$), were fasted for 24 h (F; Tcf7l2f/f, $$n = 6$$; Alb-Cre;Tcf7l2f/f, $$n = 7$$), or were fasted for 24 h and refed for either 6 h (6R; Tcf7l2f/f, $$n = 6$$; Alb-Cre;Tcf7l2f/f, $$n = 7$$) or 24 h (24 R; Tcf7l2f/f, $$n = 6$$; Alb-Cre;Tcf7l2f/f, $$n = 7$$). Levels of hepatic TG (c), hepatic glycogen (d), hepatic β-OH (e), plasma TG (f) and plasma NEFA (g) were measured. ( h–j) Ten-week-old Tcf7l2f/f and Alb-Cre;Tcf7l2f/f mice were fasted for 24 h (F; $$n = 7$$ for both genotypes), were fasted for 24 h and refed with a chow diet for 24 h (24R; Tcf7l2f/f, $$n = 7$$; Alb-Cre;Tcf7l2f/f, $$n = 6$$) or fasted for 24 h and refed an HCR for 24 h (Tcf7l2f/f, $$n = 8$$; Alb-Cre;Tcf7l2f/f, $$n = 6$$). The fasting-feeding cycle was repeated three times at 12 h intervals prior to study. A schematic diagram showing the experimental design is shown alongside qPCR analysis of glycolytic and lipogenic gene expression in mouse liver; *$p \leq 0.05$, **$p \leq 0.01$, ***$p \leq 0.001$ Alb-Cre;Tcf7l2f/f vs wild-type mice under the same conditions, analysed by unpaired Student’s t test (h). Representative western blot showing protein levels of TCF7L2, SREBP1 (full length [FL; 125 kDa] and nuclear [N; 60–70 kDa] forms, observed on the same blot), ChREBP, ACC and FAS in liver of Alb-Cre;Tcf7l2f/f (LKO) and Tcf7l2f/f (WT) mice (i). Hepatic TG levels (j). ( k) The rate of hepatic DNL in Tcf7l2f/f and Alb-Cre;Tcf7l2f/f mice subjected to fasting for 24 h ($$n = 3$$ for both genotypes) or HCR for 12 h ($$n = 4$$ for both genotypes) and injected with [14C]acetate for 1 h. HSP90, heat shock protein 90. Key in (j) also applies to (k). Data in (a) and (h) are presented as mean±SD; data in (c–g), (j) and (k) are presented as mean±SEM. * $p \leq 0.05$, **$p \leq 0.01$, ***$p \leq 0.001$, analysed by t test Dietary carbohydrates are the primary stimulus for hepatic DNL. To further elucidate the carbohydrate dependency of Tcf7l2 deficiency-induced hepatic TG accumulation, we compared normal chow refeeding with high-carbohydrate refeeding (HCR) in Alb-Cre;Tcf7l2f/f mice. Hepatic Tcf7l2 deficiency promoted expression of glycolytic and lipogenic factors and hepatic TG accumulation, which was more evident under HCR conditions compared with normal chow refeeding conditions (Fig. 4h–j). A high-carbohydrate intake markedly enhanced hepatic DNL induced by hepatic Tcf7l2 depletion (Fig. 4k). These data suggest that hepatic Tcf7l2 deficiency can aggravate liver steatosis and that the extent to which this occurs is dependent on carbohydrate intake, owing to increased hepatic DNL with carbohydrate loading.
## SREBP1c and ChREBP are core factors associated with lipid metabolism in hepatic Tcf7l2-deficient mice
To investigate the major pathways and key instigators upregulated in the liver of Alb-Cre;Tcf7l2f/f mice, we analysed mRNA-sequencing data from liver samples from refed mice. In total, 526 genes were upregulated (log2 fold-change cut-off >1.5) and were analysed using GSEA (www.gsea-msigdb.org/gsea/msigdb/mouse/annotate.jsp, accessed on 1 November 2021), based on canonical pathways gene sets derived from the Reactome pathway database (www.gsea-msigdb.org/gsea/msigdb/mouse/genesets.jsp?collection=CP:REACTOME, accessed on 1 November 2021). The lipid metabolism-related gene set was most significantly upregulated (its significance can be more intuitively and integratedly presented as a radial line graph; see Fig. 5a and ESM Table 6). Specifically, the SREBP- and ChREBP-related pathways were identified as major signalling pathways (Fig. 5a). Careful investigation of mRNA-sequencing data showed that specific target genes and commonly shared lipogenic target genes of SREBP1c and ChREBP increased in the liver of Alb-Cre;Tcf7l2f/f mice (Fig. 5b and ESM Table 7). However, despite a reduction in Srebf2 mRNA expression in the liver of Alb-Cre;Tcf7l2f/f vs wild-type mice, the expression of its target cholesterogenic genes was not altered (ESM Fig. 7a,b and ESM Table 8). SREBP1c and ChREBP are the main transcription factors for DNL [35] that regulate target gene transcription by binding to the sterol regulatory element (SRE) and carbohydrate response element (ChoRE) within their target gene promoters, respectively [36]. In the HEPG2 cell line, transfection with TCF7L2 inhibited ChREBP/MLX-induced activation of Pklr promoters containing ChoRE as well as ChREBP/MLX-induced activation of the ChoRE promoter itself (Fig. 5c). Furthermore, TCF7L2 inhibited SREBP1c-induced activation of an Fasn promoter region containing the SRE only as well as SREBP1c-induced activation of the SRE promoter itself (Fig. 5d). Liver X receptors (LXRs) are nuclear receptors that can directly regulate both SREBP1c and ChREBP [35]. However, neither KO nor overexpression of Tcf7l2 in primary hepatocytes affected lipogenic gene expression induced by the LXR agonist T0901317 (ESM Fig. 7c,d). These data suggest that TCF7L2 regulates ChREBP and SREBP1c transcriptional activities in an LXR-independent manner. Fig. 5TCF7L2 regulates hepatic lipid metabolism by targeting SREBP1c and ChREBP pathways. ( a) *Comparative analysis* of mRNA-sequencing data from liver samples from Tcf7l2f/f and Alb-Cre;Tcf7l2f/f mice refed a normal chow diet for 24 h ($$n = 3$$ per group). Reactome enrichment analysis of 526 upregulated genes (log2 fold-change cut-off >1.5) in Alb-Cre;Tcf7l2f/f mice compared with Tcf7l2f/f mice is shown. The radial line graph relates to the data for the *Reactome* gene set. In addition, data for the major pathways relating to the most significantly altered gene set (‘metabolism of lipids’; see ESM Table 6 for complete gene set) are presented. Values in brackets are overlap/input numbers. Rho, Ras homologous; SREBF, sterol regulatory element-binding transcription factor (also known as sterol regulatory element binding protein [SREBP]). ( b) Venn diagram showing SREBP1c and ChREBP target genes in liver samples from Tcf7l2f/f and Alb-Cre;Tcf7l2f/f mice refed a normal chow diet for 24 h ($$n = 3$$ per group). Heatmaps visualise hepatic expression of SREBP1c and ChREBP target genes in mice (see ESM Table 7 for list of genes). ( c, d) Luciferase (Luc) reporter assay data showing the effects of TCF7L2 expression on ChREBPα/MLX-induced activation of 4× ChoRE and Pklr promoters (c) and the effects of TCF7L2 expression on nuclear SREBP1c (nSREBP1c)-induced activation of 6× SRE and Fasn promoters (d) in the HEPG2 cell line ($$n = 3$$ per group). Data in (c) and (d) are presented as mean±SEM. ** $p \leq 0.01$, ***$p \leq 0.001$, analysed by one-way ANOVA with Tukey’s post hoc test
## TCF7L2 promotes ChREBP transcriptional activity by modulating O-GlcNAcylation and ChREBP protein stability
Among the two ChREBP isoforms (ChREBPα and ChREBPβ; both encoded by the *Mlxipl* gene) [37], the 100 kDa protein band corresponding to ChREBPα [35] was found to be significantly increased in the refed state and after 2, 4 and 8 weeks of HCD, but decreased after 22 weeks of chronic HCD (Fig. 6a,b). However, hepatic Tcf7l2 deficiency resulted in increased protein expression of ChREBPα and its target L-PK (encoded by the pklr gene), not only under acute HCR conditions but also following 22 weeks of chronic HCD (Fig. 6c,d). However, there were no changes in mRNA levels and promoter activity of Mlxipl(α) (Fig. 6e,f). Therefore, we investigated whether TCF7L2 regulates ChREBPα protein levels to modulate ChREBPα transcriptional activity. As shown in Fig. 6g, TCF7L2 inhibited ChREBPα/MLX-induced Mlxipl(β) promoter activity. In a set of experiments using the same concentration ratio of Flag-tagged ChREBPα:HA-tagged TCF7L2 used in the promoter activity assay, ChREBPα protein levels appeared to gradually decrease in a TCF7L2 expression-level-dependent manner, without changes in Mlxipl(α) mRNA expression (Fig. 6h). Further, the TCF7L2-mediated reduction in ChREBPα protein expression was restored by treatment with the proteasome inhibitor MG132 (Fig. 6i). Conversely, ChREBP ubiquitination was found to be significantly decreased in the liver of hepatic Tcf7l2-deficient mice under HCR conditions (Fig. 6j and ESM Fig. 8a). O-GlcNAcylation of ChREBP is induced by high glucose levels, blocking its ubiquitination to increase glucose utilisation and lipogenic gene transcription [38]. Loss of hepatic Tcf7l2 significantly increased O-GlcNAcylation of ChREBP under HCR conditions, whereas hepatic Tcf7l2 restoration reduced it to normal levels (Fig. 6k,l and ESM Fig. 8b,c). Consistently, CRISPR/Cas9-mediated Tcf7l2 KO in the AML12 cell line significantly promoted ChREBPα protein expression and the expression and promoter activity of its target gene Pklr in a cell-autonomous manner, under high-glucose conditions (ESM Fig. 9). Indeed, the high-glucose-stimulated promoter activity of the ChREBP target Pklr in Tcf7l2 KO cell lines was suppressed by the O-GlcNAc transferase (OGT) inhibitor OSMI-1 (Fig. 6m). These data suggest that TCF7L2 regulates ChREBP transcriptional activity via O-GlcNAcylation (Fig. 6n). Fig. 6TCF7L2 regulates ChREBP transcriptional activity by modulating O-GlcNAcylation and protein stability of ChREBP. ( a) Representative western blot showing protein levels of ChREBP in the liver of 10-week old C57BL/6N mice under ad libitum feeding (ad lib; $$n = 4$$), 24 h fasted ($$n = 4$$) or 24 h refed ($$n = 4$$) conditions. ( b) C57BL/6N mice were fed a an HCD for 2, 4 and 8 weeks ($$n = 5$$ per group; for this experiment, at ‘0 weeks of HCD feeding’, $$n = 5$$ mice that had been fed a normal chow diet [NCD] for 8 weeks were used) or they were fed an HCD for 22 weeks ($$n = 4$$; for this experiment, at ‘0 weeks of HCD feeding’, $$n = 3$$ mice that had been fed an NCD for 22 weeks were used). Representative western blot showing ChREBP protein levels in mouse livers is shown alongside the ChREBP/heat shock protein 90 (HSP90) ratio. ** $p \leq 0.01$, ***$p \leq 0.001$, analysed by one-way ANOVA with Tukey’s post hoc test. ( c) Ten-week-old Tcf7l2f/f ($$n = 8$$) and Alb-Cre;Tcf7l2f/f ($$n = 6$$) mice were subjected to HCR for 24 h. Representative western blots of protein levels of TCF7L2, ChREBP and L-PK in the liver are shown. ChREBP/HSP90 and L-PK/HSP90 ratios are also shown. ( d) Six-week-old Tcf7l2f/f ($$n = 7$$) and Alb-Cre;Tcf7l2f/f ($$n = 7$$) mice were fed an HCD for 22 weeks. Representative western blots of protein levels of TCF7L2, ChREBP and L-PK in the liver are shown. ChREBP/HSP90 and L-PK/HSP90 ratios are also shown. ( e) Ten-week-old Tcf7l2f/f ($$n = 8$$) and Alb-Cre;Tcf7l2f/f ($$n = 6$$) mice were subjected to HCR for 24 h or 6-week-old Tcf7l2f/f ($$n = 7$$) and Alb-Cre;Tcf7l2f/f ($$n = 7$$) mice were fed an HCD for 22 weeks. Quantitative PCR (qPCR) analysis of mRNA expression levels of Mlxipl(α) in the liver is shown. ( f) Luciferase reporter assay showing the effects of TCF7L2 expression and 25 mmol/l glucose and/or 10 nmol/l insulin treatment on Mlxipl(α) promoter activity in the HEPG2 cell line ($$n = 3$$ replicates). Data are representative of $$n = 3$$ independent experiments. ( g) Luciferase reporter assay showing the effects of TCF7L2 expression on ChREBPα/MLX-induced activation of the Mlxipl(β) promoter in the HEPG2 cell line ($$n = 3$$ replicates). Data are representative of $$n = 3$$ independent experiments. *** $p \leq 0.001$, assessed by one-way ANOVA with Tukey’s post hoc test. ( h) HEPG2 cells were transfected with 500 ng of Flag-tagged ChREBPα and various concentrations (250 ng, 500 ng or 1000 ng) of haemagglutinin (HA)-tagged TCF7L2 for 48 h ($$n = 3$$ replicates). A representative western blot showing changes in Flag-tagged ChREBPα protein levels according to expression levels of HA-tagged TCF7L2 is presented. qPCR analysis of mRNA levels of Tcf7l2 and Mlxipl(α) is also shown. Data are representative of $$n = 3$$ independent experiments. ( i) HEPG2 cells were transfected with HA-tagged TCF7L2 and Flag-tagged ChREBPα for 48 h and treated with 10 μmol/l MG132 or DMSO (vehicle) for 3 h ($$n = 3$$ replicates). A representative western blot showing changes in ChREBPα protein stability by TCF7L2 expression is presented. Quantification of Flag-tagged ChREBPα is also shown. Data are representative of $$n = 3$$ independent experiments. *** $p \leq 0.001$, analysed by one-way ANOVA with Tukey’s post hoc test. ( j) Tcf7l2f/f ($$n = 3$$) and Alb-Cre;Tcf7l2f/f ($$n = 3$$) mice were subjected to HCR for 24 h. Liver protein samples were immunoprecipitated with anti-ChREBP and a representative western blot of protein levels of ubiquitin (Ub), ChREBP, and TCF7L2 is shown. IP, immunoprecipitation; WCL, whole-cell lysate. Quantification data are presented in ESM Fig. 8a. ( k) Tcf7l2f/f ($$n = 3$$) and Alb-Cre;Tcf7l2f/f ($$n = 3$$) mice were subjected to HCR for 24 h. Liver protein samples were immunoprecipitated with anti-WGA and a representative western blot showing protein levels of O-GlcNAcylated ChREBP (ChREBPOG), TCF7L2 and ChREBP is shown. The blots were given various film exposure times to clarify the protein bands (short-term exposure, 1 min; long-term exposure, 5 min). IP, immunoprecipitation; WCL, whole-cell lysate. Quantification data are presented in ESM Fig. 8b. ( l) Tcf7l2f/f and Alb-Cre;Tcf7l2f/f mice were infected with Ad-gfp or adenovirus expressing Tcf7l2 (Ad-Tcf7l2) via the tail vein and then subjected to HCR for 24 h ($$n = 3$$ per group). Liver protein samples were immunoprecipitated with WGA. A representative western blot showing protein levels of O-GlcNAcylated ChREBP (ChREBPOG), TCF7L2 and ChREBP is shown. IP, immunoprecipitation; WCL, whole-cell lysate. Quantification data (including statistical significance) are presented in ESM Fig. 8c. ( m) Tcf7l2 WT (single-guide negative control [sgNC]; Tcf7l2+/+) and Tcf7l2 KO (sgTcf7l2; Tcf7l2−/−) AML12 cell lines were generated using the CRISPR-Cas9 system. Cells were transfected with the pGL4-Pklr (−191/+200) promoter and then co-treated with 25 mmol/l glucose and 40 μmol/l OSMI-1 for 24 h ($$n = 3$$ replicates). Luciferase reporter assay data showing the effects of Tcf7l2 KO and OSMI-1 on Pklr promoter activity under high-glucose conditions is shown. A representative western blot showing Tcf7l2 KO in AML12 cell lines is also shown. Data are representative of $$n = 3$$ independent experiments. * $p \leq 0.05$, **$p \leq 0.01$, ***$p \leq 0.001$, analysed by one-way ANOVA with Tukey’s post hoc test. ( n) Proposed mechanism underlying TCF7L2-induced regulation of ChREBP transcriptional activity. OG, O-linked GlcNAc modification. Key in (e) also applies to (c) and (d). Data in (b), (f), (g), (i) and (m) are presented as mean±SEM; data in (c–e) and (h) are presented as mean±SD. * $p \leq 0.05$, **$p \leq 0.01$, ***$p \leq 0.001$, analysed by t test, unless stated otherwise
## TCF7L2 regulates the transcription of miR-33-5p in the SREBP1c/miR-33-5p axis
Under refeeding conditions, hepatic Tcf7l2 deficiency significantly upregulated Srebf1c mRNA levels (Fig. 7a). However, TCF7L2 did not affect Srebf1c promoter activity induced by insulin, SREBP1c (autoregulation) or the LXR agonist T0901317 (Fig. 7b). Therefore, we investigated whether TCF7L2 regulates miRs that can trigger Srebf1c mRNA decay during the development of hepatic steatosis [39–41]. We found that hepatic Tcf7l2 deficiency decreased the hepatic expression of miR-33-5p, miR-132-3p and miR-212-3p under refeeding conditions (Fig. 7c). We noted that miR-33-5p showed apparently similar expression patterns to Tcf7l2 during periods of HCD feeding and was markedly decreased in the liver and primary hepatocytes of Alb-Cre;Tcf7l2f/f mice fed a chronic HCD (Fig. 7d and ESM Fig. 10 [comparative HCD feeding data for TCF7L2 are shown in Fig. 3c and ESM Fig. 5d,e]). miR-33-5p also had a potent inhibitory effect on the activity of the murine Srebf1 3′UTR, as determined by luciferase reporter assay (Fig. 7e). Similarly, compared with the other miRs investigated, miR-33-5p significantly reduced basal Fasn and Acaca expression and Tcf7l2 deficiency-induced expression of Srebf1c, Fasn and Acaca (Fig. 7f). Consistent with this, miR-33-5p significantly reduced Tcf7l2 deficiency-induced DNL and hepatic TG content (Fig. 7g,h). To further elucidate the relationship between TCF7L2 and miR-33-5p, we tested whether TCF7L2 could transcriptionally regulate miR-33-5p. Careful investigation of promoter sequences showed that the putative TCF-binding element (TBE) site was localised at −1249 to −1243 from the transcriptional start site of the miR-33-5p promoter (Fig. 7i). Indeed, expression of hepatic TCF7L2 enhanced activity of the miR-33-5p promoter, that included a putative TBE site (−1412 to +16), but did not affect the activity of the miR-33-5p promoter without a putative TBE site (−852 to +16; Fig. 7i). TCF7L2 specifically occupied the miR-33-5p promoter region containing the putative TBE site (−1249 to −1243; Fig. 7j). Fig. 7miR-33-5p is a transcriptional target of TCF7L2 in the SREBP1c/miR-33-5p axis. ( a) Quantitative PCR (qPCR) analysis showing Srebf1c mRNA expression in the liver of Tcf7l2f/f ($$n = 6$$) and Alb-Cre;Tcf7l2f/f ($$n = 7$$) mice under refeeding conditions. ( b) Luciferase reporter assay data showing the effects of TCF7L2 expression on Srebf1c promoter activity following induction by insulin, nuclear SREBP1c (nSREBP1c) and the LXR agonist T0901317 in the HEPG2 cell line. ( c) qPCR analysis showing expression levels of miR-33-5p, miR-132-3p and miR-212-3p in the liver of Tcf7l2f/f ($$n = 6$$) and Alb-Cre;Tcf7l2f/f ($$n = 7$$) mice under refeeding conditions. ( d) qPCR analysis showing expression levels of miR-33-5p, miR-132-3p, and miR-212-3p in primary hepatocytes from Tcf7l2f/f ($$n = 3$$) and Alb-Cre;Tcf7l2f/f ($$n = 3$$) mice fed an HCD for 16 weeks. ( e) miR binding sites in the Srebf1 3′UTR are shown. Luciferase reporter assay data showing effects of miR-33-5p, miR-132-3p and miR-212-3p mimics on Srebf1 3′UTR activity in the HEK293T cell line are also presented ($$n = 3$$ replicates). Data are representative of $$n = 3$$ independent experiments. ( f) Tcf7l2 WT (single-guide negative control [sgNC]; Tcf7l2+/+) and Tcf7l2 KO (sgTcf7l2; Tcf7l2−/−) AML12 cell lines were generated using the CRISPR-Cas9 system. qPCR analysis showing mRNA levels of Srebf1c, Fasn and Acaca in sgNC and sgTcf7l2 AML12 cells transfected with miR mimics ($$n = 3$$ replicates). Data are representative of $$n = 3$$ independent experiments. ( g) Tcf7l2 WT (sgNC; Tcf7l2+/+) and Tcf7l2 KO (sgTcf7l2; Tcf7l2−/−) AML12 cell lines were generated using the CRISPR-Cas9 system. sgNC and sgTcf7l2 AML12 cells were infected with Ad-gfp, Ad-miR-33-5p or Ad-miR-$\frac{132}{212}$-3p and treated with $3\%$ (vol./vol.) ethanol in saline (control [Con]) (Ad-gfp, $$n = 3$$ for both genotypes; Ad-miR-33-5p, $$n = 3$$ for both genotypes; Ad-miR-$\frac{132}{212}$-3p, $$n = 3$$ for both genotypes) or were infected with Ad-gfp, Ad-miR-33-5p or Ad-miR-$\frac{132}{212}$-3p and treated with 37 kBq of [14C]glucose and 10 nmol/l insulin (Ad-gfp; $$n = 5$$ for both genotypes; Ad-miR-33-5p, $$n = 6$$ for sgNC and $$n = 5$$ for sgTcf7l2; Ad-miR-$\frac{132}{212}$-3p; $$n = 5$$ for sgNC and $$n = 6$$ for sgTcf7l2) for 48 h. Incorporation of [14C]glucose into TG was measured to determine DNL rate. ( h) Tcf7l2 WT (sgNC; Tcf7l2+/+) and Tcf7l2 KO (sgTcf7l2; Tcf7l2−/−) AML12 cell lines were generated using the CRISPR-Cas9 system. Intracellular TG levels were measured in sgNC and sgTcf7l2 AML12 cells transfected with miR-33-5p mimics ($$n = 4$$ per group). ( i) Schematic diagram showing the location of the putative TBE site. Also presented is luciferase reporter assay data showing the effects of TCF7L2 expression on miR-33-5p promoter activity in the HEPG2 cell line ($$n = 3$$ replicates). Data are representative of $$n = 3$$ independent experiments. ( j) Schematic diagram showing the location of the putative TBE and primers for ChIP assay targeting the miR-33-5p promoter. The bar graph presents ChIP assay data showing occupancy of TCF7L2 over the miR-33-5p promoter in the murine hepatocyte cell line AML12 ($$n = 3$$ replicates). Data are representative of $$n = 3$$ independent experiments. A diagram showing the proposed mechanism of TCF7L2-associated miR-33-5p regulation is also shown. IP, immunoprecipitation. Key in (a) also applies to (c). Data are presented as mean±SD. * $p \leq 0.05$, **$p \leq 0.01$, ***$p \leq 0.001$, analysed by t test
## Restoration of hepatic Tcf7l2 alleviates hepatic steatosis induced by acute and chronic HCD
To further investigate the physiological role of hepatic Tcf7l2 in high-carbohydrate-induced fatty liver, we restored hepatic Tcf7l2 expression in Alb-Cre;Tcf7l2f/f mice under acute and chronic HCD conditions. The restoration of hepatic Tcf7l2 expression in mice subjected to HCR reduced the expression of glycolytic and lipogenic factors to normal levels (Fig. 8a,b). This resulted in the restoration of hepatic DNL and TG levels (Fig. 8c,d). Under chronic HCD conditions, hepatic Tcf7l2 restoration did not cause changes in body weight, adiposity or body composition, and it did not affect the weights of the liver, epididymal white adipose tissue, inguinal white adipose tissue or brown adipose tissues (Fig. 8e–i). Plasma TG and NEFA levels also remained unchanged (ESM Fig. 11a–c). However, similar to the model of acute HCD, hepatic Tcf7l2 restoration significantly recovered protein levels of key lipogenic enzymes (ACC and FAS) and mRNA levels of glycolytic Gck and lipogenic Fasn, and restored chronic HCD-induced hepatic lipid droplet formation and hepatic TG levels as shown by Oil Red O staining of liver sections (Fig. 8j–l and ESM Fig. 11d,e). Indeed, adenovirus-mediated expression of hepatic Tcf7l2 in wild-type C57BL/6N mice fed an HCD for 22 weeks decreased chronic HCD-induced hepatic lipid droplets and TG content, concomitant with reductions in DNL-associated genes, without changes in body weight (ESM Fig. 11f–j). Fig. 8Effects of restoring hepatic Tcf7l2 expression on liver steatosis induced by acute and chronic HCD feeding in Alb-Cre;Tcf7l2f/f mice. ( a–c) Eight-week-old Tcf7l2f/f ($$n = 6$$) and Alb-Cre;Tcf7l2f/f ($$n = 8$$) mice infected with Ad-gfp were fasted (F) for 24 h. Subsequently, Tcf7l2f/f ($$n = 5$$) and Alb-Cre;Tcf7l2f/f mice infected with Ad-gfp ($$n = 8$$) or Ad-Tcf7l2 ($$n = 8$$) were subjected to HCR for 24 h. A schematic diagram showing the experimental design for generation of the acute high-carbohydrate loading mouse model is shown alongside a representative western blot showing TCF7L2, SREBP1 (full length [FL] and nuclear [N] forms), ChREBP, ACC and FAS protein levels in mouse liver (a). Quantitative PCR (qPCR) analysis showing expression levels of Tcf7l2 and glycolytic (Gck and Pklr) and lipogenic (Acaca, Fasn and Scd1) genes in mouse liver (b). Hepatic TG levels in mice (c). ( d) DNL rate in primary hepatocytes isolated from Tcf7l2f/f and Alb-Cre;Tcf7l2f/f mice. Cells were infected with Ad-gfp or Ad-Tcf7l2 and treated with [14C]glucose and 10 nmol/l insulin or $3\%$ (vol./vol.) ethanol in saline (control [Con]) for 48 h. (e–i) Six-week-old Tcf7l2f/f ($$n = 7$$) and Alb-Cre;Tcf7l2f/f ($$n = 15$$) mice were fed an HCD for 22 weeks. At 21 weeks of HCD, Tcf7l2f/f mice were infected with Ad-gfp ($$n = 7$$) and Alb-Cre;Tcf7l2f/f mice were infected with Ad-gfp ($$n = 7$$) or Ad-Tcf7l2 ($$n = 8$$). A schematic diagram showing the experimental design for generation of the chronic high-carbohydrate loading mouse model is shown alongside representative images of livers, epididymal white adipose tissues (eWAT), inguinal white adipose tissues (iWAT) and brown adipose tissues (BAT) from mice (e). Body, fat and lean weights (f). Percentages of fat mass, lean mass and free body fluid (g). Liver weight (h). Weights of eWAT, iWAT and BAT (i). ( j-l) Six-week-old Tcf7l2f/f ($$n = 7$$) and Alb-Cre;Tcf7l2f/f ($$n = 8$$) mice were fed an HCD for 22 weeks. At 21 weeks of HCD, Tcf7l2f/f mice were infected with Ad-gfp ($$n = 7$$) and Alb-Cre;Tcf7l2f/f mice were infected with Ad-gfp ($$n = 4$$) or Ad-Tcf7l2 ($$n = 4$$). A representative western blot showing TCF7L2, ACC and FAS protein levels in mouse liver (j; quantification data are presented in ESM Fig. 11d). Representative H&E and Oil Red O staining of liver sections (×20 magnification; scale bars, 200 μm; k). Hepatic TG levels (l). HSP90, heat shock protein 90. Key in (b) also applies to (c), (d), (h), (i) and (l). Data in (b–d), (f–i) and (l) are presented as mean±SEM. * $p \leq 0.05$, **$p \leq 0.01$, ***$p \leq 0.001$, assessed by one-way ANOVA with Tukey’s post hoc test
### Tcf7l2f/f mice fed an HCD develop a NASH phenotype
To further emphasise the pathological role of hepatic Tcf7l2 deficiency in the liver, we explored whether the absence of hepatic Tcf7l2 exacerbated NAFLD into more severe forms. Mice lacking hepatic Tcf7l2 displayed simple liver steatosis at 16 weeks of HCD feeding and progressive fatty liver at 34 weeks of a long-term HCD (ESM Fig. 12a). In detail, Alb-Cre;Tcf7l2f/f mice fed an HCD for 34 weeks exhibited steatosis and lobular inflammation (F$\frac{4}{80}$-positive area) in the liver (ESM Fig. 12b,c). Although the data are not shown, we observed some hepatocellular ballooning with Mallory–Denk bodies in the liver. Alb-Cre;Tcf7l2f/f mice had elevated NAS, which is the sum of scores for steatosis, lobular inflammation and ballooning (ESM Fig. 12d). Additionally, liver fibrosis, as determined by the proportion of the Sirius red-stained area, was significantly increased upon hepatic Tcf7l2 deficiency (ESM Fig. 12e). Inflammation and fibrosis in the liver of Alb-Cre;Tcf7l2f/f mice was further confirmed by analysis of mRNA and protein levels (ESM Fig. 12f,g).
## Discussion
This study aimed to explore how changes in TCF7L2 expression in the liver affect NAFLD development. We found that Tcf7l2 expression was reduced in the liver of both human and diet-induced mouse models of NAFLD/NASH. To gain new insights into the relationship between reduced hepatic Tcf7l2 expression and NAFLD development, we generated a mouse model with hepatocyte-specific Tcf7l2 deficiency. Several previous studies have attempted to elucidate the role of TCF7L2 in hepatic lipid metabolism using mouse models [42–44]; however, contrasting results have been obtained. Whole-body Tcf7l2 deficiency contributed to a reduction in hepatic lipid content, whereas Tcf7l2DN caused an increase in hepatic lipid content [38–40]. Despite advanced tools for in vitro-derived 3D cell culture systems, including liver organoids that can replace animal models, there are still several limitations to reproducing the in vivo process of diet-induced NAFLD, which is influenced by varying and complex factors. Therefore, we used a mouse model capable of inducing NAFLD in vivo.
Hepatic Tcf7l2 deficiency exacerbated liver steatosis via carbohydrate dependency owing to cell-autonomous increases in DNL. During the DNL process, glucose 6-phosphate can be synthesised by or broken down into glycogen [45], and acetyl-CoA can be converted into or back from ketone bodies [44, 46]. However, hepatic glycogen and β-OH levels did not change in the Alb-Cre;Tcf7l2f/f mice used in our study, as compared with wild-type mice, indicating that the glycolytic/lipogenic pathway may be the key signalling pathway for liver TG accumulation upon Tcf7l2 deletion without interference from other signalling pathways. As a strategy to avoid the potentially redundant or compensatory effects of other TCF members following Tcf7l2 deficiency [8, 44, 47], we performed Tcf7l2DN-mediated functional knockdown studies on hepatic DNL. Adenovirus-mediated Tcf7l2DN expression promoted hepatic DNL in a cell-autonomous manner, similar to that shown in Tcf7l2f/f-derived primary hepatocytes infected with adenovirus expressing Cre.
Excessive exogenous fatty acid uptake is one of the leading contributors to intrahepatic TG accumulation in the pathogenesis of NAFLD. However, hepatic Tcf7l2 deficiency did not increase HFD-induced fat accumulation in the liver. Fatty acids typically inhibit glucose uptake [32, 33]. In the fatty acid/glucose axis, hepatic Tcf7l2 deficiency alleviated inhibition of glucose uptake by fatty acids, resulting in relatively increased glucose uptake compared with controls. Further, this promoted glucose-sensing under HFD conditions, as confirmed by the expression of pklr, a glucose-sensing marker. This might contribute to balanced basal induction of hepatic lipogenesis under HFD conditions. This might be the reason why hepatic Tcf7l2 KO did not alter hepatic lipid content despite the partially decreased fatty acid uptake observed under HFD conditions. Importantly, these data suggest that excessive fatty acid uptake is not the mechanism driving NAFLD following hepatic Tcf7l2 loss.
Hepatic Tcf7l2 expression increased under conditions that stimulate DNL by refeeding with both a standard chow diet or an HCD. Increased hepatic Tcf7l2 expression suppressed hepatic DNL. However, excessively increased intrahepatic fatty acid owing to chronic HCD feeding eventually decreased hepatic TCF7L2 expression and activity. Therefore, the early induction of hepatic TCF7L2 expression by dietary carbohydrates may play an important role in maintaining hepatic lipid homeostasis, although reduced hepatic TCF7L2 expression may indicate a transition to a pathological state.
Chronic HCD feeding reduced ChREBP protein levels, whereas hepatic Tcf7l2 deficiency maintained these. TCF7L2 modulated ChREBP protein content through O-GlcNAcylation, resulting in the regulation of ChREBP transcriptional activity as confirmed by treatment with the OGT inhibitor OSMI-1. ChREBP interacts with OGT and induces O-GlcNAcylation in liver cells [38]. However, TCF7L2 did not transcriptionally regulate OGT, as confirmed by reporter gene assay (data not shown). Therefore, TCF7L2 may be involved in the binding of OGT to ChREBP for regulation of O-GlcNAcylation. Host cell factor 1 (HCF-1) is a ChREBP-interacting DNL protein that recruits OGT to ChREBP, resulting in ChREBP O-GlcNAcylation and activation. TCF7L2 may directly target HCF-1 gene transcription [48].
miR-33-5p plays an important role in Srebf1 mRNA decay during the development of hepatic steatosis [39, 49]. In our study, hepatic miR-33-5p displayed expression patterns similar to Tcf7l2 during HCD feeding and was transcriptionally targeted by TCF7L2. Indeed, miR-33-5p reduced Tcf7l2 deficiency-induced expression and function of Srebf1c. On the other hand, although miR-132-3p and miR-212-3p (other miRs that can cause Srebf1c mRNA decay [40, 41]) were decreased in primary hepatocytes and the liver of hepatic Tcf7l2-deficient mice, and reduced Srebf1 3′UTR luciferase activity, they did not inhibit the expression and function of Srebf1c. These data suggest that miR-33-5p is an important factor for regulating the expression and function of Srebf1c in hepatic Tcf7l2 deficiency-induced liver steatosis.
## Conclusion
Although this study did not model the Tcf7l2 SNP associated with the risk of type 2 diabetes, based on its potential to utilise the excess glucose pool more efficiently, we suggest that TCF7L2 may be a promising regulator of NAFLD associated with dietary carbohydrates and diabetes. Pathologically, hepatic Tcf7l2 deficiency-induced fatty liver progressed to NASH, a severe form of NAFLD, demonstrating the potent effect of TCF7L2 on hepatic DNL. Therefore, Alb-Cre;Tcf7l2f/f mice could be a useful mouse model to investigate NAFLD development and progression under HCD conditions.
## Supplementary Information
ESM(PDF 1.66 MB)
## Authors’ relationships and activities
The authors declare that there are no relationships or activities that might bias, or be perceived to bias, their work.
## Contribution statement
K-JO, K-HB and HailK supervised the experiments. K-JO, K-HB and DSL designed the experiments. DSL, THA and HyunmiK contributed to the experimental preparation and data analysis. K-JO generated liver-specific KO mice. S-HK and C-HL helped generate mice. JKS, JWP, GK and SYO contributed to the evaluation of liver histology and body composition monitoring. S-HK, EJ, T-SH and JSK helped to generate the plasmid construct. HJC, DHH and Y-hL contributed to the preparation and analysis of human liver samples. D-SK and JJ helped with bioinformatics analysis. WKK, E-WL, B-SH and SCL contributed to data interpretation. KH and Y-HL interpreted the physiological meaning of the study. S-HK provided critical reagents, samples and comments. K-JO organised the data and wrote the manuscript. All authors made substantial contributions to editing and revising the manuscript and have approved the final version of the manuscript. K-JO is the guarantor of this work and is responsible for the integrity of the work as a whole.
## References
1. Jin T. **Current understanding on role of the Wnt signalling pathway effector TCF7L2 in glucose homeostasis**. *Endocr Rev* (2016) **37** 254-277. DOI: 10.1210/er.2015-1146
2. Migliorini A, Lickert H. **Beyond association: a functional role for Tcf7l2 in β-cell development**. *Mol Metab* (2015) **4** 365-366. DOI: 10.1016/j.molmet.2015.03.002
3. Shu L, Matveyenko AV, Kerr-Conte J, Cho JH, McIntosh CH, Maedler K. **Decreased TCF7L2 protein levels in type 2 diabetes mellitus correlate with downregulation of GIP- and GLP-1 receptors and impaired beta-cell function**. *Hum Mol Genet* (2009) **18** 2388-2399. DOI: 10.1093/hmg/ddp178
4. Nguyen-Tu MS, Martinez-Sanchez A, Leclerc I, Rutter GA, da Silva Xavier G. **Adipocyte-specific deletion of Tcf7l2 induces dysregulated lipid metabolism and impairs glucose tolerance in mice**. *Diabetologia* (2021) **64** 129-141. DOI: 10.1007/s00125-020-05292-4
5. Geoghegan G, Simcox J, Seldin MM. **Targeted deletion of Tcf7l2 in adipocytes promotes adipocyte hypertrophy and impaired glucose metabolism**. *Mol Metab* (2019) **24** 44-63. DOI: 10.1016/j.molmet.2019.03.003
6. Ip W, Shao W, Chiang YT, Jin T. **The Wnt signalling pathway effector TCF7L2 is upregulated by insulin and represses hepatic gluconeogenesis**. *Am J Physiol Endocrinol Metab* (2012) **303** E1166-E1176. DOI: 10.1152/ajpendo.00249.2012
7. Oh KJ, Park J, Kim SS, Oh H, Choi CS, Koo SH. **TCF7L2 modulates glucose homeostasis by regulating CREB- and FoxO1-dependent transcriptional pathway in liver**. *PLoS Genet* (2012) **8** e1002986. DOI: 10.1371/journal.pgen.1002986
8. Ip W, Shao W, Song Z, Chen Z, Wheeler MB, Jin T. **Liver-specific expression of dominant-negative transcription factor 7-like 2 causes progressive impairment in glucose homeostasis**. *Diabetes* (2015) **64** 1923-1932. DOI: 10.2337/db14-1329
9. Florez JC, Jablonski KA, Bayley N. **TCF7L2 polymorphisms and progression to diabetes in the Diabetes Prevention Program**. *N Engl J Med* (2006) **355** 241-250. DOI: 10.1056/NEJMoa062418
10. Grant SF, Thorleifsson G, Reynisdottir I. **Variant of transcription factor 7-like 2 (TCF7L2) gene confers risk of type 2 diabetes**. *Nat Genet* (2006) **38** 320-323. DOI: 10.1038/ng1732
11. Del Bosque-Plata L, Martínez-Martínez E, Espinoza-Camacho MÁ, Gragnoli C. **The role of TCF7L2 in type 2 diabetes**. *Diabetes* (2021) **70** 1220-1228. DOI: 10.2337/db20-0573
12. Musso G, Gambino R, Pacini G, Pagano G, Durazzo M, Cassader M. **Transcription factor 7-like 2 polymorphism modulates glucose and lipid homeostasis, adipokine profile, and hepatocyte apoptosis in NASH**. *Hepatology* (2009) **49** 426-435. DOI: 10.1002/hep.22659
13. Bril F, Cusi K. **Management of nonalcoholic fatty liver disease in patients with type 2 diabetes: a call to action**. *Diabetes Care* (2017) **40** 419-430. DOI: 10.2337/dc16-1787
14. Xia MF, Bian H, Gao X. **NAFLD and diabetes: two sides of the same coin? Rationale for gene-based personalized NAFLD treatment**. *Front Pharmacol* (2019) **10** 877. DOI: 10.3389/fphar.2019.00877
15. Kim H, Lee DS, An TH. **Metabolic spectrum of liver failure in type 2 diabetes and obesity: from NAFLD to NASH to HCC**. *Int J Mol Sci* (2021) **22** 4495. DOI: 10.3390/ijms22094495
16. Fabbrini E, Magkos F. **Hepatic steatosis as a marker of metabolic dysfunction**. *Nutrients* (2015) **7** 4995-5019. DOI: 10.3390/nu7064995
17. Diraison F, Moulin P, Beylot M. **Contribution of hepatic de novo lipogenesis and reesterification of plasma non esterified fatty acids to plasma triglyceride synthesis during non-alcoholic fatty liver disease**. *Diabetes Metab* (2003) **29** 478-485. DOI: 10.1016/s1262-3636(07)70061-7
18. Donnelly KL, Smith CI, Schwarzenberg SJ, Jessurun J, Boldt MD, Parks EJ. **Sources of fatty acids stored in liver and secreted via lipoproteins in patients with nonalcoholic fatty liver disease**. *J Clin Invest* (2005) **115** 1343-1351. DOI: 10.1172/JCI23621
19. Solinas G, Borén J, Dulloo AG. **De novo lipogenesis in metabolic homeostasis: more friend than foe?**. *Mol Metab* (2015) **4** 367-377. DOI: 10.1016/j.molmet.2015.03.004
20. Kim SH, Kim G, Han DH. **Ezetimibe ameliorates steatohepatitis via AMP activated protein kinase-TFEB-mediated activation of autophagy and NLRP3 inflammasome inhibition**. *Autophagy* (2017) **13** 1767-1781. DOI: 10.1080/15548627.2017.1356977
21. Kleiner DE, Brunt EM, Van Natta M. **Design and validation of a histological scoring system for nonalcoholic fatty liver disease**. *Hepatology* (2005) **41** 1313-1321. DOI: 10.1002/hep.20701
22. Jung E, Seong Y, Jeon B, Kwon YS, Song H. **MicroRNAs of miR-17-92 cluster increase gene expression by targeting mRNA-destabilization pathways**. *Biochim Biophys Acta Gene Regul Mech* (2018) **1861** 603-612. DOI: 10.1016/j.bbagrm.2018.06.003
23. Byun SK, An TH, Son MJ. **HDAC11 inhibits myoblast differentiation through repression of MyoD-dependent transcription**. *Mol Cells* (2017) **40** 667-676. DOI: 10.14348/molcells.2017.0116
24. Mishra AP, Siva AB, Gurunathan C, Komala Y, Lakshmi BJ. **Impaired liver regeneration and lipid homeostasis in CCl**. *Lab Anim Res* (2020) **36** 41. DOI: 10.1186/s42826-020-00076-8
25. Liang W, Menke AL, Driessen A. **Establishment of a general NAFLD scoring system for rodent models and comparison to human liver pathology**. *PLoS One* (2014) **9** e115922. DOI: 10.1371/journal.pone.0115922
26. Ran FA, Hsu PD, Wright J, Agarwala V, Scott DA, Zhang F. **Genome engineering using the CRISPR-Cas9 system**. *Nat Protoc* (2013) **8** 2281-2308. DOI: 10.1038/nprot.2013.143
27. Lee DS, Choi H, Han BS. **c-Jun regulates adipocyte differentiation via the KLF15-mediated mode**. *Biochem Biophys Res Commun* (2016) **469** 552-558. DOI: 10.1016/j.bbrc.2015.12.035
28. Miyao M, Kotani H, Ishida T. **Pivotal role of liver sinusoidal endothelial cells in NAFLD/NASH progression**. *Lab Invest* (2015) **95** 1130-1144. DOI: 10.1038/labinvest.2015.95
29. Muramatsu-Kato K, Itoh H, Kohmura-Kobayashi Y. **Undernourishment in utero primes hepatic steatosis in adult mice offspring on an obesogenic diet; involvement of endoplasmic reticulum stress**. *Sci Rep* (2015) **5** 16867. DOI: 10.1038/srep16867
30. Pompili S, Vetuschi A, Gaudio E. **Long-term abuse of a high-carbohydrate diet is as harmful as a high-fat diet for development and progression of liver injury in a mouse model of NAFLD/NASH**. *Nutrition* (2020) **75-76** 110782. DOI: 10.1016/j.nut.2020.110782
31. Duarte JA, Carvalho F, Pearson M. **A high-fat diet suppresses de novo lipogenesis and desaturation but not elongation and triglyceride synthesis in mice**. *J Lipid Res* (2014) **55** 2541-2553. DOI: 10.1194/jlr.M052308
32. Boden G, Chen X, Ruiz J, White JV, Rossetti L. **Mechanisms of fatty acid-induced inhibition of glucose uptake**. *J Clin Invest* (1994) **93** 2438-2446. DOI: 10.1172/JCI117252
33. Homko CJ, Cheung P, Boden G. **Effects of free fatty acids on glucose uptake and utilization in healthy women**. *Diabetes* (2003) **52** 487-491. DOI: 10.2337/diabetes.52.2.487
34. Turner N, Cooney GJ, Kraegen EW, Bruce CR. **Fatty acid metabolism, energy expenditure and insulin resistance in muscle**. *J Endocrinol* (2014) **220** T61-T79. DOI: 10.1530/JOE-13-0397
35. Linden AG, Li S, Choi HY. **Interplay between ChREBP and SREBP-1c coordinates postprandial glycolysis and lipogenesis in livers of mice**. *J Lipid Res* (2018) **59** 475-487. DOI: 10.1194/jlr.M081836
36. Wang Y, Viscarra J, Kim SJ, Sul HS. **Transcriptional regulation of hepatic lipogenesis**. *Nat Rev Mol Cell Biol* (2015) **16** 678-689. DOI: 10.1038/nrm4074
37. Abdul-Wahed A, Guilmeau S, Postic C. **Sweet sixteenth for ChREBP: established roles and future goals**. *Cell Metab* (2017) **26** 324-341. DOI: 10.1016/j.cmet.2017.07.004
38. Guinez C, Filhoulaud G, Rayah-Benhamed F. **O-GlcNAcylation increases ChREBP protein content and transcriptional activity in liver**. *Diabetes* (2011) **60** 1399-1413. DOI: 10.2337/db10-0452
39. Horie T, Nishino T, Baba O. **MicroRNA-33 regulates sterol regulatory element-binding protein 1 expression in mice**. *Nat Commun* (2013) **4** 2883. DOI: 10.1038/ncomms3883
40. Hu Z, Shen WJ, Cortez Y. **Hormonal regulation of microRNA expression in steroid producing cells of the ovary, testis and adrenal gland**. *PLoS One* (2013) **8** e78040. DOI: 10.1371/journal.pone.0078040
41. Li Y, Zhang J, He J, Zhou W, Xiang G, Xu R. **MicroRNA-132 cause apoptosis of glioma cells through blockade of the SREBP-1c metabolic pathway related to SIRT1**. *Biomed Pharmacother* (2016) **78** 177-184. DOI: 10.1016/j.biopha.2016.01.022
42. Yang H, Li Q, Lee JH, Shu Y. **Reduction in Tcf7l2 expression decreases diabetic susceptibility in mice**. *Int J Biol Sci* (2012) **8** 791-801. DOI: 10.7150/ijbs.4568
43. Boj SF, van Es JH, Huch M. **Diabetes risk gene and Wnt effector Tcf7l2/TCF4 controls hepatic response to perinatal and adult metabolic demand**. *Cell* (2012) **151** 1595-1607. DOI: 10.1016/j.cell.2012.10.053
44. Tian L, Shao W, Ip W, Song Z, Badakhshi Y, Jin T. **The developmental Wnt signalling pathway effector β-catenin/TCF mediates hepatic functions of the sex hormone estradiol in regulating lipid metabolism**. *PLoS Biol* (2019) **17** e3000444. DOI: 10.1371/journal.pbio.3000444
45. Hengist A, Koumanov F, Gonzalez JT. **Fructose and metabolic health: governed by hepatic glycogen status?**. *J Physiol* (2019) **597** 3573-3585. DOI: 10.1113/JP277767
46. Wakil SJ, Abu-Elheiga LA. **Fatty acid metabolism: target for metabolic syndrome**. *J Lipid Res* (2009) **50** S138-S143. DOI: 10.1194/jlr.R800079-JLR200
47. Takamoto I, Kubota N, Nakaya K. **TCF7L2 in mouse pancreatic beta cells plays a crucial role in glucose homeostasis by regulating beta cell mass**. *Diabetologia* (2014) **57** 542-553. DOI: 10.1007/s00125-013-3131-6
48. Lane EA, Choi DW, Garcia-Haro L, Levine ZG, Tedoldi M, Walker S. **HCF-1 regulates de novo lipogenesis through a nutrient-sensitive complex with ChREBP**. *Mol Cell* (2019) **75** 357-371. DOI: 10.1016/j.molcel.2019.05.019
49. Pan JH, Cha H, Tang J. **The role of microRNA-33 as a key regulator in hepatic lipogenesis signalling and a potential serological biomarker for NAFLD with excessive dietary fructose consumption in C57BL/6N mice**. *Food Funct* (2021) **12** 656-667. DOI: 10.1039/d0fo02286a
|
---
title: Scar/WAVE has Rac GTPase-independent functions during cell wound repair
authors:
- Mitsutoshi Nakamura
- Justin Hui
- Viktor Stjepić
- Susan M. Parkhurst
journal: Scientific Reports
year: 2023
pmcid: PMC10036328
doi: 10.1038/s41598-023-31973-2
license: CC BY 4.0
---
# Scar/WAVE has Rac GTPase-independent functions during cell wound repair
## Abstract
Rho family GTPases regulate both linear and branched actin dynamics by activating downstream effectors to facilitate the assembly and function of complex cellular structures such as lamellipodia and contractile actomyosin rings. Wiskott-Aldrich Syndrome (WAS) family proteins are downstream effectors of Rho family GTPases that usually function in a one-to-one correspondence to regulate branched actin nucleation. In particular, the WAS protein Scar/WAVE has been shown to exhibit one-to-one correspondence with Rac GTPase. Here we show that Rac and SCAR are recruited to cell wounds in the Drosophila repair model and are required for the proper formation and maintenance of the dynamic actomyosin ring formed at the wound periphery. Interestingly, we find that SCAR is recruited to wounds earlier than Rac and is still recruited to the wound periphery in the presence of a potent Rac inhibitor. We also show that while *Rac is* important for actin recruitment to the actomyosin ring, SCAR serves to organize the actomyosin ring and facilitate its anchoring to the overlying plasma membrane. These differing spatiotemporal recruitment patterns and wound repair phenotypes highlight the Rac-independent functions of SCAR and provide an exciting new context in which to investigate these newly uncovered SCAR functions.
## Introduction
Cells encounter physical stresses daily leading to breaks in their cortex that must be rapidly repaired to maintain cell integrity and function1–4. One of the major features of cell wound repair is the assembly of a Rho family GTPase-dependent actomyosin ring at the wound periphery that attaches the cortical cytoskeleton to the overlying plasma membrane, followed by its dynamic translocation inward to pull the cell cortex breach closed5–10.
The three major Rho family GTPases—Rho, Rac, and Cdc42—are recruited to wounds in spatially and temporally distinct patterns to carry out specific functions during the repair process7–9,11. This dynamic recruitment of Rho family GTPases to the wound site varies somewhat among cell wound repair model systems. In the Xenopus model, all three GTPases are immediately recruited to the wound where they form two concentric rings with an interior ring consisting of Rho and an outer ring containing both Cdc42 and Rac8,9,12. In the Drosophila model, Rho GTPases are also rapidly recruited to wounds in concentric rings, but they exhibit both spatial and temporal differences7,13. Rho1 accumulates first at 30 s post-wounding, and similar to that observed in the Xenopus model, becomes enriched in a ring inside of and just overlapping with the inner edge of the actin ring. Cdc42 accumulates next (30–60 s post-wounding), whereas Rac 1 and Rac 2 accumulate last (60–90 s post-wounding). While both Cdc42 and Rac recruitment overlap with the actin ring, Rac also shows an additional broad region of slightly elevated accumulation overlapping the actin halo region.
While all three Rho family GTPases are needed to achieve proper cell wound repair, their functions also vary somewhat between cell wound repair models. In the Xenopus model, RhoA is needed for cortical flow, whereas Cdc42 regulates actin ring formation, and both are involved in actin ring translocation8,9,12. Rac function in this model has not been reported. In the Drosophila model, Rac proteins are necessary for actin recruitment to the wound edge, Cdc42 is needed to stabilize the actin ring at the wound periphery, and Rho1 is involved in actomyosin ring assembly/stabilization8,13.
Rho family GTPases are known to regulate actin dynamics through their regulation of linear and branched actin nucleation factors14–19. One such family of actin nucleation factors, the Wiskott-Aldrich Syndrome (WAS) family (WASp, Scar/WAVE, WASH), works with the Arp$\frac{2}{3}$ complex to promote branched actin nucleation15,19–25. In the context of cell wound repair, WAS family proteins have been shown to contribute to actin filament orientation and nucleate branched actin to serve as a scaffold to assemble and maintain the contractile actomyosin cable at the wound periphery10,19.
As Rho family GTPases usually exhibit a one-to-one correspondence with WAS family proteins (Cdc42 > WASp; Rac > Scar/WAVE; Rho1 > Wash)22,26–36, we expected Rac and SCAR to show similar recruitment to wounds and repair phenotypes. Surprisingly, we find that SCAR is recruited to wounds prior to Rac, does not require Rac for its recruitment to wounds, and exhibits Rac-independent repair phenotypes, suggesting that Rac and SCAR are functioning outside of their usual relationship in the cell wound repair context.
## SCAR is recruited to cell wounds before Rac GTPase
In the Drosophila cell wound repair model, actin is recruited to cell wounds in a dense ring around the wound periphery and in a less dense actin halo at 60 ± 6.3 s post-wounding (Fig. 1A-A’,H)37. Wounds were generated by laser ablation on the lateral side of nuclear cycle 4–6 Drosophila syncytial embryos expressing an actin reporter (sGMCA or sChMCA;38) (see “Methods”). Consistent with its requirement for cell wound repair, *Rac is* recruited to the wound edge. Wounding embryos expressing fluorescently tagged Rac1 (ChFP-Rac1) or Rac2 (GFP-Rac2) under the control of their endogenous promoters results in the strong accumulation of these proteins beginning at 60 ± 6.3 and 78 ± 4.9 s post-wounding, respectively, in a ring encircling the wound and in a less intense concentric ring corresponding to the actin ring and halo regions (Fig. 1B,C,E,F,H; Table 1)7,13.Figure 1SCAR is recruited to cell wounds faster than Rac1 or Rac2. Confocal projection images of embryos expressing an actin reporter (sGMCA; A-A’) or co-expressing mCherry-Rac1 and sGMCA (B), GFP-Rac2 and sStMCA (C), and GFP-SCAR and sStMCA (D). Times post-wounding are indicated. UW = unwounded. Arrows indicate recruitment to the wound periphery. ( E–G) Fluorescence intensity (arbitrary units) profiles across the wound area over time for the images shown in (B–D), respectively. ( H) Dotplot of initial actin, Rac1, Rac2, and SCAR recruitment to wounds. Scale bars: 20 μm. Table 1Time when each reporter is first recruited to the wound periphery. GenotypeMean (s)SEMp-value to SCARp-value to sGMCAsChMCA606.3–nssGMCA606.3 < 0.01–Rac1606.3 < 0.01–Rac2784.9 < 0.0001–SCAR333––Statistical tests were performed using Kruskal–Wallis test.
As Rho family GTPases usually exhibit a one-to-one correspondence with WAS family proteins (Rac1 > Scar/WAVE), we expected SCAR to show similar spatial and slightly delayed temporal recruitment to wounds as Rac. Wounding embryos expressing GFP-tagged SCAR (GFP-SCAR) under the control of its endogenous promoter resulted in the recruitment of SCAR to the wound periphery (Fig. 1D,G,H; Table 1). Surprisingly, SCAR is recruited to wounds earlier (33 ± 3 s post-wounding) than that observed for Rac1/Rac2 (60 s and 78 s post-wounding, respectively) (Fig. 1E,G,H; Table 1) and only overlapping the intense actin ring encircling the wound (i.e., absent from the actin halo region). Thus surprisingly, Rac1/Rac2 and SCAR recruitment to wounds is both temporally distinct and only partially overlapping spatially, suggesting that SCAR has Rac-independent functions in the context of cell wound repair.
## SCAR recruitment to cell wounds does not require Rac activity
To further investigate the relationship between Rac and SCAR, we asked if recruitment of SCAR to cell wounds requires Rac. As Drosophila has three *Rac* genes (Rac1, Rac2, and Mtl), we modulated Rac activity by treating embryos with the potent Rac inhibitor NSC 237667,39. We examined GFP-SCAR recruitment to wounds in embryos injected with NSC 23766 (Fig. 2). Consistent with their different temporal recruitment patterns, GFP-SCAR was still recruited to wounds in these Rac inhibited embryos. The GFP-SCAR that is recruited to cell wounds does not form a well-defined actin ring, likely due to reduced actin recruitment to wounds upon Rac inhibition. Our results indicate that SCAR function is Rac-independent during cell wound repair. Figure 2SCAR is recruited to cell wounds in the absence of Rac activity. ( A-A””) Confocal projection images of buffer-injected embryos co-expressing GFP-SCAR and an actin reporter (sStMCA) at the time points indicated. ( B-B””) Fluorescence intensity (arbitrary units) profiles across the wound area in (A-A””), respectively. ( C–C””) Confocal projection images of NSC 23766-injected embryos co-expressing GFP-SCAR and an actin reporter (sStMCA) at the time points indicated. ( D-D””) Fluorescence intensity (arbitrary units) profiles across the wound area in (C–C””), respectively. ( E) Dotplot of GFP-SCAR recruitment to wounds in control (buffer-injected) and NSC 23766-injected embryos at 240 s post-wounding. Scale bars: 20 μm.
## Knockdown of Rac or SCAR results in distinct defects in wound healing dynamics
All three Rho family GTPases are required non-redundantly for cell wound repair. In particular, *Rac is* needed for the recruitment of actin to the wound edge, while Rho1 and Cdc42 are crucial for the formation and stability of the actin ring7–9,13. Interestingly, some actin organization persisted at the wound edge in embryos where Rho1 or Diaphanous (Dia; Drosophila formin protein and Rho1 downstream effector) were knocked down7, consistent with the differing roles for Rac and/or Cdc42 in the repair process.
We compared actin dynamics between embryos injected with NSC 23766 and SCAR knockdowns (Fig. 3; Table 2). In control (buffer injected) embryos, actin accumulates similarly to wildtype: in a highly enriched actomyosin ring bordering the wound edge and an elevated actin halo encircling the actin ring (Fig. 3A-C, J-L; Table 3; Video 1). As shown previously, Rac inhibition results in severely reduced recruitment of actin to the wound edge such that an actin ring is not formed, and significantly slower wound closure (control: 7.15 ± 0.25 μm2/s; NSC 23766: 5.62 ± 0.34 μm2/s, $p \leq 0.05$) (Fig. 3D–F,J–N; Table 3; Video 1)7. Rac inhibition also results in overexpansion of wounds (control: 1.74 ± 0.02 fold; NSC 23766: 2.59 ± 0.09 fold, $p \leq 0.0001$) (Fig. 3D-D’,K; Table 3). These phenotypes would be consistent with a role for branched actin nucleation factors as downstream effectors of Rac (and Cdc42) in this process. Figure 3Rac and SCAR knockdowns exhibit different cell wound repair phenotypes. ( A–I) Confocal projection images of wounds generated in embryos expressing an actin marker (sGMCA) in control (buffer only; A–C), NSC 23766 injected (potent Rac inhibitor; D–F), or SCAR RNAi knockdowns (G–I). Actin ring and halo are indicated in (A). ( A’,D’,G’) Kymographs across the wound area in (A,D,G), respectively. Wound expansion is noted with yellow lines; actomyosin ring formation with red arrows; and actomyosin ring disassembly with yellow arrows. ( B,E,H) XY projection image at 0–150 s post wounding showing cortical flow of actin to the wound edge. ( C,F,I) Vector maps from PIV (Particle Image Velocimetry) analysis depicting actin flow from 60 to 90 s for (A,B,D,E,G,H) respectively. ( J) Quantification of the wound area over time for control (buffer injected), NSC 23766 injected, and SCAR RNAi knockdowns. ( K–N) Quantification of fold wound expansion (K), wound contraction rate (L), actin ring width (M), and actin ring intensity (N). Black line and error bars represent mean ± SEM. Red dotted line and square represent mean ± $95\%$ CI from control. Kruskal–Wallis test (K–L) and Welch’s t-test (M–N) were performed with * is $p \leq 0.05$, ** is $p \leq 0.01$, *** is $p \leq 0.001$, **** is $p \leq 0.0001$, and ns is not significant. Scale bars: 20 μm. Table 2Knockdown efficiency of the SCAR RNAi line used in this study. SampleATPα SignalSCAR SignalNormalized SCAR Signal% KnockdownControl* (rep1)34,991.8630,755.07––SCAR RNAi (rep1)15,$017.691914.484460.8186\%$Control* (rep 2)39,264.1742,224.32––SCAR RNAi (rep 2)23,$228.424102.686934.9784\%$*Control refers to vermillion (unrelated gene) knockdown. Table 3Actin ring dynamics in the three genotypes shown in Fig. 3.Control*NSC 23766SCAR RNAiMeanSEMp-value to controlMeanSEMp-value to controlp-value to SCAR RNAiMeanSEMp-value to controlp-value to NSC 23766Wound Fold Expansion1.740.018–2.590.09 < 0.0001ns2.450.11 < 0.001nsContraction Rate (μm2/sec)7.150.25–5.620.34 < 0.05 < 0.00019.120.62 < 0.05 < 0.0001Actin Ring Width (μm)5.340.29–––––3.430.26 < 0.0001–Actin Ring Intensity2.430.13–––––1.340.08 < 0.0001–*Control refers to vermillion (unrelated gene) knockdown. Statistical tests were performed using Kruskal–Wallis test.
SCAR has been shown to contribute to actin filament orientation during actomyosin ring formation and translocation40, which would necessarily occur after actin recruitment to the wound periphery. Consistent with this and its Rac-independent recruitment to wounds, SCAR knockdowns exhibit different wound repair dynamics phenotypes from that of Rac inhibition. SCAR knockdowns also result in wound over-expansion (2.45 ± 0.11 fold, $p \leq 0.001$), but they exhibit a faster wound closure rate (9.12 ± 0.62 μm2/sec) than both control ($p \leq 0.05$) and Rac inhibited ($p \leq 0.0001$) embryos (Fig. 3G-G’, J-L; Table 3; Video 1). Importantly, unlike Rac inhibition, actin is recruited to the wound edge in SCAR knockdowns (Fig. 3H–I; Table 3). Despite this actin recruitment, the actomyosin ring that is formed is not as tightly organized as that in wildtype (Fig. 3G-H,M–N; Table 3), and the actomyosin ring undergoes premature disassembly (Fig. 3G’; Table 3).
## Discussion
While Rac and SCAR are both needed for proper cell wound repair, they are doing so outside of their “*Rac is* the canonical activator of SCAR” relationship in this context. We find that *Rac is* necessary for actin recruitment to cell wounds, whereas SCAR is recruited to cell wounds in the absence of Rac activity and affects actin ring organization and contraction.
Scar/WAVE forms a heteropentameric complex with the Wave Regulatory Complex (WRC) composed of Sra1/PIR121, Nap1, Abi, and HSPC30025,28,41. Scar/WAVE, along with the Arp$\frac{2}{3}$ complex, can generate branched actin networks26,27,42,43. The Scar/WAVE–WRC complex is thought to be trans-inhibited44–46. Rac binds to the WRC to relieve this trans-inhibition, allowing Scar/WAVE to work with Arp$\frac{2}{3}$ to nucleate branched actin27–29,44. Rac does not regulate Scar/WAVE by binding directly to it, but rather interacts with it through binding to the Sra1 subunit of its WRC29. Actin is recruited to the wound periphery in SCAR knockdowns, but does not assemble into a robust actomyosin ring. Thus, in the context of cell wound repair, SCAR could bypass the need for Rac by a currently unknown mechanism such that it can use its branched actin nucleation activity. In this scenario, SCAR may need to be present at the wound edge quickly to put a branched actin scaffolding in place, such that additional actin can be organized into a robust ring upon its arrival at the wound edge. Alternatively, SCAR could be using an, as yet undescribed, alternate biochemical activity when at the wound edge. For example, in addition to its branched actin nucleation activity, the related Wiskott-Aldrich Syndrome family member WASH has been shown to bundle actin, bundle microtubules, and crosslink them35.
Surprisingly, despite the disruption of the actin ring, wounds in SCAR knockdowns close faster than in controls. A recent study showed that actomyosin contraction force is dependent on F-actin architectures47. While branched actin nucleation stabilizes F-actin networks by increasing their density and connectivity, this stabilization reduces actomyosin contractility by limiting movement of the myosin motor on F-actin. The three major branched actin nucleators (WASp, SCAR, and WASH) are required to form a robust actomyosin ring during cell wound repair in the Drosophila model19. While the actomyosin ring is disrupted in SCAR knockdowns, WASp and WASH could still generate branched F-actin to support the formation of an actomyosin ring at the wound edge. This disrupted actomyosin ring could result in increased actomyosin contractility due to its reduced F-actin network density and connectivity, thereby leading to the faster wound closure in SCAR knockdowns.
While Rac has known effectors other than Scar/WAVE that could work downstream to mediate its role in actin recruitment to wounds (cf.17,48,49), less is known about Scar/WAVE regulation outside of its activation through Rac GTPases. Strikingly, our data suggest that SCAR localization and its activity are not always dependent on Rac GTPase, which raises new mechanistic questions for SCAR/WAVE function. What protein(s) regulate SCAR recruitment to the cell cortex without Rac GTPase? While active Rac GTPase regulates SCAR activity through binding to Sra1 (WRC subunit), what other proteins can change the conformation of SCAR to release its VCA domain to promote branched actin nucleation? The linear actin nucleation factor Diaphanous (Dia) has been shown to function upstream of SCAR to regulate WRC localization and activity during Drosophila myoblast fusion50,51. Consistent with this possible regulation, *Dia is* rapidly recruited to the wound edge at 30 s post wounding, similar to that of SCAR40. SCAR may also depend on another Rho family GTPase. Cip4, a Cdc42 downstream effector, has been shown to function upstream of SCAR via associating with the WRC to control Dynamin-dependent cell polarization in the Drosophila wing52. Interestingly, recent studies have also implicated other means of activating the WRC in specific contexts such as lamellipodia formation through interaction with factors including Arf-family GTPases (Arf1, Arf6), other Rho family GTPases (Cdc42, RhoG), various kinases, phospholipids, or membrane receptors25,31,53–57. The identification of these instances of Rac-independent regulation for Scar/WAVE provides exciting new entry points for investigating the upstream control of this essential branched actin nucleation promoting factor.
## Fly stocks and genetics
Flies were cultured and crossed at 25 °C on yeast-cornmeal-molasses-malt extract medium. Flies used in this study are: ChFP-Rac1 (BDSC #76266)13, GFP-Rac2 (BDSC #52286)7; SCAR-GFP40, Vermillion RNAi (BDSC #50641; TRiP.HMC03041), and SCAR RNAi (BDSC #51803; TRiP.HMC0336). RNAi lines were driven using the maternally expressed GAL4-UAS driver, Pmatalpha-GAL-VP16V37 (BDSC #7063).
An actin reporter, sGMCA (spaghetti squash driven, moesin-alpha-helical-coiled and actin binding site fused to GFP) reporter38, the mCherry fluorescent equivalent, sChMCA (BDSC #35520), or the mScarlet-i fluorescent equivalent, sStMCA (BDSC #90928)58, was used to follow wound repair dynamics of the cortical cytoskeleton.
In this study, we used an actin reporter + maternal GAL4 driver + vermilion RNAi (unrelated fly RNAi) + injection buffer as the control.
Localization patterns and mutant analyses were performed at least twice with independent genetic crosses and ≥ 10 embryos were examined. Images representing the average phenotype were selected for figures.
## Western Blotting
*To* generate embryo whole cell lysates, 10 nuclear cycles 4–6 embryos were collected, dechorionated, and then homogenized in 2X sample buffer (125 mM Tris–Cl pH 6.8, $4\%$ SDS, $0.1\%$ Bromophenol blue, $20\%$ glycerol). Western blotting was performed according to standard procedures using anti-SCAR (P1C1, 1:10)59 antibodies, with anti-ATP5A (15H4C4; 1:50,000; Abcam) for the loading control.
## Embryo handling and preparation
Nuclear cycle 4–6 embryos were collected for 30 min at 25 °C and harvested at room temperature (22 °C). Collected embryos were dechorionated by hand, mounted onto No. 1.5 coverslips coated with glue, and covered with Series 700 halocarbon oil (Halocarbon Products Corp.) as previously described37.
## Drug injections
The pan Rac GTPase inhibitor NSC 23766 (50 mM; Tocris Bioscience) was injected into NC4-6 staged Drosophila embryos, incubated at room temperature (22 °C) for 5 min, and then subjected to laser wounding. NSC23766 was prepared in injection buffer (5 mM KCl, 0.1 mM NaP pH6.8). Injection buffer alone was used as the control.
## Laser wounding
All wounds were generated using a pulsed nitrogen N2 micropoint laster (Andor Technology Ltd.) set to 435 nm and focused on the lateral surface of the embryo. An 18 × 18 μm circular region was set as the target site along the lateral midsection of the embryo, and ablation was controlled by MetaMorph software (Molecular Devices). Average ablation time was less than 3 s and time-lapse image acquisition was initiated immediately after ablation.
## Live image acquisition
All live imaging was performed at room temperature with the following microscope:Revolution WD systems (Andor Technology Ltd.) mounted on a Leica DMi8 (Leica Microsystems Inc.) with a 63x/1.4 NA objective lens under the control of MetaMorph software (Molecular devices). Images were captured using 488 nm and/or 561 nm lasers with a Yokogawa CSU-W1 confocal spinning disk head attached to an Andor iXon Ultra 897 EMCCD camera (Andor Technology Ltd.). Time-lapse images were acquired with 17–20 µm stacks/0.25 µm steps. Images were acquired every 30 s for 15 min and then every 60 s for 15 min. UltraVIEW VoX Confocal Imaging System (Perkin Elmer, Waltham, MA, USA) mounted on a Nikon Eclipse Ti (Nikon Instruments, Melville NY,USA) with a 60x/1.4 NA objective lens under the control of Volocity software(v.5.3.0, Perkin-Elmer). Images were captured using 488 nm and/or 561 nm lasers with a Yokogawa CSU-X1 confocal spinning disk head attached to a Hamamatsu C9100-13 EMCCD camera (Perkin-Elmer, Waltham,MA,USA). Time-lapse images were acquired with 17–20 µm stacks/0.25 µm steps. Images were acquired every 30 s for 15 min and then every 60 s for 15 min.
## Image processing, analysis, and quantification
Image processing was performed using FIJI software60. In all images, the top side is anterior and the bottom side is posterior of embryos. Kymographs were generated using the crop feature to select ROIs of 5.3 × 94.9 µm. Wound area was manually measured using Fiji and the values were imported into Prism 8.2.1 (GraphPad Software Inc.) to construct corresponding graphs.
For fluorescent lineplots, the mean fluorescence profile intensities were calculated from 51 equally spaced radial profiles anchored at the center of the wound, swept from 0° to 180°40. Radial profiles of 301-pixel diameter were used. Fluorescence intensity profiles were calculated and averaged using an in house code using MATLAB R2020b (MathWorks) (available at: https://github.com/FredHutch/wound_radial_lineplot), then plotted using MATLAB R2020b. For dynamic lineplots, we generated fluorescent profile plots from each timepoint and then concatenated them. The lines represent the averaged fluorescent intensity and gray area is the $95\%$ confidence interval.
Quantification of the width and average intensity of actin ring, wound expansion, and closure rate was performed as follows: the width of actin ring was calculated with two measurements, the feret diameters of the outer and inner edge of actin ring at 90 s post-wounding. Using these measurements, the width of actin ring was calculated with (outer feret diameter − inner feret dimeter)/2. The average intensity of actin ring was calculated with two measurements. Instead of measuring feret diameters, we measured area and integrated intensity in same regions as described in ring width. Using these measurements, the average intensity in the actin ring was calculated with (outer integrated intensity − inner integrated intensity)/(outer area − inner area). To calculate relative intensity for unwounded (UW) time point, average intensity at UW was measured with 50 × 50 pixels at the center of embryos and then averaged intensity of actin ring at each timepoint was divided by average intensity of UW. Wound expansion was calculated with max wound area/initial wound size. Closure rate was calculated with two time points, one is tmax that is the time of reaching maximum wound area, the other is t < half that is the time of reaching 50–$35\%$ size of max wound since the slope of wound area curve changes after t < half. Using these time points, average speed was calculated with (wound area at tmax − wound area at t < half)/tmax-t < half.
Figures were assembled in Canvas Draw 6 for Mac (Canvas GFX, Inc.).
## Statistical analysis
All statistical analysis was done using Prism 8.2.1 (GraphPad, San Diego, CA). Statistical significance was calculated using the Kruskal–Wallis or Welch’s t tests with * is $p \leq 0.05$, ** is $p \leq 0.01$, *** is $p \leq 0.001$, **** is $p \leq 0.0001$, and ns is not significant.
## Supplementary Information
Supplementary Video 1.Supplementary Legends. The online version contains supplementary material available at 10.1038/s41598-023-31973-2.
## References
1. Cooper ST, McNeil PL. **Membrane repair: Mechanisms and pathophysiology**. *Physiol. Rev.* (2015) **95** 1205-1240. DOI: 10.1152/physrev.00037.2014
2. Howard AC, McNeil AK, Xiong F, Xiong WC, McNeil PL. **A novel cellular defect in diabetes: Membrane repair failure**. *Diabetes* (2011) **60** 3034-3043. DOI: 10.2337/db11-0851
3. Galan JE, Bliska JB. **Cross-talk between bacterial pathogens and their host cells**. *Annu. Rev. Cell Dev. Biol.* (1996) **12** 221-255. DOI: 10.1146/annurev.cellbio.12.1.221
4. Velnar T, Bailey T, Smrkolj V. **The wound healing process: An overview of the cellular and molecular mechanisms**. *J. Int. Med. Res.* (2009) **37** 1528-1542. DOI: 10.1177/147323000903700531
5. Sonnemann KJ, Bement WM. **Wound repair: Toward understanding and integration of single-cell and multicellular wound responses**. *Annu. Rev. Cell Dev. Biol.* (2011) **27** 237-263. DOI: 10.1146/annurev-cellbio-092910-154251
6. Nakamura M. **Into the breach: How cells cope with wounds**. *Open Biol.* (2018) **8** 180135. DOI: 10.1098/rsob.180135
7. Abreu-Blanco MT, Verboon JM, Parkhurst SM. **Coordination of Rho family GTPase activities to orchestrate cytoskeleton responses during cell wound repair**. *Curr. Biol.* (2014) **24** 144-155. DOI: 10.1016/j.cub.2013.11.048
8. Benink HA, Bement WM. **Concentric zones of active RhoA and Cdc42 around single cell wounds**. *J. Cell Biol.* (2005) **168** 429-439. DOI: 10.1083/jcb.200411109
9. Moe A. **Cross-talk-dependent cortical patterning of Rho GTPases during cell repair**. *Mol. Biol. Cell* (2021) **32** 1417-1432. DOI: 10.1091/mbc.E20-07-0481
10. Hui J, Stjepic V, Nakamura M, Parkhurst SM. **Wrangling actin assemblies: Actin ring dynamics during cell wound repair**. *Cells* (2022). DOI: 10.3390/cells11182777
11. Simon CM, Vaughan EM, Bement WM, Edelstein-Keshet L. **Pattern formation of Rho GTPases in single cell wound healing**. *Mol. Biol. Cell* (2013) **24** 421-432. DOI: 10.1091/mbc.E12-08-0634
12. Golding AE, Visco I, Bieling P, Bement WM. **Extraction of active RhoGTPases by RhoGDI regulates spatiotemporal patterning of RhoGTPases**. *Elife* (2019) **8** 50471. DOI: 10.7554/eLife.50471
13. Nakamura M, Verboon JM, Parkhurst SM. **Prepatterning by RhoGEFs governs Rho GTPase spatiotemporal dynamics during wound repair**. *J. Cell Biol.* (2017) **216** 3959-3969. DOI: 10.1083/jcb.201704145
14. Goley ED, Welch MD. **The ARP2/3 complex: An actin nucleator comes of age**. *Nat. Rev. Mol. Cell Biol.* (2006) **7** 713-726. DOI: 10.1038/nrm2026
15. Campellone KG, Welch MD. **A nucleator arms race: Cellular control of actin assembly**. *Nat. Rev. Mol. Cell Biol.* (2010) **11** 237-251. DOI: 10.1038/nrm2867
16. Olson MF. **Rho GTPases, their post-translational modifications, disease-associated mutations and pharmacological inhibitors**. *Small GTPases* (2018) **9** 203-215. DOI: 10.1080/21541248.2016.1218407
17. Mosaddeghzadeh N, Ahmadian MR. **The RHO family GTPases: Mechanisms of regulation and signaling**. *Cells* (2021) **10** 1831. DOI: 10.3390/cells10071831
18. Narumiya S, Thumkeo D. **Rho signaling research: History, current status and future directions**. *FEBS Lett.* (2018) **592** 1763-1776. DOI: 10.1002/1873-3468.13087
19. Hui J, Nakamura M, Dubrulle J, Parkhurst SM. **Coordinated efforts of different actin filament populations are needed for optimal cell wound repair**. *Mol. Biol. Cell* (2023) **34** ar15. DOI: 10.1091/mbc.E22-05-0155
20. Millard TH, Sharp SJ, Machesky LM. **Signalling to actin assembly via the WASP (Wiskott-Aldrich syndrome protein)-family proteins and the Arp2/3 complex**. *Biochem. J.* (2004) **380** 1-17. DOI: 10.1042/BJ20040176
21. Takenawa T, Suetsugu S. **The WASP-WAVE protein network: Connecting the membrane to the cytoskeleton**. *Nat. Rev. Mol. Cell Biol.* (2007) **8** 37-48. DOI: 10.1038/nrm2069
22. Burianek LE, Soderling SH. **Under lock and key: Spatiotemporal regulation of WASP family proteins coordinates separate dynamic cellular processes**. *Semin. Cell Dev. Biol.* (2013) **24** 258-266. DOI: 10.1016/j.semcdb.2012.12.005
23. Massaad MJ, Ramesh N, Geha RS. **Wiskott-Aldrich syndrome: A comprehensive review**. *Ann. N. Y. Acad. Sci.* (2013) **1285** 26-43. DOI: 10.1111/nyas.12049
24. Gautreau AM, Fregoso FE, Simanov G, Dominguez R. **Nucleation, stabilization, and disassembly of branched actin networks**. *Trends Cell Biol.* (2021). DOI: 10.1016/j.tcb.2021.10.006
25. Kramer DA, Piper HK, Chen B. **WASP family proteins: Molecular mechanisms and implications in human disease**. *Eur. J. Cell Biol.* (2022) **101** 151244. DOI: 10.1016/j.ejcb.2022.151244
26. Machesky LM, Insall RH. **Scar1 and the related Wiskott-Aldrich syndrome protein, WASP, regulate the actin cytoskeleton through the Arp2/3 complex**. *Curr. Biol.* (1998) **8** 1347-1356. DOI: 10.1016/s0960-9822(98)00015-3
27. Miki H, Suetsugu S, Takenawa T. **WAVE, a novel WASP-family protein involved in actin reorganization induced by Rac**. *EMBO J.* (1998) **17** 6932-6941. DOI: 10.1093/emboj/17.23.6932
28. Eden S, Rohatgi R, Podtelejnikov AV, Mann M, Kirschner MW. **Mechanism of regulation of WAVE1-induced actin nucleation by Rac1 and Nck**. *Nature* (2002) **418** 790-793. DOI: 10.1038/nature00859
29. Chen B. **Rac1 GTPase activates the WAVE regulatory complex through two distinct binding sites**. *Elife* (2017) **6** 29795. DOI: 10.7554/eLife.29795
30. Schaks M. **Distinct interaction sites of Rac GTPase with WAVE regulatory complex have non-redundant functions in vivo**. *Curr. Biol.* (2018) **28** 3674-3684. DOI: 10.1016/j.cub.2018.10.002
31. Rottner K, Stradal TEB, Chen B. **WAVE regulatory complex**. *Curr. Biol.* (2021) **31** R512-R517. DOI: 10.1016/j.cub.2021.01.086
32. Ding B. **Structures reveal a key mechanism of WAVE regulatory complex activation by Rac1 GTPase**. *Nat. Commun.* (2022) **13** 5444. DOI: 10.1038/s41467-022-33174-3
33. Steffen A. **Rac function is crucial for cell migration but is not required for spreading and focal adhesion formation**. *J. Cell Sci.* (2013) **126** 4572-4588. DOI: 10.1242/jcs.118232
34. Buracco S, Claydon S, Insall R. **Control of actin dynamics during cell motility**. *F1000Res* (2019) **8** 18669. DOI: 10.12688/f1000research.18669.1
35. Liu R. **Wash functions downstream of Rho and links linear and branched actin nucleation factors**. *Development* (2009) **136** 2849-2860. DOI: 10.1242/dev.035246
36. Verboon JM, Rahe TK, Rodriguez-Mesa E, Parkhurst SM. **Wash functions downstream of Rho1 GTPase in a subset of Drosophila immune cell developmental migrations**. *Mol. Biol. Cell* (2015) **26** 1665-1674. DOI: 10.1091/mbc.E14-08-1266
37. Abreu-Blanco MT, Verboon JM, Parkhurst SM. **Cell wound repair in Drosophila occurs through three distinct phases of membrane and cytoskeletal remodeling**. *J. Cell Biol.* (2011) **193** 455-464. DOI: 10.1083/jcb.201011018
38. Kiehart DP, Galbraith CG, Edwards KA, Rickoll WL, Montague RA. **Multiple forces contribute to cell sheet morphogenesis for dorsal closure in Drosophila**. *J. Cell Biol.* (2000) **149** 471-490. DOI: 10.1083/jcb.149.2.471
39. Gao Y, Dickerson JB, Guo F, Zheng J, Zheng Y. **Rational design and characterization of a Rac GTPase-specific small molecule inhibitor**. *Proc. Natl. Acad. Sci. U.S.A.* (2004) **101** 7618-7623. DOI: 10.1073/pnas.0307512101
40. Hui J, Nakamura M, Dubrulle J, Parkhurst SM. **Cell wound repair requires the coordinated action of linear and branched actin nucleation factors**. *bioRxiv* (2022). DOI: 10.1101/2022.04.14.488397
41. Stovold CF, Millard TH, Machesky LM. **Inclusion of Scar/WAVE3 in a similar complex to Scar/WAVE1 and 2**. *BMC Cell Biol* (2005) **6** 11. DOI: 10.1186/1471-2121-6-11
42. Pollitt AY, Insall RH. **WASP and SCAR/WAVE proteins: The drivers of actin assembly**. *J. Cell Sci.* (2009) **122** 2575-2578. DOI: 10.1242/jcs.023879
43. Bear JE, Rawls JF, Saxe CL. **SCAR, a WASP-related protein, isolated as a suppressor of receptor defects in late Dictyostelium development**. *J. Cell Biol.* (1998) **142** 1325-1335. DOI: 10.1083/jcb.142.5.1325
44. Rotty JD, Wu C, Bear JE. **New insights into the regulation and cellular functions of the ARP2/3 complex**. *Nat. Rev. Mol. Cell Biol.* (2013) **14** 7-12. DOI: 10.1038/nrm3492
45. Ismail AM, Padrick SB, Chen B, Umetani J, Rosen MK. **The WAVE regulatory complex is inhibited**. *Nat. Struct. Mol. Biol.* (2009) **16** 561-563. DOI: 10.1038/nsmb.1587
46. Derivery E, Lombard B, Loew D, Gautreau A. **The Wave complex is intrinsically inactive**. *Cell Motil. Cytoskelet.* (2009) **66** 777-790. DOI: 10.1002/cm.20342
47. Muresan CG. **F-actin architecture determines constraints on myosin thick filament motion**. *Nat. Commun.* (2022) **13** 7008. DOI: 10.1038/s41467-022-34715-6
48. Takai Y, Sasaki T, Matozaki T. **Small GTP-binding proteins**. *Physiol. Rev.* (2001) **81** 153-208. DOI: 10.1152/physrev.2001.81.1.153
49. Bustelo XR, Sauzeau V, Berenjeno IM. **GTP-binding proteins of the Rho/Rac family: Regulation, effectors and functions in vivo**. *BioEssays* (2007) **29** 356-370. DOI: 10.1002/bies.20558
50. Deng S, Bothe I, Baylies M. **Diaphanous regulates SCAR complex localization during Drosophila myoblast fusion**. *Fly* (2016) **10** 178-186. DOI: 10.1080/19336934.2016.1195938
51. Deng S, Bothe I, Baylies MK. **The formin diaphanous regulates myoblast fusion through actin polymerization and Arp2/3 regulation**. *PLoS Genet.* (2015) **11** e1005381. DOI: 10.1371/journal.pgen.1005381
52. Fricke R. **Drosophila Cip4/Toca-1 integrates membrane trafficking and actin dynamics through WASP and SCAR/WAVE**. *Curr. Biol.* (2009) **19** 1429-1437. DOI: 10.1016/j.cub.2009.07.058
53. Schaks M. **RhoG and Cdc42 can contribute to Rac-dependent lamellipodia formation through WAVE regulatory complex-binding**. *Small GTPases* (2021) **12** 122-132. DOI: 10.1080/21541248.2019.1657755
54. Koronakis V. **WAVE regulatory complex activation by cooperating GTPases Arf and Rac1**. *Proc. Natl. Acad. Sci. U.S.A.* (2011) **108** 14449-14454. DOI: 10.1073/pnas.1107666108
55. Singh SP. **Cell-substrate adhesion drives Scar/WAVE activation and phosphorylation by a Ste20-family kinase, which controls pseudopod lifetime**. *PLoS Biol.* (2020) **18** e3000774. DOI: 10.1371/journal.pbio.3000774
56. Singh V, Davidson AC, Hume PJ, Koronakis V. **Arf6 Can Trigger Wave Regulatory Complex-Dependent Actin Assembly Independent of Arno**. *Int. J. Mol. Sci.* (2020) **21** 2457. DOI: 10.3390/ijms21072457
57. Buracco S. **The Scar/WAVE complex drives normal actin protrusions without the Arp2/3 complex, but proline-rich domains are required**. *bioRxiv* (2022). DOI: 10.1101/2022.05.14.491902
58. Nakamura M, Verboon JM, Prentiss CL, Parkhurst SM. **The kinesin-like protein Pavarotti functions noncanonically to regulate actin dynamics**. *J. Cell Biol.* (2020) **219** 2117. DOI: 10.1083/jcb.201912117
59. Rodriguez-Mesa E, Abreu-Blanco MT, Rosales-Nieves AE, Parkhurst SM. **Developmental expression of Drosophila Wiskott-Aldrich Syndrome family proteins**. *Dev. Dyn.* (2012) **241** 608-626. DOI: 10.1002/dvdy.23742
60. Schindelin J. **Fiji: An open-source platform for biological-image analysis**. *Nat. Methods* (2012) **9** 676-682. DOI: 10.1038/nmeth.2019
|
---
title: 'IMC-Denoise: a content aware denoising pipeline to enhance Imaging Mass Cytometry'
authors:
- Peng Lu
- Karolyn A. Oetjen
- Diane E. Bender
- Marianna B. Ruzinova
- Daniel A. C. Fisher
- Kevin G. Shim
- Russell K. Pachynski
- W. Nathaniel Brennen
- Stephen T. Oh
- Daniel C. Link
- Daniel L. J. Thorek
journal: Nature Communications
year: 2023
pmcid: PMC10036333
doi: 10.1038/s41467-023-37123-6
license: CC BY 4.0
---
# IMC-Denoise: a content aware denoising pipeline to enhance Imaging Mass Cytometry
## Abstract
Imaging Mass Cytometry (IMC) is an emerging multiplexed imaging technology for analyzing complex microenvironments using more than 40 molecularly-specific channels. However, this modality has unique data processing requirements, particularly for patient tissue specimens where signal-to-noise ratios for markers can be low, despite optimization, and pixel intensity artifacts can deteriorate image quality and downstream analysis. Here we demonstrate an automated content-aware pipeline, IMC-Denoise, to restore IMC images deploying a differential intensity map-based restoration (DIMR) algorithm for removing hot pixels and a self-supervised deep learning algorithm for shot noise image filtering (DeepSNiF). IMC-Denoise outperforms existing methods for adaptive hot pixel and background noise removal, with significant image quality improvement in modeled data and datasets from multiple pathologies. This includes in technically challenging human bone marrow; we achieve noise level reduction of $87\%$ for a 5.6-fold higher contrast-to-noise ratio, and more accurate background noise removal with approximately 2 × improved F1 score. Our approach enhances manual gating and automated phenotyping with cell-scale downstream analyses. Verified by manual annotations, spatial and density analysis for targeted cell groups reveal subtle but significant differences of cell populations in diseased bone marrow. We anticipate that IMC-Denoise will provide similar benefits across mass cytometric applications to more deeply characterize complex tissue microenvironments.
Multiplexed imaging technologies can reveal the complex cellular and molecular profiles of tissue. Here, the authors develop and implement a denoising pipeline to significantly enhance imaging mass cytometry quality and improve single-cell analyses.
## Introduction
Disease states are the result of a complex interplay of many different cell types interacting in close proximity in the context of often heterogeneous tissues. Traditional approaches to study these features at the tissue scale have been limited in the number of specific markers that can be acquired to robustly resolve distinct cell types. Flow cytometry, perhaps the most widely used technique to study cell populations and states in this milieu, requires single-cell disaggregation of the tissue resulting in complete loss of spatial context1,2. Highly multiplexed imaging provides a means to assess these events at cellular resolution in situ, with extensive protocol development in progress3, including tissue-based cyclic immunofluorescence (t-CyCIF)4, co-detection by indexing (CODEX)5, Multiplexed Ion Beam Imaging (MIBI)6,7, and Imaging Mass Cytometry (IMC)8. In IMC, tissue sections are stained with a panel of metal-conjugated antibodies, and data is acquired by UV-laser raster ablation of the section in 1-micron pixels for cytometry by time-of-flight (CyTOF) mass analyzer. This novel imaging technology allows for the detection of more than 40 antigens simultaneously to facilitate single-cell, spatially resolved, highly multiplexed analysis of solid tissues. This provides essential information on the distribution of transcripts, proteins, and protein modifications within single cells, microenvironments, and entire tissues8–17. The pixel data is processed into an image, thereby allowing the visualization of phenotypes and incorporation of spatial information in subsequent analyses. These properties make it a unique tool for the evaluation of complex biological systems.
Despite the wide applications in pre- and clinical research using this state-of-the-art multiplexed imaging technique, there exist specific technical noise sources in IMC, which include hot pixels, channel spillover and shot noise8–10,15,18,19. Hot pixels are concentrated areas of high counts which are uncorrelated with any biological structures. Putatively, these can result from deposition of metal-stained antibody aggregates. In IMC images, single hot pixels are the most common outliers, and small hot clusters with several consecutive pixels may also exist. Channel spillover refers to scenarios where the signal of a source channel contaminates a target channel or is correlated with such contamination. The spillover in IMC can occur from a variety of reasons, such as instrument properties (abundance sensitivity), isotopic impurities and oxidation. Finally, shot noise exists because of ion counting imaging processes, which are pixel-independent, signal-dependent and usually modeled as a Poisson process. Additionally, noise levels are related to multiple other factors, including variations in conjugated metal isotopes, antibody concentration and arrangement.
Together these noise sources appreciably deteriorate image quality and distort downstream analyses of IMC data. Differing from traditional fluorescence-based imaging modalities, there are low background features and no read-out noises from imaging sensors in IMC. A number of studies have attempted to address the unique imaging data features of IMC. Hot pixels can be corrected by thresholding methods10,14,15,20; however, due to the differences between marker channels and tissues, a threshold needs to be pre-set carefully. An inappropriate threshold may lead to unsatisfactory results. Post-acquisition methods10,19 and a bead-based compensation workflow18 have been proposed to correct the channel spillover phenomenon. However, spillover correction may not be necessary if the marker panel employed is well-designed and titrated; and the intensity of channel-overlapping signal is often weak18. Therefore, spillover can be neglected when using low concentrations of staining antibodies, which however further lowers signal-to-noise ratio (SNR). To account for the impact of shot noise, MAUI7,19 and a semi-automated Ilastik-based method21 have been used for background noise removal. These approaches require finely tuned parameters or manually annotated background regions, requiring preprocessing expertize. In tissues with low marker signals, highly intermixed cell populations, or difficult immunostaining defining thresholds can be time consuming with high inter-user subjectivity, which may still result in poor image quality that complicates further analyses.
In the present work we develop and apply IMC-Denoise, a content aware denoising pipeline to enhance IMC images through an automated process. To account for the two major noise sources in this modality, hot pixels and shot noise, IMC-Denoise invokes novel algorithms for differential intensity map-based restoration (DIMR) and self-supervised deep learning-based shot noise image filtering (DeepSNiF). We demonstrate the flexibility and effectiveness of the proposed pipeline on publicly available IMC datasets of pancreatic cancer10, breast cancer12, a MIBI dataset19, and deploy it on a technically challenging unique human bone marrow dataset. We benchmark our approach against existing hot pixel removal methods10,14,15,20 and other advanced biomedical imaging denoising algorithms, such as non-local means filtering (NLM)22, batch matching and 3D filtering (BM3D)23 and Noise2Void (N2V)24, which is used in IMC here for the first time. We demonstrate that the image formation model derived IMC-Denoise pipeline produces image quality enhancements that are best-in-class and leads to improved downstream analysis, with limited manual user manipulation. Qualitative improvements in images enhances their interpretation, and quantitatively improve molecularly-defined phenotyping. Results from the IMC-Denoise pipeline are suitable for further downstream analysis, such as Mesmer/DeepCell and ark-analysis25 or MCMicro26. The IMC-Denoise software package and the corresponding tutorial have been published on Github (https://github.com/PENGLU-WashU/IMC_Denoise). We provide this tool to augment studies that seek to more deeply characterize the complex and diverse tissue microenvironment.
## IMC-Denoise principle
*The* general principle of IMC-*Denoise is* schematized in Fig. 1a and Supplementary Notes 1. To account for hot pixels and shot noise, an accurate IMC imaging joint model is built as Eq. [ 1], by considering ion counting imaging as a Poisson process (Supplementary Notes 1.1).1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{{{{\bf{R}}}}}}}}={{{{{{{\mathcal{P}}}}}}}}[{{{{{{{\bf{X}}}}}}}}+{{{{{{{{\bf{X}}}}}}}}}^{{{{{{{{\rm{spillover}}}}}}}}}]+{{{{{{{\bf{Q}}}}}}}},$$\end{document}R=P[X+Xspillover]+Q,where R is the raw image, X the “clean” signal, Xspillover the spillover signals without noise, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{{{{\mathcal{P}}}}}}}}$$\end{document}P[x] the Poisson noise with mean x, and Q the hot pixels. The term Xspillover in Eq. [ 1] can be omitted if the spillover is limited, which is often the case. However, the magnitude of image degradation from hot pixel and shot noise sources is considerable, resulting in bias and errors in downstream analysis and addressed in turn, below (Supplementary Notes 1.2).Fig. 1General principle and validation of IMC-Denoise on the human bone marrow IMC dataset.a Schematic of IMC-Denoise: (i) the DIMR algorithm: after the Anscombe transformation, the difference maps calculated from the raw image are operated to form a histogram. The outliers are detected based on this histogram and removed by a 3 × 3 median filter, iteratively. ( ii) The training phase of the self-supervised DeepSNiF algorithm: In the hot pixel corrected images, several pixels are randomly selected and masked. The hot pixel corrected images before and after the masking are set as the outputs and inputs of a deep neural network, respectively. Statistics-derived I-divergence on the masked pixels combined with the Hessian norm regularization on all the pixels is set as the loss function to guarantee the optimal denoising performance. ( iii) The prediction phase of the DeepSNiF algorithm: the hot pixel corrected IMC images are fed into the trained network to account for the shot noise. b The fractions of detected hot pixels by DIMR in selected channels. c DIMR removes hot pixels in DNA intercalator channel effectively. Left: Comparison of the raw and DIMR-processed images; and the difference between the images, in which Residual corresponds to the detected hot pixels. Upper right: the corresponding histograms of the raw and DIMR-processed images. Lower right: comparisons between the raw, NTHM, MTHM and DIMR processed images. d Visual inspection of DeepSNiF and other statistics-based denoising algorithms on a Collagen III-labeled IMC image. e DeepSNiF performs significantly better than other algorithms ($$n = 12$$ independent images) on denoising Collagen III-labeled IMC images in terms of STDB and CNR. f Visual inspection of DeepSNiF denoised IMC images labeled with other markers. g DeepSNiF improves the Pearson correlations between Collagen III-labeled IMC images with low and high SNR significantly ($$n = 10$$ independent images). In e and g, box center indicates median, box edges 25th and 75th percentile, and whiskers minimum and maximum percentile; P values were calculated through two-sided Wilcoxon matched-paired test (*$P \leq 0.05$, **$P \leq 0.01$, ***$P \leq 0.001$ and ns: no significance). Scale bar: a Upper: 100 μm, lower: 125 μm. c Whole region: 75 μm, sub-region 1–3: 8 μm. d 50 μm. f 45 μm. g 100 μm.
In IMC-Denoise, the DIMR algorithm (Fig. 1ai and Supplementary Notes 1.3) builds differential maps to detect the hot pixels by comparing adjacent pixels in a 3 × 3 sliding window, as hot pixels are local maxima. The Anscombe transformation27 is applied to the raw image R followed by background removal of intensities lower than 4 (for IMC), so that the difference between adjacent pixels, Di, can be feasibly approximated as a generalized Gaussian distribution28, where i is the neighbor index in the sliding window (i ∈ {1, 2,..., 8}). Additionally, as with all biomedical imaging acquisition, in IMC datasets the tissue or background pixels should be continuous. Under these conditions, for a specific pixel p there must exist several d\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${}_{i}^{p}$$\end{document}ip close to the mean μi of its corresponding distribution Di, except in the presence of a hot pixel. To unmix outliers from normal pixels, we consequently calculate the distances between d\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${}_{i}^{p}$$\end{document}ip and μi as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\bigtriangleup }_{i}^{p}=|{d}_{i}^{p}-{\mu }_{i}|$$\end{document}△ip=∣dip−μi∣ and sort \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\bigtriangleup }_{i}^{p}$$\end{document}△ip for i ∈ {1, 2,..., 8}. Then, the d\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${}_{i}^{p}$$\end{document}ip corresponding to the first l smallest \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\bigtriangleup }_{i}^{p}$$\end{document}△ip are summed, and the results from all pixels form a new distribution, Tl. Compared to those in the distributions Di, the hot pixels move beyond the right tail of Tl, while the normal relevant pixels move towards its center (Supplementary Note 1.3.1). To robustly detect the outliers, the kernel density estimation algorithm29 is applied to Tl afterwards (Supplementary Note 1.3.2). On the fitted curve (x, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\hat{g}}_{h}$$\end{document}g^h(x)), a threshold point xT is defined so that any points x>xT are considered as outliers and filtered by a 3 × 3 median filter. Because outliers are located beyond the right tail of Tl, it is reasonable to set xT when \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\frac{d{\hat{g}}_{h}(x)}{dx}\to$$\end{document}dg^h(x)dx→0, which means the current distribution ends. Likewise, the shape of the distribution should not change from convex to concave on the right tail. Thus, it is also reasonable to set xT when \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\frac{{d}^{{{{{{{{\rm{2}}}}}}}}}{\hat{g}}_{h}(x-\bigtriangleup x)}{{dx}^{{{{{{{{\rm{2}}}}}}}}}}\ge 0$$\end{document}d2g^h(x−△x)dx2≥0 and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\frac{{d}^{{{{{{{{\rm{2}}}}}}}}}{\hat{g}}_{h}(x)}{{dx}^{{{{{{{{\rm{2}}}}}}}}}}\le$$\end{document}d2g^h(x)dx2≤0, where △x represents a small value. Because the pixel values of the raw images are discrete, △x is normally set as 1. We operate DIMR for multiple iterations to adequately remove hot pixels until no outliers are detected. The hot pixel removed images are transformed to their original scales with the direct algebraic inverse Anscombe transformation30. The DIMR algoirthm is summarized as Supplementary Algorithm 1. In the implementation, we use the median \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\widetilde{\mu }}_{i}$$\end{document}μ~i of distribution Di as a robust estimation of the mean μi. In addition, it is normally assumed that at least half of the neighbors are close to the center pixel in a 3 × 3 window so that $l = 431$,32. Validated by simulation (Supplementary Note 3.3.1), the iteration number is set as 3 to adequately remove hot pixels; for a 500 × 500 image, it takes ~0.05–0.4 s to run DIMR, depending on the hot pixel densities.
After hot pixel removal, the imaging model is simplified as: \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{{{{\bf{R}}}}}}}}={{{{{{{\mathcal{P}}}}}}}}[{{{{{{{\bf{X}}}}}}}}]$$\end{document}R=P[X], for which we have developed DeepSNiF (Fig. 1aii, iii and Supplementary Note 1.4) to account for the ion noise in IMC images. By combining Poisson statistics and detection theory, I-divergence33 is derived as the loss function to enable the maximized likelihood estimation for the denoising task. Unlike traditional imaging methods for which noise-free training label images can be generated, commonly with long exposures, the image formation process in IMC requires laser ablation. Thus, a tissue can only be imaged once in IMC. Autofluorescence artifacts in immunofluorescence (IF) images, and the tedious and potentially interfering processes for consecutive IF and IMC imaging, are further confounds. Therefore, conventional supervised denoising approaches34–36 or Noise2Noise37 are not available here.
We overcome these limitations by applying a self-supervised approach inspired by Noise2Void24 and Noise2Self38. This approach randomly masks several pixels in the DIMR-processed hot pixel-removed images by a stratified sampling strategy. Subsequently, the manipulated images are set as the inputs of the network and the hot pixel removed ones are the outputs. For this construct, the self-supervised training is approximately equivalent to a supervised learning process (Supplementary Note 1.4.1). The network follows U-Net39 structure with Res-Blocks40 to enable high quality training and prediction (Methods, Supplementary Fig. 13). Notably, the last activation function of the network is set as softplus (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\log (1+\exp (x))$$\end{document}log(1+exp(x))) to restrict non-negativity for the images. Nevertheless, the denoising performance is still sub-optimal, due to neglected information of the masked pixels and partially utilized pixels in the self-supervised strategy. To further boost DeepSNiF, the Hessian norm regularization41–43 is applied in the loss function with the continuity between biological structures a priori (Supplementary Note 1.4.2). Overall, the loss function of DeepSNiF is summarized as Eq. [ 2].2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{{{{\mathcal{L}}}}}}}}\left({{{{{{{\bf{R}}}}}}}},\,{{{{{{{{\mathcal{F}}}}}}}}}_{\!\theta }\left[\;f({{{{{{{\bf{R}}}}}}}})\right]\right)= \mathop{\sum}\limits_{p}{{{{{{{{\bf{M}}}}}}}}}_{p}\cdot \left[{r}_{p}\log \frac{{r}_{p}}{{{{{{{{{\mathcal{F}}}}}}}}}_{\!\theta }{[\;f({{{{{{{\bf{R}}}}}}}})]}_{p}}-{r}_{p}+{{{{{{{{\mathcal{F}}}}}}}}}_{\!\theta }{\left[\;f({{{{{{{\bf{R}}}}}}}})\right]}_{p}\right]/\mathop{\sum}\limits_{p}{{{{{{{{\bf{M}}}}}}}}}_{p}\\ +{\lambda }_{{{{{{{{\rm{Hessian}}}}}}}}}\mathop{\sum}\limits_{p}||{{{{{{{{\mathcal{R}}}}}}}}}_{{{{{{{{\rm{Hessian}}}}}}}}}({{{{{{{{\mathcal{F}}}}}}}}}_{\!\theta }[\;f({{{{{{{\bf{R}}}}}}}})])|{|}_{p}/\mathop{\sum}\limits_{p},$$\end{document}LR,Fθf(R)=∑pMp⋅rplogrpFθ[f(R)]p−rp+Fθf(R)p/∑pMp+λHessian∑p∣∣RHessian(Fθ[f(R)])∣∣p/∑p,where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{{{{{\mathcal{F}}}}}}}}}_{\!\theta }$$\end{document}Fθ represents the learnt weights of the network, f demonstrates the random pixel masking approach, rp the p-th pixel of the hot pixel removed training set R, Mp the pixel mask (Mp ∈ {0, 1}), \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{{{{{\mathcal{R}}}}}}}}}_{{{{{{{{\rm{Hessian}}}}}}}}}$$\end{document}RHessian the Hessian operator, λHessian the regularization parameter and p the pixel index. Here, the pixel p is masked only when Mp = 1. Note that the first term works only on the selected masked pixels, while the second regularization term utilizes all of the image information. Prior to training and prediction, the images are normalized between 0 and 1 by a percentile-normalization approach (Supplementary Note 1.4.3). The DeepSNiF algorithm is summarized as Supplementary Algorithm 2. As validated by simulation (Supplementary Note 3.4.2), λHessian is empirically set as 3e-6 to balance the trade-off between data fidelity and regularization.
## Validation of IMC image quality improvement
We initially tested our DIMR algorithm on selected markers of a human bone marrow dataset. Here, inherently high autofluorescence and tissue features (fragile haematopoietic stroma intermixed with dense cortical bone) excluded other spatial biology methods, even after substantial pre-processing. Figure 1b enumerates the proportion of hot pixels detected by DIMR for each marker. We then selected DNA intercalator and CD235a (Fig. 1c and Supplementary Fig. 14) to evaluate DIMR due to their high hot pixel density. By comparing the images and the corresponding histograms, hot pixels are effectively eliminated by DIMR.
We further compared DIMR with two recent hot pixel removal methods, neighbor-based threshold hot pixel removal method (NTHM)14,15,20 and median-based threshold hot pixel removal method (MTHM)10 with default parameters, to benchmark its performance (Table 1, Supplementary Note 2.1, Supplementary Algorithms 3 and 4). From the results, all three methods can remove spurious signal, but their performances varied from each other. To quantitatively evaluate these methods, we utilized t-CyCIF data44 to generate simulated IMC images (Supplementary Note 3.1) with a range of noise levels and hot pixel densities. The three methods were then applied on the simulated datasets, and root mean squared errors between the hot pixel-free and processed images were set as the metric to evaluate the accuracy of hot pixel removal (Supplementary Note 3.2). Note that in simulations, the thresholds of NTHM and MTHM were manually tuned to guarantee their optimal performances, while DIMR was configured automatically. Table 1Reference denoising algorithm summaryAcronymFull nameAlgorithm detailsNTHMNeighbour-based threshold hot pixel removal method14,15,20Supplementary Note 2.1Supplementary Algorithm 3MTHMMedian-based threshold hot pixel removal method10Supplementary Note 2.1Supplementary Algorithm 4N2VNoise2Void24Supplementary Note 2.2.1MN2VModified Noise2Void with Anscombe transformation27Supplementary Note 2.2.2N2TNoise2True34, only simulations usedSupplementary Note 2.2.3GAUSSGaussian filter with kernel size of 5x5 and standard deviation of 0.8Supplementary Note 2.3.1NLMNon-local means algorithm22Supplementary Note 2.3.2BM3DBatch-matching and 3D filtering algorithm23 with Anscombe transformationSupplementary Note 2.3.3 The simulation results indicate DIMR is the best performer among the three methods (Supplementary Note 3.3.2). In fact, the threshold of NTHM requires contextual adjustment as different tissues and channels may have different scales. Moreover, this method is not efficient at removing consecutive hot pixels. MTHM is not locally adaptive and may overlook hot pixels with similar intensity to that of normal pixels; or erroneously remove normal pixels located at the border between tissues and background. Use of a lower search range or threshold for MTHM may also generate false negatives. In comparison, the outlier detection of DIMR is based on overall image statistics. Therefore, no manual threshold adjustment is required for images with different intensity scales, and a higher detection sensitivity is achieved even for hot pixels with lower intensities. These features along with the simulation data results demonstrate the versatility and accuracy of DIMR. The automated DIMR approach also results in the additional benefit of moderately improved cell segmentation, the result of robust removal of artifacts caused by hot pixels (Supplementary Fig. 15).
With hot pixels removed from image data, we next benchmarked the denoising performance of DeepSNiF along with DIMR and other statistics-based methods including a Gaussian filter with standard deviation of 1 (GAUSS), NLM, BM3D, N2V, modified N2V (MN2V) and DeepSNiF with no regularization (DeepSNiF-NR) (Table 1, Supplementary Notes 2.2 and 2.3) on the simulated dataset (Supplementary Note 3.4). These comparisons were carried out on IMC images labeled with Collagen III, CD31, CD34 and CD3 from the human bone marrow dataset (Fig. 1d and Supplementary Fig. 16). First we visually assessed images with different processing approaches for their overall appearance and in particular for retention of fine cell-level details. We found all the algorithms enhanced the DIMR data even though variant performances were achieved. GAUSS lowers the noise level by sacrificing resolution. NLM is effective at background denoising but does not account adequately for the noise components of signal. BM3D improves NLM further by its cooperative denoising procedure. However, we found it tended to over-smooth foreground and distorted cell shapes. N2V always generates artifacts because of an inappropriate noise model. DeepSNiF-NR performs better than MN2V because the Anscombe transformation in MN2V may generate some bias for extremely low counts; both of which are better than GAUSS, NLM and BM3D. DeepSNiF further enhances these results by mitigating the discontinuities in the DeepSNiF-NR output, and furthermore retains cell morphology features.
We then quantitatively compared the differently processed images across a range of different characterization methods. Assessment of peak SNR (PSNR) and structural similarity (SSIM)45 (Supplementary Note 3.2) were computed from the simulated data, and the standard deviation of background (STDB) and contrast-to-noise ratio (CNR; “Methods” section) were utilized for the IMC images labeled with Collagen III. All results indicated DeepSNiF enables the optimal denoising performance among these algorithms (Fig. 1e and Supplementary Note 3.4). In particular, the noise level (STDB) decreased by $87\%$ and CNR increased by 5.6-fold after DeepSNiF (0.9938 to 0.1254 and 1.1749 to 7.8065, median value).
We further visually inspect the denoising results of IMC-Denoise on multiple datasets including human bone marrow images (Fig. 1f and Supplementary Fig. 17), human breast cancer (Supplementary Fig. 18), human pancreatic cancer (Supplementary Fig. 19) and a MIBI dataset (Supplementary Fig. 20). Image quality improvements that enhance image interpretation are apparent, for both visual inspection and quantitative assessment, in particular for low SNR channels. Two orthogonal staining approaches were pursued in order to provide further validation of these image quality improvements. Firstly, the same antibody was conjugated to two different metals and co-stained on the same tissue for detection in high and low sensitivity channels, without spillover. IMC-Denoise was employed on the low signal channel (209Bi) and was able to restore the image quality to match the high sensitivity channel (173Yb) with the Pearson correlation coefficient (PCC) improved as high as 0.16, as shown in representative images (Supplementary Fig. 21 and Fig. 1g). Similar conclusions can also be drawn from other channels with increased PCC by more than 0.48 and 0.35, respectively (Supplementary Fig. 22). Secondly, tissue sections stained with metal-conjugated antibodies (for CD3, CD4, CD61, and CD169) were probed with a fluorophore-conjugated secondary antibody for IF, individually. We then followed IF imaging by ablative-IMC (Supplementary Fig. 23). The additional handling and washing after IF imaging often leads to extremely low remaining metal isotope signal; however, enhancement in image quality can still be observed to restore the image to correlate to the IF. Specifically, the PCC quantitatively verified the image quality improvement of DeepSNiF (CD3: 0.5557–0.7939, CD4: 0.4975–0.7793, CD61: 0.9096–0.9492, and CD169: 0.4481–0.7726).
## IMC-Denoise enhances IMC background noise removal and downstream analysis
We next evaluated the ability of DeepSNiF in IMC-Denoise to remove background noise of IMC images. Visual inspection (Supplementary Fig. 24) reveal DeepSNiF enhances background noise removal of the examples effectively by a single threshold. To fully evaluate the enhancement by DeepSNiF, we manually annotated 15 images labeled with CD34 and 12 IMC images labeled with Collagen III (Fig. 2a). The single threshold-based method and semi-automated Ilastik-based method21 were applied on both DIMR and DeepSNiF-processed CD34 and Collagen III images (DIMR_thresh, DeepSNiF_thresh, DIMR_Ilastik and DeepSNiF_Ilastik, respectively), and MAUI was only applied on DIMR images (Methods). The results were compared with the manually annotated ground truths (Fig. 2b), and F1 score was set as the accuracy metric to quantitatively assess the results (Fig. 2c). To guarantee the best performance of threshold-based methods and MAUI, a wide range of parameters were tested (Supplementary Figs. 25–27). Note that in threshold-based methods, optimal thresholds from 1 to 4 were selected for individual DIMR-processed images per marker for fair comparison. Nevertheless, the single threshold 1 was selected for all the images per marker for DeepSNiF-processing, without the need of further tuning. Fig. 2IMC-Denoise enables background noise removal and enhances downstream analysis of the human bone marrow IMC dataset.a Examples of DIMR and DeepSNiF-processed IMC images labeled with CD34 and Collagen III. b Visual inspection of background removal results of DIMR and DeepSNiF-processed images, in which DIMR_thresh and DeepSNiF_thresh are binarized with the optimal thresholds (Supplementary Figs. 25 and 26), DIMR_Ilastik and DeepSNiF_Ilastik are segmented by the Ilastik software package, and MAUI results are the DIMR images processed by the MAUI software package (Supplementary Fig. 27), respectively. Manual annotated images are served as ground truths. c After DeepSNiF denoising, the background removal accuracy improves significantly in terms of F1 score, for both CD34 and Collagen III-labeled images ($$n = 15$$ independent images for CD34 and $$n = 12$$ independent images for Collagen III). Notably, DeepSNiF_Ilastik achieves the highest accuracy, while DeepSNiF_thresh performs better than all the background removal results from DIMR images. Box center indicates median, box edges 25th and 75th percentile, and whiskers minimum and maximum percentile. P values were calculated through two-sided Wilcoxon matched-paired test (**$P \leq 0.01$, ***$P \leq 0.001$, and ****$P \leq 0.0001$). d Visual inspection of DeepSNiF and DIMR_Ilastik-based denoising results on different markers-labeled IMC images. e–h Evaluations of denoising algorithms with manual gating strategies on single-cell data. The numbers in these panels are the cell percentages of the corresponding ranges. DIMR slightly enhances the single-cell analysis over raw data, while DeepSNiF further enhances the DIMR results and overall performs better than semi-automated DIMR_Ilastik-processing. Scale bar: a Top: 50 μm, bottom: 35 μm. d 107 μm.
Overlaid masks and F1 scores for both markers indicated DeepSNiF_Ilastik achieves the highest accuracy while DIMR_thresh is the weakest performer (CD34: 0.9143 to 0.4155, and Collagen III: 0.9434 to 0.5378, median value). Surprisingly, DeepSNiF_thresh is a better method for background noise removal than the semi-automated DIMR_Ilastik (CD34: 0.9040 to 0.8716, and Collagen III: 0.9345 to 0.9108, median value), and its F1 score was improved by approximately twofold compared to DIMR_thresh. We infer that DeepSNiF is capable of unmixing the signal and background, while the shot noise in DIMR images hinders the performances of the Ilastik-based method. MAUI was able to account for the background noise at the cost of false negative generation (CD34: 0.7824 and Collagen III: 0.7305, median value). Furthermore, we have also visually inspected and manually annotated marker images from Supplementary Figs. 16–18. All the results indicate DeepSNiF achieves good background removal performance (Supplementary Figs. 28–30).Indeed, the signal has been unmixed from background through DeepSNiF because we have proved a simple thresholding can remove background accurately (Fig. 2b, c, Supplementary Figs. 28b, 29b, and 30b). These findings support replacement of tedious semi-automated approaches by automated DeepSNiF.
Next, we were curious to evaluate the impact of IMC-Denoise on single-cell profiles. Using segmented cell masks, we extracted the cell intensities of CD38, MPO, CD14, CD71, CD11b, CD4, CD169, CD20, CD8a, CD15, CD3, and CD235a markers for 96232 cells in total (Methods). Please note that segmentation masks were identical for each comparison, using masks generated from DeepSNiF, so that the impact of variability in segmentation algorithms can be neglected. In Supplementary Figs. 31 and 32a, the comparison of the single-cell profiles of raw, DIMR and DeepSNiF data show that DIMR has the potential to correct false positive data, and DeepSNiF corrects all cell profiles. We have also conducted line fitting for the DIMR and DeepSNiF-processed single-cell data and calculated their PCC (Supplementary Fig. 32a). The results indicate DeepSNiF has not changed the single-cell intensity scale nor biased the overall linearity of the data. Furthermore, larger mean positive marker expressions lead to lighter corrections by DeepSNiF (Supplementary Fig. 30b). This follows from the logic that larger ion counts have lower shot noise levels.
Subsequently, we benchmarked the single-cell data from the raw, DIMR, DIMR_Ilastik and DeepSNiF-processed images (Fig. 2d). To achieve this, manual gating approaches with prior knowledge of cell markers were applied to DIMR, DIMR_Ilastik and DeepSNiF on IMC data (Fig. 2e–h). For example, among T cells (CD3-positive, CD14-negative), myeloid (CD11b, CD15) and erythroid (CD71, CD235a) markers should be absent. However, this condition may not hold because: segmentation and staining artifacts are unavoidable, and because hot pixels are present in the raw data. With the presence of shot noise, the single-cell data could be further biased (Supplementary Note 1.2.2). In Fig. 2e, f, the false positive myeloid and erythroid markers decrease slightly after DIMR correction ($0.8\%$ and $0.4\%$). DIMR_Ilastik and DeepSNiF further removed false positive myeloid ($3.86\%$ and $6.31\%$) and erythroid markers ($3.52\%$ and $5.83\%$) after the slight improvement of DIMR. Similarly, among B cells (CD20-positive), myeloid and erythroid markers (CD11b, CD15, MPO and CD235a) should be absent as well. In Fig. 2g and h, the false positive markers decrease slightly after DIMR correction ($0.06\%$ and $0.74\%$). Compared to DIMR, DIMR_Ilastik and DeepSNiF removed more false positive markers (DIMR_Ilastik: $6.41\%$ and $7.86\%$, DeepSNiF: $5.22\%$ and $8.44\%$). Overall, as expected DIMR could enhance the single-cell analysis to a limited extent. DeepSNiF and DIMR_Ilastik enable further enhancement, and overall the former achieves better performance than the latter on this task.
To test whether DeepSNiF-based segmented cell masks potentially favor DeepSNiF data, we have extracted the raw, DIMR, DIMR_Ilastik and DeepSNiF-processed single-cell data from DIMR-based cell masks. Both the single-cell data comparisons (Supplementary Fig. 33) and manual gating results (Supplementary Fig. 34) are similar to that from DeepSNiF-based cell masks. Therefore, we infer that our IMC-Denoise pipeline is also robust across different cell segmentation masks.
## DeepSNiF in IMC-Denoise enhances automated cell phenotyping
Cell phenotype annotation plays a key role in tissue microenvironment analysis. Indeed, false annotation of cell phenotypes has the potential to lead to false biological or clinical conclusions. Hot pixel removal is normally conducted before automated cell phenotyping14,15,17. Therefore, we focused on whether DeepSNiF in IMC-Denoise could impact phenotypic annotation of cell types. Here, the extracted single-cell data with DeepSNiF-based segmentation masks from the human bone marrow dataset were used for phenotypic annotation, including CD38, MPO, CD14, CD71, CD11b, CD4, CD169, CD20, CD8a, CD15, CD3 and CD235a channels. We clustered the DIMR dataset by the Phenograph algorithm46 with the Leiden community detection algorithm47 (Methods). *The* generated clusters were then annotated as immune cell subsets (B cell, CD4+ T cell, CD8+ T cell and plasma cell, monocyte/macrophages), erythroid, myeloid, and other CD4+ cells and others. To better demonstrate the modifications of DeepSNiF denoising, we then utilized a weighted KNN approach to map the DeepSNiF data into the DIMR-based clusters (Methods). The weights were acquired by calculating the Jaccard index between each DeepSNiF-processed cell profile with all of the DIMR cells (which is identical to the Jaccard graph construction of Phenograph). For visualization, the cell markers of DIMR and DeepSNiF were also compressed into two dimensions by the fast interpolation-based t-SNE algorithm48 as Supplementary Fig. 35 (Methods). The assigned phenotypes of DIMR and DeepSNiF datasets are demonstrated in Fig. 3a and the relative changes of each cell sub-population after DeepSNiF processing is shown in Fig. 3b. After DeepSNiF processing, B cells, CD8 T cells, plasma cells, CD4 T cells and other CD4+ cells decrease ($20.86\%$, $3.70\%$, $13.44\%$, $18.49\%$, and $8.23\%$, respectively), the monocytes/macrophages increase ($4.82\%$), while erythroid, myeloid and other cells remain largely unchanged. Fig. 3DeepSNiF enhances automated cell phenotyping on human bone marrow IMC data.a t-SNE plots of DIMR and DeepSNiF with cell phenotyping results. b The relative change in cell phenotypes before and after DeepSNiF. c, d Comparisons of DIMR and DeepSNiF-processed IMC images labeled with different cell markers, and the corresponding cell annotation results. The sub-panels (i)–(iv) in c and the bottom row in d correspond to the white dashed box region selection in their first panels, respectively. The white contours represent the differential phenotyping results between DIMR and DeepSNiF. e DeepSNiF enhances the sensitivity of cell phenotyping. After DeepSNiF processing, the non-specific marker signals reduce while the specific ones enrich in the cell types, respectively. The circle size indicates the positive marker percentage in a particular phenotype of DIMR, and the circle color indicates the relative changes of the positive rate for the particular markers after DeepSNiF enhancement. f DeepSNiF enhances the specificity of cell phenotyping. With DeepSNiF denoising, the ratios of specific phenotypes increase while those of non-specific phenotypes decrease in the positive markers. The relative change is the difference in percentage composition of each cell type before and after DeepSNiF enhancement. Scale bar: c 110 μm. d Top: 145 μm, bottom: 50 μm.
The phenotyping results of DIMR and DeepSNiF were mapped back into their segmentation masks and images (Fig. 3c, d and Supplementary Fig. 36). To highlight cells where DeepSNiF changes the cell phenotyping results, conflicting annotations between DIMR and DeepSNiF were labeled with white contours, and the changes were quantified for cell phenotype and marker enrichments (Methods). After DeepSNiF denoising, non-specific markers are reduced, while specific markers are enriched within the cell phenotypes (Fig. 3e). For example, we observed the positive rate increased for CD20 in B cells ($10.53\%$), CD8a in CD8 T cells ($2.32\%$), CD3 and CD4 in CD4 T cells ($6.84\%$ and $4.64\%$), CD38 in plasma cells ($6.21\%$) and CD4 in other CD4+ cells ($4.26\%$). Conversely, we observed a decrease of non-specific markers, such as CD38, MPO and CD14 in B cells ($5.24\%$, $8.11\%$, and $5.64\%$), CD3 in erythroid ($1.79\%$) and myeloid ($1.62\%$) cells, and all marker signals in “other” cells. Furthermore, the identified cell types were enriched in a marker-specific manner after DeepSNiF (Fig. 3f). For instance, we observed a post-DeepSNiF enrichment of monocytes/macrophages in CD14+ cells ($2.36\%$), CD11b+ cells ($1.70\%$) and CD169+ cells ($2.21\%$), and enrichment of B cells in CD20+ cells ($5.42\%$) and CD8 T cells in CD8a+ cells ($3.15\%$). Similarly, myeloid cells were enriched in MPO+ ($1.26\%$), and erythroid cells in CD71+, CD235a+ cells ($2.45\%$ and $1.48\%$). DeepSNiF also yielded an enriched composition of CD8 and CD4 T cells ($3.50\%$ and $2.54\%$), and reduced composition of myeloid and erythroid cells ($2.47\%$ and $4.27\%$) in CD3+ cells. However, we noticed the enrichment of erythroid cells in CD169+ cells ($1.03\%$), which may result from an artifact of the current segmentation approach due to the close relationship and irregular morphology at the boundaries between erythroids and macrophages within the bone marrow49.
Cell phenotyping by immunostaining of FFPE tissues is also inherently limited by antibody specificity and antigen retrieval protocols. In this tissue, CD38+ and CD14+ antibody staining is not strictly restricted to single lineages, and these markers can be aberrantly expressed in myeloid neoplasms included in this data set (Supplementary Fig. 37). On manual inspection, DeepSNiF improves the ability to identify co-localization of cell surface markers (Fig. 3c, d and Supplementary Fig. 36). Overall, DeepSNiF enhances the sensitivity and specificity of cell phenotyping. We have conducted similar analysis for the single-cell data from the DIMR-based segmentation masks as well (Supplementary Fig. 38). Likewise, the results are also similar to those from the DeepSNiF-based segmentation masks, which demonstrates the robustness of IMC-Denoise.
We observed that the enhancements in cell phenotyping and marker enrichments in Fig. 3e, f are related to the noise level of the IMC images. Specifically, DeepSNiF has the highest impact on CD20 and CD3 related phenotypes, improvement for CD15, MPO and CD235a related phenotypes is limited, with moderate changes for other cell classes. These findings agree with Supplementary Fig. 32b, where we plot the STD of the normalized positive marker differences between DIMR and DeepSNiF against intensity. To investigate the influence of DeepSNiF on phenotyping results more deeply, we applied a leave-one-out DeepSNiF strategy for CD20, CD3, CD71, CD235a, and MPO (“Methods” section and Supplementary Figs. 39–43). Briefly, this involved processing one marker for hot pixel removal with DIMR, e.g. CD20, and all the other markers by both DIMR and DeepSNiF. Then the same weighted KNN approach was applied on these leave-one-out DeepSNiF datasets. Similar to the conclusions from Fig. 3e, f and Supplementary Fig. 32b, CD20 and CD3 denoised by DeepSNiF improve cell phenotyping because of the high noise level of the corresponding IMC images. DeepSNiF has moderate impact on CD71 due to better IMC image quality than those of CD20 and CD3, and has minor impact on MPO and CD235a because of their good SNRs.
## DeepSNiF in IMC-Denoise enhances lymphocyte analysis
Cell-cell interactions of immune cells within the tumor microenvironment is of broad interest for many clinical pathology specimens. In myeloid malignancies, immune infiltrates are most commonly assessed by flow cytometry and are an active area of interest in therapeutic clinical trials50. However, in situ spatial context of cell-cell interaction mediated immune responses cannot be directly measured through this approach. We quantified the enhancement of lymphocyte spatial analysis for B cells, CD8+ T cells and CD4+ T cells by DeepSNiF, and compared these to a manually curated set of image annotations based on DeepSNiF-based cell masks (Fig. 4a). CD3, CD4, and CD20-stained images are more easily contaminated by shot noise than others (Supplementary Fig. 32b). Therefore, this approach can further validate the shot noise accounting ability of DeepSNiF as well. The phenotyping accuracy of DIMR and DeepSNiF as evaluated by the Jaccard score and F1 score indicate a significant improvement by DeepSNiF denoising (Fig. 4b); and DeepSNiF denoised data closely recapitulates gold-standard but laborious manual annotation. Specifically, the overall Jaccard scores improve from 0.6785, 0.8229, and 0.6781 to 0.9201, 0.8922, and 0.8860 for B cells, CD8+ T cells and CD4+ T cells, respectively. Similarly, the F1 scores improve from 0.8085, 0.9029, and 0.8082 to 0.9584, 0.9430, and 0.9396 for these cell types, respectively. We have also compared the annotation results on the DIMR-based cell masks (Supplementary Fig. 44). While there are some variations due to the differences from segmentation masks, this comparison demonstrates accuracy improvements as well. Fig. 4DeepSNiF enhances lymphocyte analysis.a Manual annotations for lymphocytes and comparisons with DIMR and DeepSNiF phenotyping results with DeepSNiF-based cell masks. The white contours represent the differential phenotyping results between the annotated and DIMR/DeepSNiF results. b Annotation evaluations of DIMR and DeepSNiF by both Jaccard and F1 scores across all the tissues ($$n = 15$$ biologically independent samples). Box center indicates median, box edges 25th and 75th percentile, and whiskers minimum and maximum percentile. P values were calculated through two-sided Wilcoxon matched-paired test (**$P \leq 0.01$ and ****$P \leq 0.0001$). c Representative images of lymphocyte markers after DeepSNiF denoising from specimens of normal (upper left), myelodysplastic syndromes (MDS, upper right), acute myeloid leukemia (AML, lower left) and AML with lymphoid aggregate (lower right) tissue samples. d Nearest distance comparisons between different cell types of normal ($$n = 5$$ biologically independent samples), MDS ($$n = 5$$ biologically independent) samples and AML ($$n = 4$$ biologically independent samples) tissues from manual, DIMR and DeepSNiF phentyping results. Tukey box center indicates median, box edges 25th and 75th percentile, and whiskers the highest and lowest values that are not outliers. Outliers (single points) are defined as values that are more than 1.5 times the interquartile range). e Cell densities comparisons of normal ($$n = 5$$ biologically independent samples), MDS ($$n = 5$$ biologically independent samples) and AML ($$n = 4$$ biologically independent samples) tissues from manual, DIMR and DeepSNiF phentyping results. The center bars define the mean, and the error bars are $95\%$ confidence interval. P values were calculated through two-sided Kolmogorov-Smirnov test (*$P \leq 0.05$, **$P \leq 0.01$, and ****$P \leq 0.0001$). f, g *Correlation analysis* between CD4 T cell and B cell, monocyte/macrophage densities per tissue from manual, DIMR and DeepSNiF phentyping results. The data from the reference group in f comes from annotated data; while that from the reference in g comes from annotated (CD4 T cells) and DeepSNiF (monocytes/macrophages) results, separately. The solid lines represent the fitting results with the data while the dashed lines represent $95\%$ confidence interval. Scale bar: a 85 μm; c 112 μm.
Subsequently, the tissues were classified as normal morphology (Normal), myelodysplastic syndromes (MDS) and acute myeloid leukemia (AML). The improvements in visual quality afforded by DeepSNiF denoising facilitated manual review of lymphocyte staining patterns for annotation annotation of lymphocyte subpopulations (Supplementary Fig. 45). B and T cell populations are scattered throughout the bone marrow cellularity in normal and malignant specimens, and lymphoid aggregates are occasionally present (Fig. 4c). To characterize the density and distance relationships between lymphocyte subpopulations, samples were analyzed in cohorts of extent of malignant blast involvement, after exclusion of the lymphoid aggregate outlier. Nearest neighbor distances between B cells, CD4 T and B cells, and CD4 and CD8 T cells were calculated for different disease tissues (Fig. 4d). Overall, the distributions from DeepSNiF are more concordant with annotated data. By contrast, those from DIMR are biased, with significant differences to the annotations due to cell misclassifications.
Automated DeepSNiF denoising reveals that as disease develops, the B cell distances become larger ($P \leq 0.01$); The distances between CD4 T and B cells in normal and MDS tissues are greater than those of AML ($P \leq 0.0001$); And CD4 and CD8 T cells in normal tissues trend towards longer distances than those in MDS ($$P \leq 0.0916$$). Interestingly, the overall distances between CD4 T and B cells in MDS tissues are greater than those of Normal samples ($P \leq 0.01$). These findings hold for DeepSNiF denoised data in distances between B cells from MDS to AML ($P \leq 0.05$), and from Normal to MDS samples for CD4 and CD8 T cells ($$P \leq 0.0929$$) and CD4 T to B cells ($P \leq 0.05$). However in non-DeepSNiF denoised DIMR data, the trends between B cells has been violated from MDS to AML by DIMR data ($$P \leq 0.4923$$), as well as those from Normal to MDS between CD4 and CD8 T cells ($$P \leq 0.4762$$), and CD4 T to B cells ($$P \leq 0.6685$$). From this point, DeepSNiF is able to correct the distorted cell spatial distributions from less accurate annotations caused by noise, which may further enhance downstream cell-specific spatial analyses.
We also calculated the cell densities per tissue of these lymphocytes (Fig. 4e). The DeepSNiF results are closer to those annotated data for B cell and CD4 T cell. By contrast, the CD8 T cell densities from DIMR, DeepSNiF and annotated data are close to each other. Additionally, we observe higher B cell and CD4 T cell densities in Normal tissues than others, and higher B cell density of MDS than that of AML. No obvious developing trend for CD8 T cells are observed as the disease status changes. Furthermore, we have analyzed the correlations between the densities from different cell types (Fig. 4f, g). Note that in the reference groups, the B and CD4 T cell densities are generated from the annotated data, while the monocyte/macrophage density comes from DeepSNiF data. This is because the relative change of the monocytes/macrophages by DeepSNiF is smaller compared to those of B and CD4 T cells (Fig. 3b) and because DeepSNiF achieves higher accuracy than DIMR for the cell phenotyping (Fig. 3e, f). From the reference group in Fig. 4f, the densities of CD4 T and B cells are negatively correlated with each other (PCC: -0.4371). Nevertheless, DIMR result indicates no correlation between the cell densities (PCC: 0.0090), which demonstrates false annotations hinder true relational definition between different cell types. Again, the negatively correlated relationship can be uncovered using automated DeepSNiF (PCC: −0.2837). Likewise, the DIMR data fails to detect the correlation between the densities of CD4 T cell and monocyte/macrophages (PCC: 0.2781; Fig. 4g). Corrected by DeepSNiF, the measured correlation (PCC: 0.4503) approximates the reference finding (PCC: 0.4310).
## Discussion
With the rise of novel multiplexed technologies for the characterization of cellular context in health and disease, IMC has emerged as a valuable tool to investigate immunophenotypes while preserving spatial information8–16. Differing from traditional multiplexed imaging approaches based upon fluorescence microscopy, IMC allows for simultaneous acquisition of more than 40 cell-specific markers with greatly suppressed channel crosstalk, and avoids tissue and marker degradation in multi-round staining protocols. Furthermore, it eliminates autofluorescence and background signal issues that are inherent in fluorescent microscopy. The high-dimensional datasets then enable complex microenvironment analysis. However, IMC suffers from unique hot pixel and shot noise features. Analyzing raw IMC data without further restoration may lead to distortions, even errors, in downstream analysis. Contemporary denoising strategies10,14,15,19,21 are usually not adaptive or effective for these particular noise conditions. For example, the parameters of some methods must be tuned manually, which is not suitable for large datasets and may cause subjective, batch, and channel-specific errors.
In this work, we propose IMC-Denoise to account for the specific technical noise present in IMC images. In this pipeline, the DIMR algorithm is first applied to adaptively remove hot pixels. It does not use a user-defined intensity threshold or range to define hot pixels, eliminating the impact of the density and intensity variations of hot pixels in different datasets or markers. Instead, it builds a histogram from the differential maps of raw images followed by an iterative outlier detection algorithm. In comparison with other methods, DIMR achieves more robust hot pixel detection capability and normal pixel preserving performance. After hot pixel removal, the DeepSNiF algorithm is proposed to restore image quality. I-divergence is derived as the optimal loss function for this denoising task. Due to the absence of noise-free IMC images and incompatibility with repeated scanning to generate training labels8, we applied a masking strategy with stratified sampling from Noise2Void24. This enabled self-supervised training for this denoising task, in which multiple pixels are randomly masked and replaced by its adjacent pixels. With the continuity of antibody signals in IMC, Hessian norm regularization41–43 is added in the loss function to boost the denoising performance. In DeepSNiF, we train a single network for a single marker, which reduces the memory allocated for training. Nevertheless, we note that DeepSNiF also works on multi-marker training (Supplementary Fig. 46). In another aspect, this demonstrates that DeepSNiF works on the markers stained for morphologically heterogeneous markers, since the variant features have been learned in the training process. In addition, monocytes/macrophages are morphologically heterogeneous so that the successful denoising of CD14/CD169 (Fig. 1f) validates the adaptability of DeepSNiF as well. In fact, the networks are able to learn all the features existing in the training images but not focus on any specific structures. As a result, markers with interstitial staining patterns (e.g. vessels, fibrosis, reticular cytoplasmic projections) can be well restored (CD31, CD34 and Collagen III in Figs. 1 and 2 and Supplementary Figs. 19 and 20). However, small areas of staining at the size of a sub-cellular synapse (e.g. 1–2 μm diameter) will not be successfully distinguished by IMC due to its relatively low resolution of 1 μm. Therefore, the network cannot learn the features of such small structures. The trained network can be employed to other datasets which share similar features (Supplementary Fig. 22).
To determine the applicability of our approach, reference denoising algorithms were utilized to rigorously evaluate IMC-Denoise on both simulated data and multiple pathological patient datasets. Compared to other methodologies, both DIMR and DeepSNiF achieve the best denoising performance, qualitatively and quantitatively. Orthogonal approaches that have not been previously tested in evaluation of IMC restoration are also used to verify the image quality improvement by IMC-Denoise. This pipeline can be further extended by existing analytical processing pipelines including Mesmer/DeepCell and ark-analysis25 or MCMicro26. If warranted, one may18 address spillover issues after hot pixel removal and shot noise filtering, as indicated in Eq. [ 1]. A related modality, MIBI6,7, shares several image formation and noise features with IMC, and the denoising pipeline deployed here may also enhance MIBI datasets.
IMC-*Denoise is* effective at removing background noise and enhancing downstream analysis of IMC data with limited, subjective, user-input. Multiple datasets processed by DIMR and DeepSNiF were compared with state-of-the-art IMC background removal methods, including single threshold binarization, semi-automated Ilastik-based21, and MAUI19, using the F1 score as the accuracy metric to evaluate the results. The qualitative and quantitative results indicate DeepSNiF can affect significant background noise removal, and is superior to tedious semi-automated approaches. In particular, DeepSNiF is capable of unmixing specific IMC staining signal from background noise. This means that even the thesholding approach for background removal is not essential after DeepSNiF denoising.
Conventional workflows typically use manual gating strategies combined with prior cell marker knowledge to identify and compare cell types in pathological samples. We used real world data and these methods to evaluate the IMC denoising algorithm for single-cell analyses, and compared to DIMR, DIMR_Ilastik, and DeepSNiF. Automated IMC-Denoise performs equally or superior to the semi-manual Ilastik-based method in downstream single cell analysis, and DeepSNiF notably enhances cell clustering and annotation. Quantitative evaluations of cell phenotyping results indicate the improvement of sensitivity and specificity by DeepSNiF denoising. Further validations with DIMR-based cell masks demonstrate the robustness of IMC-Denoise to variant cell segmentation results. For lymphocyte annotation, Jaccard and F1 scores demonstrate that DeepSNiF performs significantly better than DIMR on phenotyping of B, CD8-positive T and CD4-positive T cells. Further, spatial distribution and cell density correlation analysis indicate less accurate annotations by the data denoised solely by DIMR, leading to biased conclusions. With the data denoised by DeepSNiF, such distortions can be corrected and more accurate downstream analysis is achieved.
As noted, DeepSNiF enhances all the markers and their downstream analysis. However, the marker channels with high noise levels benefit to a larger degree. In theory, there is no maximum noise level present for denoising algorithms. Even under some extremely noisy conditions (CD20 and CD3 in Fig. 1f, markers in Supplementary Figs. 18 and 20), DeepSNiF improves the image quality. Nevertheless, lower SNR in raw images means lower specific information content and thus the quality of the restored images are lower (Supplementary Figs. 6–11). Because of the signal-dependent characteristics of shot noise, the noise components of high SNR channels contribute less to overall image quality, and thus have lower impact on downstream analysis. Empirically, we find that denoising by DeepSNiF can be omitted when the mean expressions of positive markers are larger than 7 (MPO, CD15, and CD235a), however denoising all marker channels improves performance and is not computationally intensive.
Limitations of IMC-Denoise include the inability to remove large hot pixel clusters, as DIMR cannot discriminate these larger areas of outliers from signal (Supplementary Fig. 47). Further the self-supervised DeepSNiF algorithm cannot reach the accuracy of supervised denoising methods due to unavailability of ground truths (Supplementary Figs. 8 and 9). Nevertheless, DIMR can remove single hot pixels and small hot clusters of several consecutive pixels, and DeepSNiF performs better than other unsupervised and self-supervised denoising methods on IMC datasets. To conclude, we have developed the content aware IMC-Denoise for improved IMC image quality and quantitative accuracy. Predicated on a novel combination of differential map-based and self-supervised CNN-based algorithms, this open source pipeline removes hot pixels and effectively suppresses shot noise in multiplexed IMC data. Multiple image and cell-based analyses from different IMC datasets verified the enhancements brought by this approach, with the ability to resolve significant cellular phenotypic and spatial information approximating manual annotation. We have provided tutorials to help users implement IMC-Denoise (Supplementary Note 4 and https://github.com/PENGLU-WashU/IMC_Denoise). We expect IMC-Denoise to become a widely used pipeline in IMC analysis due to its adaptability, effectiveness and flexibility.
## Human bone marrow dataset
Sections were cut in 4-6 μm thickness from formalin-fixed paraffin-embedded (FFPE) blocks of ethylenediaminetetraacetic acid (EDTA)-decalicifed bone marrow trephine biopsy specimens. Three patients demonstrated normal morphology, and 4 patients were diagnosed with myelodysplastic syndromes (MDS), with additional later timepoints obtained at disease progression including acute myeloid leukemia (AML). Use of specimens for secondary analysis in this study was approved by the Washington University in St. Louis Institutional Review Board (#201912110). Informed consent was waived, per IRB-approved protocol.
## Tissue staining and IMC data acquisition
Descriptions of cell markers and isotope tags are provided in Supplementary Tables 2–5. Staining was performed according to Fluidigm IMC recommendations for FFPE as follows. Briefly, tissue sections were dewaxed in xylene and rehydrated in a graded series of alcohol. Epitope retrieval was conducted in a water bath at 96 °C in Tris-EDTA buffer at pH 9 for 30 minutes, then cooled and washed in metal-free PBS. Blocking with Superblock (ThermoFisher) plus $5\%$ FcX TruBlock (Biolegend) was followed by staining with antibody cocktail prepared in $0.5\%$ BSA and metal-free PBS overnight at 4 °C. Sections were washed in $0.02\%$ TritonX100 followed by metal-free PBS, then nuclear staining was performed using 1:200 or 1:300 dilution of Intercalator-Ir (125 μm, Fluidigm) solution for 30 min, followed by ddH2O for 5 min. Slides were air-dried before IMC measurement.
The abundance of bound antibody was quantified using the Hyperion imaging system (Fluidigm) controlled by CyTOF Software (version 7.0.8493), with UV-laser set at 200 Hz. Count data were then converted to tiff image stacks for further analysis using MCD Viewer (version 1.0.560.6, Fluidigm) or imctools (Bodenmiller lab, https://github.com/BodenmillerGroup/imctools).
## Tissue staining and IF data acquisition
For IF staining, tissue was prepared using the same protocal in IMC staining, then stained overnight at 4 °C with a single metal-conjugated primary antibody (CD3, CD4, CD169, or CD61 in Supplemental Table 5). The single-stained tissue was washed, then stained with secondary antibody (donkey anti-rabbit AF647 or goat anti-mouse AF750, Invitrogen, 2 mg/mL diluted 1:400 in $0.5\%$ BSA in PBS) at room temperature 1 h, followed by additional washing in PBS and DAPI (1 μg/mL) staining. Slides were mounted with SlowFade Glass antifade reagent (ThermoFisher) and # 1 $\frac{1}{2}$ coverslips. Images were acquired using Leica DMi8 inverted widefield microscope with Lumencor SOLA SE U-nIR light engine, DAPI/FITC/TRITC/Cy5/Cy7 filters, DFC9000 GT sCMOS camera, PL APO 20x/0.80 objective and LAS X software (version 3.7.3.23245). After image acquisition, coverslips were removed with gentle agitation in PBS, then Ir-intercalator staining, washing and drying performed as above for subsequent *Hyperion data* acquisition.
## Human pancreatic, breast cancer IMC datasets, and MIBI dataset
We applied the human pancreatic10, breast cancer12 IMC datasets, and a MIBI dataset19 to verify the flexibility of IMC-Denoise. All of these datasets are publicly available. The links are provided in the corresponding papers. In breast cancer dataset, CD3, CD20, CD45, CD68, c-Myc, EGFR, EpCAM, Ki-67, Rabbit IgG H L, Slug, Twist, and vWF were selected; in pancreatic cancer dataset, CD3, CD4, CD8, CD11b, CD14, CD31, CD44, CD45, CD45RO, CD56, Foxp3, and pS6 were selected; and in the MIBI dataset, CD3, CD4, CD8, CD11b, CD11c, CD14, CD20, CD31, CD45, CD68, CD206, and HLA-DR were selected. The two IMC datasets were processed by both DIMR and DeepSNiF. The MIBI dataset is only processed by DeepSNiF because no hot pixels are observed, and the hot clusters observed in MIBI images can be removed by the MAUI software package19. Details on software implementation can be found in the relevant sections below.
## Neural network implementation
The DeepSNiF neural network follows the U-Net architecture39 with Res-block modules40, in which the input and output images share the same size (Supplementary Fig. 13). U-Net architecture is widely used for image deblurring and denoising24,34. *In* general, the network is composed of an encoder and a decoder. Starting with the input, the encoder path gradually condenses the spatial information into high-level feature maps with growing depths; the decoder path reverses this process by recombining the information into feature maps with gradually increased lateral details. The information in adjacent feature maps transfers by convolving with 3 × 3 convolutional filters. The down-sampling and up-sampling are used in encoder and decoder for compressing and reconstructing features, performed here by 2 × 2 max-pooling and 2 × 2 up-sampling operations, respectively. Res-blocks are applied to facilitate efficient training. Each res-block contains a convolution layer, batch normalization and the rectified linear unit (ReLU) nonlinear activation, in which the batch normalization layer aims to speed up training process, ReLU could provide non-linearity in the network. Drop out layers are also added with 0.5 dropout rate after the central two res-blocks to mitigate overfitting. The skip connections link low-level features and high-level features by concatenating their feature maps. We use the softplus function (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\log (1+\exp (x))$$\end{document}log(1+exp(x))) as the activation function of the final layer and Eq. [ 2] as the loss function so that the output of the network is guaranteed to be non-negative.
The hot pixel-removed images are split into multiple 64 × 64 patches. Then, the patches are rotated by 90°, 180°, and 270°, and flipped as a data augmentation approach. In IMC images, foreground objects of interest might be distributed sparsely. In this case, the model might overfit the background areas and fail to learn the structure of foreground objects if the entire image is used indiscriminately for training. Therefore, patches from the background regions are excluded from training. In IMC images, pixels with intensity value 0 are considered as background. Afterwards, we define the background pixel ratio r as ratio of the number of background pixels and that of total pixels in the patch. Patches are considered as the background regions if r ≤ ρ, where ρ is the threshold and set from 0.2 to 0.99 for different channels and datasets. We applied a smaller ρ for the datasets less sparse images and vice versa. For good generalization ability of the network, we recommend at least 5000 patches for training. Before training, all the generated patches were percentile normalized (99.9–99.999, Supplementary Note 1.3.3). The percentile of 99.9 was applied for those training sets with extremely bright markers and larger percentile with relatively homogeneous intensity distributions. To balance the training efficiency and accuracy, $0.2\%$ pixels of each patch are masked and replaced by their neighbors using a stratified sampling strategy24. Finally, $85\%$ of the patches are set as training set and the rest as validation set.
All models were trained using Keras51 (version 2.3.1) on a single NVIDIA Quadro RTX 6000 GPU with 24 GB of VRAM. Adam optimizer52 was applied as the optimization algorithm with a initial learning rate of 0.001 for 200 epochs and batch size of 128. Learning rate is multiplied by 0.6 if validation loss does not improve for 20 epoches. The training details for all the datasets are summarized as Supplementary Tables 6–11. Note that the training datasets for N2V, MN2V, and DeepSNiF-NR are the same as those for DeepSNiF, and the training time of N2V, MN2V, and DeepSNiF-NR is approximately equal to that of DeepSNiF.
## Neural network inference details
Given a trained denoising model, we denoise full-size IMC images to avoid edge stitching effects. In order to achieve end-to-end prediction, we pad pixels around each image so their width and height are the multiples of 16 with reference to the network architecture (Supplementary Fig. 13). The padding pixels are the replications of the border pixels. Before prediction, the IMC images are normalized by the pre-calculated maximum of the corresponding channels in the training set. The outputs of the network are re-scaled and set as the denoised images. Given the trained denoising model, inference is fast. We are able to denoise IMC images with pixels of 1000 by 1000 less than 1 second per image on a single NVIDIA Quadro RTX 6000 GPU.
## Semi-automated Ilastik-based background noise removal
The semi-automated strategy in21 utilizes Ilastik segmentation53 to remove background noise in IMC images. An expert annotates signal or background regions of IMC images, and then Ilastik trains a random forest classifier for background noise removal. To achieve good denoising quality, large areas of background require manual labeling, which is labor-intensive. Furthermore, low image quality may affect the accuracy of this method as well. After background removal, the images are binarized to solve batch effect issues. Then the single cell information is calculated by counting the positive signal frequency rather than the mean intensity of every single cell. Here, we only utilized Ilastik (version 1.3.2post1) for background noise removal of IMC images, and still applied the mean intensities as the single cell profiles. To better reveal the enhancement by DeepSNiF, we applied the same labels for the trainings of DIMR and DeepSNiF-processed images.
## MAUI
MAUI software package7,19 includes spillover correction, noise removal and aggregate removal. All three steps require expert observation, which is also labor-intensive. Here, we only benchmarked the noise removal method in MAUI with our DeepSNiF algorithm. Briefly, it calculates the distances between a non-zero pixel and its K nearest non-zero neighbors, then builds a histogram based on the summations of the distances for all the non-zero pixels. Next, a threshold is manually selected to remove the pixels with larger summations by observing the distribution of the histogram. This method is based on the assumption that noisy regions look more sparse than normal regions. MAUI was implemented by the software package from https://github.com/angelolab/MAUI. The parameter K and the threshold were manually tuned to guarantee the best performance of MAUI (Supplementary Fig. 27).
## Pixel classification and cell segmentation
In single cell segmentation, the pixels in each image were defined as belonging to the nucleus, cytoplasm, or background compartment using the pixel classification module of Ilastik53 (version 1.3.2post1) as described in https://github.com/BodenmillerGroup/ImcSegmentationPipeline. In our experiments, the DIMR and DeepSNiF-processed images were both set as inputs for cell mask segmentation. The DeepSNiF-based cell masks were primarily used for further analysis, while those based on DIMR were validated for robustness in some cases. The Random Forest classifier was trained on the channels including CD38, MPO, CD14, CD71, CD11b, CD4, CD20, CD8a, CD15, Ki-67, CD3, CD45RO CD235a, Histone-H3, and Iridium. This allowed for the generation of maps that integrate for each pixel the probability of belonging to each of three compartments. Based on the trained classifier, probability maps were generated for the whole dataset and exported as tiff files in batch mode.
Subsequently, CellProfiler54 (version 3.1.8) was used to define cell masks for marker expression quantification. To define cell borders, nuclei were first identified as primary objects based on ilastik probability maps and expanded through the cytoplasm compartment until either a neighboring cell or the background compartment was reached. Cell masks were generated for identification of single cells and used to extract single-cell information from IMC images.
## Single-cell marker profile extraction and line fitting
We used HistoCAT55 (version 1.7.6) to extract single-cell marker profiles based on the IMC images and their segmentation masks. All the data were not transformed and used directly.
We conducted bisqaure line fitting for the extracted DIMR and DeepSNiF-processed single cell data with customized MATLAB (R2021a, MathWork) scripts.
## Positive cell identification
For initial identification of marker-positive cells, we modified the method described in14. Briefly, univariate Gaussian mixture models with scikit-learn56 (version 1.0.2) were used to estimate the positive thresholds of each marker. Before threshold estimation, all data were 99th-percentile normalized so that the impact of extremely bright cells can be eliminated.
For each channel, we performed model selection with models with 6–15 mixtures for DIMR data, in order to estimate the positive threshold accurately. We selected the model on the basis of the Davies-Bouldin index57 and identified a positive threshold for a given channel by considering both the distributions of cell profiles and the overall IMC image intensities. The estimated positive thresholds of single cell data from DIMR and DeepSNiF-based cell segmentations are summarized in Supplementary Tables 12 and 13.
## Cell-type annotation
A subset of markers extracted from DeepSNiF-based cell segmentation masks of the human bone marrow dataset were utilized for cell phenotypic annotation, including CD38, MPO, CD14, CD71, CD11b, CD4, CD169, CD20, CD8a, CD15, CD3, and CD235a. Before analysis, data were 99th-percentile normalized followed by Z-score normalization. Then the DIMR data was clustered by the Phenograph algorithm with 20 nearest neighbors of each cell46 with the Leiden community detection algorithm47 with resolution of 6.0, which resulted in over clustering with 117 clusters. *The* generated clusters were manually labeled with a broad ontogeny and the channels that were most abundant in each cluster (Supplementary Fig. 35), resulting in 9 cell types, including immune cell subsets (B cell, CD4+ T cell, CD8+ T cell, and plasma cell, monocyte/macrophages), erythroid, myeloid, and other CD4+ cells and others.
The DeepSNiF data clustering and annotation utilized a weighted K-nearest neighbor (KNN) approach ($K = 20$) to map the DeepSNiF data into the DIMR clusters. It first constructs a Jaccard graph between each cell from DeepSNiF and all the DIMR cells, and then maps the DeepSNiF data into the DIMR clusters with the shortest weighted distance. The leave-one-out DeepSNiF data was also annotated with the same approach. The Phenograph with Leiden algorithms were implemented by the software packages from https://github.com/jacoblevine/PhenoGraph and https://github.com/vtraag/leidenalg. DeepSNiF data annotation was implemented with customized python scripts. The same markers extracted from DIMR-based cell masks were also analyzed with the same approach (Supplementary Fig. 38).
Notably, multiple strategies were applied to reduce the noise impact during DIMR clustering: [1] Z-score normalization is consistent for handling different sources of noise in multiplexed cell data, including low intensity signal, high background signal, segmentation noise, and imaging artifacts, as verified by58; [2] the Jaccard graph construction in *Phenograph is* robust to noise, which is verified in46; and [3] over-clustering could improve the clustering accuracy58. Besides, we didn’t annotate the DeepSNiF and leave-one-out DeepSNiF data with the same approach of DIMR because [1] The community detection results by Leiden algorithm is random so that it is very difficult to compare the annotations from different data; and [2] the weighted KNN method for DeepSNiF and leave-one-out DeepSNiF clustering could clearly reveal the differences before and after the processing.
The manual cell-type annotation in Fig. 4 was based on the DeepSNiF-based cell segmentation masks. Briefly, DIMR and DeepSNiF images were overlaid with the cell masks in FIJI59. In some extremely noisy cases, the DIMR images were denoised by Gaussian filters to improve the annotation accuracy. Based on the signal in each cell mask, the cells were classified as B, CD8 T, CD4 T cells and other cells. Some positive signals were identified as hot clusters and discarded. The annotation results were manually recorded. To test the impact of segmentation masks on annotation results, we have also annotated the lymphocytes on DIMR-based cell masks (Supplementary Fig. 44).
## Fast interpolation-based t-SNE algorithm
For visualization, high-dimensional single-cell data of DIMR and DeepSNiF were reduced to two dimensions using the nonlinear dimensionality reduction algorithm fast interpolation-based t-SNE48. This algorithm was implemented by the software package in https://github.com/KlugerLab/FIt-SNE. Before the analysis, data were 99th-percentile normalized followed by Z-score normalization. The t-SNE parameters with perplexity of 50 and theta of 0.5 were used. The random seeds for the individual runs were recorded.
## Enrichment calculation of positive cell markers after DeepSNiF and leave-one-out DeepSNiF
To evaluate the effect of positive cell marker enrichment after DeepSNiF and leave-one-out DeepSNiF, the cell-type annotations before and after the processing were selected, and the percentage of positive markers on each cell types was calculated. The relative change was then defined as the difference between the percentage of positive markers after and before the processing.
## Enrichment calculation of cell types after DeepSNiF and leave-one-out DeepSNiF
To evaluate the effect of cell type enrichment after DeepSNiF and leave-one-out DeepSNiF, the positive cells for a given marker before and after the processing were selected, and the percentage of each cell type based on cell-type annotation was calculated. The relative change was then defined as the difference between the percentage of cell-type composition after and before the processing.
## Accuracy metrics
In simulation, the accuracy metrics including root mean squared error, peak SNR and structural similarity45 are used to access the image qualities because of the availability of ground truths. They are defined in Supplementary Note 3.2 in detail.
For the real experimental data, five types of metrics were used for quantitative evaluations. The standard deviation of background (STDB) and contrast-to-noise ratio (CNR) were used to evaluate the noise level and contrast of IMC images. CNR is defined as Eq. [ 3],3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{{{{\rm{CNR}}}}}}}}=({C}_{{{{{{{{\rm{sig}}}}}}}}}-{C}_{{{{{{{{\rm{bg}}}}}}}}})/{\sigma }_{{{{{{{{\rm{bg}}}}}}}}}$$\end{document}CNR=(Csig−Cbg)/σbgwhere Csig and Cbg are the mean of the signal and background and σbg is the STDB. In this metric, the signal and background regions of IMC images are manually annotated.
Pearson’s correlation coefficient (PCC) was used as the metric to reflect the similarity between two groups of data. The PCC between measured data Y and the reference *Yref is* defined as Eq. [ 4],4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{{{{\rm{PCC}}}}}}}}({{{{{{{\bf{Y}}}}}}}},{{{{{{{{\bf{Y}}}}}}}}}_{{{{{{{{\rm{ref}}}}}}}}})=\frac{E[({{{{{{{\bf{Y}}}}}}}}-{\mu }_{{{{{{{{\bf{Y}}}}}}}}})({{{{{{{{\bf{Y}}}}}}}}}_{{{{{{{{\rm{ref}}}}}}}}}-{\mu }_{{{{{{{{{\bf{Y}}}}}}}}}_{{{{{{{{\rm{ref}}}}}}}}}})]}{{\sigma }_{{{{{{{{\bf{Y}}}}}}}}}{\sigma }_{{{{{{{{{\bf{Y}}}}}}}}}_{{{{{{{{\rm{ref}}}}}}}}}}},$$\end{document}PCC(Y,Yref)=E[(Y−μY)(Yref−μYref)]σYσYref,where μY and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mu }_{{{{{{{{{\bf{Y}}}}}}}}}_{{{{{{{{\rm{ref}}}}}}}}}}$$\end{document}μYref are the mean values of images Y and Yref, respectively; σY and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\sigma }_{{{{{{{{{\bf{Y}}}}}}}}}_{{{{{{{{\rm{ref}}}}}}}}}}$$\end{document}σYref are the standard deviations of Y and Yref, respectively; and E represents arithmetic mean.
Furthermore, F1 score was used to evaluate the accuracy of background noise removal. F1 score and Jaccard score were used to evaluate the accuracy of cell annotation of B, CD8 T and CD4 T cells, which can be formulated as Eqs. [ 5] and [6], respectively.5\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{{{{\rm{F}}}}}}}}1\,{{{{{{{\rm{score}}}}}}}}=\frac{2{{{{{{{\rm{TP}}}}}}}}}{2{{{{{{{\rm{TP}}}}}}}}+{{{{{{{\rm{FP}}}}}}}}+{{{{{{{\rm{FN}}}}}}}}},$$\end{document}F1score=2TP2TP+FP+FN,6\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{{{{\rm{Jaccard}}}}}}}}\,\,{{{{{{{\rm{score}}}}}}}}=\frac{{{{{{{{\rm{TP}}}}}}}}}{{{{{{{{\rm{TP}}}}}}}}+{{{{{{{\rm{FP}}}}}}}}+{{{{{{{\rm{FN}}}}}}}}},$$\end{document}Jaccardscore=TPTP+FP+FN,where TP, FP and FN are the pixel number of true positives, false positives and false negatives, respectively. All of the evaluation process was implemented with customized MATLAB (R2021a, MathWork) scripts. RMSE, PSNR, SSIM, PCC, F1 score, and Jaccard score were computed using MATLAB built-in functions.
## Statistical analysis
Other than specially stated, quantitative data are presented as box-and-whisker plots (center line, median; limits, $75\%$ and $25\%$; whiskers, maximum and minimum). The two-sided Wilcoxon matched-paired test was used for the statistical significance determination of repeated measurements. The two-sided Kolmogorov-Smirnov test was used for the statistical significance determination of different distributions in Fig. 4d. All the statistical tests are implemented with Prism 9 (GraphPad Software Inc.). Statistical significance at $P \leq 0.05$, 0.01, 0.001, and 0.0001 are denoted by *, **, *** and ****, respectively. “ ns” means “no significance”.
## Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.
## Supplementary information
Supplementary Information Peer Review File Reporting Summary The online version contains supplementary material available at 10.1038/s41467-023-37123-6.
## Source data
Source data
## Peer review information
Nature Communications thanks Eric Gamazon, Febe van Maldegem and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Peer reviewer reports are available.
## References
1. Saeys Y, Van Gassen S, Lambrecht BN. **Computational flow cytometry: helping to make sense of high-dimensional immunology data**. *Nat. Rev. Immunol.* (2016.0) **16** 449-462. DOI: 10.1038/nri.2016.56
2. Adan A, Alizada G, Kiraz Y, Baran Y, Nalbant A. **Flow cytometry: basic principles and applications**. *Crit. Rev. Biotechnol.* (2017.0) **37** 163-176. DOI: 10.3109/07388551.2015.1128876
3. Hickey JW. **Spatial mapping of protein composition and tissue organization: a primer for multiplexed antibody-based imaging**. *Nat. Methods* (2022.0) **19** 284-295. DOI: 10.1038/s41592-021-01316-y
4. Lin JR. **Highly multiplexed immunofluorescence imaging of human tissues and tumors using t-CyCIF and conventional optical microscopes**. *Elife* (2018.0) **7** e31657. DOI: 10.7554/eLife.31657
5. Goltsev Y. **Deep profiling of mouse splenic architecture with codex multiplexed imaging**. *Cell* (2018.0) **174** 968-981. DOI: 10.1016/j.cell.2018.07.010
6. Angelo M. **Multiplexed ion beam imaging of human breast tumors**. *Nat. Med.* (2014.0) **20** 436-442. DOI: 10.1038/nm.3488
7. Keren L. **A structured tumor-immune microenvironment in triple negative breast cancer revealed by multiplexed ion beam imaging**. *Cell* (2018.0) **174** 1373-1387. DOI: 10.1016/j.cell.2018.08.039
8. Giesen C. **Highly multiplexed imaging of tumor tissues with subcellular resolution by mass cytometry**. *Nat. Methods* (2014.0) **11** 417-422. DOI: 10.1038/nmeth.2869
9. Baharlou H, Canete NP, Cunningham AL, Harman AN, Patrick E. **Mass cytometry imaging for the study of human diseases–applications and data analysis strategies**. *Front. Immunol.* (2019.0) **10** 2657. DOI: 10.3389/fimmu.2019.02657
10. Wang YJ. **Multiplexed in situ imaging mass cytometry analysis of the human endocrine pancreas and immune system in type 1 diabetes**. *Cell Metab.* (2019.0) **29** 769-783. DOI: 10.1016/j.cmet.2019.01.003
11. Damond N. **A map of human type 1 diabetes progression by imaging mass cytometry**. *Cell Metab.* (2019.0) **29** 755-768. DOI: 10.1016/j.cmet.2018.11.014
12. Jackson HW. **The single-cell pathology landscape of breast cancer**. *Nature* (2020.0) **578** 615-620. DOI: 10.1038/s41586-019-1876-x
13. Ali HR. **Imaging mass cytometry and multiplatform genomics define the phenogenomic landscape of breast cancer**. *Nat. Cancer* (2020.0) **1** 163-175. DOI: 10.1038/s43018-020-0026-6
14. Rendeiro AF. **The spatial landscape of lung pathology during COVID-19 progression**. *Nature* (2021.0) **593** 564-569. DOI: 10.1038/s41586-021-03475-6
15. Wu M. **Single-cell analysis of the human pancreas in type 2 diabetes using multi-spectral imaging mass cytometry**. *Cell Rep.* (2021.0) **37** 109919. DOI: 10.1016/j.celrep.2021.109919
16. Moldoveanu L. **Spatially mapping the immune landscape of melanoma using imaging mass cytometry**. *Sci. Immunol.* (2022.0) **7** eabi5072. DOI: 10.1126/sciimmunol.abi5072
17. Kuett L. **Three-dimensional imaging mass cytometry for highly multiplexed molecular and cellular mapping of tissues and the tumor microenvironment**. *Nat. Cancer* (2022.0) **3** 122-133. DOI: 10.1038/s43018-021-00301-w
18. Chevrier S. **Compensation of signal spillover in suspension and imaging mass cytometry**. *Cell Syst.* (2018.0) **6** 612-620. DOI: 10.1016/j.cels.2018.02.010
19. Baranski A. **MAUI (MBI analysis user interface)–an image processing pipeline for multiplexed mass based imaging**. *PLoS Comput. Biol.* (2021.0) **17** e1008887. DOI: 10.1371/journal.pcbi.1008887
20. 20.Zanotelli, V. & Bodenmiller, B. IMC segmentation pipeline: a pixel classification based multiplexed image segmentation pipeline. Zenodo10.5281/zenodo.3841960 (2017).
21. Ijsselsteijn ME, Somarakis A, Lelieveldt BP, Höllt T, de Miranda NF. **Semi-automated background removal limits data loss and normalizes imaging mass cytometry data**. *Cytometry A* (2021.0) **99** 1187-1197. DOI: 10.1002/cyto.a.24480
22. 22.Buades, A., Coll, B. & Morel, J. M. A non-local algorithm for image denoising. Proceedings of the 2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition.Vol. 2, p. 60–65 (IEEE Computer Society, 2005). .
23. Dabov K, Foi A, Katkovnik V, Egiazarian K. **Image denoising by sparse 3-D transform-domain collaborative filtering**. *IEEE Trans. Image Process.* (2007.0) **16** 2080-2095. DOI: 10.1109/TIP.2007.901238
24. 24.Krull, A., Buchholz, T. O. & Jug, F. Noise2void-learning denoising from single noisy images. 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), p. 2129–2137 (IEEE, 2019).
25. Greenwald NF. **Whole-cell segmentation of tissue images with human-level performance using large-scale data annotation and deep learning**. *Nat. Biotechnol.* (2022.0) **40** 555-565. DOI: 10.1038/s41587-021-01094-0
26. Schapiro D. **MCMICRO: A scalable, modular image-processing pipeline for multiplexed tissue imaging**. *Nat. Methods* (2022.0) **19** 311-315. DOI: 10.1038/s41592-021-01308-y
27. Anscombe FJ. **The transformation of poisson, binomial and negative-binomial data**. *Biometrika* (1948.0) **35** 246-254. DOI: 10.1093/biomet/35.3-4.246
28. 28.Russell, M. F. T. B. C. & Freeman, W. T. Exploiting the sparse derivative prior for super-resolution and image demosaicing. IEEE SCTV, p. 1–28 (Institute of Electrical and Electronics Engineers (IEEE), 2003).
29. 29.Silverman, B. W. Density Estimation For Statistics And Data Analysis (Routledge, 2018).
30. Makitalo M, Foi A. **Optimal inversion of the Anscombe transformation in low-count Poisson image denoising**. *IEEE Trans. Image Process.* (2010.0) **20** 99-109. DOI: 10.1109/TIP.2010.2056693
31. Garnett R, Huegerich T, Chui C, He W. **A universal noise removal algorithm with an impulse detector**. *IEEE Trans. Image Process.* (2005.0) **11** 1747-1754. DOI: 10.1109/TIP.2005.857261
32. Dong Y, Chan RH, Xu S. **A detection statistic for random-valued impulse noise**. *IEEE Trans. Image Process.* (2007.0) **16** 1112-1120. DOI: 10.1109/TIP.2006.891348
33. Finesso L, Spreij P. **Nonnegative matrix factorization and I-divergence alternating minimization**. *Linear Algebra Appl.* (2006.0) **416** 270-287. DOI: 10.1016/j.laa.2005.11.012
34. Weigert M. **Content-aware image restoration: pushing the limits of fluorescence microscopy**. *Nat. Methods* (2018.0) **15** 1090-1097. DOI: 10.1038/s41592-018-0216-7
35. Wang Z, Xie Y, Ji S. **Global voxel transformer networks for augmented microscopy**. *Nat. Mach. Intell.* (2021.0) **3** 161-171. DOI: 10.1038/s42256-020-00283-x
36. Chen J. **Three-dimensional residual channel attention networks denoise and sharpen fluorescence microscopy image volumes**. *Nat. Methods* (2021.0) **18** 678-687. DOI: 10.1038/s41592-021-01155-x
37. 37.Lehtinen, J. et al. Noise2Noise: Learning image restoration without clean data. Proceedings of the 35th International Conference on Machine Learning, p. 4620–4631 (Machine Learning Research Press, 2018).
38. 38.Batson, J. & Royer, L. Noise2self: Blind denoising by self-supervision. Proceedings of Machine Learning Research, p. 524–533 (Machine Learning Research Press, 2019).
39. 39.Ronneberger, O., Fischer, P. & Brox, T. U-net: convolutional networks for biomedical image segmentation. Medical Image Computing and Computer-Assisted Intervention, p. 234–241 (Springer, 2015).
40. 40.He, K., Zhang, X., Ren, S. & Sun, J. Identity mappings in deep residual networks. In Computer Vision – ECCV 2016. Lecture Notes in Computer Science, (eds Leibe, B., Matas, J., Sebe, N. & Welling, M.) Vol 9908 p. 630–645 (Springer, Cham., 2016) 10.1007/978-3-319-46493-0_38.
41. Lu P. **Blind image restoration enhances digital autoradiographic imaging of radiopharmaceutical tissue distribution**. *J. Nucl. Med.* (2022.0) **63** 591-597. DOI: 10.2967/jnumed.121.262270
42. Huang X. **Fast, long-term, super-resolution imaging with hessian structured illumination microscopy**. *Nat. Biotechnol.* (2018.0) **36** 451-459. DOI: 10.1038/nbt.4115
43. Zhao W. **Sparse deconvolution improves the resolution of live-cell super-resolution fluorescence microscopy**. *Nat. Biotechnol.* (2022.0) **40** 606-617. DOI: 10.1038/s41587-021-01092-2
44. Rashid R. **Highly multiplexed immunofluorescence images and single-cell data of immune markers in tonsil and lung cancer**. *Sci. Data* (2019.0) **6** 1-10. DOI: 10.1038/s41597-019-0332-y
45. Wang Z, Bovik AC, Sheikh HR, Simoncelli EP. **Image quality assessment: from error visibility to structural similarity**. *IEEE Trans. Image Process.* (2004.0) **13** 600-612. DOI: 10.1109/TIP.2003.819861
46. Levine JH. **Data-driven phenotypic dissection of aml reveals progenitor-like cells that correlate with prognosis**. *Cell* (2015.0) **162** 184-197. DOI: 10.1016/j.cell.2015.05.047
47. Traag VA, Waltman L, Van Eck NJ. **From Louvain to Leiden: guaranteeing well-connected communities**. *Sci. Rep.* (2019.0) **9** 1-12. DOI: 10.1038/s41598-019-41695-z
48. Linderman GC. **Fast interpolation-based t-SNE for improved visualization of single-cell RNA-seq data**. *Nat. Methods* (2019.0) **16** 243-245. DOI: 10.1038/s41592-018-0308-4
49. Chou A. **CD169+ macrophages provide a niche promoting erythropoiesis under homeostasis and stress**. *Nat. Med.* (2013.0) **19** 429-436. DOI: 10.1038/nm.3057
50. Williams P. **The distribution of T-cell subsets and the expression of immune checkpoint receptors and ligands in patients with newly diagnosed and relapsed acute myeloid leukemia**. *Cancer* (2019.0) **125** 1470-1481. DOI: 10.1002/cncr.31896
51. 51.Chollet, F. Keras. GitHubhttps://github.com/fchollet/keras (2015).
52. 52.Kingma, D. P. & Adam, J. Ba. Adam: A method for stochastic optimization. Preprint at arXiv https://arxiv.org/abs/1412.6980 (2014).
53. 53.Sommer, C., Straehle, C., Koethe, U. & Hamprecht, F. A. Ilastik: Interactive learning and segmentation toolkit. ISBI, p. 230–233 (2011).
54. Kamentsky L. **Improved structure, function and compatibility for CellProfiler: modular high-throughput image analysis software**. *Bioinformatics* (2011.0) **27** 1179-1180. DOI: 10.1093/bioinformatics/btr095
55. Schapiro D. **histoCAT: analysis of cell phenotypes and interactions in multiplex image cytometry data**. *Nat. Methods* (2017.0) **14** 873-876. DOI: 10.1038/nmeth.4391
56. Pedregosa F. **Scikit-learn: machine learning in Python**. *JMLR* (2011.0) **12** 2825-2830
57. Davies DL, Bouldin DW. **A cluster separation measure**. *IEEE Trans. Pattern Anal. Mach. Intell.* (1979.0) **2** 224-227. DOI: 10.1109/TPAMI.1979.4766909
58. Hickey JW, Tan Y, Nolan GP, Goltsev Y. **Strategies for accurate cell type identification in CODEX multiplexed imaging data**. *Front. Immunol.* (2021.0) **12** 727626-727639. DOI: 10.3389/fimmu.2021.727626
59. Schindelin J. **Fiji: an open-source platform for biological-image analysis**. *Nat. Methods* (2012.0) **9** 676-682. DOI: 10.1038/nmeth.2019
60. 60.Lu, P. et al. Human bone marrow IMC dataset and simulation dataset. Zenodo10.5281/zenodo.6533905 (2022).
61. 61.Lu, P. et al. Results of IMC-Denoise: a content aware pipeline to enhance Imaging Mass Cytometry. Zenodo10.5281/zenodo.7336448 (2022).
62. 62.Lu, P. et al. IMC-Denoise: a content aware denoising pipeline to enhance Imaging Mass Cytometry. Zenodo10.5281/zenodo.7559428 (2023).
|
---
title: Availability and readiness of primary healthcare facilities for the management
of non-communicable diseases in different districts of Punjab, Pakistan
authors:
- Sadaf Rashid
- Humaira Mahmood
- Asma Asma Iftikhar
- Nimrah komal
- Zikria Butt
- Hassan Mumtaz
- Duha Shellah
journal: Frontiers in Public Health
year: 2023
pmcid: PMC10036340
doi: 10.3389/fpubh.2023.1037946
license: CC BY 4.0
---
# Availability and readiness of primary healthcare facilities for the management of non-communicable diseases in different districts of Punjab, Pakistan
## Abstract
### Introduction
Non-communicable diseases (NCDs) and their effects are rising quickly. NCDs such as cardiovascular illnesses, diabetes, cancer, and chronic lung diseases cause $60\%$ of global deaths; of which, $80\%$ occur in developing countries. In established health systems, primary healthcare handles most of the NCD care.
### Methodology
This is a mixed-method study conducted to analyze the health service availability and readiness toward NCDs using the SARA tool. It included 25 basic health units (BHUs) of Punjab, which were selected through random sampling. Quantitative data were collected using the SARA tools, while qualitative data were collected through in-depth interviews with healthcare providers working at the BHUs.
### Results
There was a problem of load shedding of both electricity and water in $52\%$ of the BHUs, which leads to the poor availability of healthcare services. Only eight ($32\%$) out of 25 BHUs provide the diagnosis or management of NCDs. The service availability was the highest for diabetes mellitus ($72\%$), followed by cardiovascular disease ($52\%$) and then chronic respiratory disease ($40\%$). No services were available for cancer at the BHU level.
### Conclusion
This study raises issues and questions about the primary healthcare system in Punjab in two areas: first, the overall performance system, and second, the readiness of basic healthcare institutions to treat NCDs. The data show that there are many persisting primary healthcare (PHC) deficiencies. The study found a major training and resource deficit (guidelines and promotional materials). Therefore, it is important to include NCD prevention and control training in district training activities. NCDs are underrecognized in primary healthcare (PHC).
## Introduction
The incidence of non-communicable diseases (NCDs) as well as the damage they cause is increasing at a rapid pace. Recently, NCDs such as cardiovascular diseases, diabetes, cancer, and chronic lung diseases are responsible for approximately $60\%$ of all deaths that occur across the globe, with $80\%$ of these deaths taking place in developing counties [1]. Primary healthcare (PHC) is vital to achieve UHC and decrease the prevalence of NCDs [2].
The majority of the care for NCDs is handled by primary care in highly developed health systems; however, the majority of primary care institutions in less-developed settings is not equipped to diagnose and treat NCDs. Only $6\%$ of low-income nations compared with $85\%$ of high-income countries have the requisite equipment to take six basic primary care assessments [3]. Developing countries face an increase in NCDs from 1990 to 2010, with a $14\%$ decrease in communicable diseases. As of now, NCD is the leading cause of death; therefore, health systems must be strengthened and prepared to tackle the disease burden of NCDs [4].
To reduce NCDs' impact on individuals and society, all sectors, including health, must work together. Low-cost options exist to minimize modifiable risk factors (primarily cigarette use, bad food, physical inactivity, and hazardous alcohol use) and map the NCD epidemic and risk variables. Primary care can provide high-impact NCD medicines for early detection and treatment. Creating healthy public policies and reorienting health systems can have the largest impact. The prevention and control of NCDs are less effective in low-income countries. High-income countries cover four times more NCD services than low-income countries. The WHO's Worldwide Action Plan for the Prevention and Control of NCDs 2013–2020 aims to reduce premature deaths from NCDs by $25\%$ by 2025 through nine voluntary worldwide targets, including tobacco use, harmful alcohol use, unhealthy eating, and physical inactivity [5].
The WHO recommendations of essential medicines should be available to up to $80\%$ of people as part of the global action plan for NCDs. The WHO has developed the service availability and readiness assessment (SARA) tool for collecting data on equipment, health workforce, and readiness and serves as an accountability and evaluative tool [6].
Our study aims to assess the availability of equipment, the workforce of primary healthcare services for their readiness toward the increasing burden of NCDs, and perceptions of healthcare providers toward the provision and upgrading of services.
## Methods
This study is a cross-sectional mixed method. Quantitative data were collected from 25 basic health units of different districts of Punjab. The sample size estimation was carried out after a thorough literature search using the SARA tool, and the estimate was based on a similar study conducted in Bangladesh [6]. BHUs were chosen through the multistage sampling technique. In the first stage, eight districts were chosen by the lottery method, and in the second stage, three BHUs were randomly chosen (the lottery method) from sampled districts. An additional BHU was chosen for pilot testing of feasibility. The WHO's service availability and readiness assessment (SARA) tool was used for collecting quantitative data. The SARA tool general service readiness was measured by responses on the basic amenities and equipment: standard precautions for infection prevention, diagnostic capacity, availability of basic tests (hemoglobin, blood glucose, malaria diagnostic capacity, urine dipstick for protein, urine dipstick for glucose, HIV diagnostic capacity, syphilis RDT, and urine pregnancy test), and availability of essential medicines. Service-specific readiness for NCDs was measured through the availability of diagnostic and management services, pieces of training, and guidelines for the treatment of common NCDs (diabetes, cardiovascular disease, chronic respiratory disease, and cervical cancer).
A pilot study was conducted to look into the feasibility of data collection. The participants took almost 10–15 min to complete the informed consent and questionnaire. An incomplete questionnaire was excluded from the final analysis.
For collecting qualitative data, one focus group discussion (FGD) () with 10 medical officers currently working in BHUs and one in-depth interview with MS were conducted. Purposive sampling was used. Detailed notes were taken, and interviews were recorded with the permission of the participants. An interview guide was used to probe problems regarding the provision of services, suggestions for improvement, and public–private partnerships. Data were collected till the saturation of response. It was later transcribed, and themes and subthemes were identified through the coding of the response.
## Demographic description
In this study, 25 BHUs from the rural setups were included to analyze the readiness of primary healthcare facilities for the management of NCDs in different districts of Punjab. Two BHUs ($8\%$) were taken from each Basti Alamgir, Basti Malook, Kotla Faqir, Bhatia, and Darapur districts. Three BHUs ($12\%$) were taken from each Jandala, Multan, and Sagri districts. One BHU ($4\%$) was taken from each Jalalabad, Khairabad, and Mubarikpur.
## Service availability at BHUs of Punjab district
General health infrastructure and health service providers at BHUs. Service availability was analyzed by accessing the infrastructure and health workforce available at the BHUs included in the study. The availability of overnight inpatient beds was not present in 19 ($76\%$) BHUs, while one ($4\%$) BHUs have one inpatient bed availability, and five ($20\%$) BHUs have three inpatient overnight beds. Almost all the BHUs available in our study had dedicated maternity beds. Ten BHUs ($40\%$) have one dedicated maternity bed, and 15 BHUs have two dedicated maternity beds.
All the BHUs included in our study have only one general practitioner each, and none of the BHUs have any medical specialist, pharmacist, or lab technologist. Almost 15 ($60\%$) BHUs do not have any nursing facilities. Five ($20\%$) BHUs have only one nurse available, and five ($20\%$) BHUs have two nurses available at their unit. Midwives were present in all the BHUs, five ($20\%$) BHUs have two midwives, five ($20\%$) BHUs have three midwives, five ($20\%$) BHUs have four midwives, and 10 ($40\%$) BHUs have five midwives each. Community health workers were present in all BHUs included in our study.
## Communication and transport
Service readiness was analyzed by accessing the availability of modes of communication, access to ambulances in need of emergency situations, power supply and water supply resources, rooms with privacy, and adequate sanitation amenities. Ten ($40\%$) BHUs have 24 h available functioning landline facility, whereas 15 ($60\%$) BHUs do not have this facility. Twelve BHUs ($48\%$) have functioning mobile phones dedicated to the facility, while 13 ($52\%$) BHUs do not have this facility. None of the BHUs have functioning short-wave radio for radio calls.
A total of 11 ($44\%$) BHUs have functioning computers available in the facility, while 14 ($56\%$) BHUs lack this facility. Notably, 19 ($76\%$) BHUs have email access or Internet within the facility, while six ($24\%$) BHUs did not have any Internet access. In regard to ambulance access, 10 ($40\%$) BHUs have a functioning ambulance for the transport of patients stationed at the facility, while 15 ($60\%$) BHUs do not have this facility. In total, 17($68\%$) BHUs have the access to ambulances/emergency vehicles stationed at nearby facilities, while eight ($32\%$) BHUs did not have such facilities. In regard to fuel for the ambulance, 10 ($40\%$) BHUs have the fuel for the ambulance/emergency vehicles available on the day of the interview, while 15 ($60\%$) BHUs do not have the fuel facility.
## Power supply, water supply, sanitation, and privacy
The power supply of all the BHUs was analyzed by checking the sources of electricity at the BHUs and their hours of availability. It was observed that all the BHUs were having a central electrical supply from the national grid. Seven ($28\%$) BHUs were using only standalone electric medical devices/appliances (e.g., epi cold room, refrigerator, and suction apparatus), while the remaining 18 ($72\%$) were fulfilling all the electrical needs of the facility from the main supply. A total of 12($48\%$) BHUs do not have the facility of a functional generator if the main source of supply is not functional, and the remaining 13 ($52\%$) do not know whether the unit has a functional generator or not. Three out of the 25 BHUs included in the study have a generator (fuel or battery-operated generator) as a secondary or backup source of electricity, while 18 have a solar system as a backup ($72\%$). Electricity is always available in 12 BHUs ($48\%$), while it is frequently available in 13 ($52\%$) BHUs. In regard to the water supply, 20 ($80\%$) BHUs have a 5- to 8-h water supply, and five ($20\%$) BHUs have a 24-h water supply. Five ($20\%$) BHUs used supply water as the most commonly used source of water for the facility at the time of the interview, and 25 BHUs used tube well or borehole water as the most commonly used source of water for the facility at the time of the interview. Seven ($28\%$) BHUs have auditory privacy, and 18 ($72\%$) BHUs have both auditory and visual privacies. All of the 25 BHUs have a flush toilet (latrine) within the premises in functioning condition and accessible for general outpatient client use.
## Reuse of equipment and healthcare waste management
Processing equipment for reuse is an important parameter for the good hygiene of a health unit. We analyzed the presence of some processing equipment in the 25 BHUs included in our study. Four ($16\%$) BHUs do not have any electric autoclave, while three ($72\%$) have a non-functioning electric autoclave, and 18 ($72\%$) have a properly functioning electric autoclave. In total, 17($68\%$) out of 25 BHUs do not have the facility of a non-electric autoclave, while three ($12\%$) have a non-functioning non-electric autoclave, and five ($20\%$) BHUs have a fully functional non-electric autoclave. Seven ($28\%$) BHUs do not have an electric dry heat sterilizer, while three ($72\%$) have a non-functioning electric dry heat sterilizer, and 15 ($60\%$) have a properly functioning electric dry heat sterilizer.
Five ($20\%$) out of 25 BHUs do not have the facility of electric boilers or steamers, while 20 ($80\%$) BHUs have fully functional electric boilers or steamers. Non-electric pots with covers for boiling and steaming were not available in 17 ($68\%$) BHUs, while eight ($32\%$) have a fully functional non-electric pot. The heat source for non-electric equipment was not available in 24 ($96\%$) BHUs, while only one BHU has its availability. Of note, 11 BHUs used two-chamber burn incinerators for sharp waste disposal, 10 BHUs used open burning in a pit or protected ground, and four BHUs used dump without burning in a covered pit.
## Basic equipment
The readiness of basic equipment is a very important parameter in accessing the quality of primary healthcare. Out of the 25 BHUs, 22 ($88\%$) have a properly functioning adult weighing scale, while only three ($12\%$) BHUs do not have this basic equipment. A total of 17 ($68\%$) BHUs have child weighing scales, while eight ($32\%$) do not have this facility. Notably, 18($72\%$) BHUs have the infant weighing scale, while seven ($28\%$) do not have this basic equipment. Stadiometer was available at 20 ($80\%$) BHUs and not available at five ($20\%$) BHUs. A thermometer, stethoscope, and blood pressure apparatus were present in all the included BHUs. Nine ($36\%$) BHUs do not have the facility of oxygen concentrators, while 16 ($64\%$) BHUs were equipped with this device. All of the 25 BHUs included in the study were having oxygen cylinders. The central oxygen supply was not available in 17 ($68\%$) BHUs, while eight ($32\%$) BHUs were equipped with this facility. In total, 12 ($48\%$) out of the 25 BHUs have the flowmeter with oxygen, while 13 ($52\%$) BHUs were deprived of it. Oxygen delivery apparatus was available in 22 ($88\%$) BHUs and absent in three ($12\%$) BHUs. Intravenous infusion kits were functioning properly in 20 ($80\%$) BHUs, while it was not available in five ($20\%$) BHUs.
## Laboratory diagnostic facility
Laboratory testing is one of the principal tools in the diagnosis of a particular disease. The availability of these testing facilities at the level of BHUs is one of the important parameters in a primary healthcare management system. In our study, 17 ($68\%$) out of the 25 included BHUs, and an onsite diagnostic testing facility was available. The facility for blood glucose tests using a glucometer was available at the BHUs. A total of 12 ($48\%$) BHUs had urine glucose dipstick testing and hemoglobin testing was taking place at 22 ($88\%$) BHUs. General microscopy was taking place at six ($24\%$) BHUs. Stool routine examination was taking place at eight ($32\%$) BHUs. ABO blood grouping, rhesus blood grouping, urine microscopy testing, and blood group serology testing facilities were available at seven ($28\%$) BHUs.
None of the BHUs were providing urine protein dipstick testing, urine ketone dipstick testing, special renal function test, liver function tests cross matching, serum electrolyte testing, and CSF/body fluid testing. The availability of general and special laboratory equipment along with laboratory reagents was analyzed. It was reported that six out of 25 BHUs have a fully functional light microscope. The availability of glass slides and cover slips was $20\%$ (five out of 25 BHUs). Out of 25 BHUs, seven ($28\%$) BHUs have a fully functional refrigerator, 13 ($52\%$) BHUs have a non-functioning refrigerator, and five ($20\%$) BHUs do not have a refrigerator facility.
Seven ($28\%$) BHUs have the facility of a glucometer, five ($20\%$) BHUs do not have the facility of the glucometer, and 13 ($52\%$) BHUs have a non-functioning glucometer. Glucometer test strips were available at seven ($28\%$) BHUs, while in 13 ($52\%$) BHUs non functioning glucometer test strips were available, and five ($20\%$) BHUs lacked this facility. Colorimeter or hemoglobinometer was available at four ($16\%$) BHUs, five ($20\%$) BHUs have non-functioning colorimeters, while 16 ($64\%$) do not have the facility of a colorimeter.
The incubator was available in a non-functioning condition in nine ($36\%$) BHUs, and 16 ($64\%$) BHUs do not have the facility. The centrifuge was available in a functioning condition at five ($20\%$) BHUs. White blood counting chamber was present in a functioning condition at six ($24\%$) BHUs.
## Imaging diagnostic facility
The availability of imaging diagnostic facilities was analyzed in all the BHUs included in this study. Out of 25 BHUs, only six ($24\%$) have the facility of imaging diagnostics. None of the BHUs were performing CT scans or MRIs as they do not have the equipment. A total of 19 ($76\%$) out of 25 BHUs do not have the facility of x-rays and ultrasound imaging. Two ($8\%$) have x-rays and ultrasound machines but they are not in a functioning condition, while four ($16\%$) BHUs have the proper functioning x-rays and ultrasound machine setups. Two ($8\%$) BHUs have the facility of a properly functioning ECG machine, while three ($12\%$) BHUs have a non-functional ECG machine.
## Indicators for non-communicable disease service availability and readiness
Service availability and service readiness at the primary healthcare levels or at the level of basic health units for the treatment, management, and prevention of NCDs: NCDs are emerging at a greater pace in Pakistan. The included 25 BHUs in the current study were accessed to check how many services these BHUs provide against NCDs. The availability of treatment and management of diabetes mellitus, cardiovascular disease, and respiratory disease was assessed. Eight ($32\%$) out of 25 BHUs provide the diagnosis or management of NCDs.
## Diabetes mellitus
Diagnosis and management of diabetes mellitus in patients were available at 18 ($72\%$) BHUs. A total of 15($60\%$) BHUs have national guidelines for the diagnosis and management of diabetes mellitus. Eight ($32\%$) BHUs have attempted the diagnosis and management of diabetes mellitus in the last 2 years. Service readiness for the treatment and management of diabetes mellitus was analyzed by accessing the equipment, diagnostic facilities, and medicines that are required for the management and treatment of diabetes mellitus. It was reported that all BHUs have the basic equipment including a stethoscope and the blood pressure apparatus, while 22 ($88\%$) BHUs have the facility of an adult weighing scale, and 20 ($80\%$) BHUs have a fully functional stadiometer. Onsite testing of blood glucose was present in all the BHUs. A urine glucose dipstick test was available at 12 ($48\%$) BHUs.
Medicines and commodities related to diabetes mellitus include gliclazide tablet or glipizide tablet (not available at any of the BHUs), metformin (available at all BHUs), insulin regular (available at seven ($28\%$) BHUs), glucose $50\%$ injection (available at six ($24\%$) BHUs), and glibenclamide cap/tab (available at 12 ($48\%$) BHUs).
## Cardiovascular diseases
In total, 13($52\%$) BHUs have the facility to diagnose and manage cardiovascular diseases such as hypertension in patients. Three ($12\%$) BHUs have attempted the diagnosis and management of cardiovascular diseases in the last 2 years. Eight ($32\%$) have the national guidelines for cardiovascular disease diagnosis and management. Service readiness against the treatment and management of cardiovascular disease was analyzed by accessing the equipment, diagnostic facilities, and medicines that are required for the management and treatment of diabetes mellitus. It was reported that all BHUs have the basic equipment including a stethoscope, the blood pressure apparatus, and oxygen cylinders, while only 22 ($88\%$) BHUs have the facility of an adult weighing scale. A properly functioning ECG machine was available at only two ($8\%$) BHUs. Medicines and commodities related to cardiovascular disease include metformin, aspirin, and beta-blockers (available at all BHUs), and ACE inhibitors and calcium channel blockers were not available at any of the BHUs.
## Chronic respiratory diseases
A total of 10 ($40\%$) BHUs diagnose and manage chronic respiratory diseases in patients. Eight ($32\%$) BHUs have national guidelines for chronic respiratory disease diagnosis and management. Three ($12\%$) BHUs have attempted diagnosis and management of CRD in the last 2 years. Peak flow meters were present in seven ($28\%$) BHUs. Service readiness against the treatment and management of cardiovascular disease was analyzed by accessing the equipment, diagnostic facilities, and medicines that are required for the management and treatment of chronic respiratory disease. Peak flow meters were present in seven ($28\%$) BHUs. Stethoscopes, oxygen cylinders, and blood pressure apparatuses were present in all the BHUs. Oxygen concentrators were present in 16 ($64\%$) BHUs.
Medicines and commodities for chronic respiratory diseases include prednisolone and epinephrine injection (both available at six ($24\%$) BHUs), and hydrocortisone injections were present in all the BHUs. None of the BHUs have salbutamol and beclomethasone inhaler.
## Cervical cancer
None of the BHUs provide the facility to diagnose cervical cancer in patients. These BHUs neither have the national guidelines for cervical cancer prevention and control nor training in cervical cancer prevention and control. The availability of auto-disable syringes was present in 15 ($60\%$) BHUs.
Two focal group discussions having (8–12) participants including the general practitioners, nurses, medical officers, and community health workers vaccinators were done to qualitatively analyze the primary healthcare services and availability at the BHUs of Punjab district for the NCDs. The collected data were analyzed through thematic analysis.
## Health service availability for the treatment and management of NCDs at BHUs
Available guidelines for the treatment and management of NCDs include diabetes mellitus, cardiovascular disease, chronic respiratory disease, and cancer: Regarding the service, only the practitioner were available for the management and treatment of NCDs at primary health levels that includes the BHUs, a detailed interview was conducted with the group of the service provider at the BHUs that infer that almost all the BHUs included in our study do not have the availability of medical specials, general doctors, and practitioners were only there. The management and treatment of NCDs were only offered at $24\%$ of the BHUs, and the reasons that the staff shared were the lack of specialized staff at the BHUs that include medical specialists, pharmacists, and medical technologists. The staff further adds that there is less accountability of the staff at the BHU level.
During the interview, it was also reported that treatment and management services were available at maximum BHUs, but they still lack the availability of a diabetic specialist.
They said that “BHUs lack professionals. It's hard to find doctors with NCD expertise to operate in community health stations.” They further added that “The budget for NCD primary healthcare is quite limited; funding is mostly from national goal programs, but they've been slashed.” Services related to cardiovascular diseases were not present in half of the BHUs due to the non-availability of specialists and lack of staff. The interview also reported that national guidelines for NCDs were not present in all the BHUs. A smaller number of BHUs have received such guidelines, that is, the main reason quoted by the staff and healthcare workers at the BHUs was the lack of interest of the government officials toward the BHUs and the lack of awareness regarding the NCDs. Even a small number of the staff had an awareness of the national strategies regarding the guidelines against the NCDs.
## Problems regarding the service available at the BHUs
The main reported problem during the in-depth interview appeared to be the lack of proper infrastructure of the BHUs, either general or medical. Maximum BHUs lack the trained staff for the NCDs. This results in weak monitoring of the NCDs at the primary healthcare level (BHUs).
## Suggestions for improvement
The provision of a good infrastructure including a good building with proper ventilation and sanitation along with an efficient group of specialists including a medical specialist, a lab technologist, and a pharmacist along with the midwives and community health workers can improve the level of service available at the level of BHUs.
It was suggested during the interview that service readiness can be improved for the public if there is an availability of an ambulance at all the BHUs and if there is the provision of a backup supply for electricity in case of load shedding. For a better mode of communication, they suggested a 24 h availability of the landline connection.
Suggestions gathered during the interviews for the problems related to laboratory and imaging diagnostics were mainly the repair of the non-functioning equipment and machines used for imaging and laboratory testing. The BHUs that do not have the imaging and the laboratory facility suggested induction of these machineries at the BHUs. The staff also suggested a good supply of medicine for the NCDs so that they are easily accessible to the public.
## Service readiness in terms of power supply, water supply, communication, and transport
Problems regarding power supply, water supply, communication, and transport: The main concern of the health service providers at the BHUs was the availability of landline connections for better communication among the public and healthcare providers. Many of the BHUs do not even have an ambulance facility. The workers also shared the problems of load shedding of electricity at maximum BHUs and also shared their problems that their routine is very much disturbed at the BHUs as they are unable to perform their duties without electricity as many BHUs do not have an electricity backup, for example, a generator. During the interview, water shortage was also reported at maximum BHUs.
## Service readiness in terms of imaging, laboratory diagnostics, and medicines for NCDs
Problems regarding imaging, laboratory diagnostics, and medicines at BHUs: Regarding imaging diagnostic services, maximum BHUs do not have the facility to provision of imaging and laboratory diagnostics. Moreover, those BHUs that have this facility have non-functional x-ray and ultrasound machines. Onsite blood testing was available at the BHUs, but not all the basic tests are performed at all the BHUs as they lack the equipment for special tests.
They said that, “Our BHU setup do[sic] not stock any medications. What we do have are medications designed for use in emergencies.” In regard to medicine some BHUs lacked the basic medicines for the NCDs, the reasons shared during the interview for these problems were mainly the negligence of the government official toward BHUs.
## Discussion
The study explored the service readiness of primary healthcare facilities for the management of NCDs in different districts of Punjab. Our study showed that non-communicable diseases are a burden on the economy and healthcare system of Pakistan. The current study analyzed the readiness of the basic health units toward the NCDs.
Non-communicable diseases are a burden on the economy and healthcare system of Pakistan. The current study analyzed the readiness of the basic health units toward the NCDs. It just had a general practitioner, nurse, midwives, and some community health workers. Researchers in Bangladesh, Haiti, Malawi, Nepal, and Tanzania found that a few institutions were entirely “equipped” to perform any one NCD service [7].
The purpose of this study was to assess the existing readiness of facilities in basic health units to provide care for NCDs. The evaluation focused on analyzing two aspects of the NCD-specific services: the availability of the services and the readiness of the services for patients with NCDs. The service availability was the highest for diabetes mellitus ($72\%$), followed by cardiovascular disease ($52\%$) and chronic respiratory disease ($40\%$). No services were available for cancer at the BHU level. The good prevalence score for CVD was the greatest ($22.6\%$), followed by the readiness score for diabetes ($17.2\%$). Other NCDs in the current study include diabetes, cardiovascular diseases, chronic respiratory diseases, and cancer [8].
Onsite testing of blood glucose was present in all the BHUs. Glibenclamide cap/tab was 19–$52\%$ available. A study found limited insulin availability (9–$16\%$ depending on insulin type), whereas another study found $34\%$. Captopril availability ranged from 13 to $48\%$, calcium channel blockers from 29 to $57\%$, and beta-blockers from 15 to $50\%$ [9]. The study in India found that the availability of all essential technology and medications in primary care ranged from $1.1\%$ in rural public facilities to $9.0\%$ in urban private facilities for the management of three NCDs. At present, neither private nor public primary care facilities, nor public secondary care facilities are fully prepared to effectively manage the burden of NCDs in India [10].
According to Nhsrc.pk 2019 report [11], $43\%$ of doctors and $98\%$ of support staff were present at the facilities (BHUs in Islamabad). Only $21\%$ of facilities had blood sugar testing. Metformin, sulphonylureas, and insulin were not present in 65, 79, and $93\%$ of the facilities. Thiazides and statins were not accessible anywhere, and beta-blockers were available in $29\%$. A total of $36\%$ of the institutions possessed a computer, but only one utilized it. Only $14\%$ of buildings had Internet. In rural areas, a model of reform at one BHU provides a framework for consolidation.
Concerns about the availability of NCD pharmaceuticals have been raised as a result of the findings of a second study, particularly in the public sector and rural areas. This has led to an inadequate supply of relatively affordable NCD medications. Metformin, glibenclamide, and other ACE inhibitors are no longer protected by patents and are available from multiple sources. In low- and middle-income countries, there is a concern regarding the availability of insulin and the price of insulin, with syringes and other diabetes goods adding to the burden of treatment [12].
## Limitations
We had some limitations while conducting the study. The first limitation is based on a small sample because of limited resources. The second limitation is based on the population. This study population may not be representative of all BHUs in Pakistan as it is covering only one province. Finally, the third limitation is based on the non-availability of the staff on the day of the interview at some BHUs.
## Conclusion
Improving efforts to prevent and control NCDs through BHUs (primary healthcare) in Punjab requires increased political commitment and financial investment at all levels of the health system. A major gap also exists in the primary healthcare (PHC) setting for the identification and treatment of NCDs. These findings indicate a sizable need that has to be filled, especially in terms of the provision of education and accessibility of tools.
## Recommendations and future research
A reorientation of the national health system is necessary to address the NCD burden and enhance NCD services at the primary healthcare center level. Workflow adjustments, duty distribution among primary care teams, recording patient information, and including community health workers in patient follow-up are all issues that plague primary care settings in India. The team's hierarchy inhibited quality improvements and team-based treatments. More studies on organizational behavior in primary care facilities in India are needed to help advance the state of primary healthcare in the country.
## Data availability statement
The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.
## Ethics statement
The studies involving human participants were reviewed and approved by National University of Medical Sciences. The patients/participants provided their written informed consent to participate in this study.
## Author contributions
All authors listed have made a substantial, direct, and intellectual contribution to the work and approved it for publication.
## Conflict of interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher's note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## References
1. Bawazir AA, Al-Surimi K, Suwaidan SD, AlShehri AM, AlFarhan AI, Aboulfotouh MA. **Capacity and readiness of primary health care centers for implementation of the basic strategy for prevention and control of non-communicable diseases in Saudi Arabia**. *Saudi Med J.* (2019.0) **40** 614-8. DOI: 10.15537/smj.2019.6.24164
2. Tesema AG, Abimbola S, Mulugeta A, Ajisegiri WS, Narasimhan P, Joshi R. **Health system capacity and readiness for delivery of integrated non-communicable disease services in primary health care: a qualitative analysis of the Ethiopian experience**. *PLOS Global Public Health* (2021.0) **1** e0000026. DOI: 10.1371/journal.pgph.0000026
3. Varghese C, Nongkynrih B, Onakpoya I, McCall M, Barkley S, Collins TE. **Better health and wellbeing for billion more people: integrating non-communicable diseases in primary care**. *BMJ.* (2019.0) **364** l327. DOI: 10.1136/bmj.l327
4. Kostova D, Husain MJ, Sugerman D, Hong Y, Saraiya M, Keltz J. **Synergies between communicable and non-communicable disease programs to enhance global health security**. *Emerg Infect Dis.* (2017.0) **23** 13. DOI: 10.3201/eid2313.170581
5. Islam MR, Laskar SP, Macer D. **A study on service availability and readiness assessment of non-communicable diseases using the WHO tool for Gazipur district in Bangladesh**. *Bangladesh J Bioethics* (2016.0) **7** 1-13. DOI: 10.3329/bioethics.v7i2.30785
6. Armstrong-Hough M, Kishore SP, Byakika S, Mutungi G, Nunez-Smith M, Schwartz JI. **Disparities in availability of essential medicines to treat non-communicable diseases in Uganda: a poisson analysis using the service availability and readiness assessment**. *PLoS ONE.* (2018.0) **13** e0192332. DOI: 10.1371/journal.pone.0192332
7. Moucheraud C. **Service readiness for non-communicable diseases was low in five countries in 2013–15**. *Health Affairs.* (2018.0) **37** 1321-30. DOI: 10.1377/hlthaff.2018.0151
8. Alam W, Nujhat S, Parajuli A, Banyira L, Mohsen WA, Sutradhar I. **Readiness of primary health-care facilities for the management of non-communicable diseases in rural Bangladesh: a mixed methods study**. *Lancet Global Health* (2020.0) **8** S17. DOI: 10.1016/S2214-109X(20)30158-3
9. Robertson J, Macé C, Forte G, de Joncheere K, Beran D. **Medicines availability for non-communicable diseases: the case for standardized monitoring**. *Global Health.* (2015.0) **11** 18. DOI: 10.1186/s12992-015-0105-0
10. Krishnan A, Mathur P, Kulothungan V, Salve HR, Leburu S, Amarchand R. **Preparedness of primary and secondary health facilities in India to address major non-communicable diseases: results of a national non-communicable disease monitoring survey (NNMS)**. *BMC Health Serv Res.* (2021.0) **21** 757. DOI: 10.1186/s12913-021-06530-0
11. 11.Nhsrc.pk. Available online at: https://phkh.nhsrc.pk/sites/default/files/2019-06/Rapid%20Assessment%20of%20NCDs%20Preparedness%20ICT%202017.pdf. *Nhsrc.pk*
12. Beran D, Yudkin JS. **Looking beyond the issue of access to insulin: what is needed for proper diabetes care in resource poor settings**. *Diabetes Res Clin Pract.* (2010.0) **88** 217-21. DOI: 10.1016/j.diabres.2010.03.029
|
---
title: 'Association of rheumatoid arthritis with bronchial asthma and asthma-related
comorbidities: A population-based national surveillance study'
authors:
- Jung Gon Kim
- Jiyeon Kang
- Joo-Hyun Lee
- Hyeon-Kyoung Koo
journal: Frontiers in Medicine
year: 2023
pmcid: PMC10036351
doi: 10.3389/fmed.2023.1006290
license: CC BY 4.0
---
# Association of rheumatoid arthritis with bronchial asthma and asthma-related comorbidities: A population-based national surveillance study
## Abstract
### Background
The aim of this study was to investigate the impact of rheumatoid arthritis (RA) on the prevalence of bronchial asthma and asthma-related comorbidities. We also aimed to identify the influence of RA on interrelationship between asthma and asthma-related comorbidities.
### Methods
From the Korean National Health and Nutrition Examination Survey, participants >40 years of age who completed questionnaires and spirometry tests were enrolled. Patient data on RA, asthma, allergic rhinitis, atopic dermatitis, chronic obstructive pulmonary disease (COPD), sinusitis, otitis media, and body mass index (BMI) were collected. Logistic regression and network analyses were performed.
### Results
A total of 14,272 subjects were enrolled, among which, 334 ($2.4\%$) had RA. RA was significantly associated with asthma (OR 2.32; $95\%$ CI 1.51–3.57), allergic rhinitis (OR 1.51; $95\%$ CI 1.08–2.10), and sinusitis (OR 1.64; $95\%$ CI 1.08–2.50). The network analysis of total patients revealed a positive interrelationship between asthma and allergic rhinitis, sinusitis, otitis media, atopic dermatitis, BMI, and RA. The interrelationship between asthma and sinusitis was stronger in the RA group. Of note, the relationship between asthma and BMI was distinctively found only in the RA group ($r = 0.214$, $P \leq 0.05$). In patients with asthma, the prevalence of obesity was $64\%$ in the presence of RA, and $40\%$ in the absence of RA ($$P \leq 0.034$$).
### Conclusion
This study supports the positive association of RA with asthma, allergic rhinitis, and sinusitis. Our analysis suggests a notable interrelationship between the presence of asthma and higher BMI values in patients with RA, indicating that asthma is more obesity-related in patients with RA.
## Introduction
Rheumatoid arthritis (RA) is a chronic inflammatory disease characterized by polyarthritis that causes the joint deformity. RA develops as a consequence of complex interactions between genetic and environmental factors [1]. Although various immune components can participate in pathogenesis, RA is generally believed to be a T cell-driven disease [2], especially involving pathogenic T helper 1 (Th1) and T helper 17 cells [3, 4]. However, a recent study reported the expansion of T helper 2 (Th2) cells in the blood of treatment-resistant patients with RA, which suggested the contribution of Th2 cells to the RA pathogenesis [5].
Asthma, a Th2 cell disorder, is characterized by airway inflammation but can also propagate systemically [6]. Accordingly, the relationship between asthma and systemic inflammatory disorders was previously elaborated (7–10). In addition to differences in immunologic mechanisms underlying RA and asthma, there are contrasting epidemiological patterns between the two conditions; RA tends to affect middle-aged to elderly females, while asthma is more prevalent in younger individuals [1, 11]. Despite those differences, a large-scale prospective cohort study and meta-analysis revealed that asthma was an independent risk factor for RA [8]. However, the association of RA with other asthma-related Th2 allergic diseases, including allergic rhinitis and atopic dermatitis, is controversial (12–16). Chronic obstructive pulmonary disease (COPD) and sinusitis, which could complicate asthma, were reported to be associated with RA (17–19). Nevertheless, their relationship with RA needs further confirmation. Briefly, the association of RA with bronchial asthma is considered to be robust, but not with asthma-related comorbidities.
Previously, the concept of an “allergic march” was suggested to describe the sequential development of allergic disorders (i.e., atopic dermatitis, asthma, and allergic rhinitis) [20]. In addition, the term “asthma-related comorbidities” was used to refer to the co-occurrence of asthma-related immunologic, inflammatory, or metabolic diseases such as allergic rhinitis, sinusitis, and obesity [19]. Immune and metabolic conditions are perturbed in RA [21], but the impact of RA on the relationship between asthma and its comorbidities has rarely been investigated.
This study aimed to reveal the association of RA with asthma and asthma-related comorbidities in a comprehensive manner. In addition, we investigated the comorbidity pattern of asthma as well as degree of interrelatedness between asthma and comorbidities in patients with and without RA. Herein, we conducted a population-based study using data from the Korean National Health and Nutrition Examination Survey (KNHANES).
## Study population and definition of variables
The KNHANES is a collection of nationally representative, cross-sectional, population-based health and nutritional surveys by the Korean Disease Control and Prevention Agency (KDCA) since 1998. KNAHNES includes a health interview, physical examination, laboratory tests, and nutritional questionnaire. The present study analyzed the 7 and 8th KNHANES (2016–2018, KNHANES-VII; 2019–2021, KNHANES-VIII) data from 2017 to 2019. Study participants had to meet all of the following inclusion criteria: [1] over 40 years old, [2] completed the questionnaires, and [3] completed spirometry tests. The maximum eligible age was 80 years. Medical data were obtained from the questionnaire or physical examination, which were performed by trained investigators following standardized procedures. Spirometry was performed for subjects >40 years of age using standardized equipment (model 1022; SensorMedics Corp, BD, Franklin Lakes, NJ, USA) according to guidelines of the American Thoracic Society/European Respiratory Society [22]. Diagnosis of RA, asthma, allergic rhinitis, atopic dermatitis, sinusitis, and otitis media was defined based on the answers of self-reported questionnaire asking “Have you been diagnosed with the disease by a doctor?” ( Yes/No) or “Do you take medicine or treatment for the disease?” Asthma patients were divided into two groups according to presence of airflow limitation [forced expiratory volume in 1 s (FEV1: L)/forced vital capacity (FVC: L) < 0.7] by spirometry test. COPD was defined as a spirometry result of airway obstruction (FEV1/FVC < 0.7) among adult >40 years of age without history of asthma according to the Global Initiative for Chronic Obstructive Lung Disease (GOLD) guidelines [23]. Obesity was defined as a body mass index (BMI) greater than 25 kg/m2 based on the World Health Organization’s recommendations for the Asian population [24]. The entire individual participated voluntarily and provided written informed consent. The KNHANES protocol was approved by the institutional review board of the KDCA. The data is available under following web site address.1
## Statistical analysis
For continuous variables, the subjects’ characteristics are presented as means and standard deviations, and for categorical variables, as relative frequencies. Continuous variables were compared using a t-test and categorical variables were compared using the chi-square test. For the multivariable logistic regression analysis of RA, two models were analyzed. Model 1 was adjusted for age, sex, body mass index (BMI), and current smoking status. It is widely recognized that individuals with allergic diseases, such as asthma, frequently suffer from other allergic and non-allergic conditions simultaneously. In order to determine the independent effect of comorbidities on RA, model 2 was further adjusted for the presence of asthma in addition to Model 1. Since cigarette smoking is known to be a risk factor for asthma, COPD, and RA, we also performed a sensitivity analysis to exclude the possible confounding effect of smoking by excluding subjects currently smoking [25]. Additionally, to evaluate the probable interactions between independent variables on this association, an interaction term was added to Model 1. For the correlation matrix, correlation between variables were performed using the cor function in the stats package. For the correlation network, each variable is represented as a specific node. The size of the node reflects the prevalence, and the color of the node refers to the category of variables: demographics (yellow), correctable features (green), and comorbidities (coral). The links between nodes indicated statistically significant associations ($P \leq 0.05$). The thickness of the links correlated with the strength of their association (correlation coefficient), and color indicated the direction of association: blue for positive and pink for negative. The igraph package was used to visualize the correlation networks. To compare the strength of correlations between given variables in the RA and non-RA groups, we compared two correlations based on independent groups using Zou’s method [26], with an alpha level of 0.05 and a confidence level of 0.95. The cocor package was used to compare the correlation strength of the paired variables between patients with and without RA. To avoid bias, propensity score matching was performed between RA and non-RA groups with a 1:1 ratio based on age and sex using the MatchIt package [27, 28]. Subsequently, a network analysis was conducted on the matched population. All statistical analyses were performed using R software (version 4.1.3).
## Characteristics of the participants
A total of 14,272 subjects aged >40 years who completed the questionnaires and spirometry tests from 2017 to 2019 were finally included. The patient inclusion process is depicted in Supplementary Figure 1. Among them, 334 ($2.4\%$) had RA, 380 ($2.8\%$) had asthma, 1,436 ($11.0\%$) had allergic rhinitis, 693 ($5.3\%$) had sinusitis, 545 ($4.2\%$) had otitis media, 177 ($1.4\%$) had atopic dermatitis, and 1,395 ($12.5\%$) had COPD. The demographics and clinical characteristics of participants with and without RA are compared in Table 1. In the RA group, the mean age and proportion of women were higher, but the number of current smokers was lower. The prevalence of asthma was higher in the RA group ($7.5\%$ vs. $2.8\%$, $P \leq 0.001$), but that of other allergic diseases and COPD were not different between the two groups. The comorbidity patterns of asthma patients with and without RA are compared in Table 2. The prevalence of obesity was significantly higher in the asthma patients having concomitant RA ($64.0\%$ vs. $40.2\%$, $$P \leq 0.034$$). The prevalence of comorbidities is also illustrated with Venn diagrams comparing the groups with and without RA (Supplementary Figure 2).
## Multivariable analysis for RA
Since the distribution of age, sex, and current smoking status were different between the groups (Table 1), those confounding variables, in addition to BMI, were adjusted for multivariable logistic regression (Model 1). The detailed results of the multivariable analysis are summarized in Figure 1. RA was significantly associated with asthma [odds ratio (OR) 2.32; $95\%$ confidence interval (CI) 1.51–3.57], but not with COPD (OR 0.69; $95\%$ CI 0.45–1.06). The association between asthma and RA was relevant in asthma patients without airflow limitation but not in those with airflow limitation. RA was also associated with the presence of allergic rhinitis (OR 1.51; $95\%$ CI 1.08–2.10) and sinusitis (OR 1.64; $95\%$ CI 1.08–2.50). However, atopic dermatitis or otitis media was not significantly associated with RA. Considering the inter-correlation between allergic diseases, Model 1 was further adjusted for the presence of asthma (Model 2), and RA was still significantly associated with allergic rhinitis (OR 1.44; $95\%$ CI 1.03–2.02) and sinusitis (OR 1.57; $95\%$ CI 1.03–2.40), independent of asthma.
**FIGURE 1:** *The odds ratio for rheumatoid arthritis in multivariable logistic regression analysis was adjusted by age, sex, current smoking status, and body mass index. The analysis was conducted using data from 13,718 patients with information about rheumatoid arthritis. Asthma w/o AL, asthma without airflow limitation; Asthma w AL, asthma with airflow limitation; COPD, chronic obstructive pulmonary disease; OR, odds ratio; CI, confidence interval; RA, rheumatoid arthritis. *P < 0.05 and **P < 0.001.*
Sensitivity analysis was performed by excluding subjects currently smoking, and the results were quite similar (Supplementary Table 1). Furthermore, to analyze the effect of current smoking on this association, the interaction term between smoking and allergic diseases was added to Model 1. There was no significant interaction between current smoking status and above three diseases in association for RA (P for asthma = 0.76, P for allergic rhinitis = 0.66, and P for sinusitis = 0.44). To analyze the effect of sex on this association, the interaction term between sex and allergic diseases was added to Model 1; however, their interactions for RA were not significant (P for asthma = 0.56, P for allergic rhinitis = 0.83, and P for sinusitis = 0.48). The interaction between asthma and allergic rhinitis ($$P \leq 0.62$$) or sinusitis ($$P \leq 0.12$$) was not significant.
## Network analysis of the association between asthma and asthma-related comorbidities
To understand the correlation between demographics, asthma, asthma-related comorbidities, and RA, the correlation coefficients for these variables were calculated (Supplementary Table 2). Asthma was positively correlated with all disease entities in our analysis to varying degrees, except for COPD. The strongest correlations were observed between sinusitis and allergic rhinitis, sinusitis and otitis media, and asthma and allergic rhinitis.
To visualize these correlations more intuitively, a correlation network was constructed (Supplementary Figure 3). The network analysis exhibited a positive relationship between asthma and allergic rhinitis, sinusitis, otitis media, atopic dermatitis, BMI, and RA. To evaluate the difference in the pattern of the correlation network between the groups with and without RA, correlation networks were reconstructed separately for each group (Figure 2). The correlation coefficients for the groups with and without RA are summarized in Supplementary Tables 3, 4, respectively. In the RA network, a positive correlation was found between asthma and BMI, whereas the correlation network of the group without RA showed no significant relationship between asthma and BMI. In the RA group, asthma was associated with comorbidities, such as allergic rhinitis and sinusitis, but not with otitis media or atopic dermatitis. Interrelations between asthma, sinusitis and allergic rhinitis were observed in both groups, however, the strength of the correlation between asthma and sinusitis was significantly stronger in patients with RA (Supplementary Table 5). To avoid bias driven by demographic differences, we matched the two groups by age and sex (Supplementary Table 6). We also obtained separate correlation coefficients and networks for the two groups after matching. Similar to the analysis of the total population, significant associations of asthma with BMI and sinusitis were found only in the RA group (Supplementary Figure 4).
**FIGURE 2:** *Correlation networks of the group with rheumatoid arthritis (N = 334) and those without rheumatoid arthritis (N = 13,384). Links between nodes indicate the existence of statistically significant association (P < 0.05). The thickness of links correlates with the strength of their association (correlation coefficient), and color means the direction of association: blue for positive and pink for negative. COPD, chronic obstructive pulmonary disease; BMI, body mass index.*
## Discussion
We found that the prevalence of asthma was higher in RA patients than in non-RA patients ($7.5\%$ vs. $2.8\%$, $P \leq 0.001$). After multivariable adjustment, asthma, allergic rhinitis, and sinusitis were positively associated with RA. Among them, asthma had the strongest association with RA (OR 2.32; $95\%$ CI 1.51–3.57). These associations were consistently found in logistic regression analyses using interaction terms or sensitivity analysis, supporting the unbiased nature of the results. The comorbidity pattern was different in asthma patients according to the presence of RA; obesity was more common in patients with asthma and concomitant RA. According to network analysis, the interconnection between asthma and sinusitis was stronger in patients with RA. Moreover, asthma and BMI were positively correlated only in RA patients. Our data suggest that asthma in RA patients is more obesity-related, which could be particularly meaningful since treatment refractoriness of asthma is partly attributable to obesity [29].
Among allergic diseases, asthma has been proven to be associated with RA [10]. Longitudinal analysis revealed a higher likelihood of developing RA in patients with asthma [17]. Conversely, a higher rate of asthma development has been observed in patients with RA [9]. Lung inflammation was noted as one of the major pathogenic mechanisms involved in the development of RA. Mucosal inflammation of the respiratory system upregulates citrullination, which can provoke the formation of anti-citrullinated antibodies, the major contributors to RA pathogenesis [30, 31]. Asthma, which is caused by Th2 cell-rich airway inflammation, is also known to enhance citrullination reactions, and a higher positive rate of anti-citrullinated antibodies has been demonstrated in the peripheral blood of asthma patients. Consistently, the presence of asthma elevated the risk of seropositive RA but not the risk of overall RA [32]. Another finding of our study was that their association was only relevant in the group without airflow limitation. There is a possibility that the group of patients with asthma with airflow limitation might have included some patients with COPD. As asthma and COPD commonly display overlapping features [33], it is often difficult to differentiate them without bronchodilator response tests or bronchoprovocation tests [34]. Nevertheless, the proportion of patients with airflow limitation in asthma that we reported was comparable to a previous report [$38\%$ ($\frac{120}{380}$) vs. $31\%$ ($\frac{74}{239}$), $$P \leq 0.87$$] [35]. The overall prevalence of asthma and COPD in our data was also consistent with Korean epidemiologic data [36, 37]. The exact mechanism is unclear for why RA is not associated with asthma with airflow limitation but could be attributable to the heterogeneity of this population.
The association between RA and COPD has remained controversial. Some studies reported positive associations between these diseases, independent of the smoking effect [17, 38]. An increase in citrullinated proteins was detected in the lungs of patients with COPD as well [39]. Contrarily, other studies have shown insignificant relationships across different stages of lung function [40, 41]. No relationship between COPD and RA was observed in the present study. This could be due to the heterogeneous definitions of COPD in each study and the variance in adjustment for the smoking effect.
Positive association between RA and allergic rhinitis had been reported by some studies [15]. Our data showed that the association with allergic rhinitis was weaker than that with asthma, which is consistent with the results of previous publications [42]. This is probably attributable to the absence of mucosal inflammation in the lower respiratory tract in allergic rhinitis, which is the major location for citrullination. In the present study, sinusitis was found to be positively associated with RA. A recent study reported positive association of sinusitis and pharyngitis with RA, implicating the contribution of upper airway mucosal inflammation to RA pathogenesis [18]. Previous publications showed a lower prevalence of atopic dermatitis in RA patients than in healthy controls. This was attributed to the Th1-dominant autoimmunity of RA, which may counteract the Th2 immunity of atopic dermatitis [43]. A representative study suggesting a higher incidence of RA in patients with atopic dermatitis performed an analysis of participants aged 40 years or younger, most of whom were children or adolescents [16]. Our analysis, however, included participants 40 years or older. Since a recent report highlighted the different characteristics between pediatric and adult atopic dermatitis, the findings from the younger and the older participants could be different [44].
One of the striking findings of our study is the association of higher BMI with asthma in patients with RA. Among the heterogenous phenotypes of asthma, obesity-related asthma is a subgroup that is difficult to manage. Due to the increased production of inflammatory cytokines from adipose tissue [45], obesity-related asthma tends to have a Th1-skewed response to inflammatory stimuli and non-atopic features, such as a low fraction of exhaled nitric oxide [46]. Our results indicated that there was no significant correlation between asthma and atopic dermatitis in patients with RA. On the other hand, a more pronounced relationship was discovered between asthma and sinusitis in those with RA. Taking into account that atopic dermatitis is mediated by Th2 immunity and that Th1 immunity plays dominant role in the subtype of sinusitis [47], these overall findings suggest that asthma in RA might be more Th1-skewed. Unfortunately, our analysis lacked immune profiling of the subjects. Therefore, further basic or translational studies are needed to confirm the association.
Significant relationships between asthma, allergic rhinitis, and sinusitis were observed in both groups with and without RA. The concept, “allergic march,” suggests a strong relationship between allergic diseases, and was also supported by several reports [20]. Sinusitis often develops as a complication of allergic rhinitis, and the co-occurrence of asthma and sinusitis is frequently observed [48]. Thus, these interrelations are consistent with traditional concepts. However, the strengthened relationship between asthma and sinusitis in patients with RA is a novel finding of the present study.
The limitations of this study include followings. First, this was a population-based cross-sectional study that did not include detailed information on disease activity, the diagnostic criteria applied, serologic status, medication history, other autoimmune diseases, other environmental risk factors, or the number of cigarettes smoked. In addition, laboratory findings such as blood eosinophil count, C-reactive protein level, and bronchodilator response on spirometry or methacholine bronchoprovocation test for asthma severity were not measured. This is attributable to the nature of the secondary data we used. Second, atopic dermatitis was infrequent because we included participants aged >40 years. The design for this study population was intended to adjust for age-related confounding effects, but atopic dermatitis is less frequent in individuals over 40 [49]. Thus, the relationship between RA and atopic dermatitis may have been underestimated. Lastly, causal relationships cannot be found in the significant associations because of the cross-sectional design of the study and need to be confirmed with further longitudinal studies.
In conclusion, RA was significantly positively associated with asthma (OR 2.32; $95\%$ CI 1.51–3.57), allergic rhinitis (OR 1.51; $95\%$ CI 1.08–2.10), and sinusitis (OR 1.64; $95\%$ CI 1.08–2.50), but not with COPD, atopic dermatitis, or otitis media. The network analysis exhibited interrelationships between asthma and sinusitis was more accentuated in the RA group compared to the non-RA group. Most strikingly, network analysis revealed a distinct positive link between asthma and higher BMI values in patients with RA.
## Data availability statement
Publicly available datasets were analyzed in this study. This data can be found here: https://www.data.go.kr/data/15076556/fileData.do.
## Ethics statement
The KNHANES protocols were reviewed and approved by the Research Ethics Review Board of National Center for Health Statistics. The patients/participants provided their written informed consent to participate in this study.
## Author contributions
H-KK had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. H-KK and JGK contributed to the study concept and design and drafted the manuscript and contributed to the statistical analysis. H-KK, JGK, J-HL, and JK contributed to the acquisition, analysis, or interpretation of the data, and contributed to the critical revision of the manuscript for important intellectual content. All authors read and approved the final manuscript.
## Conflict of interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher’s note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## Supplementary material
The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmed.2023.1006290/full#supplementary-material
## References
1. Smolen J, Aletaha D, Barton A, Burmester G, Emery P, Firestein G. **Rheumatoid arthritis.**. (2018) **4**. DOI: 10.1038/nrdp.2018.1
2. Weyand C, Goronzy J. **The immunology of rheumatoid arthritis.**. (2021) **22** 10-8. DOI: 10.1038/s41590-020-00816-x
3. Cope AP. **T cells in rheumatoid arthritis.**. (2008) **10**. DOI: 10.1186/ar2412
4. Annunziato F, Cosmi L, Liotta F, Maggi E, Romagnani S. **Type 17 T helper cells-origins, features and possible roles in rheumatic disease.**. (2009) **5** 325-31. DOI: 10.1038/nrrheum.2009.80
5. Slauenwhite D, McAlpine S, Hanly J, Malik A, Haidl I, Marshall J. **Association of a type 2-polarized T cell phenotype with methotrexate nonresponse in patients with rheumatoid arthritis.**. (2020) **72** 1091-102. DOI: 10.1002/art.41223
6. Akdis C, Arkwright P, Brüggen M, Busse W, Gadina M, Guttman-Yassky E. **Type 2 immunity in the skin and lungs.**. (2020) **75** 1582-605. DOI: 10.1111/all.14318
7. Kuenzig M, Bishay K, Leigh R, Kaplan G, Benchimol E. **Co-occurrence of asthma and the inflammatory bowel diseases: a systematic review and meta-analysis.**. (2018) **9**. DOI: 10.1038/s41424-018-0054-z
8. Charoenngam N, Ponvilawan B, Rittiphairoj T, Tornsatitkul S, Wattanachayakul P, Rujirachun P. **Patients with asthma have a higher risk of rheumatoid arthritis: a systematic review and meta-analysis.**. (2020) **50** 968-76. DOI: 10.1016/j.semarthrit.2020.07.015
9. Kim S, Min C, Oh D, Choi H. **Increased risk of asthma in patients with rheumatoid arthritis: a longitudinal follow-up study using a national sample cohort.**. (2019) **9**. DOI: 10.1038/s41598-019-43481-3
10. Kronzer V, Crowson C, Sparks J, Vassallo R, Davis J. **Investigating asthma, allergic disease, passive smoke exposure, and risk of rheumatoid arthritis.**. (2019) **71** 1217-24. DOI: 10.1002/art.40858
11. Dharmage S, Perret J, Custovic A. **Epidemiology of asthma in children and adults.**. (2019) **7**. DOI: 10.3389/fped.2019.00246
12. Tirosh A, Mandel D, Mimouni F, Zimlichman E, Shochat T, Kochba I. **Autoimmune diseases in asthma.**. (2006) **144** 877-83. DOI: 10.7326/0003-4819-144-12-200606200-00004
13. Rolfes M, Juhn Y, Wi C, Sheen Y. **Asthma and the risk of rheumatoid arthritis: an insight into the heterogeneity and phenotypes of asthma.**. (2017) **80** 113-35. DOI: 10.4046/trd.2017.80.2.113
14. Hajdarbegovic E, Thio B, Nijsten T. **Lower lifetime prevalence of atopy in rheumatoid arthritis.**. (2014) **34** 847-8. DOI: 10.1007/s00296-013-2785-1
15. Lai N, Tsai T, Koo M, Lu M. **Association of rheumatoid arthritis with allergic diseases: a nationwide population-based cohort study.**. (2015) **36** 99-103. DOI: 10.2500/aap.2015.36.3871
16. Schmitt J, Schwarz K, Baurecht H, Hotze M, Fölster-Holst R, Rodríguez E. **Atopic dermatitis is associated with an increased risk for rheumatoid arthritis and inflammatory bowel disease, and a decreased risk for type 1 diabetes.**. (2016) **137** 130-6. DOI: 10.1016/j.jaci.2015.06.029
17. Ford J, Liu X, Chu S, Lu B, Cho M, Silverman E. **Asthma, chronic obstructive pulmonary disease, and subsequent risk for incident rheumatoid arthritis among women: a prospective cohort study.**. (2020) **72** 704-13. DOI: 10.1002/art.41194
18. Kronzer V, Huang W, Zaccardelli A, Crowson C, Davis J, Vassallo R. **Association of sinusitis and upper respiratory tract diseases with incident rheumatoid arthritis: a case-control study.**. (2022) **49** 358-64. DOI: 10.3899/jrheum.210580
19. Boulet L, Boulay M. **Asthma-related comorbidities.**. (2011) **5** 377-93. DOI: 10.1586/ers.11.34
20. Galli S, Tsai M, Piliponsky A. **The development of allergic inflammation.**. (2008) **454** 445-54. DOI: 10.1038/nature07204
21. Agnihotri P, Monu S, Ramani D, Chakraborty M, Saquib S. **Differential metabolome in rheumatoid arthritis: a brief perspective.**. (2021) **23**. DOI: 10.1007/s11926-021-00989-w
22. **Standardization of spirometry, 1994 update.**. (1995) **152** 1107-36. DOI: 10.1164/ajrccm.152.3.7663792
23. Vestbo J, Hurd S, Agustí A, Jones P, Vogelmeier C, Anzueto A. **Global strategy for the diagnosis, management, and prevention of chronic obstructive pulmonary disease: gold executive summary.**. (2013) **187** 347-65. DOI: 10.1164/rccm.201204-0596PP
24. **Appropriate body-mass index for Asian populations and its implications for policy and intervention strategies.**. (2004) **363** 157-63. DOI: 10.1016/s0140-6736(03)15268-3
25. Tamimi A, Serdarevic D, Hanania N. **The effects of cigarette smoke on airway inflammation in asthma and COPD: therapeutic implications.**. (2012) **106** 319-28. DOI: 10.1016/j.rmed.2011.11.003
26. Zou G. **Toward using confidence intervals to compare correlations.**. (2007) **12** 399-413. DOI: 10.1037/1082-989x.12.4.399
27. Sheen Y, Rolfes M, Wi C, Crowson C, Pendegraft R, King K. **Association of asthma with rheumatoid arthritis: a population-based case-control study.**. (2018) **6** 219-26. DOI: 10.1016/j.jaip.2017.06.022
28. Mason C, Sabariego C, Thắng D, Weber J. **Can propensity score matching be applied to cross-sectional data to evaluate community-based rehabilitation? Results of a survey implementing the who’s community-based rehabilitation indicators in Vietnam.**. (2019) **9**. DOI: 10.1136/bmjopen-2018-022544
29. Pite H, Aguiar L, Morello J, Monteiro E, Alves A, Bourbon M. **Metabolic dysfunction and asthma: current perspectives.**. (2020) **13** 237-47. DOI: 10.2147/jaa.S208823
30. Reynisdottir G, Karimi R, Joshua V, Olsen H, Hensvold A, Harju A. **Structural changes and antibody enrichment in the lungs are early features of anti-citrullinated protein antibody-positive rheumatoid arthritis.**. (2014) **66** 31-9. DOI: 10.1002/art.38201
31. Klareskog L, Catrina A. **Autoimmunity: lungs and citrullination.**. (2015) **11** 261-2. DOI: 10.1038/nrrheum.2015.38
32. Zaccardelli A, Liu X, Ford J, Cui J, Lu B, Chu S. **Asthma and elevation of anti-citrullinated protein antibodies prior to the onset of rheumatoid arthritis.**. (2019) **21**. DOI: 10.1186/s13075-019-2035-3
33. Cosío B, Dacal D, Pérez de Llano L. **Asthma-copd overlap: identification and optimal treatment.**. (2018) **12**. DOI: 10.1177/1753466618805662
34. Kitaguchi Y, Yasuo M, Hanaoka M. **Comparison of pulmonary function in patients with COPD, asthma-COPD overlap syndrome, and asthma with airflow limitation.**. (2016) **11** 991-7. PMID: 27274220
35. Huang S, Vasquez M, Halonen M, Martinez F, Guerra S. **Asthma, airflow limitation and mortality risk in the general population.**. (2015) **45** 338-46. DOI: 10.1183/09031936.00108514
36. Lee E, Kim A, Ye Y, Choi S, Park H. **Increasing prevalence and mortality of asthma with age in Korea, 2002–2015: a nationwide, population-based study.**. (2020) **12** 467-84. PMID: 32141260
37. Rhee C. **High prevalence of chronic obstructive pulmonary disease in Korea.**. (2016) **31** 651-2. DOI: 10.3904/kjim.2016.196
38. Bieber V, Cohen A, Freud T, Agmon-Levin N, Gertel S, Amital H. **Autoimmune smoke and fire–coexisting rheumatoid arthritis and chronic obstructive pulmonary disease: a cross-sectional analysis.**. (2013) **56** 261-6. DOI: 10.1007/s12026-013-8395-x
39. Lugli E, Correia R, Fischer R, Lundberg K, Bracke K, Montgomery A. **Expression of citrulline and homocitrulline residues in the lungs of non-smokers and smokers: implications for autoimmunity in rheumatoid arthritis.**. (2015) **17**. DOI: 10.1186/s13075-015-0520-x
40. Bergström U, Jacobsson L, Nilsson J, Berglund G, Turesson C. **Pulmonary dysfunction, smoking, socioeconomic status and the risk of developing rheumatoid arthritis.**. (2011) **50** 2005-13. DOI: 10.1093/rheumatology/ker258
41. de Roos A, Cooper G, Alavanja M, Sandler D. **Personal and family medical history correlates of rheumatoid arthritis.**. (2008) **18** 433-9. DOI: 10.1016/j.annepidem.2007.12.011
42. Jeong H, Jung S, Cho S. **Association between rheumatoid arthritis and respiratory allergic diseases in Korean adults: a propensity score matched case-control study.**. (2018) **2018**. DOI: 10.1155/2018/3798124
43. Rudwaleit M, Andermann B, Alten R, Sörensen H, Listing J, Zink A. **Atopic disorders in ankylosing spondylitis and rheumatoid arthritis.**. (2002) **61** 968-74. DOI: 10.1136/ard.61.11.968
44. Ramírez-Marín H, Silverberg J. **Differences between pediatric and adult atopic dermatitis.**. (2022) **39** 345-53. DOI: 10.1111/pde.14971
45. Mohanan S, Tapp H, McWilliams A, Dulin M. **Obesity and asthma: pathophysiology and implications for diagnosis and management in primary care.**. (2014) **239** 1531-40. DOI: 10.1177/1535370214525302
46. Peters U, Dixon A, Forno E. **Obesity and asthma.**. (2018) **141** 1169-79. DOI: 10.1016/j.jaci.2018.02.004
47. Akdis C, Bachert C, Cingi C, Dykewicz M, Hellings P, Naclerio R. **Endotypes and phenotypes of chronic rhinosinusitis: a PRACTALL document of the European academy of allergy and clinical immunology and the American academy of allergy, asthma & immunology.**. (2013) **131** 1479-90. DOI: 10.1016/j.jaci.2013.02.036
48. Skoner D. **Complications of allergic rhinitis.**. (2000) **105** S605-9. DOI: 10.1067/mai.2000.106150
49. Barbarot S, Auziere S, Gadkari A, Girolomoni G, Puig L, Simpson E. **Epidemiology of atopic dermatitis in adults: results from an international survey.**. (2018) **73** 1284-93. DOI: 10.1111/all.13401
|
---
title: 'Anemia in pregnancy and sleep of 6-month-old infants: A prospective cohort
study'
authors:
- Lei Zhang
- Shuangshuang Ma
- Feicai Dai
- Qiong Li
- Lin Wu
- Lijun Yu
- Tianqin Xie
- Dao-min Zhu
- Peng Zhu
journal: Frontiers in Nutrition
year: 2023
pmcid: PMC10036361
doi: 10.3389/fnut.2023.1049219
license: CC BY 4.0
---
# Anemia in pregnancy and sleep of 6-month-old infants: A prospective cohort study
## Abstract
### Objective
Anemia has been reported to adversely influence sleep in infants. However, the association between anemia in pregnancy and infant sleep remains unclear. We aimed to examine the association between maternal anemia in pregnancy and sleep parameters of 6-month-old infants.
### Methods
We enrolled 2,410 mother-infant pairs between 2018 and 2021 in Hefei. Data on maternal hemoglobin concentration were collected at 24–28 gestational weeks from the electronic medical records of the hospitals. Nocturnal and daytime sleep duration, number of night awakenings, nocturnal wakefulness, and sleep latency of infants aged 6 months were measured using the Brief Infant Sleep Questionnaire with five items. A restricted cubic spline model was used to examine the relationship between maternal hemoglobin concentration and infant nocturnal sleep duration after adjusting for potential confounders.
### Results
In our study, 807 ($33.5\%$) mothers had anemia during pregnancy. Compared to infants born to mothers without anemia, infants born to mothers with anemia in pregnancy had shorter nocturnal sleep duration [mean (SD), 560.29 (79.57) mins vs. 574.27 (75.36) mins] at the age of 6 months. Subgroup analysis showed consistent significant differences in nocturnal sleep duration between infant born to anemic and non-anemic mothers, except in case of stratification by preterm birth [mean difference (mins), 2.03 ($95\%$ CI, −20.01, −24.07)] and pre-pregnancy obesity [mean difference (mins), −0.85 ($95\%$ CI, −16.86, −15.16)]. A J-shaped nonlinear correlation curve was observed between maternal hemoglobin concentration and infant nocturnal sleep duration. Compared with mothers without daily iron supplementation, mothers who had daily iron supplementation had higher hemoglobin concentrations [mean (SD), 112.39 (11.33) g/L vs. 110.66 (10.65) g/L] at delivery and their infants had longer nocturnal sleep duration [mean (SD), 565.99 (82.46) mins vs. 553.66 (76.03) mins].
### Conclusion
Anemia in pregnancy may have an adverse influence on the sleep of 6-mon-old infants, and the relationship between maternal hemoglobin concentration and nocturnal sleep duration is nonlinear.
## Introduction
Anemia in pregnancy, characterized by hemoglobin (Hb) level of < 110 g/l, is a complication caused by maternal malnutrition, it is one of the most critical health conditions and an urgent public health priority worldwide [1, 2]. A recent meta-analysis of the global prevalence of anemia in pregnant women indicated that the overall prevalence of anemia in pregnancy was $36.8\%$ [3]. There is broad agreement that anemia in pregnancy is associated with adverse fetal growth and infant development outcomes, such as fetal malformation, low birth weight, and autism spectrum disorder (4–6). It is worth noting that sleep disorders seem to be more common in infants with developmental problems, especially in the first few years of life. Additionally, development of infant circadian system begins in utero and continues throughout early-life; a damaged intrauterine environment, as a consequence of low Hb concentration, may disrupt infant sleep development in utero and after birth [7]. Therefore, anemia in pregnancy may have a potential correlation with infant sleep. Previous studies have demonstrated associations between infant anemia and sleep, such as associations between wakefulness duration and sleep state, during infancy [8, 9]. However, investigations on the associations of maternal anemia in pregnancy with infant sleep are still limited.
Sleep development is an active, complex neurophysiological process, and infant sleep development has specific time sequences, similar to other organs or functions [10]. The sixth month is a key time point of the sleep development process, and trends of nocturnal sleep duration (NSD), defined as 19:00–7:00, showed rapid changes over the first 6 months before stabilizing to a plateau [11]. After birth, NSD gradually increases and daytime sleep duration (DSD) gradually decreases, eventually leading to a stable sleep pattern, but there is still no accepted recommended range for NSD, DSD, or other sleep parameters. Previous studies on sleep in 6-month-old infants revealed that NSD was 491–572 min and DSD was 186–272 min [12, 13]. In addition, half of the infants achieved self-regulated sleep at 6 months of age, and their electrical patterns of non-rapid eye movement and rapid eye movement (REM) sleep progressively resembled those seen in adults [14]. Changes in the intrauterine environment caused by anemia, such as chronic intrauterine hypoxia, may impair the development of infant sleep. Sleep is interlinked throughout the lifespan, and poor sleep at 6 months of age increases the risk of sleep disorders later in life [15]. Moreover, previous studies have suggested that children with higher proportions of sleep at night perform better on executive functions, whereas children with shorter NSDs are at higher risk of obesity and physical inactivity [16, 17]. Therefore, understanding the relationship between maternal anemia in pregnancy and infant sleep at 6 months of age is critical for developing preventive strategies to promote healthy infant development.
In this prospective cohort study, we aimed to determine the potential impact of maternal anemia in pregnancy on infant sleep parameters at 6 months of age. We further examined the nonlinear relationship between maternal Hb concentration and infant NSD using a restricted cubic spline model.
## Participants and study design
This was a prospective birth cohort study. From March 2018 to June 2021, 4,216 pregnant women at 16–23 gestational weeks were recruited from three hospitals including the Hefei First People’s Hospital, the Hefei Maternal and Child Care Hospital, and the First Affiliated Hospital of Anhui Medical University. The inclusion criteria for this study included the following: pregnant women, aged 18–45 years, 16–23 gestational weeks, no communication difficulties and lived in Hefei, single gestation, and pregnancy without assisted reproductive technology.
Exclusion criteria were the following: missing maternal blood samples, stillbirth, and birth defects. Finally, our study included 2,410 mother-infant pairs with complete data (Supplemental Figure 1). Ethics committee approval for the study was obtained from the ethics committee of Anhui Medical University (No. 20180092), and all patients provided written informed consent before participating in the study.
During recruitment, pregnant women were required to complete structured questionnaires regarding demographic characteristics, food frequency intake, health status, and iron supplementation. After recruitment, we obtained maternal Hb concentration data from electronic medical records of hospitals at 24–28 gestational weeks. At delivery, information on maternal Hb concentration at delivery and newborn birth outcomes was collected from electronic medical records. In the sixth month after delivery, we assessed the infants’ sleep; infants’ sleep parameters, at 6 months of age, were obtained from their parents via a questionnaire.
## Secondary data
The diagnosis of anemia in pregnancy was based on the Hb concentration according to the WHO standard [18]. Hb < 110 g/l was defined as anemia, and pregnant women were classified into two groups according to Hb concentration (Hb < 110 g/l, and Hb ≥ 110 g/l). Maternal Hb concentration was measured using an Auto Hematology Analyzer BC6800plus (Shenzhen Mindray Bio-Medical Electronics Co., Ltd., China) using venous blood samples from the hospitals.
## Measurement of infants’ sleep parameters
Data on infants’ sleep parameters were collected at 6 months of age. Infant sleep was assessed using a face-to-face survey based on the Brief Infant Sleep Questionnaire (BISQ). The BISQ includes 11 questions on daytime and nocturnal sleep patterns and behaviors and is widely used in the assessment of infant sleep parameters [19]. The validity and reliability of the BISQ ($r = 0.82$–0.95) have been widely reported [20]. During the investigation, mothers completed the items of the translated BISQ concerning the infants’ sleep parameters. The following five items were used to calculate the infants’ sleep parameters: “How long does your baby sleep at night (19:00–7:00),” “How long does your baby sleep in the daytime (7:00–19:00),” “How long does your baby awake at night,” “How long does your baby need to fall asleep,” as well as “How many times your baby wakes up each night on average?”
## Measurement of potential confounders
The characteristics of mothers during pregnancy were collected through face-to-face interviews, including maternal age at delivery, years of education > 12 years, family income < 10,000 yuan/month, and pre-pregnancy BMI. Maternal systolic blood pressure (SBP) and diastolic blood pressure (DBP) were obtained from the electronic medical records. The past month’s maternal diet (fruit, dessert, vegetable, and bean product intake) was assessed using an adapted food frequency questionnaire [21]. With responses ranging from “never,” “one to two times a week,” “three to six times a week” to “1 time a day or more.” The characteristics of infants were collected from electronic medical records, including gestational age at birth, birth weight and gender, preterm birth, small for gestational age (SGA), and large for gestational age (LGA). LGA (birth weight > 90th percentile of newborns at the same gestational age) and SGA (birth weight < 10th percentile of newborns at the same gestational age) were defined as the gestational age- and sex-specific international reference for fetal growth [22]. Breastfeeding was defined as receiving breast milk and no other food or drink during the first 6 months after birth. In the end, the following potential confounders were adjusted in the models due to the previous literature and the effect on anemia and NSD, include maternal age at delivery, education levels, family income, pre-pregnancy BMI [23], SBP, DBP, maternal diet (fruits, dessert, vegetables, bean products) [24], infant gender, preterm birth, SGA, LGA, breastfeeding [25].
## Statistical analysis
The Shapiro–Wilk test was used to confirm the normality of the distribution of variables. Continuous variables with a normal distribution are expressed as mean (SD), and categorical variables are expressed as numbers (percentages). For variables with non-normal distribution, the results are expressed as medians (interquartile range). Characteristics of mothers and infants were compared between mothers with anemia and mothers without anemia using Student’s t-test for continuous variables and Chi-square analysis for categorical variables. Comparison of sleep parameters between infants born to mothers with anemia and mothers without anemia was performed using covariance analysis. We further conducted subgroup analyses for the primary outcome stratified by mother (education level and family income) and infants characteristics (gender, preterm birth, SGA, LGA, and breastfeeding). In addition, a nonlinear model was fitted with restricted cubic spline curves to examine the nonlinear association between maternal Hb and NSD in 6-month-old infants after adjusting for potential confounders. We also examined the difference in infant sleep between mothers with anemia and without daily iron supplementation using covariance analysis. All statistical analyses were performed using the SPSS statistical software (SPSS Statistics 21.0; SPSS Inc.) and R (version 4.0.2, R Foundation for Statistical Computing). $p \leq 0.05$ was considered statistically significant, and all tests were two-sided.
## Characteristics of the study population
The characteristics of 2,410 mother-infant pairs are presented in total and by maternal anemia in Table 1. Of the mothers, 807($33.5\%$) had anemia during pregnancy, 1,618 ($67.1\%$) had received education for more than 12 years, and 1,262 ($52.4\%$) had a family income lower than 10,000 yuan every month. The mean (SD) age at delivery was 30.93 (4.29) years, and the mean (SD) pre-pregnancy BMI was 21.46 (2.92) kg/m2. Of the infants included in the study, 1,258 ($52.2\%$) were male, 1,152 ($47.8\%$) were female, 225 ($9.3\%$) were preterm birth, and 1,354 ($56.2\%$) were breastfeeding. Compared with mothers without anemia, mothers with anemia in pregnancy had lower pre-BMI [mean (SD), 21.17 (2.60) kg/m2 vs. 21.61 (3.06) kg/m2], SBP [mean (SD), 108.62 (9.58) mmHg vs. 111.61 (10.22) mmHg], and DBP [mean (SD), 67.19 (7.08) mmHg vs. 70.04 (7.66) mmHg]. Compared with children born to mothers without anemia, those born to mothers with anemia had a higher birth weight [mean (SD), 3.49 (0.59) kg vs. 3.43 (0.58) kg] and lower SGA percentage (6.7 vs. $9.5\%$).
**Table 1**
| Variables | Total (n = 2,410) | Anemia (n = 807) | Non-anemia (n = 1,603) | p Valuea |
| --- | --- | --- | --- | --- |
| Maternal characteristics | | | | |
| Age at delivery, years, mean (± SD) | 30.93 ± 4.29 | 31.09 ± 4.29 | 30.8 ± 4.29 | 0.190 |
| Educational years > 12 years, n (%) | 1,618(67.1) | 525(65.1) | 1,093(68.2) | 0.123 |
| Family income < 10,000 yuan/mon, n (%) | 1,262(52.4) | 429(53.2) | 833(52.0) | 0.579 |
| Pre-pregnancy BMI, Kg/m2, mean (± SD) | 21.46(2.92) | 21.17 ± 2.6 | 21.61 ± 3.06 | <0.001 |
| SBP, mmHg, mean (± SD) | 110.60(10.11) | 108.62 ± 9.58 | 111.61 ± 10.22 | <0.001 |
| DBP, mmHg, mean (± SD) | 69.08(7.59) | 67.19 ± 7.08 | 70.04 ± 7.66 | <0.001 |
| Daily iron supplement, n (%) | 801(33.2) | 434(53.8) | 367(22.9) | <0.001 |
| Fruits frequency ≥ 3 time/week, n (%) | 2,270(94.2) | 761(94.3) | 1,509(94.1) | 0.963 |
| Dessert frequency ≥ 3 time/week, n (%) | 420(17.4) | 147(18.2) | 273(17.0) | 0.477 |
| Vegetables frequency ≥ 3 time/week, n (%) | 2,335(96.9) | 783(97.0) | 1,552(96.8) | 0.907 |
| Bean products frequency ≥ 3 time/week, n (%) | 1,189(49.3) | 377(46.7) | 812(50.7) | 0.064 |
| Infant characteristics | | | | |
| Male, n (%) | 1,258(52.2) | 409(50.7) | 849(53.0) | 0.290 |
| Gestational age at birth, week, mean (± SD) | 39.03(1.37) | 38.98 ± 2.15 | 39.01 ± 1.75 | 0.704 |
| Birth weight, Kg, mean (± SD) | 3.41(0.46) | 3.49 ± 0.59 | 3.43 ± 0.58 | 0.016 |
| Preterm birth, n (%) | 225(9.3) | 75(9.3) | 150(9.4) | 0.959 |
| SGA, n (%) | 207(8.6) | 54(6.7) | 153(9.5) | 0.018 |
| LGA, n (%) | 335(13.9) | 117(14.5) | 218(13.6) | 0.547 |
| Breastfeeding, n (%) | 1,354(56.2) | 465(57.6) | 889(55.5) | 0.313 |
## Nonlinear association between maternal Hb concentration and infant NSD
Using a nonlinear regression model, we found that the association between maternal Hb concentration and infant NSD was J-shaped (nonlinear $p \leq 0.001$) after adjusting for confounders (Figure 1). A flattened increase in NSD was observed when maternal Hb concentration was < 100 g/l, whereas a dramatical increase in NSD was observed when maternal Hb concentration was > 100 g/l.
**Figure 1:** *Association between maternal hemoglobin concentration and infant nocturnal sleep duration in the restricted cubic spline model (n = 2,410). The restricted cubic spline model adjusted for maternal age at delivery, education levels, family income, pre-pregnancy BMI, SBP, DBP, daily iron supplementation, maternal diet (fruits, dessert, vegetables, bean products), infant gender, preterm birth, SGA, LGA, breastfeeding. SGA: small for gestational age; LGA: large for gestational age. BMI: Body mass index; SBP: systolic blood pressure; DBP: diastolic blood pressure; SGA: small for gestational age; LGA: large for gestational age.*
## Comparison of sleep parameters between infants born to mothers with anemia vs. those born to mothers without anemia
We compared six types of infant sleep parameters between the two groups, including NSD, daytime sleep duration, number of night awakenings, nocturnal wakefulness, sleep latency, and total sleep duration (Table 2). Infants born to mothers with anemia had significantly shorter NSD than infants born to mothers without anemia [mean (SD), 560.29 (79.57) mins vs. 574.27 (75.36) mins]. However, there were no significant differences in the other five infant sleep parameters between the two groups.
**Table 2**
| Sleep parameters* | Total (n = 2,410) | Anemia (n = 807) | Non-anemia (n = 1,603) | p Valuea |
| --- | --- | --- | --- | --- |
| Nocturnal sleep duration (min) | 569.59(77.06) | 560.29(79.57) | 574.27(75.36) | <0.001 |
| Daytime sleep duration (min) | 233.22(93.24) | 237.59(97.78) | 231.02(90.82) | 0.265 |
| Number of night awakening (n) | 2.00(2.00) | 2.00(2.00) | 2.00(2.00) | 0.954 |
| Nocturnal wakefulness (min) | 30.00(40.00) | 30.00(40.00) | 30.00(40.00) | 0.837 |
| Sleep latency (min) | 30.00(10.00) | 30.00(10.00) | 30.00(10.00) | 0.922 |
| Total sleep duration (min) | 802.45(117.37) | 798.40(120.25) | 804.48(115.88) | 0.114 |
## Subgroup analyses on differences in infant NSD
We examined whether the associations differed by potential effect modifiers including maternal education level (> 12 years, ≤ 12 years), family income (≥ 10,000 yuan/month, < 10,000 yuan/month), infant gender (male, female), preterm birth (yes, no), SGA (yes, no), LGA (yes, no), and breastfeeding (yes, no) in the subgroup analysis (Supplementary Table 1). Compared with infants born to mothers without anemia, those born to mothers with anemia had a 13.98 min($95\%$ CI, 7.48–20.48) shorter NSD (Figure 2). After grouping by maternal education level, family income, infant gender, SGA, LGA, and breastfeeding, infants born to mothers with anemia had a shorter NSD than infants born to mothers without anemia, in all subgroups. However, the difference in NSD did not exist in infants with preterm birth [mean difference (min), 2.03 ($95\%$ CI, −20.01, –24.07)], and infants born to mothers with pre-pregnancy obesity [mean difference (min), −0.85 ($95\%$ CI, −16.86, –15.16)].
**Figure 2:** *Mean difference of nocturnal sleep duration in infants born to mothers with anemia vs. mothers without anemia (n = 1,347, Mean difference ± SD). Infants of the two groups were divided into subgroups according to maternal education levels, family income, pre-pregnancy obesity, infant gender, preterm birth, SGA, LGA, breastfeeding. The covariance analysis model adjusted for maternal age at delivery, education levels (if not stratified), family income (if not stratified), pre-pregnancy BMI (if not stratified), SBP, DBP, daily iron supplementation, maternal diet (fruits, dessert, vegetables, and bean products), infant gender, preterm birth, SGA (if not stratified), LGA (if not stratified), breastfeeding (if not stratified). SGA: small for gestational age; LGA: large for gestational age. BMI: body mass index; SBP: systolic blood pressure; DBP: diastolic blood pressure; SGA: small for gestational age; LGA: large for gestational age.*
## Comparison of maternal Hb concentration and infant sleep parameters between anemic mothers with daily iron supplementation vs. those without daily iron supplementation
We compared maternal Hb concentration and infant sleep parameters between the two groups and found that daily iron supplementation during pregnancy had a positive influence on maternal Hb concentration and infant sleep parameters. We found that mothers in the daily iron supplementation group had higher Hb concentrations [mean (SD): 112.39 (11.33) g/L vs. 110.66 (10.65) g/L] at delivery (Table 3). Infants in the daily iron supple-mentation group had significantly longer NSD than those without the daily iron supplementation group [mean (SD), 565.99 (82.46) mins vs. 553.66 (76.03) mins]. However, we did not find any significant difference in other infant sleep parameters.
**Table 3**
| Maternal Hb concentration and infant sleep parameters* | Daily iron supplementation (n = 434) | Non-daily iron supplementation (n = 373) | p Valuea |
| --- | --- | --- | --- |
| Hb concentration in the second trimester (g/L) | 103.11(6.30) | 103.61(4.91) | 0.214 |
| Hb concentration at delivery (g/L) | 112.39(11.33) | 110.66(10.65) | 0.039# |
| Nocturnal sleep duration (min) | 565.99(82.46) | 553.66(76.03) | 0.028# |
| Daytime sleep duration (min) | 235.23(95.64) | 240.34(99.34) | 0.458 |
| Number of night awakenings (n) | 2.00(2.00) | 2.00(2.00) | 0.910 |
| Nocturnal wakefulness (min) | 30.00(40.00) | 30.00(40.00) | 0.753 |
| Sleep latency (min) | 30.00(10.00) | 30.00(10.00) | 0.902 |
| Total sleep duration (min) | 801.02(122.69) | 793.84(117.42) | 0.398 |
## Discussion
Our study adds to the scarce literature on the association between anemia in pregnancy and infant sleep. In this prospective cohort study, we also observed a J-shaped nonlinear association between maternal Hb concentration and infant NSD. We found that infants born to mothers with anemia in pregnancy had shorter NSD at 6 months of age, and the differences were consistent across subgroups except in preterm birth infants and infants born to mothers with pre-pregnancy obesity. Daily iron supplementation may have a positive influence on maternal anemia and infant NSD.
Consistent with our study, a study conducted in Japan investigated a nonlinear relationship between maternal Hb concentration and infant sleep, and a U-shaped correlation curve between maternal Hb level and the risk of infant sleep at 22:00 or later was reported [23]. The nonlinear relationship between maternal Hb concentration and infant sleep might be a possible cause for the conflicting findings reported in literature. We observed a J-shaped nonlinear correlation curve between maternal Hb concentration and infant NSD. The underlying mechanism of this nonlinear association is unclear, but we speculate that it may be related to health behaviors and medical adherence in pregnant women. When mothers have moderate or severe anemia, they are more likely to adopt a healthier lifestyle or stricter medical practices to ensure that their fetus receives adequate nutrition. These changes may attenuate the adverse effects of anemia to some extent. In our study, infants born to mothers with anemia (especially those with Hb concentration of 90–110 g/l) had a shorter NSD than infants born to mothers without anemia. It is noteworthy that $97.7\%$ of anemia in pregnancy is mild anemia (Hb concentration 100–109 g/l) or moderate anemia (Hb concentration 90–99 g/l) worldwide [3]. Therefore, it might be a tremendous opportunity to improve infant early-life sleep development and decrease the risk of numerous disorders originating from sleep by providing timely interventions for anemia in pregnancy. Furthermore, follow-up studies are needed to investigate the effect of interventions toward maternal anemia on infants’ sleep development.
Our findings suggest that anemia in pregnancy may be associated with shorter NSD in infants. In line with our study, a study in Chile suggested that iron deficiency anemia (IDA) is associated with altered infants’ sleep states; children with IDA in infancy showed shorter REM sleep episodes in the last third [9]. Similar negative association between Hb levels and sleep durations was also found in the English Longitudinal Study of Ageing [26]. Contrary to our findings, a study in Nepal and Zanzibar reported that IDA infants slept longer at night than non-IDA infants [27]; however, as it was an intervention study, the effects on sleep could have been confounded by effects of administering iron or folic acid supplements. In contrast, iron deficiency has been reported to only account for $75\%$ of maternal anemia [28], and the association between anemia and sleep may be moderated. Therefore, there is an urgent need to conduct more studies with sufficient samples, to determine the effect of anemia in pregnancy on infant sleep development.
Notably, NSD did not differ between two groups when the infants in both had preterm birth or had mothers with pre-pregnancy obesity. One reason may be that the sample sizes of preterm birth infants and mothers with pre-pregnancy obesity were too small, which may have increased the risk of type II errors. Furthermore, recent studies have shown that maternal pre-pregnancy obesity and preterm birth are closely corelated with infant neurodevelopment, and this association may have confounding effects on assessment of the real-world association between maternal anemia and infant sleep. Our results showed that infants whose mothers had daily iron supplementation exhibited a longer NSD. Iron supplementation has been recommended for health care during pregnancy and helps prevent and treat anemia in pregnancy. The severity of fatigue symptoms, daytime sleepiness, sleep quality, and sleep-related breathing disorders are mitigated after iron supplementation in non-pregnant participants with anemia [29, 30]. However, evidence of the potential impact of iron supplementation during pregnancy on early infant sleep is limited. Our results suggest that daily iron supplementation in pregnancy may optimize the NSD in infants.
Several possible mechanisms may explain the association between maternal anemia and infant sleep. Anemia can cause reduced peripheral oxygen saturation or hypoxia, both of which are key drivers of hypothalamic–pituitary–adrenal (HPA) axis alterations [31, 32]. Animal experiments also elucidated that intermittent hypoxia during gestation results in differential alterations in the HPA axis and behavior of the offspring [33]. Therefore, anemia in pregnancy may affect infants’ HPA axis function and lead to sleep problems. Iron deficiency is considered to be one of the most important risk factors for infant sleep disorders. Animal studies have shown that the hippocampus is very sensitive to a lack of iron during early development [34], and iron deficiency during pregnancy may damage fetal hippocampal development and lead to sleep disorders in infancy. Although numerous studies on human have been carried out, findings regarding the relationship between maternal iron deficiency and infant sleep have been inconsistent [8, 35]. One reason may be that iron supplementation, which is universally used to prevent anemia during pregnancy, alleviates the negative effects of iron deficiency. Perinatal complications and infant anemia caused by anemia are also considered to be possible mechanisms between anemia in pregnancy and infant sleep (36–38). Further well-designed studies are required to understand these mechanisms.
Our study has several advantages. First, this is the first study to report nonlinear associations between maternal Hb concentration and NSD in 6-month-old infants. Second, the prospective study design revealed clear time-ordered relationships between maternal anemia and infant sleep. Third, statistical analyses were performed using multiple covariates. Our study has some limitations. Information on infants’ sleep parameters has been reported by parents, and recall bias may exist. Objective measures, such as actigraphy and polysomnography readings, could be used in future studies. Performing actigraphy is not feasible sometimes due to logistical or financial reasons, while the BISQ has been validated against actigraphy and is widely used in infant sleep measurement [39]. Second, this was an observational study, and thus, we could not establish a causal relationship between anemia and infant sleep. Third, we did not collect the data on infant Hb concentration. The close association of infant anemia with maternal anemia and infant sleep might be an important mediator in this study. Therefore, we could not eliminate interference of such confounding factors from infant anemia, and caution is required when interpreting the results of this study. Last, we collect maternal Hb concentrations at 24–28 gestational weeks and at delivery, and maternal anemia condition during the whole pregnancy was not clear. Future studies should collect *Hb data* at more time points to evaluate the effects of anemia duration and period on infant sleep.
## Conclusion
In this prospective birth cohort study, we found that maternal anemia in pregnancy was associated with shorter NSD in infants aged 6 months. The relationship between maternal Hb concentration and infant NSD was nonlinear. Daily iron supplementation may have a positive influence on maternal anemia and infant NSD. Our study highlights the dangers of anemia in pregnancy and the potential benefits of preventing it on infant sleep. In the future, clinical medical facilities should include prevention, screening, and treatment of anemia as a priority in maternal health care, and iron supplementation might optimize infant sleep and decrease the risk of infant sleep disorders. Policies are also needed to reduce the prevalence of anemia during pregnancy, eliminate potential risk factors, and provide support for maternal and child health. Further research is needed to replicate these preliminary findings with objective recordings and to examine the potential association between infant anemia at birth and sleep.
## Data availability statement
The original contributions presented in the study are included in the article/Supplementary material, further inquiries can be directed to the corresponding authors.
## Ethics statement
The studies involving human participants were reviewed and approved by Ethics Committees of Anhui Medical University. The patients/participants provided their written informed consent to participate in this study. Written informed consent was obtained from the individual(s) for the publication of any potentially identifiable images or data included in this article.
## Author contributions
LZ, SM, FD, QL, LW, LY, TX, and PZ: data curation. LZ and SM: formal analysis and writing original draft. LZ, SM, FD, QL, LW, LY, and TX: investigation. LZ, SM, DZ, and PZ: methodology. DZ: project administration. LZ: visualization. All authors contributed to the article and approved the submitted version.
## Funding
This study was funded by National Natural Science Foundation of China [81872631, 82173531], National Key R&D Program of China (2022YFC2702901) and Key Projects of Excellent Young Talents Fund in universities of Anhui Province (gxyqZD2018025).
## Conflict of interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher’s note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## Supplementary material
The Supplementary material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fnut.2023.1049219/full#supplementary-material
## References
1. Stevens GA, Paciorek CJ, Flores-Urrutia MC, Borghi E, Namaste S, Wirth JP. **National, regional, and global estimates of anaemia by severity in women and children for 2000-19: a pooled analysis of population-representative data**. *Lancet Glob Health* (2022) **10** e627-39. DOI: 10.1016/S2214-109X(22)00084-5
2. Abd Rahman R, Idris IB, Isa ZM, Rahman RA, Mahdy ZA. **The prevalence and risk factors of iron deficiency anemia among pregnant women in Malaysia: a systematic review**. *Front Nutr* (2022) **9** 847693. DOI: 10.3389/fnut.2022.847693
3. Karami M, Chaleshgar M, Salari N, Akbari H, Mohammadi M. **Global prevalence of anemia in pregnant women: a comprehensive systematic review and meta-analysis**. *Matern Child Health J* (2022) **26** 1473-87. DOI: 10.1007/s10995-022-03450-1
4. Shi H, Chen L, Wang Y, Sun M, Guo Y, Ma S. **Severity of anemia during pregnancy and adverse maternal and fetal outcomes**. *JAMA Netw Open* (2022) **5** e2147046. DOI: 10.1001/jamanetworkopen.2021.47046
5. Liu D, Li S, Zhang B, Kang Y, Cheng Y, Zeng L. **Maternal hemoglobin concentrations and birth weight, low birth weight (LBW), and small for gestational age (SGA): findings from a prospective study in Northwest China**. *Nutrients* (2022) **14** 858. DOI: 10.3390/nu14040858
6. Wiegersma AM, Dalman C, Lee BK, Karlsson H, Gardner RM. **Association of Prenatal Maternal Anemia with Neurodevelopmental Disorders**. *JAMA Psychiat* (2019) **76** 1294-304. DOI: 10.1001/jamapsychiatry.2019.2309
7. Wong SD, Wright KP, Spencer RL, Vetter C, Hicks LM, Jenni OG. **Development of the circadian system in early life: maternal and environmental factors**. *J Physiol Anthropol* (2022) **41** 22. DOI: 10.1186/s40101-022-00294-0
8. Angulo-Barroso RM, Peirano P, Algarin C, Kaciroti N, Lozoff B. **Motor activity and intra-individual variability according to sleep-wake states in preschool-aged children with iron-deficiency anemia in infancy**. *Early Hum Dev* (2013) **89** 1025-31. DOI: 10.1016/j.earlhumdev.2013.08.014
9. Peirano PD, Algarin CR, Chamorro RA, Reyes SC, Duran SA, Garrido MI. **Sleep alterations and iron deficiency anemia in infancy**. *Sleep Med* (2010) **11** 637-42. DOI: 10.1016/j.sleep.2010.03.014
10. Lefter R, Cojocariu RO, Ciobica A, Balmus IM, Mavroudis I, Kis A. **Interactions between sleep and emotions in humans and animal models**. *Medicina (Kaunas)* (2022) **58** 274. DOI: 10.3390/medicina58020274
11. Lin QM, Spruyt K, Leng Y, Jiang YR, Wang GH, Dong SM. **Cross-cultural disparities of subjective sleep parameters and their age-related trends over the first three years of human life: a systematic review and meta-analysis**. *Sleep Med Rev* (2019) **48** 101203. DOI: 10.1016/j.smrv.2019.07.006
12. Huang YS, Paiva T, Hsu JF, Kuo MC, Guilleminault C. **Sleep and breathing in premature infants at 6 months post-natal age**. *BMC Pediatr* (2014) **14** 303. DOI: 10.1186/s12887-014-0303-6
13. TIKOTZKY L, de MARCAS GALI, HAR-TOOV J, DOLLBERG S, BAR-HAIM Y, SADEH A. **Sleep and physical growth in infants during the first 6 months**. *J Sleep Res* (2010) **19** 103-10. DOI: 10.1111/j.1365-2869.2009.00772.x
14. Bathory E, Tomopoulos S. **Sleep regulation, physiology and development, sleep duration and patterns, and sleep hygiene in infants, toddlers, and preschool-age children**. *Curr Probl Pediatr Adolesc Health Care* (2017) **47** 29-42. DOI: 10.1016/j.cppeds.2016.12.001
15. Sivertsen B, Harvey AG, Reichborn-Kjennerud T, Ystrom E, Hysing M. **Sleep problems and depressive symptoms in toddlers and 8-year-old children: a longitudinal study**. *J Sleep Res* (2021) **30** e13150. DOI: 10.1111/jsr.13150
16. Bernier A, Beauchamp MH, Bouvette-Turcot AA, Carlson SM, Carrier J. **Sleep and cognition in preschool years: specific links to executive functioning**. *Child Dev* (2013) **84** 1542-53. DOI: 10.1111/cdev.12063
17. Hager ER, Calamaro CJ, Bentley LM, Hurley KM, Wang Y, Black MM. **Nighttime sleep duration and sleep behaviors among toddlers from low-income families: associations with obesogenic behaviors and obesity and the role of parenting**. *Child Obes* (2016) **12** 392-400. DOI: 10.1089/chi.2015.0252
18. 18.WHO. Haemoglobin Concentrations for the Diagnosis of Anaemia and Assessment of Severity. Vitamin and Mineral Nutrition Information System. Geneva: World Health Organization (2011).. *Haemoglobin Concentrations for the Diagnosis of Anaemia and Assessment of Severity. Vitamin and Mineral Nutrition Information System* (2011)
19. Sadeh A. **A brief screening questionnaire for infant sleep problems: validation and findings for an internet sample**. *Pediatrics* (2004) **113** e570-7. DOI: 10.1542/peds.113.6.e570
20. Ma S, Yin X, Tao R, Jiang X, Xie J, Li P. **Association of maternal prenatal depression and anxiety with toddler sleep: the China-Anhui birth cohort study**. *Arch Womens Ment Health* (2022) **25** 431-9. DOI: 10.1007/s00737-021-01200-w
21. Yin WJ, Yu LJ, Wu L, Zhang L, Li Q, Dai FC. **Adequate 25(OH)D moderates the relationship between dietary inflammatory potential and cardiovascular health risk during the second trimester of pregnancy**. *Front Nutr* (2022) **9** 952652. DOI: 10.3389/fnut.2022.952652
22. Villar J, Cheikh Ismail L, Victora CG, Ohuma EO, Bertino E, Altman DG. **International standards for newborn weight, length, and head circumference by gestational age and sex: the newborn cross-sectional study of the INTERGROWTH-21st project**. *Lancet* (2014) **384** 857-68. DOI: 10.1016/S0140-6736(14)60932-6
23. Nakahara K, Michikawa T, Morokuma S, Hamada N, Ogawa M, Kato K. **Association of maternal hemoglobin levels during pregnancy with sleep and developmental problems in 1-year-old infants: a cohort study**. *Health Sci Rep* (2022) **5** e552. DOI: 10.1002/hsr2.552
24. Sugimori N, Hamazaki K, Matsumura K, Kasamatsu H, Tsuchida A, Inadera H. **Association between mothers' fish intake during pregnancy and infants' sleep duration: a nationwide longitudinal study-the Japan environment and Children's study (JECS)**. *Eur J Nutr* (2022) **61** 679-86. DOI: 10.1007/s00394-021-02671-4
25. Abdul Jafar NK, Tham EKH, Pang WW, Fok D, Chua MC, Teoh OH. **Association between breastfeeding and sleep patterns in infants and preschool children**. *Am J Clin Nutr* (2021) **114** 1986-96. DOI: 10.1093/ajcn/nqab297
26. Jackowska M, Kumari M, Steptoe A. **Sleep and biomarkers in the English longitudinal study of ageing: associations with C-reactive protein, fibrinogen, dehydroepiandrosterone sulfate and hemoglobin**. *Psychoneuroendocrinology* (2013) **38** 1484-93. DOI: 10.1016/j.psyneuen.2012.12.015
27. Kordas K, Siegel EH, Olney DK, Katz J, Tielsch JM, Chwaya HM. **Maternal reports of sleep in 6-18 month-old infants from Nepal and Zanzibar: association with iron deficiency anemia and stunting**. *Early Hum Dev* (2008) **84** 389-98. DOI: 10.1016/j.earlhumdev.2007.10.007
28. Sifakis S, Pharmakides G. **Anemia in pregnancy**. *Ann N Y Acad Sci* (2000) **900** 125-36. DOI: 10.1111/j.1749-6632.2000.tb06223.x
29. Macher S, Herster C, Holter M, Moritz M, Matzhold EM, Stojakovic T. **The effect of parenteral or Oral iron supplementation on fatigue, sleep, quality of life and restless legs syndrome in iron-deficient blood donors: a secondary analysis of the IronWoMan RCT**. *Nutrients* (2020) **12** 1313. DOI: 10.3390/nu12051313
30. Zilberman M, Silverberg DS, Bits I, Steinbruch S, Wexler D, Sheps D. **Improvement of anemia with erythropoietin and intravenous iron reduces sleep-related breathing disorders and improves daytime sleepiness in anemic patients with congestive heart failure**. *Am Heart J* (2007) **154** 870-6. DOI: 10.1016/j.ahj.2007.07.034
31. Shen J, Miao X, Vu C, Xu B, Gonzalez-Zacarias C, Nederveen AJ. **Anemia increases oxygen extraction fraction in deep brain structures but not in the cerebral cortex**. *Front Physiol* (2022) **13** 896006. DOI: 10.3389/fphys.2022.896006
32. Ghiciuc CM, Dima Cozma LC, Bercea RM, Lupusoru CE, Mihaescu T, Szalontay A. **Restoring the salivary cortisol awakening response through nasal continuous positive airway pressure therapy in obstructive sleep apnea**. *Chronobiol Int* (2013) **30** 1024-31. DOI: 10.3109/07420528.2013.795155
33. Fan JM, Chen XQ, Jin H, Du JZ. **Gestational hypoxia alone or combined with restraint sensitizes the hypothalamic-pituitary-adrenal axis and induces anxiety-like behavior in adult male rat offspring**. *Neuroscience* (2009) **159** 1363-73. DOI: 10.1016/j.neuroscience.2009.02.009
34. Lozoff B, Georgieff MK. **Iron deficiency and brain development**. *Semin Pediatr Neurol* (2006) **13** 158-65. DOI: 10.1016/j.spen.2006.08.004
35. Shah HE, Bhawnani N, Ethirajulu A, Alkasabera A, Onyali CB, Anim-Koranteng C. **Iron deficiency-induced changes in the hippocampus, corpus striatum, and monoamines levels that Lead to anxiety, depression, sleep disorders, and psychotic disorders**. *Cureus* (2021) **13** e18138. DOI: 10.7759/cureus.18138
36. Jung J, Rahman MM, Rahman MS, Swe KT, Islam MR, Rahman MO. **Effects of hemoglobin levels during pregnancy on adverse maternal and infant outcomes: a systematic review and meta-analysis**. *Ann N Y Acad Sci* (2019) **1450** 69-82. DOI: 10.1111/nyas.14112
37. Young MF, Oaks BM, Tandon S, Martorell R, Dewey KG, Wendt AS. **Maternal hemoglobin concentrations across pregnancy and maternal and child health: a systematic review and meta-analysis**. *Ann N Y Acad Sci* (2019) **1450** 47-68. DOI: 10.1111/nyas.14093
38. Zhang Y, Jin L, Liu JM, Ye R, Ren A. **Maternal haemoglobin concentrations before and during pregnancy as determinants of the concentrations of children at 3-5 years of age: a large follow-up study**. *Eur J Clin Nutr* (2019) **73** 1102-9. DOI: 10.1038/s41430-018-0284-x
39. Perkin MR, Bahnson HT, Logan K, Marrs T, Radulovic S, Craven J. **Association of Early Introduction of solids with infant sleep: a secondary analysis of a randomized clinical trial**. *JAMA Pediatr* (2018) **172** e180739. DOI: 10.1001/jamapediatrics.2018.0739
|
---
title: 'COVID-19 and unfavorable changes in mental health unrelated to changes in
physical activity, sedentary time, and health behaviors among Swedish adolescents:
A longitudinal study'
authors:
- Gisela Nyberg
- Björg Helgadóttir
- Karin Kjellenberg
- Örjan Ekblom
journal: Frontiers in Public Health
year: 2023
pmcid: PMC10036362
doi: 10.3389/fpubh.2023.1115789
license: CC BY 4.0
---
# COVID-19 and unfavorable changes in mental health unrelated to changes in physical activity, sedentary time, and health behaviors among Swedish adolescents: A longitudinal study
## Abstract
### Background
The COVID-19 pandemic has had major impact on the daily lives of adolescents. This study examined whether mental health outcomes had changed over the pandemic, and if such changes were related to changes in physical activity (PA), sedentary time, sleep, screen time, and participation in organized sports.
### Materials and methods
In this longitudinal study, data were collected in autumn 2019 with follow-up measurements in spring 2021. In total, 558 schools were invited and 34 schools around Stockholm with a variation in socioeconomic background were included. Physical activity and sedentary time were measured for seven consecutive days by accelerometry (Actigraph). Anxiety, health-related quality of life (HRQoL), psychosomatic health, stress, sleep duration, screen time, and organized sports participation were self-reported in questionnaires. Linear models were applied to estimate associations between changes in mental health outcomes and exposures.
### Results
From the baseline sample of 1,139 participants, 585 ($55\%$ girls), mean (SD) age 14.9 (0.3) years, participated in the follow-up. Between 2019 and 2021, there was a decrease in HRQoL [mean difference −1.7 (−2.3, −1.2), $p \leq 0.001$], increase in psychosomatic health problems [mean difference 1.8 (1.3, 2.3), $p \leq 0.001$], and an increase in the number of participants with high stress [from 94 ($28\%$) to 139 ($42\%$), $p \leq 0.001$]. Weekly light PA and sleep duration decreased and weekly sedentary time and screen time increased unrelated to changes in mental health outcomes. An increase in sleep duration during weekdays was significantly related to both a decrease in anxiety (B = −0.71, CI: −1.36, −0.06) and an increase in HRQoL ($B = 1.00$, CI: 0.51, 1.49).
### Conclusion
During the COVID-19 pandemic, mental health appears to have been impaired in Swedish adolescents, but unrelated to changes in PA, sedentary time, screen time, or participation in organized sports. However, increased sleep duration on weekdays was related to less anxiety and better HRQoL. The results may help policy makers and other stakeholders comprehend the differential effects of the COVID-19 pandemic on mental health outcomes and help guiding the planning of policy actions.
### Trial registration
ISRCTN15689873.
## Introduction
There has been an increase in self-reported mental health and psychosomatic health problems in recent years among Swedish adolescents [1, 2]. Globally, psychiatric conditions were the single largest contributor to the overall burden of disease in ages under 20 years in high income countries in 2019 [3]. In addition, mental health problems at young age have been associated with an increased risk for prolonged and more severe forms of mental illnesses later in life [4]. Importantly, physical activity (PA) has been found to be associated with better mental health outcomes [5]. However, the majority of adolescents do not meet the PA recommendation of 60 min of daily PA at moderate to vigorous intensity (MVPA) [6, 7]. In addition, sedentary behavior, in particular screen time, has been associated with poor mental health [5].
The COVID-19 pandemic restrictions may have had an influence on current and future mental health among adolescents. It is therefore important to assess the impact of the pandemic on mental health so relevant interventions can be developed to mitigate its effects. In Sweden, restrictions related to the pandemic were minimal compared to those in many other countries. There was no lockdown but there were recommendations to keep physical distance, avoid crowding and large gatherings and non-essential traveling, and to stay home and avoid close contact with others when ill. Organized physical activities continued, albeit with adapted forms of participation, but matches and competitions were not allowed for adolescents. A large proportion of the physical education classes were performed outdoors. There were no school closures for adolescents but most schools had distance learning to some degree.
Global findings on changes in mental health during the pandemic are a cause of concern, suggesting an impact on mental health among adolescents [8, 9]. In addition, studies indicate that PA has decreased and sedentary behavior has increased during the pandemic [10, 11]. Currently, few longitudinal studies investigating the effects of COVID-19 on adolescents' mental health include baseline data from just before the pandemic. In addition, most previous studies measuring PA have used self-reported measurements [11]. To our knowledge, no longitudinal studies have yet been published that have data on mental health outcomes in relation to detailed device-measured PA, sedentary time, and health behaviors just before and during the pandemic.
The first phase of this study was performed between September and December 2019, thus providing unique data from just before the pandemic started. The data include multiple measures of mental health outcomes, device-measured PA, sedentary time, and health behaviors with a mean (SD) follow-up 18.4 (1.0) months later during the pandemic. The aim of this study was therefore to examine potential changes in mental health outcomes [anxiety, health-related quality of life (HRQoL), psychosomatic problems, and stress] and if such changes were related to changes in PA, sedentary time, and health behaviors (sleep, screen time, and participation in organized sports) before and during the COVID-19 pandemic among Swedish boys and girls.
## Study population
This longitudinal study used data from the study “Physical activity for healthy brain functions in school youth” [12] collected in September–December 2019 (baseline, $$n = 1$$,139) and in a follow-up in the same sample ($$n = 585$$) in April–June 2021.
Schools situated within a 2-to-3-h drive from Stockholm, Sweden were invited to participate in the study. Schools with a sports profile (having additional scheduled PA every week), with fewer than 15 students in each class, or with a student population that did not speak Swedish, were excluded. In total, 558 schools with students in grade seven (aged 13–14 years) were invited to participate in the study and 84 schools agreed to participate at baseline (see flowchart, Figure 1). For feasibility reasons, the inclusion stopped after 40 schools had been included. The schools represented a variation in type of municipality (urban and rural), socioeconomic background (parental education), and school organization (independent and public schools). One to four classes participated from each school. All the students in the participating classes were invited to participate in the study. In total, 1,556 students were invited and 1,139 accepted ($73\%$ participation rate). In total, $12\%$ of the schools were geographically located in rural areas and $88\%$ in urban areas. The proportion of parents with high education, on school level, was $60\%$. Eleven schools ($32\%$) were independents schools and 23 schools ($68\%$) were public schools. In Sweden, parents can choose between independent and public schools without paying any fees. At follow-up, schools were contacted again and asked to assist in the data collection. Many schools then had distance education and a high absence due to sickness. A total of 28 schools agreed to participate in the follow-up measures ($82\%$) and the students in the remaining six schools were invited to participate by post or e-mail. In total, 39 participants had moved; 1,100 students were invited, and 585 students agreed to participate ($51\%$ participation rate).
**Figure 1:** *Flow-chart of the schools and participants in the study.*
## Data collection
At baseline, the measurements were conducted at GIH by trained researchers. Physical activity and sedentary time were measured by accelerometry. Anxiety, health-related quality of life (HRQoL), psychosomatic health, stress, sleep duration, screen time, and sports participation were self-reported in questionnaires. Further methodological details can be found elsewhere [12]. At follow-up, accelerometers were sent to schools ($$n = 982$$) or by post to individual students ($$n = 157$$). The teacher or student was asked to report the first accelerometer wear-day and the participant wore the accelerometer for seven consecutive days. The accelerometers were sent back in pre-paid envelopes. Information on how to log in to complete the web-based questionnaire was distributed by the teachers or by post. The link was also sent to the e-mail address that the participant had provided at baseline. The teachers provided information regarding the degree of distance learning during the accelerometer wear period. Information from schools that declined to assist in the data collection was provided by the participating students in the questionnaire or by e-mail. The participants received a gift card as a compensation (in total 15€).
At both measurements, the students gave written consent. The study was conducted in accordance with the Declaration of Helsinki and the protocol was approved by the Ethical Review Agency in Stockholm, Sweden (Dnr: 2019-03579 and Dnr: 2021-01235). Trial registration: ISRCTN15689873.
## Mental health
Anxiety was measured using a short version of the Spence Children's Anxiety Scale (SCAS-S). The questionnaire includes 19 items and each item is rated on a four-point scale with the options of “never,” “sometimes,” “often,” and “always.” The scores are summed with higher values reflecting more symptoms of anxiety [13].
Health-related quality of life was measured using the questionnaire Kidscreen-10 [14]. The scale includes 10 items about how the students have felt during the last week (for example, “fit and well”), and how often. Answers range from “not at all” to “extremely” or from “never” to “always” on a five-point scale. The scores are summed with higher values indicating better well-being.
Psychosomatic health was assessed using the questionnaire Psychosomatic Problems Scale (PSP) [15]. The scale includes eight items with questions to the students about how they have felt during the last week, for example if they have had a “headache” or “stomach ache.” Each item is rated on a 5-point Likert-scale, with the options “never,” “seldom,” “sometimes,” “often,” and “always.” The scores are summed with higher values indicating more psychosomatic problems.
Stress was measured with the Single Item Stress Question (SISQ) [16]. Participants responded using a 5-point Likert scale with the options “not at all,” “only a little,” “to some extent,” “rather much,” and “very much.” Changes in the mental health variables were calculated by subtracting the baseline values from the follow-up values. Change in stress was derived by including only those who had low stress at baseline (not at all, only a little, to some extent) and categorizing them as continuing to have low stress or changing to having high stress (rather much, very much).
## Physical activity and time spent sedentary
Time spent in PA and sedentary was collected during 7 days using hip-worn accelerometers (model GT3X+, Actigraph, LCC, Pensacola, FL, USA). The same method for data collection and data processing was used at baseline and at follow-up described in more detail elsewhere [12]. The program ActiLife, version 6.13.3 was used to process data from the accelerometers. Individual wake and sleep times from the questionnaire were used to define time awake. Days with wear-time of at least 500 min were defined as valid days and at least three valid days, where at least one of those days was a weekend day, were included in the weekly analyses. The total wear-time from the monitors was categorized into sedentary intensity (0–100 counts per min), light PA intensity (101–2,295 counts per min), and MVPA (>2,295 counts per min) [17]. The results are presented as minutes within the respective intensity, as averages per day over the whole week. Change in PA and sedentary time was calculated by subtracting the baseline values from the follow-up values.
## Sleep
Sleep duration was calculated using bed-time and time for getting up reported by the participants in the questionnaire for weekdays and weekends. The response alternatives were in 30-min intervals. Change in sleep duration was calculated in hours.
## Screen time
The amount of screen time outside school hours (not including schoolwork) on weekdays and weekends was self-reported by the participants. The response alternatives were: no time at all, < 1 h, 1–2 h, 3–4 h, 5–6 h, and 7 h or more, and were categorized into ≤ 2 h, 3–4 h, and ≥5 h. The responses from baseline and follow-up were coded to reflect either an increase or decrease in screen time or no change in screen time.
## Organized sports participation
The participants reported in the questionnaire if they were active in any sports organization (yes/no). The answers from baseline and follow-up were combined to reflect change in organized sports: [1] no change, i.e., either participated at both baseline and follow-up or did not participate at both baseline and follow-up, [2] dropped out, i.e., participated in organized sports at baseline but not at follow-up, and [3] started, i.e., did not participate in organized sports at baseline but participated at follow-up.
## Parental education and country of birth
Register data on parental education was collected from the Statistics of Sweden and dichotomized into ≤ 12 years (low) and >12 years (high) of education. The highest level of education attained by either of the parents was used. The participant's and their parents' country of birth were reported by the participants in the questionnaire. Parents' country of birth was classified as “both born in Sweden,” “one born outside Sweden and one born in Sweden,” and “both parents born outside Sweden.”
## Anthropometry
Baseline data was used on body mass index (BMI). Underweight, normal weight, overweight and obesity were defined according to the International Obesity Task Force recommendations, with different cut-off values depending on age and sex [18].
## Distance learning
The amount of distance learning was self-reported at follow-up by the participants by asking them “Did you have distance learning during January to April 2021?” The response alternatives were “no, not at all,” “yes partially distance,” and “yes fully distance.” The two last alternatives were combined as very few ($$n = 27$$) reported full distance learning.
## Statistical analysis
Descriptive statistics are presented as percentages or means with standard deviations. Differences between girls and boys for the background variables were tested with chi-square tests and independent t-tests for categorical and continuous variables, respectively. The same analyses were used for comparing those who participated both at baseline and follow-up and those who did not participate at follow-up. Changes in mental health outcomes, as well as PA, sedentary time, sleep, screen time, and organized sports were analyzed with McNemars change test for categorical variables and paired t-tests for continuous variables. Relationships between the exposure variables and the continuous outcome variables (change in anxiety, HRQoL, and psychosomatic health) were analyzed in separate models with ANCOVAs (analysis of covariance), reported as non-standardized betas with $95\%$ confidence intervals. The binary outcome (change in stress) was modeled using multivariable logistic regression models, reported as odds ratios with $95\%$ confidence intervals. As the outcome variables showed minimal influence of clusters (0–$1\%$), all the models were on one level. The independent variables included in the models were checked for multi-collinearity, but none was detected. Homoscedasticity assumptions were checked through plotting of residuals against predicted values and all residuals were visually checked for normality of distribution. All models were adjusted for gender, parental education, parental country of birth as well as the baseline value of the outcome variable. Wear time was added in the initial analyses and did not alter the results and was therefore not included in the final analyses. No imputations for missing data or loss to follow-up were performed. All statistical tests were two-sided with a significance threshold of $P \leq 0.05.$ *The data* analyses were performed in IBM SPSS Statistics, version 27 (IBM Corp., Armonk, N.Y., USA).
## Sample characteristics
From the baseline sample of 1,139 participants, 585 ($51\%$) participated in the follow-up measurements. The mean (SD) age was 14.9 (0.3) years and 263 were boys ($45\%$) and 321 girls ($55\%$), one participant did not reveal gender. There were significant differences between the baseline and follow-up in gender composition, the proportions of parental country of birth, and the proportions of stress, as shown in Table 1. There were no significant differences between the baseline and follow-up samples in the other variables.
**Table 1**
| Variables measured at baseline | Participants | Drop-outs | p-Valuea |
| --- | --- | --- | --- |
| No. (%) with datab | 585 (51.4) | 554 (48.6) | |
| Gender [No. (%)] | Gender [No. (%)] | Gender [No. (%)] | Gender [No. (%)] |
| Boys | 263 (45.0) | 295 (53.2) | 0.006 |
| Girls | 321 (55.0) | 259 (46.8) | |
| Parental education [No. (%)] | Parental education [No. (%)] | Parental education [No. (%)] | Parental education [No. (%)] |
| Low ( ≤ 12 years) | 389 (68.1) | 341 (64.2) | 0.17 |
| High (>12 years) | 182 (31.9) | 190 (35.8) | |
| Participant country of birth [No. (%)] | Participant country of birth [No. (%)] | Participant country of birth [No. (%)] | Participant country of birth [No. (%)] |
| Sweden | 501 (86.1) | 466 (85.2) | 0.86 |
| Europe | 24 (4.1) | 22 (4.0) | |
| Outside Europe | 57 (9.8) | 59 (10.8) | |
| Parental country of birth [No. (%)] | Parental country of birth [No. (%)] | Parental country of birth [No. (%)] | Parental country of birth [No. (%)] |
| Both parents born in Sweden | 357 (62.5) | 299 (55.8) | 0.008 |
| One born outside Sweden | 69 (12.1) | 99 (18.5) | |
| Both born outside Sweden | 145 (25.4) | 138 (25.7) | |
| BMI status [No. (%)] | BMI status [No. (%)] | BMI status [No. (%)] | BMI status [No. (%)] |
| Underweight and normal weight | 467 (80.2) | 437 (79.0) | 0.61 |
| Overweight and obesity | 115 (19.8) | 116 (21.0) | |
| Mental health [mean (SD)] | Mental health [mean (SD)] | Mental health [mean (SD)] | Mental health [mean (SD)] |
| Anxiety (SCAS-S) (Missing n = 66) | 13.9 (7.7) | 13.4 (8.1) | 0.23 |
| HRQoL (Kidscreen) (Missing n = 42) | 39.6 (5.2) | 39.6 (5.6) | 0.96 |
| Psychosomatic problems (PSP) (Missing n = 43) | 17.3 (5.2) | 17.3 (5.7) | 0.88 |
| Stress [No. (%)] | Stress [No. (%)] | Stress [No. (%)] | Stress [No. (%)] |
| Not at all | 83 (14.4) | 123 (23.1) | 0.002 |
| Slightly | 247 (42.7) | 201 (37.7) | |
| Somewhat | 131 (22.7) | 101 (18.9) | |
| Quite much | 88 (15.2) | 72 (13.5) | |
| Very much | 29 (5.0) | 36 (6.8) | |
| Moderate to vigorous physical activity [mean (SD)] | Moderate to vigorous physical activity [mean (SD)] | Moderate to vigorous physical activity [mean (SD)] | Moderate to vigorous physical activity [mean (SD)] |
| Week (min/day) (Missing n = 236) | 52.3 (19.1) | 51.6 (18.9) | 0.63 |
| Light physical activity [mean (SD)] | Light physical activity [mean (SD)] | Light physical activity [mean (SD)] | Light physical activity [mean (SD)] |
| Week (min/day) (Missing n = 236) | 140.3 (29.1) | 136.7 (31.4) | 0.07 |
| Sedentary [mean (SD)] | Sedentary [mean (SD)] | Sedentary [mean (SD)] | Sedentary [mean (SD)] |
| Week (min/day) (Missing n = 236) | 603.7 (62.5) | 600.1 (70.9) | 0.43 |
| Sleep [mean (SD)] | Sleep [mean (SD)] | Sleep [mean (SD)] | Sleep [mean (SD)] |
| Weekdays (h) (Missing n = 17) | 8.6 (0.9) | 8.5 (1.0) | 0.47 |
| Weekends (h) (Missing n = 17) | 9.8 (1.4) | 10.0 (1.4) | 0.19 |
| Screen time (weekdays) [No. (%)] | | | 0.78 |
| ≤ 2 h | 190 (32.7) | 170 (31.2) | |
| 3–4 h | 260 (44.8) | 255 (46.8) | |
| ≥5 h | 131 (22.5) | 120 (22.0) | |
| Screen time (weekends) [No. (%)] | | | 0.44 |
| ≤ 2 h | 84 (14.5) | 94 (17.2) | |
| 3–4 h | 213 (36.9) | 199 (36.5) | |
| ≥5 h | 281 (48.6) | 252 (46.2) | |
## Change in mental health outcomes
Between 2019 and 2021, there was a decrease in HRQoL [mean difference −1.7 (−2.3, −1.2), $p \leq 0.001$], increase in psychosomatic health problems [mean difference 1.8 (1.3, 2.3), $p \leq 0.001$], and an increase in the number of participants with high stress [from 94 ($28\%$) to 139 ($42\%$), $p \leq 0.001$], in the total sample and also among both boys and girls when stratified (see Table 2). There was no significant change in anxiety in the total sample but among girls there was a significant increase [mean difference 1.7 (0.6, 2.7), $$p \leq 0.001$$], and among boys a significant decrease [mean difference −1.2 (−2.1, −0.1), $$p \leq 0.005$$], between baseline and follow-up measurements.
**Table 2**
| Unnamed: 0 | Unnamed: 1 | All participants | All participants.1 | Unnamed: 4 | Unnamed: 5 | Unnamed: 6 | Girls | Girls.1 | Unnamed: 9 | Unnamed: 10 | Unnamed: 11 | Boys | Boys.1 | Unnamed: 14 | Unnamed: 15 |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| | | Baseline | Follow-up | | | | Baseline | Follow-up | | | | Baseline | Follow-up | | |
| | | (2019) | (2021) | | | | (2019) | (2021) | | | | (2019) | (2021) | | |
| | n | | | Mean difference (95% CI) | p- value | n | | | Mean difference (95% CI) | p -value | n | | | Mean difference (95% CI) | p -value |
| Anxietya, Mean (SD) | 457 | 13.9 (7.9) | 14.2 (9.6) | 0.3 (−0.4, 1.0) | 0.35 | 248 | 16.1 (8.0) | 17.8 (9.3) | 1.7 (0.6, 2.7) | 0.001 | 208 | 11.2 (6.9) | 9.9 (8.1) | −1.2 (−2.1, −0.1) | 0.005 |
| HRQoLa Mean (SD) | 438 | 39.8 (5.2) | 38.1 (5.2) | −1.7 (−2.3, −1.2) | < 0.001 | 233 | 38.5 (5.1) | 36.6 (5.9) | −2.0 (−2.7, −1.2) | < 0.001 | 204 | 41.3 (4.9) | 39.8 (5.7) | −1.5 (−2.3, −0.6) | < 0.001 |
| Psychosomatic problemsa Mean (SD) | 501 | 9.3 (5.3) | 11.1 (6.7) | 1.8 (1.3, 2.3) | < 0.001 | 277 | 10.7 (5.3) | 13.4 (6.6) | 2.7 (2.0, 3.4) | < 0.001 | 223 | 7.5 (4.6) | 8.2 (5.8) | 0.7 (0.0, 1.4) | 0.049 |
| Stressb | | | | | | | | | | | | | | | |
| Low stress [No. (%)] | 333 | 239 (71.8) | 194 (58.3) | | < 0.001 | 178 | 101 (56.7) | 69 (38.8) | | < 0.001 | 154 | 138 (89.6) | 125 (81.2) | | < 0.001 |
| High stress [No. (%)] | | 94 (28.2) | 139 (41.7) | | | | 77 (43.3) | 109 (61.2) | | | | 16 (10.4) | 29 (18.8) | | |
## Change in physical activity, sedentary time, and health behaviors
In total, 264 participants had valid accelerometer registrations for the whole week, both at baseline and follow-up. There were no significant differences between participants with and without valid accelerometer registrations in anxiety ($$p \leq 0.31$$), HRQoL ($$p \leq 0.11$$), psychosomatic problems ($$p \leq 0.11$$), and stress ($$p \leq 0.44$$). Accelerometer wear-time at baseline compared to follow-up was higher for all participants, mean (SD) 803 [56] min/791 [71] min, $$p \leq 0.007$$ and girls 806 [55] min/793 [77] min, $$p \leq 0.02.$$
Between baseline and follow-up measurements, there was a significant decrease of an average of 20 min/day in light PA for the whole sample ($p \leq 0.001$), for girls (~18 min, $p \leq 0.001$), and for boys (~23 min, $p \leq 0.001$) but there was no significant change in MVPA, as shown in Table 3. There was a significant increase of an average of ~8 min in sedentary time for the whole group ($$p \leq 0.04$$) and for boys (~16 min, $$p \leq 0.03$$).
**Table 3**
| Unnamed: 0 | Unnamed: 1 | All participants | All participants.1 | Unnamed: 4 | Unnamed: 5 | Girls | Girls.1 | Unnamed: 8 | Unnamed: 9 | Boys | Boys.1 | Unnamed: 12 |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| | | Baseline | Follow-up | | | Baseline | Follow-up | | | Baseline | Follow-up | |
| | | (2019) | (2021) | | | (2019) | (2021) | | | (2019) | (2021) | |
| | n | | | p -value | n | | | p -value | n | | | p- value |
| MVPA in min a | MVPA in min a | MVPA in min a | MVPA in min a | MVPA in min a | MVPA in min a | MVPA in min a | MVPA in min a | MVPA in min a | MVPA in min a | MVPA in min a | MVPA in min a | MVPA in min a |
| Week, mean (SD) | 264 | 51.7 (19.0) | 51.2 (23.5) | 0.75 | 166 | 50.0 (18.1) | 49.9 (22.6) | 0.97 | 97 | 54.8 (20.1) | 53.8 (24.7) | 0.69 |
| Light PA in min a | Light PA in min a | Light PA in min a | Light PA in min a | Light PA in min a | Light PA in min a | Light PA in min a | Light PA in min a | Light PA in min a | Light PA in min a | Light PA in min a | Light PA in min a | Light PA in min a |
| Week, mean (SD) | 264 | 139.9 (28.4) | 119.9 (31.5) | < 0.001 | 166 | 138.5 (27.4) | 120.2 (29.7) | < 0.001 | 97 | 142.5 (30.1) | 119.9 (34.1) | < 0.001 |
| Sedentary in min a | Sedentary in min a | Sedentary in min a | Sedentary in min a | Sedentary in min a | Sedentary in min a | Sedentary in min a | Sedentary in min a | Sedentary in min a | Sedentary in min a | Sedentary in min a | Sedentary in min a | Sedentary in min a |
| Week, mean (SD) | 264 | 611.3 (60.8) | 619.6 (70.7) | 0.04 | 166 | 617.8 (59.3) | 622.4 (69.3) | 0.37 | 97 | 599.7 (62.0) | 615.6 (73.4) | 0.03 |
| Sleep in h a | Sleep in h a | Sleep in h a | Sleep in h a | Sleep in h a | Sleep in h a | Sleep in h a | Sleep in h a | Sleep in h a | Sleep in h a | Sleep in h a | Sleep in h a | Sleep in h a |
| Weekdays, mean (SD) | 557 | 8.6 (0.9) | 8.1 (1.0) | < 0.001 | 310 | 8.5 (0.9) | 8.0 (1.0) | < 0.001 | 246 | 8.6 (0.9) | 8.3 (0.9) | < 0.001 |
| Weekends, mean (SD) | 558 | 9.8 (1.4) | 9.5 (1.3) | < 0.001 | 311 | 9.9 (1.3) | 9.5 (1.3) | < 0.001 | 246 | 9.8 (1.5) | 9.5 (1.2) | 0.02 |
| Screen time during weekdaysb [No. (%)] | Screen time during weekdaysb [No. (%)] | Screen time during weekdaysb [No. (%)] | Screen time during weekdaysb [No. (%)] | Screen time during weekdaysb [No. (%)] | Screen time during weekdaysb [No. (%)] | Screen time during weekdaysb [No. (%)] | Screen time during weekdaysb [No. (%)] | Screen time during weekdaysb [No. (%)] | Screen time during weekdaysb [No. (%)] | Screen time during weekdaysb [No. (%)] | Screen time during weekdaysb [No. (%)] | Screen time during weekdaysb [No. (%)] |
| < 1 h per day | 556 | 39 (7.0) | 12 (2.2) | < 0.001 | 310 | 19 (6.1) | 5 (1.6) | < 0.001 | 245 | 20 (8.2) | 7 (2.9) | < 0.001 |
| 1–2 h per day | 556 | 144 (25.9) | 88 (15.8) | < 0.001 | 310 | 77 (24.8) | 41 (13.2) | < 0.001 | 245 | 66 (26.9) | 47 (19.2) | < 0.001 |
| 3–4 h per day | 556 | 245 (44.1) | 231 (41.5) | < 0.001 | 310 | 144 (46.5) | 132 (42.6) | < 0.001 | 245 | 101 (41.2) | 99 (40.4) | < 0.001 |
| 5–6 h per day | 556 | 101 (18.2) | 177 (31.8) | < 0.001 | 310 | 55 (17.7) | 110 (35.5) | < 0.001 | 245 | 46 (18.8) | 66 (26.9) | < 0.001 |
| 7 h or more | 556 | 27 (4.9) | 48 (8.6) | < 0.001 | 310 | 15 (4.8) | 22 (7.1) | < 0.001 | 245 | 12 (4.9) | 26 (10.6) | < 0.001 |
| Screen time during weekendsb [No. (%)] | Screen time during weekendsb [No. (%)] | Screen time during weekendsb [No. (%)] | Screen time during weekendsb [No. (%)] | Screen time during weekendsb [No. (%)] | Screen time during weekendsb [No. (%)] | Screen time during weekendsb [No. (%)] | Screen time during weekendsb [No. (%)] | Screen time during weekendsb [No. (%)] | Screen time during weekendsb [No. (%)] | Screen time during weekendsb [No. (%)] | Screen time during weekendsb [No. (%)] | Screen time during weekendsb [No. (%)] |
| < 1 h per day | 553 | 18 (3.3) | 10 (1.8) | < 0.001 | 310 | 8 (2.6) | 6 (1.9) | < 0.001 | 242 | 10 (4.1) | 4 (1.7) | < 0.001 |
| 1–2 h per day | 553 | 63 (11.4) | 29 (5.2) | < 0.001 | 310 | 32 (10.3) | 14 (4.5) | < 0.001 | 242 | 31 (12.8) | 15 (6.2) | < 0.001 |
| 3–4 h per day | 553 | 201 (36.3) | 140 (25.3) | < 0.001 | 310 | 123 (39.7) | 73 (23.5) | < 0.001 | 242 | 77 (31.8) | 66 (27.3) | < 0.001 |
| 5–6 h per day | 553 | 174 (31.5) | 214 (38.7) | < 0.001 | 310 | 100 (32.3) | 127 (41.0) | < 0.001 | 242 | 74 (30.6) | 87 (36.0) | < 0.001 |
| 7 h or more | 553 | 97 (17.5) | 160 (28.9) | < 0.001 | 310 | 47 (15.2) | 90 (29.0) | < 0.001 | 242 | 50 (20.7) | 70 (28.9) | < 0.001 |
| Organized sport participationb [No. (%)] | Organized sport participationb [No. (%)] | Organized sport participationb [No. (%)] | Organized sport participationb [No. (%)] | Organized sport participationb [No. (%)] | Organized sport participationb [No. (%)] | Organized sport participationb [No. (%)] | Organized sport participationb [No. (%)] | Organized sport participationb [No. (%)] | Organized sport participationb [No. (%)] | Organized sport participationb [No. (%)] | Organized sport participationb [No. (%)] | Organized sport participationb [No. (%)] |
| No | 515 | 134 (26.0) | 128 (24.9) | 0.54 | 288 | 73 (25.3) | 65 (22.6) | 0.26 | 227 | 61 (26.9) | 63 (27.8) | 0.77 |
| Yes | | 381 (74.0) | 387 (75.1) | | | 215 (74.7) | 223 (77.4) | | | 166 (73.1) | 164 (72.2) | |
| Distance learning spring 2021c [No. (%)] | Distance learning spring 2021c [No. (%)] | Distance learning spring 2021c [No. (%)] | Distance learning spring 2021c [No. (%)] | Distance learning spring 2021c [No. (%)] | Distance learning spring 2021c [No. (%)] | Distance learning spring 2021c [No. (%)] | Distance learning spring 2021c [No. (%)] | Distance learning spring 2021c [No. (%)] | Distance learning spring 2021c [No. (%)] | Distance learning spring 2021c [No. (%)] | Distance learning spring 2021c [No. (%)] | Distance learning spring 2021c [No. (%)] |
| No | 551 | | 118 (21.4) | | 306 | | 66 (21.6) | | 244 | | 52 (21.3) | |
| Yes | | | 433 (78.6) | | | | 240 (78.4) | | | | 192 (78.7) | |
Sleep duration decreased significantly during weekdays and weekends for the whole group (~0.5 h/night, $p \leq 0.001$)/(~0.3 h/night, $p \leq 0.001$), and also significantly among girls and boys when stratified (see Table 3).
Screen time increased significantly between baseline and follow-up: for example, the proportion of participants reporting ≥5 h of screen time during weekdays increased from $23.1\%$ to $40.4\%$ (girls $22.5\%$ to $42.6\%$ and boys $23.7\%$ to $37.5\%$) and during weekends from $49\%$ to $67.6\%$ (girls $47.5\%$ to $70\%$ and boys $51.3\%$ to $64.9\%$).
There was no significant difference between baseline and follow-up in organized sports participation.
At follow-up measurements, 433 participants ($78.6\%$) reported that they had some amount of distance learning during the spring period.
## Relationships between changes in mental health outcomes and changes in physical activity, sedentary time, and health behaviors
Results from the multivariable logistic regression are presented in Table 4 and showed that an increase in sleep during weekdays was significantly related to both a decrease in anxiety (B = −0.71, CI: −1.36, −0.06) and an increase in HRQoL ($B = 1.00$, CI: 0.51, 1.49). No significant relationships were found between changes in mental health outcomes and changes in MVPA, light PA, sedentary time, organized sports participation, and screen time.
**Table 4**
| Unnamed: 0 | Change in anxiety (SCAS-S) | Change in anxiety (SCAS-S).1 | Change in HRQoL (Kidscreen) | Change in HRQoL (Kidscreen).1 | Change in psychosomatic problems (PSP) | Change in psychosomatic problems (PSP).1 | Change in stress (SISQ) | Change in stress (SISQ).1 |
| --- | --- | --- | --- | --- | --- | --- | --- | --- |
| | β (95% CI) | p | β (95% CI) | p | β (95% CI) | p | OR (95% CI) | p |
| Change in MVPA | Change in MVPA | Change in MVPA | Change in MVPA | Change in MVPA | Change in MVPA | Change in MVPA | Change in MVPA | Change in MVPA |
| Week (min/day) | −0.02 (−0.05, 0.02) | 0.41 | 0.01 (−0.02, 0.04) | 0.53 | −0.00 (−0.03, 0.02) | 0.84 | 0.99 (0.98, 1.01) | 0.46 |
| Change in light PA | Change in light PA | Change in light PA | Change in light PA | Change in light PA | Change in light PA | Change in light PA | Change in light PA | Change in light PA |
| Week (min/day) | −0.01 (−0.04, 0.02) | 0.74 | 0.02 (−0.00, 0.04) | 0.09 | 0.00 (−0.02, 0.02) | 0.98 | 1.00 (0.99, 1.01) | 0.99 |
| Change in sedentary | Change in sedentary | Change in sedentary | Change in sedentary | Change in sedentary | Change in sedentary | Change in sedentary | Change in sedentary | Change in sedentary |
| Week (min/day) | −0.00 (−0.02, 0.01) | 0.58 | −0.00 (−0.01, 0.01) | 0.53 | −0.00 (−0.01, 0.01) | 0.68 | 1.00 (0.99, 1.00) | 0.69 |
| Change in sleep | Change in sleep | Change in sleep | Change in sleep | Change in sleep | Change in sleep | Change in sleep | Change in sleep | Change in sleep |
| Weekdays (h) | –0.71 (–1.36, –0.06) | 0.03 | 1.00 (0.51, 1.49) | < 0.001 | −0.40 (−0.89, 0.09) | 0.10 | 0.95 (0.74, 1.22) | 0.68 |
| Weekends (h) | −0.23 (−0.66, 0.21) | 0.30 | 0.26 (−0.06, 0.59) | 0.11 | −0.24 (−0.55, 0.08) | 0.14 | 0.88 (0.75, 1.03) | 0.10 |
| Change in organized sport participation | Change in organized sport participation | Change in organized sport participation | Change in organized sport participation | Change in organized sport participation | Change in organized sport participation | Change in organized sport participation | Change in organized sport participation | Change in organized sport participation |
| No change | REF | | REF | | REF | | REF | |
| Dropped out | 1.94 (−0.42, 4.30) | 0.11 | −1.76 (−3.72, 0.19) | 0.08 | 1.56 (−0.20, 3.31) | 0.08 | 0.88 (0.37, 2.10) | 0.77 |
| Started | −0.01 (−2.30, 2.27) | 0.99 | −0.67 (−2.57, 1.23) | 0.49 | −0.80 (−2.49, 0.89) | 0.35 | 0.96 (0.41, 2.34) | 0.93 |
| Change in screen time during weekdays | Change in screen time during weekdays | Change in screen time during weekdays | Change in screen time during weekdays | Change in screen time during weekdays | Change in screen time during weekdays | Change in screen time during weekdays | Change in screen time during weekdays | Change in screen time during weekdays |
| No change | REF | | REF | | REF | | REF | |
| Increase | −0.17 (−1.60, 1.27) | 0.82 | 0.63 (−0.44, 1.70) | 0.25 | 0.03 (−1.03, 1.09) | 0.96 | 1.00 (0.59, 1.68) | 0.99 |
| Decrease | 0.12 (−1.84, 2.09) | 0.90 | 0.39 (−1.15, 1.93) | 0.62 | 0.31 (−1.20, 1.82) | 0.68 | 1.82 (0.90, 3.70) | 0.10 |
| Change in screen time during weekends | Change in screen time during weekends | Change in screen time during weekends | Change in screen time during weekends | Change in screen time during weekends | Change in screen time during weekends | Change in screen time during weekends | Change in screen time during weekends | Change in screen time during weekends |
| No change | REF | | REF | | REF | | REF | |
| Increase | −0.71 (−2.15, 0.73) | 0.33 | 0.67 (−0.44, 1.78) | 0.23 | −0.72 (−1.80, 0.35) | 0.19 | 0.81 (0.49, 1.36) | 0.43 |
| Decrease | −0.30 (−2.30, 1.71) | 0.77 | 0.02 (−1.52, 1.57) | 0.98 | −1.32 (−2.80, 0.17) | 0.08 | 1.19 (0.59, 2.43) | 0.63 |
| Distance learning spring 2021 a | Distance learning spring 2021 a | Distance learning spring 2021 a | Distance learning spring 2021 a | Distance learning spring 2021 a | Distance learning spring 2021 a | Distance learning spring 2021 a | Distance learning spring 2021 a | Distance learning spring 2021 a |
| No | REF | | REF | | REF | | REF | |
| Yes | −0.71 (−2.31, 0.89) | 0.39 | 0.61 (−0.67, 1.88) | 0.35 | −0.06 (−1.26, 1.15) | 0.93 | 0.79 (0.45, 1.41) | 0.43 |
## Discussion
To the best of our knowledge, this is the first longitudinal study with data on mental health outcomes in relation to detailed device-measured PA, sedentary time, and health behaviors, just before and during the pandemic. Our findings show that self-reported HRQoL, psychosomatic health, and stress have worsened in adolescents in Sweden during the COVID-19 pandemic but unrelated to changes in PA, sedentary time, screen time, or participation in organized sports. Nonetheless, an increase in sleep duration during weekdays was related to better mental health.
The change in mental health in this study is in line with studies in two systematic reviews showing that the COVID-19 pandemic had impacted the mental health of adolescents [8, 9]. Studies from Norway ($$n = 3$$,752) [19] and Australia ($$n = 248$$) [20] also found that anxiety and depressive symptoms increased between baseline and follow-up in adolescents. In contrast, one Swedish study found no longitudinal changes in mental health, PA or health behaviors among adolescents ($$n = 584$$) [21]. Similar results have also been reported in two studies of adolescents in China where no deterioration in adolescents' mental health was found [22, 23]. Studies in England have shown conflicting findings: some data indicated increased depressive symptoms and other data indicated no changes in anxiety or wellbeing [24]. The overall variability of the results between studies might relate to different ages and cultural contexts of participants. In addition, differences in the scope of pandemic restrictions, different study designs, outcome measures, heterogeneity in methods, and length of follow-up periods may have played a role.
The observational design prevented any control group. Therefore, the mean change in the prevalence of mental health in this study cannot be fully attributed to the COVID-19 pandemic, as no comparable data are available. However, the main question was to investigate if any such changes were related to changes in PA variables. The main result from these analyses were that only few such relations were found. To what extent these findings can be relevant for other situations, i.e., without a pandemic, is difficult to assess. However, as we identify temporal changes between baseline and follow-up but only few relations to changes in PA, it can be hypothesized that smaller changes in mental health (possibly found in non-pandemic situations) would also be unrelated to changes in PA.
Several studies have shown that the impact on mental health outcomes has been worse for females compared to males [8, 20, 23]. In our study sample, the prevalence of mental health problems was higher in girls than in boys but the decrease in well-being as well as impaired psychosomatic health and stress was evident in both boys and girls. However, there was a difference in anxiety where anxiety levels decreased in boys and increased in girls during the pandemic.
The findings in this study showed no decline in MVPA. However, there was a decline in light PA and an increase in sedentary time. In addition, there was an increase in screen time and no change in participation in organized sports. This might imply that Swedish adolescents have not stopped participating in organized activities outside school hours but may be moving less in their everyday lives since they spent more time at home due to distance education. Studies from other countries showed declines in PA levels and increases in sedentary behavior during the pandemic [10, 11].
This study also showed a decrease in sleep duration during the pandemic. Furthermore, sleep duration during weekdays was related to mental health outcomes, suggesting that an increase in sleep duration during weekdays was related to better HRQoL and less anxiety. Other studies that have measured sleep before and during the pandemic showed mixed results, including increased sleep duration [11], fewer sleep problems [23], and no changes in sleep [21] among adolescents.
In a study with Chinese children and adolescents, it was reported that longer screen time before and during the pandemic was associated with a higher risk of psychological symptoms [25]. In the present study the results showed no relationships between changes in mental health and changes in PA, sedentary time, screen time, or participation in organized sports, which suggest that there were other factors underlying the negative changes in mental health outcomes. Physical activity did not decline to the same extent in Sweden compared to other countries, possibly explained by Sweden's milder pandemic restrictions. One Swedish study showed that during the pandemic both children and adolescents expressed worries, for example about the disease and death among their relatives, about the future, or missing out on their youth and employment [26]. Swedish adolescents further reported increased conflicts with parents, less time spent with peers, and poorer control over their everyday life [27]. Similar data are available from Australia [20].
The major strength of this study is the longitudinal design with baseline data collected just before the pandemic in late 2019 and follow-up in the spring of 2021. Another strength is the detailed device-measured PA in relation to several mental health outcomes. One limitation was the loss of valid accelerometer measurements due to insufficient wear time. However, there were no significant differences between participants with and without valid accelerometer measurements in mental health outcomes. Another limitation of the study is that the sample was non-representative and therefore caution with generalizability should be taken. In addition, the drop-out rate between baseline and follow-up might also have impacted the generalizability of the study. However, the recruitment of schools was based on a variation in type of municipality and socioeconomic background of parents and attrition bias may be limited as the results showed few differences in variables between participants and drop-outs. Another limitation, as mentioned above, is that the effects on mental health outcomes might have been confounded by an expected age-related increase in mental health problems within this group and this could have led to a slight overestimation of the changes in mental health outcomes.
## Conclusion
The results suggest that the COVID-19 pandemic has impaired the mental health of Swedish adolescents and that the decrease in mental health was not related to changes in PA, sedentary time, screen time, or participation in organized sports. However, increased sleep duration during weekdays was related to positive changes in anxiety and HRQoL. The results of this study may help policy makers and other stakeholders comprehend the differential effects of the COVID-19 pandemic on mental health outcomes and help guiding the planning of policy actions.
## Data availability statement
The datasets are not available for download to protect the confidentiality of the participants. The data are held at The Swedish School of Sport and Health Sciences. Anonymous data for meta-analyses can be provided upon request from the corresponding author. Requests to access the datasets should be directed to gisela.nyberg@gih.se.
## Ethics statement
The studies involving human participants were reviewed and approved by the Ethical Review Agency in Stockholm, Sweden (Dnr: 2019-03579 and Dnr: 2021-01235). Written informed consent to participate in this study was provided by the participants' legal guardian/next of kin.
## Author contributions
GN, BH, and KK conducted the data collection. BH and KK cleaned, processed, accessed, and verified the data. GN, BH, and ÖE selected the design of the statistical model. BH performed the analyses. GN drafted the manuscript. All authors contributed to the design of the study and the interpretation of the results. All authors contributed to the article and approved the submitted version.
## Conflict of interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher's note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## References
1. 1.FORTE. Psykiskt Välbefinnande, Psykiska Besvär och Psykiatriska Tillstånd hos Barn och Unga – Begrepp, Mätmetoder och Förekomst (Mental Well-Being, Mental Distress and Mental Disorders among Children, and Young Adults. Terminology, Measurement Methods and Prevalence – An Overview). Stockholm: FORTE (2021).. *Psykiskt Välbefinnande, Psykiska Besvär och Psykiatriska Tillstånd hos Barn och Unga – Begrepp, Mätmetoder och Förekomst (Mental Well-Being, Mental Distress and Mental Disorders among Children, and Young Adults. Terminology, Measurement Methods and Prevalence – An Overview)* (2021)
2. **Stockholm: Public Health Agency of Sweden**. (2018)
3. 3.Institute for Health Metrics Evaluation. Global Burden of Disease (GBD). (2021). Available online at: http://www.healthdata.org/gbd. (2021)
4. 4.The National Board of Health and Welfare. Utvecklingen av Psykisk Ohälsa Bland Barn och Unga Vuxna - Till och Med 2016 (Development of Mental Ill-Health Among Children and Youth Until 2016). Stockholm: National Board of Health and Welfare (2017).. *Utvecklingen av Psykisk Ohälsa Bland Barn och Unga Vuxna - Till och Med 2016 (Development of Mental Ill-Health Among Children and Youth Until 2016)* (2017)
5. Chaput JP, Willumsen J, Bull F, Chou R, Ekelund U, Firth J. **2020 WHO guidelines on physical activity and sedentary behaviour for children and adolescents aged 5–17 years: summary of the evidence**. *Int J Behav Nutr Phys Act.* (2020) **17** 141. DOI: 10.1186/s12966-020-01037-z
6. Guthold R, Stevens GA, Riley LM, Bull FC. **Global trends in insufficient physical activity among adolescents: a pooled analysis of 298 population-based surveys with 16 million participants**. *Lancet Child Adolesc Health.* (2020) **4** 23-35. DOI: 10.1016/S2352-4642(19)30323-2
7. Nyberg G, Kjellenberg K, Froberg A, Lindroos AK. **A national survey showed low levels of physical activity in a representative sample of Swedish adolescents**. *Acta Paediatr.* (2020) **109** 2342-53. DOI: 10.1111/apa.15251
8. Samji H, Wu J, Ladak A, Vossen C, Stewart E, Dove N. **Review: Mental health impacts of the COVID-19 pandemic on children and youth - a systematic review**. *Child Adolesc Ment Health.* (2022) **27** 173-89. DOI: 10.1111/camh.12501
9. Newlove-Delgado T, Russell AE, Mathews F, Cross L, Bryant E, Gudka R. **Annual Research Review: The impact of Covid-19 on psychopathology in children and young people worldwide: systematic review of studies with pre- and within-pandemic data**. *J Child Psychol Psychiatry.* (2022) **2022** 1-30. DOI: 10.1111/jcpp.13716
10. Stockwell S, Trott M, Tully M, Shin J, Barnett Y, Butler L. **Changes in physical activity and sedentary behaviours from before to during the COVID-19 pandemic lockdown: a systematic review**. *BMJ Open Sport Exerc Med.* (2021) **7** e000960. DOI: 10.1136/bmjsem-2020-000960
11. Paterson DC, Ramage K, Moore SA, Riazi N, Tremblay MS, Faulkner G. **Exploring the impact of COVID-19 on the movement behaviors of children and youth: a scoping review of evidence after the first year**. *J Sport Health Sci.* (2021) **10** 675-89. DOI: 10.1016/j.jshs.2021.07.001
12. Nyberg G, Ekblom O, Kjellenberg K, Wang R, Larsson H, Thedin Jakobsson B. **Associations between the school environment and physical activity pattern during school time in swedish adolescents**. *Int J Environ Res Public Health.* (2021) **18** 10239. DOI: 10.3390/ijerph181910239
13. Ahlen J, Vigerland S, Ghaderi A. **Development of the Spence Children's Anxiety Scale-Short Version (SCAS-S)**. *J Psychopathol Behav Assess.* (2018) **40** 288-304. DOI: 10.1007/s10862-017-9637-3
14. Ravens-Sieberer U, Erhart M, Rajmil L, Herdman M, Auquier P, Bruil J. **Reliability, construct and criterion validity of the KIDSCREEN-10 score: a short measure for children and adolescents' well-being and health-related quality of life**. *Qual Life Res.* (2010) **19** 1487-500. DOI: 10.1007/s11136-010-9706-5
15. Hagquist C. **Psychometric properties of the psychosomatic problems scale: a rasch analysis on adolescent data**. *Soc Indic Res.* (2008) **86** 511-23. DOI: 10.1007/s11205-007-9186-3
16. Elo A-L, Leppänen A, Jahkola A. **Validity of a single-item measure of stress symptoms**. *Scand J Work Environ Health.* (2003) **29** 444-51. DOI: 10.5271/sjweh.752
17. Evenson KR, Catellier DJ, Gill K, Ondrak KS, McMurray RG. **Calibration of two objective measures of physical activity for children**. *J Sports Sci.* (2008) **26** 1557-65. DOI: 10.1080/02640410802334196
18. Cole TJ, Lobstein T. **Extended international (IOTF) body mass index cut-offs for thinness, overweight and obesity**. *Pediatr Obes.* (2012) **7** 284-94. DOI: 10.1111/j.2047-6310.2012.00064.x
19. Hafstad GS, Saetren SS, Wentzel-Larsen T, Augusti EM. **Adolescents' symptoms of anxiety and depression before and during the Covid-19 outbreak - a prospective population-based study of teenagers in Norway**. *Lancet Reg Health Eur.* (2021) **5** 100093. DOI: 10.1016/j.lanepe.2021.100093
20. Magson NR, Freeman JYA, Rapee RM, Richardson CE, Oar EL, Fardouly J. **Risk and protective factors for prospective changes in adolescent mental health during the COVID-19 pandemic**. *J Youth Adolesc.* (2021) **50** 44-57. DOI: 10.1007/s10964-020-01332-9
21. Chen Y, Osika W, Henriksson G, Dahlstrand J, Friberg P. **Impact of COVID-19 pandemic on mental health and health behaviors in Swedish adolescents**. *Scand J Public Health.* (2022) **50** 26-32. DOI: 10.1177/14034948211021724
22. Lu P, Yang L, Wang C, Xia G, Xiang H, Chen G. **Mental health of new undergraduate students before and after COVID-19 in China**. *Sci Rep.* (2021) **11** 18783. DOI: 10.1038/s41598-021-98140-3
23. Li Y, Zhou Y, Ru T, Niu J, He M, Zhou G. **How does the COVID-19 affect mental health and sleep among Chinese adolescents: a longitudinal follow-up study**. *Sleep Med.* (2021) **85** 246-58. DOI: 10.1016/j.sleep.2021.07.008
24. Newlove-Delgado T, McManus S, Sadler K, Thandi S, Vizard T, Cartwright C. **Child mental health in England before and during the COVID-19 lockdown**. *Lancet Psychiatry.* (2021) **8** 353-4. DOI: 10.1016/S2215-0366(20)30570-8
25. Xiang M, Liu Y, Yamamoto S, Mizoue T, Kuwahara K. **Association of changes of lifestyle behaviors before and during the COVID-19 pandemic with mental health: a longitudinal study in children and adolescents**. *Int J Behav Nutr Phys Act.* (2022) **19** 92. DOI: 10.1186/s12966-022-01327-8
26. Sarkadi A, Sahlin Torp L, Perez-Aronsson A, Warner G. **Children's expressions of worry during the COVID-19 pandemic in Sweden**. *J Pediatr Psychol.* (2021) **46** 939-49. DOI: 10.1093/jpepsy/jsab060
27. Kapetanovic S, Gurdal S, Ander B, Sorbring E. **Reported changes in adolescent psychosocial functioning during the COVID-19 outbreak**. *Adolescents.* (2021) **1** 10-20. DOI: 10.3390/adolescents1010002
|
---
title: 'The association between visceral adiposity index and decreased renal function:
A population-based study'
authors:
- Zheng Qin
- Xinyang Chen
- Jiantong Sun
- Luojia Jiang
journal: Frontiers in Nutrition
year: 2023
pmcid: PMC10036366
doi: 10.3389/fnut.2023.1076301
license: CC BY 4.0
---
# The association between visceral adiposity index and decreased renal function: A population-based study
## Abstract
### Aims
We aimed to investigate the association of visceral adiposity index (VAI) with decreased renal function in US adults.
### Design and methods
Cross-sectional data were analyzed for 35,018 adults in the National Health and Nutrition Examination Survey (NHANES) 2005–2018. VAI was determined using waist circumference, body mass index (BMI), triglycerides (TGs) and high-density lipoprotein-cholesterol. Albuminuria was defined as urinary albumin-to-creatinine ratio (ACR) >30 mg/g. A low estimated-glomerular filtration rate (eGFR) was defined as an eGFR lower than 60 ml/min/1.73 m2. Chronic kidney disease (CKD) was defined as either albuminuria or low-eGFR. A multivariable logistic regression analysis was utilized to explore the relationship of VAI with albuminuria, low-eGFR and CKD. Subgroup analysis and interaction tests were also conducted.
### Results
A total of 35,018 participants were enrolled with albuminuria, low-eGFR, and CKD prevalence rates of 5.18, 6.42, and $10.62\%$, respectively, which increased with the higher VAI tertiles. After full adjustment, a positive association of VAI with albuminuria (OR = 1.03, $95\%$ CI: 1.00, 1.06) and CKD (OR = 1.04, $95\%$ CI: 1.02, 1.06) was observed. Participants in the highest VAI tertile had a significantly $30\%$ increased risk for albuminuria (OR = 1.30, $95\%$ CI: 1.07, 1.58) and a $27\%$ increased risk for CKD (OR = 1.27, $95\%$ CI: 1.08, 1.49) compared with those in the lowest VAI tertile. No statistically significant association between VAI and low-eGFR was detected. Subgroup analysis and the interaction term indicated that there was no significant difference among different stratifications.
### Conclusion
Visceral adiposity accumulation evaluating by VAI was associated with increased likelihood of the decline in renal function.
## 1. Introduction
The damage of renal function could be reflected in albuminuria, a decrease in the estimated-glomerular filtration rate (eGFR) and even the development of chronic kidney disease (CKD) [1]. Approximately 5–$19\%$ of the general population suffer from albuminuria, which is a putative marker of an impaired glomerular filtration barrier and abnormal urinary albumin excretion (2–4). Decreased renal function is usually defined as eGFR <60 mL/min/1.73 m2 [5]. Both albuminuria and low-eGFR are not only markers of early kidney disease, but also predictors of CKD progression and cardiovascular disease [6, 7]. CKD is a chronic condition presented as kidney structural or functional abnormalities caused by multiple factors with a global prevalence of 10.5–$13.1\%$, which represents a significant global health burden as well [8, 9]. Globally, the incidence of CKD increased by $89\%$, prevalence increased by $87\%$, death due to CKD increased by $98\%$, and disability-adjusted-life-years (DALYs) increased by $62\%$ from 1990 to 2016 [10].
Visceral Adiposity index (VAI), which is also known as visceral fat grade, has proven to be a reliable indicator of visceral fat accumulation and dysfunction in adipose tissue [11, 12]. It was calculated using anthropometric [waist circumference [13], body mass index (BMI)] and metabolic parameters [triglyceride (TG) and high density lipoprotein-cholesterol (HDL-C) concentrations] to evaluate visceral obesity function, which has been broadly used in previous studies [14]. The higher the index value, the higher the content of visceral fat. Compared to other traditional body assessment parameters including BMI, WC, and waist-to-height ratio (WHtR), VAI can accurately distinguish visceral adiposity from subcutaneous adiposity. Magnetic Resonance Imaging (MRI) and Computed Tomography (CT) are precise and reliable to identify visceral fat, however, these machine-based measurements can be costly and difficult to conduct for some individuals. Thus, using VAI, a mathematical model that takes into account both anthropometric and metabolic parameters to evaluate the adipose distribution may be a better tool for assessing the impacts of visceral adiposity on clinical outcomes. Many studies have confirmed the positive predictive value of VAI for insulin insensitivity and diabetes (15–18). Moreover, VAI is significantly associated with cardiovascular diseases, such as hypertension and arterial atherosclerosis (19–22). The correlation of VAI with non-alcoholic fatty liver disease (NAFLD) [23, 24], metabolic syndrome (MS) [25], and some tumors [26, 27] has also been widely reported. More particularly, VAI may also be a superior predictor for kidney disease. Several studies have suggested a relationship between visceral fat and renal function. An epidemiological study in Netherlands revealed that visceral fat was associated with microalbuminuria in women while liver fat was not, which was supported by Mendelian randomization [28]. Another retrospective study including 14,529 male and 10,561 female adults in China demonstrated independent predictive power of visceral obesity on renal damage in all ages and both genders, except men under 45 years of age [29]. Moreover, there is evidence to suggest that obesity, especially central obesity, is an important risk factor for CKD (30–33). However, the association has not been studied in a nationally representative sample of US adults.
Hence, we investigated whether VAI was associated with decreased renal function among National Health and Nutrition Examination Survey (NHANES) subjects. Our hypothesis was that an elevated VAI would increase the risk of reduced kidney function.
## 2.1. Survey description
National Health and Nutrition Examination *Survey is* a national population-based cross-sectional cohort study to assess the nutrition and health status of non-institutionalized civilian populations in the US [34]. For recruitment of a representative sample of the US population, a stratified, multistage and probability sampling design was used.
The National Center for Health Statistics (NCHS) Research Ethics Review Board approved all NHANES protocols. All survey participants provided written informed consent. A detailed description of the NHANES study and its data is available online at https://www.cdc.gov/nchs/nhanes/.
## 2.2. Study population
We used the NHANES survey cycles 2005–2018 because these surveys have provided complete data to calculate the VAI, urinary albumin: creatinine ratio (ACR) and eGFR using the same protocols. An in-home interview and a physical examination were conducted at a mobile examination center to collect blood and urine samples.
The analysis included participants with complete information about VAI and renal function. There were initially 70,190 participants enrolled in the study. After excluding participants aged <18 years ($$n = 28047$$), pregnant ($$n = 737$$), missing data about ACR ($$n = 2646$$), VAI (total, $$n = 3741$$; BMI, $$n = 445$$; TG, $$n = 2124$$; WC, $$n = 1163$$; HDL-C, $$n = 9$$), and eGFR ($$n = 1$$), our final analysis included 35,018 eligible participants (Figure 1).
**FIGURE 1:** *Flowchart of the sample selection from NHANES 2005–2018.*
## 2.3. Definition of visceral adiposity index, albuminuria, low-eGFR and chronic kidney disease
Visceral adiposity index is a gender-specific index that estimates visceral adiposity by combining WC, BMI, TG, and HDL-C. Higher VAI scores indicate more estimated visceral fat. The VAI for each participant was calculated by using the following formulas [19]. For males: VAI = WC/(39.68 + (1.88 × BMI)) × (TG/1.03) × (1.31/HDL-C); For females: VAI = WC/(36.58 + (1.89 × BMI)) × (TG/0.81) × (1.52/HDL-C). TG and HDL-C were calculated in mmol/L, and WC was calculated in cm in the formulas. The VAI was analyzed as a continuous variable and further analysis was conducted by grouping participants according to their VAI tertiles.
A single, spot urine sample was used to determine urinary albumin and creatinine using a solid-phase fluorescent immunoassay and a modified Jaffe kinetic method. Based on the urinary ACR, we defined albuminuria as ACR >30 mg/g. Serum creatinine was measured using the Jaffe rate method and calibrated isotope dilution mass spectrometry. The CKD Epidemiology Collaboration (CKD-EPI) creatinine equation was used to calculate eGFR based on data about gender, race, age, and SCr [35]. Low-eGFR was defined as an eGFR lower than 60 ml/min/1.73 m2. CKD is characterized by albuminuria or a low-eGFR as defined by Kidney Disease: Improving Global Outcomes 2012 [36].
In our study, VAI were designed as an exposure variable, albuminuria, low-eGFR and CKD were treated as outcome variables.
## 2.4. Selection of covariates
Several potential covariates that may confound the association between VAI and decreased renal function were summarized in our analysis, including gender, age, race, education level, smoking status, BMI, serum creatinine, serum uric acid, total cholesterol, ALT, AST, hypertension and diabetes. According to BMI, participants were considered normal weight, overweight and obese when their BMI fell between <25, 25–29.9, and ≥30 kg/m2. In addition, we also treated gender (male/female), age (<60/≥60 years), BMI (normal weight/overweight/obesity), hypertension (yes/no), diabetes (yes/no) as stratified factors to conduct subgroup analysis and pre-specified effect modifiers to evaluate the interaction effect. All details regarding these variables are available on the website at www.cdc.gov/nchs/nhanes/.
## 2.5. Statistical analysis
According to NHANES analytic guidelines, statistical analyses were performed with appropriate sampling weights and accounting for complex multistage cluster surveys. Means with standard error (SE) was calculated for continuous variables, and proportions were calculated for categorical variables. Participants grouped by VAI tertiles were compared using a weighted Student’s t-test (for continuous variables) or a weighted chi-square test (for categorical variables). Three different models were analyzed using multivariable logistic regression to determine the effects of VAI on outcome variables (albuminuria, low eGFR, and CKD). In model 1, no covariates were adjusted. Model 2 was adjusted for gender, age and race. Model 3 was adjusted for gender, age, race, education level, BMI, ALT, AST, serum creatinine, serum uric acid, total cholesterol, hypertension, diabetes and smoking status. To reduce the potential bias and enhance the reliability of our study, first, we conducted all analysis with the consideration of NHANES sampling weights to make our samples more representative and reduce the selection bias. In addition, we adjusted for many confounding covariates to reduce the confounding bias. We also treated VAI as tertiles to evaluate the robustness in sensitivity analysis. Subgroup analyses of the associations of VAI with albuminuria, low-eGFR and CKD were conducted with stratified factors, including gender (male/female), age (<60/≥60 years), BMI (normal weight/overweight/obesity), hypertension (yes/no), and diabetes (yes/no). In addition, the stratified factors were also treated as pre-specified potential modifiers, with an interaction term added to measure heterogeneity among subgroups. Input of missing values was done by median for continuous variables or mode for categorical variables. All analyses were preformed using R version 3.4.3 (The R Foundation)1 and Empower software (X&Y Solutions, Inc., Boston MA, USA).2 The level of statistical significance was set at $P \leq 0.05.$
## 3.1. Participants characteristics at baseline
There were 35,018 participants enrolled in this study, whose average age was 46.44 ± 0.23 years. Among them, $49.20\%$ were male and $50.80\%$ were female. The prevalence rates of albuminuria, low-eGFR and CKD was 5.18, 6.42, and $10.62\%$, respectively. Participants in higher VAI tertiles had increased rates of albuminuria, low-eGFR, CKD as well. In the lowest VAI tertile participants, $3.93\%$ had albuminuria, $4.23\%$ had low-eGFR and $7.65\%$ had CKD. In the middle VAI tertile participants, $5.17\%$ had albuminuria, $6.26\%$ had low-eGFR and $10.54\%$ had CKD. Participants in the highest VAI tertile showed the highest rates of albuminuria ($6.44\%$), low-eGFR ($8.78\%$) and CKD ($13.67\%$). Age, gender, smoking status, BMI, diabetes, hypertension, serum creatinine, serum uric acid, total cholesterol, HDL-C, ALT, AST, waist circumference, TGs, urinary albumin and ACR were significantly different among the VAI tertiles (all $P \leq 0.05$). Compared to the lowest VAI group, participants with increased VAI group were significantly more likely to have hypertension, diabetes, elevated BMI, serum uric acid, total cholesterol, ALT, AST, waist circumference, TGs, urinary albumin, ACR and decreased HDL-C levels (all $P \leq 0.05$). There was no statistically significant difference between tertiles in race, education level or urinary creatinine (all $P \leq 0.05$) (Table 1).
**TABLE 1**
| VAI | Overall | Tertile 1 | Tertile 2 | Tertile 3 | P for trend |
| --- | --- | --- | --- | --- | --- |
| Age (year) | 46.44 ± 0.23 | 43.35 ± 0.32 | 46.87 ± 0.24 | 49.09 ± 0.27 | <0.0001 |
| Gender, % (SE) | Gender, % (SE) | Gender, % (SE) | Gender, % (SE) | Gender, % (SE) | Gender, % (SE) |
| Male | 49.20 (0.27) | 49.33 (0.54) | 46.90 (0.62) | 51.40 (0.62) | 0.0308 |
| Female | 50.80 (0.27) | 50.67 (0.54) | 53.10 (0.62) | 48.60 (0.62) | |
| Race, % (SE) | Race, % (SE) | Race, % (SE) | Race, % (SE) | Race, % (SE) | Race, % (SE) |
| Mexican American | 8.76 (0.68) | 6.48 (0.51) | 9.05 (0.73) | 10.76 (0.90) | 0.6185 |
| Other Hispanic | 5.63 (0.44) | 4.90 (0.43) | 5.71 (0.45) | 6.27 (0.53) | |
| Non-Hispanic white | 67.29 (1.27) | 65.33 (1.25) | 67.36 (1.37) | 69.19 (1.43) | |
| Non-Hispanic black | 10.68 (0.66) | 15.71 (0.93) | 10.35 (0.64) | 5.98 (0.45) | |
| Other Races | 7.64 (0.38) | 7.58 (0.43) | 7.53 (0.46) | 7.80 (0.48) | |
| Education level, % (SE) | Education level, % (SE) | Education level, % (SE) | Education level, % (SE) | Education level, % (SE) | Education level, % (SE) |
| Less than high school | 16.28 (0.55) | 13.24 (0.53) | 16.22 (0.70) | 19.38 (0.69) | 0.1927 |
| High school or GED | 23.61 (0.47) | 21.64 (0.68) | 23.21 (0.59) | 25.98 (0.68) | |
| Above high school | 60.06 (0.81) | 65.10 (0.92) | 60.52 (0.92) | 54.56 (0.92) | |
| Others | 0.05 (0.01) | 0.03 (0.01) | 0.05 (0.02) | 0.07 (0.03) | |
| Smoking status, % (SE) | Smoking status, % (SE) | Smoking status, % (SE) | Smoking status, % (SE) | Smoking status, % (SE) | Smoking status, % (SE) |
| Never | 55.37 (0.55) | 59.59 (0.79) | 55.90 (0.71) | 50.71 (0.70) | <0.0001 |
| Former | 24.44 (0.41) | 22.31 (0.59) | 24.34 (0.60) | 26.64 (0.64) | |
| Current | 20.18 (0.44) | 18.10 (0.55) | 19.76 (0.61) | 22.65 (0.60) | |
| BMI (Kg/m2) | 28.84 ± 0.08 | 25.94 ± 0.09 | 29.12 ± 0.11 | 31.45 ± 0.09 | <0.0001 |
| Diabetes, % (SE) | 8.95 (0.22) | 4.44 (0.23) | 7.94 (0.32) | 14.50 (0.47) | <0.0001 |
| Hypertension, % (SE) | 30.54 (0.47) | 20.99 (0.57) | 30.33 (0.65) | 40.32 (0.68) | <0.0001 |
| SCr (μmol/L) | 78.21 ± 0.21 | 77.13 ± 0.33 | 77.95 ± 0.30 | 79.55 ± 0.32 | <0.0001 |
| Serum uric acid (μmol/L) | 322.21 ± 0.74 | 300.57 ± 0.98 | 321.35 ± 1.16 | 344.75 ± 1.22 | <0.0001 |
| TC (mmol/L) | 5.01 ± 0.01 | 4.76 ± 0.02 | 4.97 ± 0.02 | 5.30 ± 0.02 | <0.0001 |
| HDL-C (mmol/L) | 1.38 ± 0.01 | 1.70 ± 0.01 | 1.36 ± 0.00 | 1.08 ± 0.00 | <0.0001 |
| ALT (U/L) | 25.27 ± 0.14 | 22.26 ± 0.22 | 24.56 ± 0.17 | 29.00 ± 0.26 | <0.0001 |
| AST (U/L) | 25.24 ± 0.11 | 24.96 ± 0.20 | 24.45 ± 0.15 | 26.32 ± 0.20 | <0.0001 |
| Triglycerides (mmol/L) | 1.71 ± 0.01 | 0.79 ± 0.00 | 1.38 ± 0.00 | 2.98 ± 0.02 | <0.0001 |
| Waist circumference (cm) | 98.69 ± 0.22 | 90.29 ± 0.23 | 99.34 ± 0.25 | 106.46 ± 0.23 | <0.0001 |
| Albumin, urine (mg/L) | 32.26 ± 1.26 | 23.05 ± 1.29 | 26.35 ± 1.50 | 47.46 ± 3.47 | <0.0001 |
| Creatinine, urine (mg/dL) | 122.32 ± 0.83 | 121.67 ± 1.39 | 121.80 ± 1.11 | 123.49 ± 1.05 | 0.2675 |
| ACR (mg/g) | 31.30 ± 1.27 | 20.83 ± 1.23 | 26.55 ± 1.76 | 46.60 ± 3.36 | <0.0001 |
| eGFR (mL/min/1.73 m2) | 94.82 ± 0.31 | 98.53 ± 0.38 | 94.32 ± 0.36 | 91.60 ± 0.36 | <0.0001 |
| Albuminuria, % (SE) | 5.18 (0.17) | 3.93 (0.21) | 5.17 (0.26) | 6.44 (0.37) | <0.0001 |
| Low-eGFR, % (SE) | 6.42 (0.22) | 4.23 (0.28) | 6.26 (0.28) | 8.78 (0.38) | <0.0001 |
| CKD, % (SE) | 10.62 (0.26) | 7.65 (0.33) | 10.54 (0.37) | 13.67 (0.50) | <0.0001 |
## 3.2. A positive association of visceral adiposity index and albuminuria
Our results showed that a higher VAI was associated with an increased risk of albuminuria. This association was significant both in our crude model (OR = 1.04, $95\%$ CI: 1.03, 1.06) and minimally adjusted model (OR = 1.06, $95\%$ CI: 1.04, 1.09). After full adjustment, a positive association between VAI and albuminuria still remained stable (OR = 1.03, $95\%$ CI: 1.00, 1.06), indicating that each unit of VAI score was associated with a $4\%$ increase in albuminuria risk. When treating VAI as tertiles, participants in the highest VAI tertile had a significantly $30\%$ increased risk of albuminuria compared with those in the lowest tertile (OR = 1.30, $95\%$ CI: 1.07, 1.58) (Table 2).
**TABLE 2**
| Visceral adiposity index group | Albuminuria OR1 (95% CI2), P-value | Low-eGFR OR (95% CI), P-value | CKD OR (95% CI), P-value |
| --- | --- | --- | --- |
| Crude model (Model 1)3 | Crude model (Model 1)3 | Crude model (Model 1)3 | Crude model (Model 1)3 |
| Continuous | 1.04 (1.03, 1.06), <0.0001 | 1.04 (1.03, 1.05), <0.0001 | 1.04 (1.03, 1.06), <0.0001 |
| Categories | Categories | Categories | Categories |
| Tertile 1 | Reference | Reference | Reference |
| Tertile 2 | 1.33 (1.16, 1.52), 0.0001 | 1.51 (1.31, 1.75), <0.0001 | 1.42 (1.27, 1.59), <0.0001 |
| Tertile 3 | 1.68 (1.43, 1.98), <0.0001 | 2.18 (1.88, 2.53), <0.0001 | 1.91 (1.69, 2.16), <0.0001 |
| P for trend | <0.0001 | <0.0001 | <0.0001 |
| Minimally adjusted model (Model 2)4 | Minimally adjusted model (Model 2)4 | Minimally adjusted model (Model 2)4 | Minimally adjusted model (Model 2)4 |
| Continuous | 1.06 (1.04, 1.09), <0.0001 | 1.06 (1.04, 1.08), <0.0001 | 1.05 (1.03, 1.07), < 0.0001 |
| Categories | Categories | Categories | Categories |
| Tertile 1 | Reference | Reference | Reference |
| Tertile 2 | 1.20 (1.04, 1.38), 0.0142 | 1.30 (1.10, 1.54), 0.0028 | 1.19 (1.05, 1.35), 0.0087 |
| Tertile 3 | 1.61 (1.35, 1.92), <0.0001 | 2.01 (1.70, 2.37), <0.0001 | 1.59 (1.39, 1.83), <0.0001 |
| P for trend | <0.0001 | <0.0001 | <0.0001 |
| Fully adjusted model (Model 3)5 | Fully adjusted model (Model 3)5 | Fully adjusted model (Model 3)5 | Fully adjusted model (Model 3)5 |
| Continuous | 1.03 (1.00, 1.06), 0.0336 | 1.00 (0.93, 1.08), 0.9325 | 1.04 (1.02, 1.06), 0.0005 |
| Categories | Categories | Categories | Categories |
| Tertile 1 | Reference | Reference | Reference |
| Tertile 2 | 1.14 (0.98, 1.34), 0.0952 | 1.65 (0.82, 3.31), 0.1633 | 1.09 (0.95, 1.25), 0.2474 |
| Tertile 3 | 1.30 (1.07, 1.58), 0.0108 | 1.24 (0.56, 2.72), 0.5932 | 1.27 (1.08, 1.49), 0.0056 |
| P for trend | 0.0118 | 0.8004 | 0.0050 |
## 3.3. Visceral adiposity index and low-eGFR
We also estimated the association of VAI with low-eGFR in three different models. We found a significant positive association between VAI and low-eGFR both in both the crude (Model 1: OR = 1.04, $95\%$ CI: 1.03, 1.05) and minimally adjusted models (Model 2: OR = 1.06, $95\%$ CI: 1.04, 1.08). Despite this, this positive relationship did not reach statistical significance in fully adjusted analysis (OR = 1.00, $95\%$ CI: 0.93, 1.08). Additionally, they did not show any statistically significant association when VAI was treated as tertiles (Table 2).
## 3.4. A positive association of visceral adiposity index and chronic kidney disease
For CKD, we also found a positive association between VAI and the increased likelihood of CKD with statistical significance. In our crude model and minimally adjusted model, participants with a higher VAI tended to show an increased risk of CKD (Model 1: OR = 1.04, $95\%$ CI: 1.03, 1.06; Model 2: OR = 1.05, $95\%$ CI: 1.03, 1.07). A unit higher of VAI increased CKD risk by $4\%$ after full adjustment (Model 3: OR = 1.04, $95\%$ CI: 1.02, 1.06). Even after treating VAI as tertiles, there was still a statistically significant association. A significant $27\%$ higher risk was experienced by participants in the highest VAI tertile compared to those in the lowest VAI tertile (OR = 1.27, $95\%$ CI: 1.08, 1.49) (Table 2).
## 3.5. Subgroup analysis
Our results indicated that the associations of VAI level with decreased renal function were not consistent. A significant relationship of VAI with albuminuria was detected in female (OR = 1.03), age ≥60 years (OR = 1.05), normal weight (OR = 1.06) and obese (OR = 1.04), hypertension (OR = 1.04) and non-diabetes subjects (OR = 1.03), respectively. In the interaction test, VAI and albuminuria were not significantly associated with each stratification (Figure 2).
**FIGURE 2:** *Subgroup analysis for the association between VAI and albuminuria.*
For the association between VAI and low-eGFR, consistent previous results, we still did not observe any statistically significant relationship (Figure 3).
**FIGURE 3:** *Subgroup analysis for the association between VAI and low-eGFR.*
For CKD, a positive association was found in females (OR = 1.04), both age <60 (OR = 1.03) and ≥60 years (OR = 1.05), normal weight (OR = 1.08) and obese (OR = 1.04), both hypertension (OR = 1.05) and non-hypertension (OR = 1.03), both diabetes (OR = 1.04) and non-diabetes (OR = 1.04) subjects (Figure 4). In addition, there was no significant difference suggested by the interaction test among different stratifications, indicating that this positive association was not significantly influenced by gender, age, BMI, hypertension and diabetes on this positive association (all P for interaction >0.05).
**FIGURE 4:** *Subgroup analysis for the association between VAI and CKD.*
## 4. Discussion
Our cross-sectional study with 35,018 participants found that participants with higher VAI were more likely to have albuminuria and CKD. Subgroup analysis and interaction tests revealed similar associations across different populations. Based on our results, visceral adiposity accumulation should be considered, and the management of visceral fat distribution may alleviate the decrease in renal function. Our results also indicated the negative effects of the accumulation of visceral adiposity on renal health. Since the calculation of VAI was simple and cheap, and it could distinguish visceral adiposity from subcutaneous adiposity accurately, individuals should pay more attention to this mathematical parameter for their health.
Our study assessed the association between VAI and decreased renal function. The impact of the VAI on renal function has been previously evaluated in the different regions and populations (37–43). A study conducted in 6,693 non-diabetic participants in Iran observed that VAI seems to be an independent predictor of renal function decline only in males [37]. A cross-sectional study that enrolled 4,947 participants in Korea confirmed that the VAI was a good predictor of the pathogenesis of CKD in men but not in women [38]. However, Dai and colleagues has previously demonstrated that VAI was significantly associated with CKD in women in rural population of northeast China [39]. A recent study of 400 individuals aged 50–90 years also reported a stronger correlation of VAI with CKD in Taiwan, China, especially for middle-aged and elderly females [40]. Similarly, Huang et al. reported a consistent result based on a cross-sectional study that included 2,142 individuals in South China [41]. Furthermore, a population-based study of 15,159 participants conducted by Bamba et al. demonstrated that VAI can be a predictor of incident CKD in both males and females in Japan [42]. Wen et al. also showed a clear association between the VAI and urinary albumin excretion in Chinese type 2 diabetic patients regardless of the gender [43]. Similarly, we highlighted the negative effects of visceral adiposity accumulation on renal health and our present study was consistent with those reports of Bamba et al. and Wen et al., suggesting that higher VAI indicated increased likelihood of albuminuria and CKD. As for different results reported before, we think the variance of population characteristics, including race, region, sample size, CKD definition, and eGFR-related calculation methods may contribute to the discrepancy among these studies. In addition, the results without adjustment for established risk factors for CKD, such as blood pressure, plasma glucose, and serum low density lipoprotein-cholesterol, might lead to misleading conclusions as well [44]. VAI can distinguish visceral adiposity from subcutaneous adiposity accurately compared to some other body assessment parameters, such as BMI, WC, and WHtR. For the application of VAI in clinical practice, we think individuals can measure the VAI index of each patient and stratify patients’ risk according to the VAI. According to different risk stratification of patients, more targeted health management for patients could be conducted. In addition, considering the negative effect of visceral fat on renal health, individuals can take the initiative to change their lifestyle, use drugs and other methods to reduce their visceral obesity.
The association between obesity and an increased risk of incident CKD [45, 46], end-stage renal disease [47, 48], and mortality [49, 50] has already been demonstrated by previous studies. In addition to VAI, other visceral adiposity accumulation indicators, such as lipid accumulation product (LAP), BMI, WC, WHtR, and waist-to-hip ratio (WHR), also have clinical predictive effects. BMI is the most classic indicator in the assessment of adiposity [37]. A meta-analysis including 39 cohorts covering 630,677 participants revealed that higher BMI was associated with an increased risk of low-eGFR (hazard ratio, HR = 1.02, $95\%$ CI: 1.01–1.03) and albuminuria (HR = 1.02, $95\%$ CI: 1.00–1.04) [51]. However, due to the deficiency in distinguishing between fat and muscle as well as between subcutaneous and visceral fat tissues, BMI may lead to bias in measuring the effects of obesity on health outcomes [52]. In a prospective study of a Korean population, Oh H et al. reported that WC, not BMI, could predict a decline in renal function. Simultaneously, WHR and WHtR were reported to be associated with renal function decline as well [53]. Moreover, Elsayed et al. revealed that the assessment of CKD risk should use WHR rather than BMI as an anthropomorphic measure of obesity [54]. Unfortunately, both WC and WHR have limited accuracy in distinguishing between visceral adipose tissue (VAT) and subcutaneous adipose tissue (SAT) [55]. Previous studies have shown that LAP and VAI are superior to BMI, WC, WHtR and WHR in the evaluation of renal function decline in clinical practice [37, 39]. Furthermore, Mousapour et al. suggested that while LAP and VAI outperform BMI, WC, WHtR and WHR, VAI could be an independent predictor of renal function decline in non-diabetic males [37]. Similarly, our study focused on the association between VAI and decreased renal function, and detected a positive relationship of higher VAI with increased likelihood of albuminuria and CKD. Since VAI was a reliable parameter of visceral fat, our results highlighted the negative effects of visceral obesity on renal health.
Several potential mechanisms may explain the association of VAI and decreased kidney function. Visceral adiposity has a positive relationship with the development of inflammation, oxidative stress, endothelial dysfunction, and atherosclerosis, resulting in glomerulosclerosis and tubulointerstitial fibrosis (56–58). Thus, it may lead to a decrease in kidney function. Adiposity accumulation could induce pro-inflammatory pathways, including interleukin-6 (IL-6), tumor necrosis factor-alpha (TNF-α), and transforming growth factor-beta (TGF-β), as well as augment the production of reactive oxygen species [59, 60]. Furthermore, this accumulation could also activate the renin-angiotensin-aldosterone system (RAAS), causing hypertension and increasing insulin resistance, which are recognized renal injury factors [61, 62]. Additionally, the distribution of central fat might increase the glomerular filtration rate relative to effective renal plasma flow, leading to an increased filtration fraction, and ultimately to glomerular hyper-perfusion, hypertension and even functional loss [63]. In addition, it was noting that participants in the third tertile of VAI seems to have worse conditions, such as higher age, obesity, hypertension, diabetes and disorder of lipid metabolism, these comorbidities may contribute to the increased risk of decreased renal function.
Although we detected a significantly positive association of VAI and CKD, it may be mediated by some other diseases and conditions. Using same NHANES dataset, Ciardullo et al. found that participant with MS (both obese and non-obese) showed higher prevalence of albuminuria and reduced eGFR compared with those without obesity or MS, while there was no significant difference in those with MS but without obese. They suggested that MS and but not increased body fat alone was related with CKD [64]. Indeed, the association between MS and CKD was controversial. Some studies did not find increased risk of CKD in metabolically healthy obese individuals compared to metabolically healthy non-obese individuals, while others found a residual increase in CKD risk that remained (65–69). A possible explanation of these inconsistencies is related to different definitions of CKD and MS. Some studies did not include albuminuria as an indicator of CKD and the MS diagnostic criteria varied among different studies as well. In addition to MS, NAFLD and related fibrosis may also influence the relationship between VAI and CKD. The correlation of VAI with NAFLD has been reported before [23, 24]. Meanwhile, liver fibrosis, but not steatosis, has been proven to associated with reduced kidney function independently [70]. In a meta-analysis enrolled seven cross-sectional studies, Ciardullo et al. reported that elevated liver stiffness was associated with higher likelihood of kidney outcomes including albuminuria and CKD among patients with NAFLD [71]. Taken together, at least part of the association between VAI and CKD may be mediated by MS, NAFLD and associated fibrosis, etc. In our analysis, VAI is calculated considering TG and HDL levels, which are MS parameters. Even VAI representing visceral fat accumulation, perhaps some metabolism-related factor could have influenced our results as well. Thus, more large-scale prospective studies are still needed to valid our findings.
This study has several strengths. First, the sample selection and sample size are representative and sufficient. To our knowledge, the present study included the largest number of samples on this topic. Additionally, we adjusted for confounding covariates to reduce the confounding bias. Therefore, more reliable conclusions can be obtained. However, the limitations should also be noted. Our cross-sectional study design did not permit us to establish a causal relationship. Prospective studies with larger sample sizes are needed to clarify this issue. In addition, even though some potential covariates have been adjusted, other potentially confounding variables could not be completely excluded, such as the use of drugs including diuretics and steroids, etc. The use of a single spot of urine to evaluate albuminuria is a limitation as well. Although it is a validated method, is not the best option compared to 24 h urine.
## 5. Conclusion
Elevated VAI levels were independently associated with albuminuria and CKD, which highlights the importance of managing decreased renal function in patients with visceral adipose accumulation. However, the validity of our findings needs to be further confirmed by large-scale prospective studies.
## Data availability statement
Publicly available datasets were analyzed in this study. This data can be found here: https://www.cdc.gov/nchs/nhanes/.
## Ethics statement
The studies involving human participants were reviewed and approved by the National Center for Health Statistics (NCHS) Research Ethics Review Board. The patients/participants provided their written informed consent to participate in this study.
## Author contributions
ZQ: software, data analysis, and writing—original draft. XC: writing—original draft, formal analysis, and methodology. JS: data analysis. LJ: conceptualization, funding acquisition, and writing—reviewing and editing. All authors contributed to the article and approved the submitted version.
## Conflict of interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher’s note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## References
1. Gillen DL, Worcester EM, Coe FL. **Decreased renal function among adults with a history of nephrolithiasis: a study of NHANES III.**. (2005) **67** 685-90. DOI: 10.1111/j.1523-1755.2005.67128.x
2. Böhm M, Thoenes M, Danchin N, Bramlage P, La Puerta P, Volpe M. **Association of cardiovascular risk factors with microalbuminuria in hypertensive individuals: the i-SEARCH global study.**. (2007) **25** 2317-24. DOI: 10.1097/HJH.0b013e3282ef1c5f
3. Romagnani P, Remuzzi G, Glassock R, Levin A, Jager KJ, Tonelli M. **Chronic kidney disease.**. (2017) **3**. DOI: 10.1038/nrdp.2017.88
4. Comper WD, Hilliard LM, Nikolic-Paterson DJ, Russo LM. **Disease-dependent mechanisms of albuminuria.**. (2008) **295** F1589-600. DOI: 10.1152/ajprenal.00142.2008
5. Inker LA, Eneanya ND, Coresh J, Tighiouart H, Wang D, Sang Y. **New Creatinine- and Cystatin C-based equations to estimate GFR without race.**. (2021) **385** 1737-49. DOI: 10.1056/NEJMoa2102953
6. Matsushita K, van der Velde M, Astor BC, Woodward M, Levey AS, de Jong PE. **Association of estimated glomerular filtration rate and albuminuria with all-cause and cardiovascular mortality in general population cohorts: a collaborative meta-analysis.**. (2010) **375** 2073-81. PMID: 20483451
7. Perkovic V, Verdon C, Ninomiya T, Barzi F, Cass A, Patel A. **The relationship between proteinuria and coronary risk: a systematic review and meta-analysis.**. (2008) **5**. DOI: 10.1371/journal.pmed.0050207
8. Levey AS, Coresh J. **Chronic kidney disease.**. (2012) **379** 165-80. PMID: 21840587
9. ópez-Novoa JML, Martínez-Salgado C, Rodríguez-Peña AB, ópez-Hernández FJL. **Common pathophysiological mechanisms of chronic kidney disease: therapeutic perspectives.**. (2010) **128** 61-81. PMID: 20600306
10. Cong J, Cheng B, Liu J, He P. **RTEF-1 inhibits vascular smooth muscle cell calcification through regulating Wnt/β-catenin signaling pathway.**. (2021) **109** 203-14. DOI: 10.1007/s00223-021-00833-4
11. Li R, Li Q, Cui M, Ying Z, Li L, Zhong T. **Visceral adiposity index, lipid accumulation product and intracranial atherosclerotic stenosis in middle-aged and elderly Chinese.**. (2017) **7**. DOI: 10.1038/s41598-017-07811-7
12. Roriz AK, Passos LC, de Oliveira CC, Eickemberg M, Moreira Pde A, Sampaio LR. **Evaluation of the accuracy of anthropometric clinical indicators of visceral fat in adults and elderly.**. (2014) **9**. DOI: 10.1371/journal.pone.0103499
13. Zdziechowska M, Gluba-Brzozka A, Poliwczak AR, Franczyk B, Kidawa M, Zielinska M. **Serum NGAL, KIM-1, IL-18, L-FABP: new biomarkers in the diagnostics of acute kidney injury (AKI) following invasive cardiology procedures.**. (2020) **52** 2135-43. DOI: 10.1007/s11255-020-02530-x
14. Amato MC, Giordano C, Pitrone M, Galluzzo A. **Cut-off points of the visceral adiposity index (VAI) identifying a visceral adipose dysfunction associated with cardiometabolic risk in a Caucasian Sicilian population.**. (2011) **10**. DOI: 10.1186/1476-511X-10-183
15. Tsou MT, Chang YC, Hsu CP, Kuo YC, Yun CH, Huang WH. **Visceral adiposity index outperforms conventional anthropometric assessments as predictor of diabetes mellitus in elderly Chinese: a population-based study.**. (2021) **18**. DOI: 10.1186/s12986-021-00608-6
16. Wei J, Liu X, Xue H, Wang Y, Shi Z. **Comparisons of visceral adiposity index, body shape index, body mass index and waist circumference and their associations with diabetes mellitus in adults.**. (2019) **11**. DOI: 10.3390/nu11071580
17. Ahn N, Baumeister SE, Amann U, Rathmann W, Peters A, Huth C. **Visceral adiposity index (VAI), lipid accumulation product (LAP), and product of triglycerides and glucose (TyG) to discriminate prediabetes and diabetes.**. (2019) **9**. DOI: 10.1038/s41598-019-46187-8
18. Han M, Qin P, Li Q, Qie R, Liu L, Zhao Y. **Chinese visceral adiposity index: a reliable indicator of visceral fat function associated with risk of type 2 diabetes.**. (2021) **37**. DOI: 10.1002/dmrr.3370
19. Amato MC, Giordano C, Galia M, Criscimanna A, Vitabile S, Midiri M. **Visceral adiposity index: a reliable indicator of visceral fat function associated with cardiometabolic risk.**. (2010) **33** 920-2. DOI: 10.2337/dc09-1825
20. Leite NN, Cota BC, Gotine A, Rocha D, Pereira PF, Hermsdorff HHM. **Visceral adiposity index is positively associated with blood pressure: a systematic review.**. (2021) **15** 546-56. DOI: 10.1016/j.orcp.2021.10.001
21. Yu Y, Zhang FL, Yan XL, Zhang P, Guo ZN, Yang Y. **Visceral adiposity index and cervical arterial atherosclerosis in northeast China: a population based cross-sectional survey.**. (2021) **28** 161-71. DOI: 10.1111/ene.14513
22. Li J, Zhu J, Tan Z, Yu Y, Luo L, Zhou W. **Visceral adiposity index is associated with arterial stiffness in hypertensive adults with normal-weight: the china H-type hypertension registry study.**. (2021) **18**. DOI: 10.1186/s12986-021-00617-5
23. Ismaiel A, Jaaouani A, Leucuta DC, Popa SL, Dumitrascu DL. **The visceral adiposity index in non-alcoholic fatty liver disease and liver fibrosis-systematic review and meta-analysis.**. (2021) **9**. DOI: 10.3390/biomedicines9121890
24. Vural Keskinler M, Mutlu HH, Sirin A, Erkalma Senates B, Colak Y, Tuncer I. **Visceral adiposity index as a practical tool in patients with biopsy-proven nonalcoholic fatty liver disease/nonalcoholic steatohepatitis.**. (2021) **19** 26-31. DOI: 10.1089/met.2020.0054
25. Bijari M, Jangjoo S, Emami N, Raji S, Mottaghi M, Moallem R. **The accuracy of visceral adiposity index for the screening of metabolic syndrome: a systematic review and meta-analysis.**. (2021) **2021**. DOI: 10.1155/2021/6684627
26. Cardoso-Peña E, Soto Pina AE, Villanueva Á, López Chavez G, Ramírez Martínez P, Ramírez Montoya H. **Visceral adiposity index in breast cancer survivors: a case-control study.**. (2020) **2020**
27. Okamura T, Hashimoto Y, Hamaguchi M, Obora A, Kojima T, Fukui M. **Visceral adiposity index is a predictor of incident colorectal cancer: a population-based longitudinal study.**. (2020) **7**. DOI: 10.1136/bmjgast-2020-000400
28. Dekkers IA, de Vries APJ, Smit RAJ, Rosendaal FR, Rabelink TJ, Lamb HJ. **The separate contributions of visceral fat and liver fat to chronic kidney disease-related renal outcomes.**. (2020) **30** 286-95. DOI: 10.1053/j.jrn.2019.09.002
29. Tsao YC, Chen JY, Yeh WC, Li WC. **Gender- and age-specific associations between visceral obesity and renal function impairment.**. (2019) **12** 67-77. DOI: 10.1159/000496626
30. Wang J, Jin X, Chen K, Yan W, Wang A, Zhu B. **Visceral adiposity index is closely associated with urinary albumin-creatinine ratio in the Chinese population with prediabetes.**. (2021) **37**
31. Moh MC, Sum CF, Tavintharan S, Ang K, Kwan PY, Lee SBM. **Gain in adiposity over 3 years is associated with progressive renal decline in multi-ethnic South-east Asians with type 2 diabetes.**. (2019) **11** 316-25. DOI: 10.1111/1753-0407.12848
32. Wu IW, Liao PJ, Ting MK, Chen SW, Yang NI, Hsu KH. **Combination of thigh circumference and indices of central obesity helps predict incident chronic kidney disease: a 14-year prospective cohort study using a three-dimensional body laser scanner.**. (2021) **32** 405-13. DOI: 10.1053/j.jrn.2021.06.009
33. Sun K, Lin D, Li F, Qi Y, Feng W, Ren M. **Visceral adiposity index is associated with increased urinary albumin excretion: a population-based study.**. (2019) **38** 1332-8. DOI: 10.1016/j.clnu.2018.05.025
34. Borrud L, Chiappa MM, Burt VL, Gahche J, Zipf G, Johnson CL. **National health and nutrition examination survey: national youth fitness survey plan, operations, and analysis, 2012.**. (2014) 1-24. PMID: 24709592
35. Levey AS, Stevens LA, Schmid CH, Zhang YP, Castro AF, Feldman HI. **A new equation to estimate glomerular filtration rate.**. (2009) **150** 604-12. PMID: 19414839
36. **Definition and classification of CKD.**. (2013) **3** 19-62. DOI: 10.1038/kisup.2012.64
37. Mousapour P, Barzin M, Valizadeh M, Mahdavi M, Azizi F, Hosseinpanah F. **Predictive performance of lipid accumulation product and visceral adiposity index for renal function decline in non-diabetic adults, an 8.6-year follow-up.**. (2020) **24** 225-34. DOI: 10.1007/s10157-019-01813-7
38. Seong JM, Lee JH, Gi MY, Son YH, Moon AE, Park CE. **Gender difference in the association of chronic kidney disease with visceral adiposity index and lipid accumulation product index in Korean adults: Korean National Health and Nutrition Examination Survey.**. (2021) **53** 1417-25. DOI: 10.1007/s11255-020-02735-0
39. Dai D, Chang Y, Chen Y, Chen S, Yu S, Guo X. **Visceral adiposity index and lipid accumulation product index: two alternate body indices to identify chronic kidney disease among the rural population in Northeast China.**. (2016) **13**. DOI: 10.3390/ijerph13121231
40. Chen IJ, Hsu LT, Lu MC, Chen YJ, Tsou MT, Chen JY. **Gender differences in the association between obesity indices and chronic kidney disease among middle-aged and elderly taiwanese population: a community-based cross-sectional study.**. (2021) **12**. DOI: 10.3389/fendo.2021.737586
41. Huang J, Zhou C, Li Y, Zhu S, Liu A, Shao X. **Visceral adiposity index, hypertriglyceridemic waist phenotype and chronic kidney disease in a southern Chinese population: a cross-sectional study.**. (2015) **47** 1387-96. DOI: 10.1007/s11255-015-1040-y
42. Bamba R, Okamura T, Hashimoto Y, Hamaguchi M, Obora A, Kojima T. **The visceral adiposity index is a predictor of incident chronic kidney disease: a population-based longitudinal study.**. (2020) **45** 407-18. DOI: 10.1159/000506461
43. Wen J, Yuan H. **Independent association between the visceral adiposity index and microalbuminuria in patients with newly diagnosed type 2 diabetes.**. (2020) **36**. DOI: 10.1002/dmrr.3198
44. Li M, Hu L, Hu L, Huang X, Liu X, Zhou W. **Visceral Adiposity index is inversely associated with renal function in normal-weight adults with hypertension: the China H-type hypertension registry study.**. (2021) **151** 1394-400. DOI: 10.1093/jn/nxab022
45. Foster MC, Hwang SJ, Larson MG, Lichtman JH, Parikh NI, Vasan RS. **Overweight, obesity, and the development of stage 3 CKD: the framingham heart study.**. (2008) **52** 39-48. DOI: 10.1053/j.ajkd.2008.03.003
46. Fox CS, Larson MG, Leip EP, Culleton B, Wilson PW, Levy D. **Predictors of new-onset kidney disease in a community-based population.**. (2004) **291** 844-50. DOI: 10.1001/jama.291.7.844
47. Vivante A, Golan E, Tzur D, Leiba A, Tirosh A, Skorecki K. **Body mass index in 1.2 million adolescents and risk for end-stage renal disease.**. (2012) **172** 1644-50. DOI: 10.1001/2013.jamainternmed.85
48. Iseki K, Ikemiya Y, Kinjo K, Inoue T, Iseki C, Takishita S. **Body mass index and the risk of development of end-stage renal disease in a screened cohort.**. (2004) **65** 1870-6. DOI: 10.1111/j.1523-1755.2004.00582.x
49. Berrington de Gonzalez A, Hartge P, Cerhan JR, Flint AJ, Hannan L, MacInnis RJ. **Body-mass index and mortality among 1.46 million white adults.**. (2010) **363** 2211-9. DOI: 10.1056/NEJMoa1000367
50. Visscher TL, Seidell JC, Menotti A, Blackburn H, Nissinen A, Feskens EJ. **Underweight and overweight in relation to mortality among men aged 40-59 and 50-69 years: the Seven Countries Study.**. (2000) **151** 660-6. DOI: 10.1093/oxfordjournals.aje.a010260
51. Garofalo C, Borrelli S, Minutolo R, Chiodini P, De Nicola L, Conte G. **A systematic review and meta-analysis suggests obesity predicts onset of chronic kidney disease in the general population.**. (2017) **91** 1224-35. DOI: 10.1016/j.kint.2016.12.013
52. Rothman KJ. **BMI-related errors in the measurement of obesity.**. (2008) **32** S56-9
53. Oh H, Quan SA, Jeong JY, Jang SN, Lee JE, Kim DH. **Waist circumference, not body mass index, is associated with renal function decline in korean population: hallym aging study.**. (2013) **8**. DOI: 10.1371/journal.pone.0059071
54. Elsayed EF, Sarnak MJ, Tighiouart H, Griffith JL, Kurth T, Salem DN. **Waist-to-hip ratio, body mass index, and subsequent kidney disease and death.**. (2008) **52** 29-38. DOI: 10.1053/j.ajkd.2008.02.363
55. Chen YC, Lai SH, Tsai YW, Chang SS. **Visceral adiposity index as a predictor of chronic kidney disease in a relatively healthy population in Taiwan.**. (2018) **28** 91-100. DOI: 10.1053/j.jrn.2017.07.006
56. Chen S, Liu H, Liu X, Li Y, Li M, Liang Y. **Central obesity, C-reactive protein and chronic kidney disease: a community-based cross-sectional study in southern China.**. (2013) **37** 392-401. DOI: 10.1159/000355718
57. Tchernof A, Després JP. **Pathophysiology of human visceral obesity: an update.**. (2013) **93** 359-404. DOI: 10.1152/physrev.00033.2011
58. França AK, Dos Santos AM, Salgado JV, Hortegal EV, da Silva AA, Salgado Filho N. **Estimated visceral adipose tissue, but not body mass index, is associated with reductions in glomerular filtration rate based on cystatin C in the early stages of chronic kidney disease.**. (2014) **2014**
59. Zhang XL, Topley N, Ito T, Phillips A. **Interleukin-6 regulation of transforming growth factor (TGF)-beta receptor compartmentalization and turnover enhances TGF-beta1 signaling.**. (2005) **280** 12239-45. DOI: 10.1074/jbc.M413284200
60. Glass CK, Olefsky JM. **Inflammation and lipid signaling in the etiology of insulin resistance.**. (2012) **15** 635-45. PMID: 22560216
61. Rüster C, Wolf G. **The role of the renin-angiotensin-aldosterone system in obesity-related renal diseases.**. (2013) **33** 44-53. PMID: 23374893
62. Brands MW, Hall JE, Van Vliet BN, Alonso-Galicia M, Herrera GA, Zappe D. **Obesity and hypertension: roles of hyperinsulinemia, sympathetic nervous system and intrarenal mechanisms.**. (1995) **125** 1725s-31s. PMID: 7782935
63. Kwakernaak AJ, Toering TJ, Navis G. **Body mass index and body fat distribution as renal risk factors: a focus on the role of renal haemodynamics.**. (2013) **28** iv42-9. DOI: 10.1093/ndt/gft331
64. Ciardullo S, Ballabeni C, Trevisan R, Perseghin G. **Metabolic syndrome, and not obesity, is associated with chronic kidney disease.**. (2021) **52** 666-72. PMID: 34569517
65. Jung CH, Lee MJ, Kang YM, Hwang JY, Kim EH, Park JY. **The risk of chronic kidney disease in a metabolically healthy obese population.**. (2015) **88** 843-50. DOI: 10.1038/ki.2015.183
66. Chang Y, Ryu S, Choi Y, Zhang Y, Cho J, Kwon MJ. **Metabolically healthy obesity and development of chronic kidney disease: a cohort study.**. (2016) **164** 305-12. PMID: 26857595
67. Chang AR, Surapaneni A, Kirchner HL, Young A, Kramer HJ, Carey DJ. **Metabolically healthy obesity and risk of kidney function decline.**. (2018) **26** 762-8. PMID: 29498223
68. Yun HR, Kim H, Park JT, Chang TI, Yoo TH, Kang SW. **Obesity, metabolic abnormality, and progression of CKD.**. (2018) **72** 400-10. PMID: 29728317
69. Stefansson VTN, Schei J, Solbu MD, Jenssen TG, Melsom T, Eriksen BO. **Metabolic syndrome but not obesity measures are risk factors for accelerated age-related glomerular filtration rate decline in the general population.**. (2018) **93** 1183-90. PMID: 29395334
70. Ciardullo S, Ballabeni C, Trevisan R, Perseghin G. **Liver fibrosis assessed by transient elastography is independently associated with albuminuria in the general United States population.**. (2021) **53** 866-72. DOI: 10.1016/j.dld.2021.02.010
71. Ciardullo S, Ballabeni C, Trevisan R, Perseghin G. **Liver stiffness, albuminuria and chronic kidney disease in patients with NAFLD: a systematic review and meta-analysis.**. (2022) **12**. DOI: 10.3390/biom12010105
|
---
title: Cobalt containing glass fibres and their synergistic effect on the HIF-1 pathway
for wound healing applications
authors:
- Anu K. Solanki
- Hélène Autefage
- Antonio R. Rodriguez
- Shweta Agarwal
- Joaquin Penide
- Muzamir Mahat
- Thomas Whittaker
- Amy Nommeots-Nomm
- Elena Littmann
- David J. Payne
- Anthony D. Metcalfe
- Felix Quintero
- Juan Pou
- Molly M. Stevens
- Julian R. Jones
journal: Frontiers in Bioengineering and Biotechnology
year: 2023
pmcid: PMC10036384
doi: 10.3389/fbioe.2023.1125060
license: CC BY 4.0
---
# Cobalt containing glass fibres and their synergistic effect on the HIF-1 pathway for wound healing applications
## Abstract
Introduction and Methods: Chronic wounds are a major healthcare problem, but their healing may be improved by developing biomaterials which can stimulate angiogenesis, e.g. by activating the Hypoxia Inducible Factor (HIF) pathway. Here, novel glass fibres were produced by laser spinning. The hypothesis was that silicate glass fibres that deliver cobalt ions will activate the HIF pathway and promote the expression of angiogenic genes. The glass composition was designed to biodegrade and release ions, but not form a hydroxyapatite layer in body fluid.
Results and Discussion: Dissolution studies demonstrated that hydroxyapatite did not form. When keratinocyte cells were exposed to conditioned media from the cobalt-containing glass fibres, significantly higher amounts of HIF-1α and Vascular Endothelial Growth Factor (VEGF) were measured compared to when the cells were exposed to media with equivalent amounts of cobalt chloride. This was attributed to a synergistic effect of the combination of cobalt and other therapeutic ions released from the glass. The effect was also much greater than the sum of HIF-1α and VEGF expression when the cells were cultured with cobalt ions and with dissolution products from the Co-free glass, and was proven to not be due to a rise in pH. The ability of the glass fibres to activate the HIF-1 pathway and promote VEGF expression shows the potential for their use in chronic wound dressings.
## Introduction
Chronic wounds are described as those that do not heal in a timely manner, or that reoccur frequently (Singh et al., 2004; Fonder et al., 2008). When a patient suffers from an underlying pathological condition which impairs healing, such as diabetes, the normal wound healing cascade of hemostasis, inflammation, proliferation, and remodeling can be disrupted. Stalling of healing often occurs in the inflammation phase, resulting in delay of the proliferation phase and wound repair (Boateng et al., 2008; Broderick, 2009). Chronic wounds do not heal through conventional treatment and are often painful, exudating and odorous, which causes a significant reduction in a patient’s quality of life (Franks and Morgan, 2003; Green et al., 2013), and also leads to a huge economic burden to the healthcare system (Sen et al., 2009; Olsson et al., 2019). The number of patients suffering from chronic wounds is expected to grow with an increasingly ageing population and an increased incidence of diabetes (Posnett and Franks, 2008; Frykberg and Banks, 2015).
Healing can be promoted by influencing the hypoxia inducible factor (HIF) pathway, which plays a key role in regulating wound healing and is usually activated under hypoxic conditions (Botusan et al., 2008; Duscher et al., 2015). The HIF-1 pathway is activated through the stabilisation of HIF-1α, which under normal oxygen pressure is constitutively expressed and degraded (Semenza, 2004; Dery et al., 2005). Hypoxia, or hypoxia mimicking agents, stabilise HIF-1α leading to the formation of the HIF-1 complex, which promotes the expression of genes involved in adapting to hypoxia (Gleadle et al., 1995; Semenza, 2004; Esakkimuthukumar et al., 2022). This includes genes that play a role in glucose metabolism and angiogenesis, such as Vascular Endothelial Growth Factor (VEGF). High glucose levels can destabilise HIF-1α in hypoxic conditions, but local stabilisation of HIF-1α has improved wound healing in diabetic mice (Botusan et al., 2008; Zhang et al., 2022a; Zhang et al., 2022b). Wound healing was also impaired in mice with fibroblasts that do not express HIF-1α, which was thought to be due to the decreased expression of VEGF (Duscher et al., 2015). This indicates the HIF pathway could be an important target in developing biomaterials to improve the healing of chronic wounds, particularly in diabetic patients (Zhang et al., 2022a).
Bioactive glasses have recently found clinical application in healing chronic wounds. They were initially investigated for bone repair (Hench et al., 1971; Greenspan, 2019) because they bond to bone via formation of a hydroxycarbonate apatite (HCA) layer and they can stimulate bone growth through the release of calcium ions and silica species (Xynos et al., 2001). Due to their amorphous structure, glasses of low silicate content (low network connectivity) can release other ions that can give therapeutic properties, so their use in soft tissue applications is growing (Miguez-Pacheco et al., 2015; Baino et al., 2016; Kargozar et al., 2019), including wound healing (Naseri et al., 2017).
A material for wound healing must be able to be shaped to fit a wound and fibrous scaffolds have been developed for wound healing applications (Han and Ceilley, 2017). Borate-based glass in the form of cotton-like fibres, named MIRRAGEN (ETS Woundcare, Rolla, MO), recently attained FDA approval for chronic wound treatment. The borate glass fibres improved the healing of full thickness wounds in healthy rats (Zhao et al., 2015) and in human clinical trials (Jung et al., 2012). In vivo, borate glasses have led to an increase in blood vessels in comparison to silicate fibres when applied to a wound (Zhou et al., 2016) or implanted subcutaneously (Lin et al., 2014), but they also dissolve rapidly, forming HCA. Incorporating copper, which is able to promote angiogenesis, into borate glass fibres further increased VEGF expression and cell migration in vitro, and improved wound healing in vivo (Zhao et al., 2015). However, the in vitro viability and migration of cells exposed to glass fibres is highly dependent on the dose of fibres and the culture conditions (Yang et al., 2015). Recently, sol-gel borate fibres were produced and were found to accelerate the migration of keratinocytes in vitro (Naseri et al., 2022).
Silicate glasses may give greater control of dissolution rate and have been investigated for their potential in wound healing (Lin et al., 2012; Yu et al., 2016), but few studies have investigated the effects of silicate glass fibres on wound cells. Here, the aim is to develop fibrous scaffolds of homogeneous thickness that can deliver ions that can stabilise HIF-1α, without promoting mineralisation.
Cobalt is known to stabilise HIF-1α (Yuan et al., 2003) and has been recently incorporated into wound dressings (Shi et al., 2019). Silicate based bioactive glasses containing cobalt (not in the form of fibres) have been produced (Vyas et al., 2015; Kargozar et al., 2018) and been shown to promote VEGF expression by endothelial cells (Quinlan et al., 2015), fibroblasts (Solanki et al., 2021), mesenchymal stem cells (Azevedo et al., 2015), bone marrow stromal cells (Wu et al., 2012; Littmann et al., 2018; Chen et al., 2020), and human umbilical vein endothelial cells (HUVECs) (Kargozar et al., 2017). Cobalt-containing borate glasses also increased HIF-1α, VEGF protein secretion, ALP activity in hBMSC culture (Deng et al., 2019). Cobalt was also incorporated into silicate based sol-gel glass (Barrioni et al., 2018a; Barrioni et al., 2018b; Kermani et al., 2020), provoking VEGF expression from fibroblasts (Dziadek et al., 2018) and HUVECs (de Laia et al., 2021), and into inorganic/organic hybrid scaffolds (de Laia et al., 2020). Interestingly, the original 45S5 Bioglass composition has also been shown increase the expression of VEGF in fibroblasts (Day et al., 2004; Day, 2005; Gorustovich et al., 2010; Gerhardt et al., 2011) and endothelial cells (Leu and Leach, 2008) at specific concentrations. However, the mechanism by which this occurs is not clear. There have been no studies conducted with keratinocytes, which play a crucial role in wound healing.
Fibrous bioactive glass scaffolds can be produced by melt-spinning or electrospinning of the sol-gel process. Electrospun sol-gel glass fibres that only contained silica and calcium oxide in their composition (SiO2-CaO) increased VEGF expression from human dermal fibroblast cell line (CD-18CO) (Norris et al., 2020). Melt-derived bioactive glasses of low silicate content often crystallise during conventional fibre drawing or melt-spinning. However, laser spinning has been shown to be a suitable technique to produce fibres of 45S5 Bioglass with mean fibre diameters of around 200–300 nm (Quintero et al., 2009). Here, the aim was to produce laser spun Co-containing bioactive glasses.
Our use of the term “bioactive glass” does not refer to HCA layer formation but rather to a glass that will stimulate a beneficial biological response through its dissolution ions (Rahaman et al., 2011; Jones, 2013). As HCA layer formation is not required for wound healing applications, and apatite deposits have been shown to inhibit the healing of leg ulcers (Tokoro et al., 2009). We previously developed cobalt doped glass compositions designed not form an HCA layer, even in simulated body fluid (SBF) (Solanki et al., 2021). One of these glass compositions (55 mol% SiO2, 20 mol% Na2O, 10 mol% K2O, 10 mol% MgO, 5 mol% CoO) released cobalt into SBF without forming an HCA layer over 21 days. Glasses containing cobalt did stabilise HIF-1α and provoked a significantly higher expression of VEGF in primary human fibroblasts, which was not seen in Co-free controls. This cobalt containing glass composition was chosen for the fibres produced in this study and is referred to as 5Co, and 0Co is the equivalent composition without cobalt (55 mol% SiO2, 20 mol% Na2O, 10 mol% K2O, 15 mol% MgO).
Herein, the hypothesis was that bioactive glass fibres that can deliver cobalt ions, without forming an HCA layer, activate the HIF-1 pathway and therefore have potential for use in a dressing for chronic wounds. The materials were tested using keratinocytes because epidermal VEGF production is required for permeability barrier homeostasis and is thought to stimulate dermal angiogenesis (Elias et al., 2008; Detmar, 2000). In normal human skin, VEGF is expressed and secreted not only by platelets and fibroblasts, but also by epidermal keratinocytes. HIF plays an important role in cytokine production by keratinocytes and in neutrophil recruitment to the skin (Leire et al., 2013). Additionally, HIF-1α is known to stimulate a broad range of effects, and its high level of expression in the basal layer of keratinocytes in the epidermis likely reflects an important role in local and systemic adaptation and sensing to environmental stresses during healing e.g., epidermal oxygenation (Rezvani et al., 2011).
Glass fibres were compared to glass particles to investigate any changes to the glass after processing into fibres. Glass compositions both with and without cobalt were compared to understand if activation of the HIF pathway was solely due to the cobalt released from the glass, or if the other ions released also played a role.
## Preparation of bioactive glasses
SiO2 (Prince Minerals), CoCO3, Na2CO3, MgO, and K2CO3 (Sigma Aldrich) were mixed and melted in a platinum-gold crucible for 1.5 h at 1,400°C. Nominal glass compositions both with (5Co) and without cobalt (0Co) are reported in Table 1. The molten glass was quenched into deionised water and the frit was dried overnight at 120°C. Glass plates were prepared by remelting the frit for 30–45 min at 1,400°C and pouring into graphite moulds preheated to 500°C. The plates were left to anneal for 15 min at 500°C before allowing to cool to room temperature. Glass particles were prepared by grinding the frit in a planetary ball mill (Fritsch Pulversitte 7) for 6 min at 500 rpm (no annealing step) to produce glass particles with a D(0.9) of between 60 and 80 μm.
**TABLE 1**
| Unnamed: 0 | 5Co-Nom | 5Co-P | 5Co-F | 0Co-Nom | 0Co-P | OCo-F |
| --- | --- | --- | --- | --- | --- | --- |
| SiO 2 | 55.0 | 51.4 ± 0.2 | 56.4 ± 0.3 | 55.0 | 47.0 ± 0.2 | 51.3 ± 0.2 |
| Na 2 O | 20.0 | 21.6 ± 0.1 | 19.1 ± 0.2 | 20.0 | 23.7 ± 0.1 | 21.5 ± 0.1 |
| K 2 O | 10.0 | 11.4 ± 0.0 | 10.1 ± 0.0 | 10.0 | 12.6 ± 0.1 | 11.3 ± 0.0 |
| MgO | 10.0 | 10.6 ± 0.3 | 9.8 ± 0.2 | 15.0 | 16.7 ± 0.2 | 15.9 ± 0.2 |
| CoO | 5.0 | 5.0 ± 0.1 | 4.6 ± 0.1 | 0.0 | 0.0 ± 0.2 | 0.0 ± 0.0 |
## Making glass fibres
Figure 1A shows a schematic of glass fibre (Figure 1C) production from the glass plates (Figure 1B) by laser spinning, as described previously (Quintero et al., 2009; Quintero et al., 2014). A high-power laser beam is used to melt a small volume of a precursor glass monolith. At the same time, a supersonic gas jet is applied to elongate the molten material by the action of the viscous friction at the gas/melt interface. This jet also produces the fast cooling and solidification of the fibres, hence producing amorphous nanofibers during the process. A CO2 laser (Rofin Sinar DC035) was used, emitting with wavelength of 10.6 μm and radiant power of 2.5 kW. The laser beam was focused 13 mm above the surface of the glass plate using a lens with focal length of 190.5 mm. Compressed air was supplied to the co-axial nozzle and the off-axis supersonic nozzle at a pressure of approximately 5 and 10 bar, respectively. The plate was moved with velocity relative to the laser beam of 300 mm min−1 to produce the glass fibres.
**FIGURE 1:** *(A) Schematic of the laser spinning process using a glass precursor plate (B) photograph of an example of a bioactive glass precursor plate (scale bar is 1 cm) (C) mesh of fibres produced (D) XPS core line spectra of cobalt (Co2p) showing no difference in oxidation state of Co between the cobalt glass fibres and particles, and the core line spectra of silica (Si2p) (E) showing no difference in oxidation state of Si between the 5Co or 0Co glass fibres (5Co-F and 0Co-F) or particles (5Co-P and 0Co-P).*
## Measuring glass compositions
The composition of the glass fibres and particles was measured by Inductively Coupled Plasma Optical Emission Spectroscopy (ICP-OES) (Thermo Scientific, iCAP6000 Series ICP). 100 mg of glass fibres or particles were mixed with 400 mg lithium metaborate (Alfa Aesar) and heated to 950°C for 30 min in a platinum crucible to form a fused pellet. Once cooled, the pellet was dissolved in 50 ml of 2 M HNO3. This was diluted 1:10 and ICP-OES used to measure the elemental concentration of the solution, and the composition calculated.
## Ion release into DMEM
Glass fibres and particles of both compositions were incubated in Dulbecco’s Modified Eagle Medium (DMEM, LifeTechnologies) at a ratio of 1.5 mg ml−1 to evaluate the ion release (Macon et al., 2015). 10 ml DMEM was added to 15 mg of fibres or particles and incubated at 37°C in an orbital shaker under agitation at 120 rpm. At each time point (30 min, 1, 2, 4, 8, 24 h, 3, and 7 days) 0.5 ml of DMEM was collected and diluted 1:20 with deionised water to quantify the ionic concentration by ICP-OES, and replaced with 0.5 ml of fresh DMEM. At 7 days, samples were collected using filter paper, rinsed with deionised water and acetone, and dried overnight.
Fourier Transform Infrared Spectroscopy (FTIR), X-ray diffraction (XRD) and Scanning Electron Microscopy (SEM) were conducted on fibres and particles, before and after incubation in DMEM. FTIR spectra were collected with a Nicolet iS10 FTIR in Attenuated Total Reflectance mode from 400–2000 cm−1 with a resolution of 0.4 cm−1. XRD was carried out on a PANalytical XRD measuring between 5 and 80° 2θ using a step size of 0.0334225. SEM images were taken after coating samples in chromium, using a LEO Gemini 1525 FEGSEM and an accelerating voltage of 5 kV.
## XPS
Surface compositions and chemical states were obtained after drying the samples at 120°C for 1 h and conducting X-ray photoelectron spectroscopy (XPS) using a Thermo K Alpha + spectrometer (Thermo Scientific) operating at around 10−8 to 10−9 Torr with a Al Kα radiation source (1,486.7 eV). High resolution spectra were collected for cobalt (Co 2p) and silica (Si 2p) to assess the oxidation states with the peaks normalised to the carbon peak (285.0 eV).
## FIB/TEM
Samples were prepared using a focussed ion beam (FIB) operated at 30 kV (FIB, FEI Helios NanoLab 600) for analysis by Transmission Electron Microscopy (TEM) and Energy Dispersive Spectroscopy (EDS). Briefly, a 15 μm × 2 μm site was coated with 1.5 μm of platinum. Two trenches were made on either side of the platinum layer and the base of the region was cut, using currents of 2.8 and 6.4 nA respectively. The sample was attached to the omniprobe manipulator, lifted out, and attached to a TEM lift-out three post copper grid (Agar Scientific). Samples were thinned to a width of around 100 nm using currents between 0.46 and 2.8 nA. Finally, the sample was polished with a gallium ion beam operated at 2 kV to remove possible artefacts introduced by milling. The FIB samples were imaged in the TEM (JEOL JEM 2100F) operating at 200 kV using dark field Scanning Transmission Electron Microscopy (STEM) mode. Elemental maps were obtained by EDS (Oxford Instruments INCA EDS 80 mm X-Max detector system with light-element (Z > 5) analysis and STEM capability).
## Cell culture
Immortalised human epidermal keratinocyte (HaCaT; RRID:CVCL_0038) cells were cultured in DMEM, supplemented with $10\%$ Fetal Bovine Serum (FBS), and maintained in an incubator kept at 37°C and $5\%$ CO2, which was also used as control medium. The cell line present in this study were obtained from Dr Vania Braga, Imperial College London, who has curated the cells since obtaining the cell line from the group of NE Fusenig, German Cancer Research Center, Heidelberg, Germany, who first isolated and characterised the cells, demonstrating their full epidermal differentiation capacity (Boukamp et al., 1988).
Bioactive glass conditioned medium was prepared immediately before use. Glass fibres were sterilised by dry heat at 120°C for 2–3 h and incubated with DMEM (without serum) at a ratio of 6.67 mg ml−1. The medium was conditioned with glass fibres for 24 h, after which it was removed, sterile filtered through a 0.2 µm syringe filter and transferred to T25 flasks with vented caps. The conditioned medium was kept an incubator at 37°C with $5\%$ CO2 overnight to allow equilibration of the pH prior to use. Conditioned medium from the glass particles was made in the same way, but the particles were not sterilised at 120°C. For CoCl2 controls, a 10 mM stock solution of CoCl2 was made up in distilled water and added to DMEM to give a final concentration of 300 µM before equilibrating the pH. When producing conditioned medium from either the glass fibres or glass particles, five groups were investigated: 1—DMEM control; 2—DMEM +300 μM CoCl2; 3—5Co conditioned DMEM; 4—0Co conditioned DMEM; and 5—0Co conditioned media +300 µM CoCl2. When investigating the high pH media, NaOH was added dropwise to DMEM to give a pH of 8.5 and 9.5, after which the media was sterile filtered and the pH allowed to equilibrate overnight before use in T25 flasks with vented caps.
## In-cell western for HIF-1α
To determine the amount of HIF-1α after exposure to conditioned media, 4 × 104 cells were seeded per well of a 48 well plate and left for 2–3 days to reach confluence. The medium was removed, and wells washed once with PBS before adding 400 µL conditioned medium to each well. After 4 h, the conditioned medium was removed, wells washed with PBS and cells fixed in a $3.7\%$ formaldehyde solution. An In-Cell Western blotting (ICW) was conducted using a protocol modified from Cell Signalling Technology. Briefly, wells were permeabilised with $0.25\%$ Triton 100X for 5 min, washed, and blocked with blocking buffer ($3\%$ w/v Bovine Serum Albumin (BSA) in PBS/$0.1\%$ v/v Tween 20) for 1 h. The mouse monoclonal primary antibody for HIF-1α (Abcam, ab16066) was diluted 1:400 in blocking buffer and incubated for 1.5 h at room temperature. A donkey anti-mouse IR secondary antibody (Licor, 925–32212) and DRAQ5 (New England Biolabs, 4084S) were diluted 1:2000 in blocking buffer and incubated for 1 h, before the plate was read using the Licor Odyssey.
## ELISA for VEGF
To measure the amount of VEGF secreted by cells exposed to conditioned media, 1 × 104 cells were seeded per well of a 96 well plate and left for 1 day to attach. Conditioned media was supplemented with $10\%$ FBS prior to equilibrating the pH, and 100 µL was added to each well. This was changed for serum free conditioned media 24 h before collecting the supernatant and cell lysate at 1, 3 and 7 days. Supernatants were kept at −80°C before an ELISA was carried out using a human VEGF antibody pair kit (LifeTechnologies, CHG0113) according to the recommended protocol, and a plate reader (Perkin Elmer Enspire) used to measure the absorbance at a wavelength of 450 nm with a reference wavelength of 650 nm. Cell lysates were prepared by collecting the cell monolayer in 200 µL 1 v/v% triton 100X in PBS, and freeze thawing 3 times. The DNA was quantified from the cell lysates using a florescent dye which binds DNA (bisbenzimidazole, Hoechst 33258), with a standard curve prepared using a DNA standard (LifeTechnologies). The fluorescence was measured at 360 nm excitation and 460 nm emission using a plate reader (Perkin Elmer Enspire).
## Nanoparticle measurement in the conditioned media
While the glass particle and glass fibre conditioned medium was expected to contain ionic dissolution products of the glasses, nanoparticles may also have been lost from the glasses and therefore given to the cells. The concentration and size distribution of any nanoparticles in the conditioned media was therefore measured using a Nanosight NS300 (Malvern). Conditioned media was diluted in fresh, confirmed particle-free 18.2 MΩ water until the concentration of particles was within the linear range of the instrument (1–10 × 108 particles ml−1). Five 60-s measurements were taken per sample at Camera Level 15 and Detection Threshold 5. The FTLA (Finite Track Length Adjusted) size distribution smoothing algorithm was disabled to avoid generation of artefact peaks.
## Statistics
All results are expressed at the mean ± standard deviation. Statistical significance between the different groups was determined using a one or two way ANOVA and Tukey’s test.
## Producing glass fibres
Glass fibres were successfully formed by laser spinning, which produced an intertwined mesh of fibres (Figure 1C), with diameters ranging from ∼0.5 μm—10 μm. As shown in Table 1, the compositions of the fibres (5Co-F and 0Co-F) as measured by ICP-OES were close to the measured compositions of the glass particles (5Co-P and 0Co-P) and to the nominal compositions. This is in agreement with previous work which showed that the laser spinning process does not significantly change the composition of bioactive glasses (Quintero et al., 2009).
## XPS analysis of fibres
XPS was conducted to determine if the oxidation states of cobalt or silica varied between the glass fibres and glass particles. The core line spectra of cobalt (Co2p) is shown in Figure 1D, and is comprised of a spin split orbital associated with a shake-up satellite. The spin split orbital is between 781.13 and 781.28 eV for Co2p $\frac{3}{2}$, and 796.83–796.98 eV for Co2p $\frac{1}{2}$ with splitting energy at 15.7 eV. The shake up satellites are between 786.33—786.78 eV and 802.73—802.98 eV (Ho et al., 1990; Tan et al., 1991; Khodakov et al., 2002; Zhang et al., 2009). The core line spectra of Si2p are shown in Figure 1E and the peak at 102.6 ± 0.3 eV is attributed to Si4+ (Barr, 1983). There was little difference between the fibres and the particles, suggesting that the oxidation state of both cobalt and silica are unchanged by the laser spinning process.
## Ion release into DMEM medium
The ion release from the glass fibres and particles of 5Co and 0Co into DMEM over 7 days is shown in Figure 2. The 5Co-P gave a burst release of cobalt in the first 30 min, and by 1 h the cobalt concentration reached a plateau between 5.7 and 8.5 μg ml−1. A burst release of cobalt was not seen from 5Co-F, instead there was a continuous increase for up to 7 days without a plateau being reached. At 24 h, the Co release from the fibres and particles was similar, but by 7 days the Co concentration reached approximately 12 µg ml−1 for 5Co-F, compared to around 7.5 µg ml−1 for 5Co-P. A burst release of Co was observed in the particles because fluid was accessible all around the particles, allowing rapid ion exchange (Co ions for H+ in the DMEM) and formation of the silica rich layer. In the fibres, ion exchange was initially slower than the particles, but release was more sustained as the distance for Co ion diffusion was less than for the particles, due to the fibre thickness being at least an order of magnitude smaller than the particle diameters.
**FIGURE 2:** *Elemental concentration in DMEM after incubation with 5Co and 0Co glass fibres (5Co-F and 0Co-F) and particles (5Co-P and 0Co-P) at a concentration of 1.5 mg ml−1 as a function of time of immersion. Data is represented as the mean ± standard deviation with N = 3.*
At 30 min, the concentration of the silica released from the fibres was lower than that of the glass particles, with around 5 μg ml−1 measured for the fibres, and around 13 μg ml−1 for the particles, but by 8 h the silica release from the fibres and particles was similar. For up to 7 days, there was a steady increase in the amount of silica released without a plateau being reached, and the final concentration was between 52 and 80 μg ml−1.
For both the particles and fibres in both compositions, the concentration of calcium and phosphate decreased in a similar manner. At 24 h, the concentration remained steady at around 65–70 μg ml−1 for calcium and approximately 27 μg ml−1 for phosphorous. After this time, a decrease in both calcium and phosphorus was measured for all samples, suggesting that there was some precipitation of calcium phosphate on the glass surface.
## Characterisation of the glasses after incubation in DMEM media
XRD and FTIR of the glass fibres and particles (Figure 3; Supplementary Figure S1) showed that glass with (Figure 3) and without cobalt (Supplementary Figure S1) behaved in a similar manner. Prior to incubation in media, FTIR spectra obtained from the particles and fibres looked similar with bands at approximately 1,000, 920 and 450 cm−1 corresponding to the Si-O(s), the Si-O(s) associated with a modifying cation, and the Si-O(b) respectively (Sanders et al., 1974; Serra et al., 2003; Liu et al., 2013).
**FIGURE 3:** *FTIR (A) and XRD (B) of the glasses containing cobalt before and after incubation in DMEM for 7 days: 5Co fibres (5Co-F) and particles (5Co-P); SEM images of 5Co-F before (C) and after (D) incubation in DMEM for 7 days, and 5Co-P before (E) and after (F) incubation in DMEM at a concentration of 1.5 mg ml−1. Scale bar is 10 μm, and * indicates cracks forming in the fibres and particles after incubation in DMEM.*
After incubation in medium, no differences were seen between the glass fibres and particles. The FTIR spectra showed a decrease in the Si-O(s) associated with a modifying cation, as the modifiers left the glass and accumulated in the medium (Filgueiras et al., 1993), as measured by ICP-OES (Figure 2). The band corresponding to the Si-O(s) at around 1,000 cm−1 became more prominent, suggesting the formation of a silica rich layer (Serra et al., 2003). All samples were amorphous both before and after incubation in DMEM for 7 days, but in the cobalt containing samples the XRD patterns showed a second amorphous halo at higher values of °2θ. There was no evidence of HCA formation.
SEM images in Figure 3D and SI Supplementary Figure S1D show how the glass fibres and particles degraded. Prior to incubation (Figure 3C and SI Supplementary Figure S1C), the fibres of both 0Co and 5Co looked similar in morphology. The fibres had a wide range of diameters, and some particulates were seen adhering to the glass fibres which were due to glass dust forming during the laser spinning process. After incubation in DMEM for 7 days, no difference was seen between the 5Co (Figure 3D) and 0Co (Supplementary Figure S1D) compositions. For both 5Co-F and 0Co-F, a thin layer of precipitate was present and many of the fibres had cracked, as indicated in Figure 3D. Similar cracks have been reported to be due to the formation of a silica rich layer (Dietrich et al., 2009). The precipitated layer was seen to have peeled away from the fibre, revealing an underlying smoother surface, and often showed that the fibre was cracked in multiple places through the cross section. A precipitated layer was also seen on the glass particles after incubation in media. Most of the particles had cracked, and although the precipitated layer was observed to come away from the particles, this occurred less frequently compared to the fibres.
## Elemental analysis of glasses by TEM
TEM-EDS is an important tool for understanding nanoscale changes to glass surface during degradation. SEM images showed little differences between 5Co or 0Co glasses and ion release was similar, except for the release of cobalt. Therefore, the 5Co composition was chosen to assess the changes to the glass fibres and particles at the nanoscale using TEM-EDS (Figure 4). Prior to incubation in DMEM, the fibres and the particles did not show any reaction layers and were free of precipitate (Supplementary Figure S2). After 7 days in DMEM (Figure 4A), a layer of precipitate was seen around the fibres, as seen under the SEM, and it bridged a crack in the glass fibre, suggesting that, for some fibres at least, the precipitated layer formed before the fibre cracks. EDS of the precipitated layer showed that it contained calcium, phosphate, magnesium, cobalt and sodium (higher resolution image shown in Supplementary Figure S3). Elemental mapping of the cross-section of the fibres showed homogeneous distribution of Si, Co and Mg throughout, with no Ca, and reduced concentration of Na and K towards the glass surface, due to dissolution (ion exchange).
**FIGURE 4:** *Darkfield TEM images and EDX spectra of a 5Co fibre cross section (A) and 5Co particle (B) showing several reaction layers with differing elemental concentrations after incubation in DMEM for 7 days. Scale bar for (A) is 1 µm for (B) is 250 nm.*
Figure 4B shows several reaction layers close to the top surface of a glass particle, after incubation in DMEM for 7 days, with the top layer containing similar elements as the precipitated layer on the fibres. Subsequent layers had a similar composition to each other and to the centre of the glass fibres. Although the relative composition of the glass particle was consistent throughout the depth, except for the top layer, there was a clear difference in the contrast of the particle between layers ★3 and ★4, with layer ★3 appearing to be much darker, which is indicative of a lower density. This observation was not made for the glass fibres. K and Na were detected on the surface of both the glass fibres and glass particles, indicating that there may also be small amounts of residual salts from the DMEM media, such as NaCl and KCl present on the glass surface.
## HIF-1α stabilisation and VEGF expression of fibres
To determine whether the glass conditioned media could activate the HIF pathway in cells that are known to play a key role in wound healing, keratinocytes were used to measure the stabilisation of HIF-1α and expression of VEGF protein (Figure 5). Media contained either cobalt from cobalt chloride or as dissolution products of the glasses. Significantly higher HIF-1α was measured after keratinocytes were exposed to cobalt containing media when compared to the DMEM only control, regardless of whether the cobalt was added as CoCl2 or released from 5Co-F. Surprisingly, the amount of HIF-1α measured after exposure to 5Co-F conditioned medium was significantly higher than that produced by cells exposed to media with 300 μM CoCl2, even though the amount of cobalt present in the media was similar (17.0 ± 0.6 μg ml−1 for 300 µM CoCl2 and 14.7 ± 1.2 μg ml−1 for 5Co-F). To determine whether the cobalt ions released from the glass fibres had the same effect as adding CoCl2 to the dissolution products of a cobalt free glass, the effect of 0Co-F conditioned media supplemented with 300 µM CoCl2 was also investigated. Importantly, HIF-1α measured in cells incubated with 0Co-F + CoCl2 did not match that of cells incubated with 5Co-F, even though their Co concentration in solution (measured by ICP) was similar.
**FIGURE 5:** *Results of in vitro studies of keratinocytes exposed to conditioned media from the glass fibres: Increase in HIF-1α measured after 4 h (A); VEGF expression (B); and total DNA (C). Data is represented as mean ± standard deviation with N = 4. For Fig B and C the 0Co-F + CoCl2 condition is represented as N = 3 for Day 1 and 3, and n = 2 for Day 7. Statistical significance was determined with a (A) one way ANOVA or (B) and (C) two way ANOVA, with a Tukey means comparison test with *p < 0.05 compared to DMEM, #p < 0.05 compared to 300 μM CoCl2, Δ p < 0.05 compared to 0Co-F and ⋄p < 0.05 compared to 0Co-F + CoCl2.*
The expression of VEGF showed a similar trend to the HIF-1α measurements. For all conditions, an increase in VEGF expression was seen between Day 1 and 7. The VEGF expression was significantly higher for CoCl2 and all glass conditioned media compared to the DMEM control. Incubating cells in 5Co-F conditioned medium led to a significantly higher VEGF release compared to 300 μM CoCl2; and incubating cells in 0Co-F + CoCl2 did not replicate the effect of 5Co-F: VEGF protein expression in the 0Co-F + CoCl2 group was found to be significantly lower than that of 5Co-F treatment but significantly higher than the 300 µM CoCl2 condition. The only difference in ion concentration between the two groups was elevation Mg and Si concentration in 0Co-F + CoCl2 compared to 5Co-F conditioned media (Figure 2). The DNA measured for all conditions was similar, with no significant differences observed with time or between conditions, indicating that the differences in protein expression were not due to changes in cell number.
## Investigating the effect of conditioned media from 5Co-F
Investigating the HIF-1α stabilisation and VEGF expression showed that conditioned medium from 5Co-F produced an effect that could not be replicated by an equivalent amount of CoCl2 or by supplementing 0Co-F conditioned medium with CoCl2. As glass dissolution can cause pH increase, through ion exchange (e.g. Na+ of the glass for H+ from the medium), a pH change could also have an effect on the in vitro cellular response in addition to the ionic concentration of conditioned medium. The pH of DMEM increased after incubation with bioactive glasses (Figure 6) to a maximum value of 9.5 and was allowed to equilibrate overnight, but a slight increase in pH was still seen in the bioactive glass conditioned medium. Therefore, the effect of an elevated pH on HIF-1α was investigated. The pH of the medium was increased to pH 8.5 and 9.5, using NaOH, to mimic the burst rise seen when incubating bioactive glasses in cell culture medium, and the effect of an increased pH in combination with CoCl2 was also investigated as shown in Supplementary Figure S4. After equilibration of the high pH media in a $5\%$ CO2 incubator overnight, the pH decreased to levels similar to the DMEM and CoCl2 controls. Although the measured pH was always slightly higher than the controls, the difference was not statistically significant. After allowing the pH to equilibrate overnight, keratinocytes were incubated in the medium and the HIF-1α measured, as shown in Figure 6B. Only the presence of CoCl2 led to an increase in HIF-1α and no difference was seen between the addition of CoCl2 and addition of CoCl2 in combination with pH adjusted to a higher value prior to overnight equilibration.
**FIGURE 6:** *Effect of pH on HIF-1α expression: pH of DMEM media before and after equilibration overnight in a 5% CO2 incubator (A) and HIF-1α measured when exposing keratinocytes to media of elevated pH of 8.5 (with and without cobalt chloride) and 9.5 after equilibration overnight (B). Data is represented as mean ± standard deviation with N = 3. Statistical significance was determined with a (A) two way ANOVA or (B) one way ANOVA, with a Tukey means comparison test with *p < 0.05 compared to DMEM, #p < 0.05 compared to 300µM CoCl2, □ p < 0.05 compared to 8.5, ○ p < 0.05 compared to 8.5 + CoCl2 and ✕p < 0.05 compared to 9.5.*
HIF-1α was also measured in cells incubated with conditioned media from the glass particles (5Co-P and 0Co-P, Figure 7A) to understand whether the effects of 5Co-F were due to the glass composition, or differences between the fibres and particles. As seen with the glass fibre conditioned media, an increased amount of HIF-1α was measured with 300 μM CoCl2, 5Co-P, and 0Co-P + CoCl2 when compared to the DMEM control, with 5Co-P being significantly higher than 300 µM CoCl2. However, unlike with the glass fibre conditioned media, the amount of HIF-1α measured for 0Co-P + CoCl2 was the same as 5Co-P. A smaller significant increase in HIF-1α was also measured for the 0Co-P conditioned media compared to DMEM, which was not seen in the conditioned medium prepared with glass fibres.
**FIGURE 7:** *Increase in HIF-1α measured when keratinocytes are exposed to conditioned media from glass powder for 4 h. Data are represented as mean ± standard deviation with N = 3 (A). Co and Si concentrations in DMEM for the glass fibre and glass particle conditioned media prepared at a ratio of 6.67 mg ml−1; data are represented as mean ± standard deviation with N = 7 for the Fibre conditions and N = 3 for the Particle conditions (B) and size distribution of nanoparticles measured, by Nanoparticle Tracking Analysis, in 5Co and 0Co glass particle conditioned media (5Co-P and 0Co-P) (C). Statistical significance was determined with a one way ANOVA, with a Tukey means comparison test with *p < 0.05 compared to DMEM, #p < 0.05 compared to 300 μM CoCl2, Δ p < 0.05 compared to 0Co-P in (A) and *p < 0.05 compared to DMEM in both the fibre and particle experiment, #p < 0.05 compared to 300 µM CoCl2 controls in both the fibre and particles experiment, ○ p < 0.05 compared to the glass particle conditioned media for the same condition in (B).*
The elemental concentration of the conditioned media from both the fibres and particles was characterised by ICP-OES and the Si and Co. concentrations are shown in Figure 7B. The 5Co-P medium contained a slightly higher concentration of cobalt (20.2 ± 0.6 μg ml−1) compared to in the 5Co-F conditioned medium (14.7 ± 1.2 μg ml−1). The glass particle conditioned media also contained higher concentrations of magnesium, and calcium and phosphorous, as seen in Supplementary Figure S4. This suggests there may have been less precipitation of a cobalt and magnesium substituted calcium phosphate layer on the glass particles after 24 h, compared to the glass fibres. The most striking difference observed was the difference in the amount of Si, with a concentration of around 150 μg ml−1 Si measured for the glass particle conditioned media compared to approximately 80 μg ml−1 in the glass fibre conditioned media. A concentration of 150 μg ml−1 silica in solution is above the reported solubility limit in water (116 μg ml−1 at 25°C (Gunnarsson and Arnorsson, 2000)). When the solubility limit is exceeded, silica can polymerise and form nanoparticles, although this is highly dependent on the pH, temperature and ionic strength of the aqueous solution (Tobler et al., 2009). The presence of nanoparticles was investigated, using Nanoparticle Tracking Analysis, to try and explain the high concentration of silica measured by ICP-OES. Nanoparticles were detected in the glass particle conditioned media, for both 5Co and 0Co, and the size distribution of these particles is shown in Figure 7C. Nanoparticles in the 5Co-P conditioned media had a modal size of 117 ± 6 nm and the concentration of particles was 8.5 ± 0.9 × 109 ml−1, whereas for 0Co-P conditioned media the modal size was 102 ± 4 nm and concentration was 9.2 ± 1.0 × 109 ml−1. In contrast, a reliable measurement of nanoparticles could not be obtained from the glass fibre conditioned media, for either 5Co or 0Co, or the DMEM control, as the concentration was below the detection limit of 5 × 107 particles ml−1.
## Discussion
The aim of this study was to investigate the use of glass fibres as a device to activate the HIF pathway and therefore promote the healing of chronic wounds. The aim would be to use the fibres to fill a wound, which would then be covered with a conventional dressing. The fibres would be a temporary scaffold for wound regeneration biodegrading and releasing their active ions. The materials properties of both fibres and particles, with and without cobalt, were compared and the in vitro response of keratinocytes to conditioned media from the glasses was characterised.
The fibre spinning process retained the amorphous structure of the glasses (Figure 3). The ion release characteristics during incubation in DMEM for the glass fibres and particles, at a concentration of 1.5 mg ml−1, either with or without cobalt, were similar (Figure 2), indicating that the laser spinning process did not significantly change the glass. After incubation in DMEM, a calcium phosphate layer containing cobalt and magnesium formed on the surface of the fibres and particles (Figure 4; Supplementary Figure S3) but it was not crystalline (Figure 3). Cobalt has previously been shown to substitute into calcium phosphate on bioactive glass (Hoppe et al., 2014), and both cobalt and magnesium have previously been seen to substitute into the calcium phosphate layer when incubating 5Co particles in SBF for 21 days (Solanki et al., 2021). When extracts were prepared for cell culture, ion release was higher for particles compared to the fibres (Figure 7). The conditioned media were prepared with a higher concentration of glass (6.67 mg ml−1) in the media, compared to the degradation study (Figure 2). The increased Si content in the media conditioned with the glass particles (150 mg ml−1), compared to that prepared with the fibres (80 mg ml−1) and the subsequent discovery of nanoparticles (mean of 98–123 nm) was unexpected and highlights the need for proper analysis of conditioned media before carrying out cell culture. The glass fibre conditioned media did not contain any nanoparticles in a measurable concentration.
The in vitro cell response to bioactive glasses that contain Co has previously been characterised in mesenchymal stem cells (Azevedo et al., 2015), bone marrow stromal cells (Wu et al., 2012), osteoblasts, fibroblasts (Solanki et al., 2021) and endothelial cells (Quinlan et al., 2015; Solanki et al., 2021). Cobalt is known to stabilise HIF-1α in environments with normal oxygen levels (Yuan et al., 2003), and cobalt doped bioactive glasses have been shown to cause an increase in proteins such as VEGF (Wu et al., 2012; Azevedo et al., 2015; Quinlan et al., 2015) which can promote angiogenesis. Here, the in vitro cell response of keratinocytes to conditioned media from cobalt doped bioactive glasses was investigated for the first time, as these cells are crucial in the wound healing process. In the current study, incubating keratinocytes in 5Co-F conditioned media led to a significantly higher measurement of HIF-1α compared to the equivalent amount of CoCl2 (Figure 6). When adding CoCl2 to 0Co-F medium, the response did not match that of 5Co-F, suggesting there is a synergistic effect of the glass dissolution products which is more than a simple combination of ions released from the glass and cobalt from CoCl2. The trends seen in the measurement of HIF-1α were reflected in the VEGF expression. 5Co-F conditioned media caused a significant increase in VEGF expression when compared all other conditions, and this effect could not be replicated by the addition of CoCl2 to 0Co-F conditioned media. Interestingly, 0Co-F caused an increase in VEGF expression, even though cobalt was not present, and an increase in HIF-1α was not measured with this condition. The ability of bioactive glasses such as 45S5, which do not contain cobalt, to cause an increase in VEGF production has previously been reported in fibroblasts (Day et al., 2004; Day, 2005; Gerhardt et al., 2011) and endothelial cells (Leu and Leach, 2008), but the mechanism by which it occurs is not clear (Ranmuthu et al., 2020). The effect is attributed to silicon and calcium ion release, for example conditioned media from bioactive glass nanoparticles of SiO2-CaO-P2O5 compositions stimulated up-regulated expression of the VEGF, basic fibroblast growth factor, their receptors, and endothelial nitric oxide synthase from HUVECs, resulting in enhanced tube formation in vitro (Mao et al., 2015). Shi et al. found that bioactive dissolution ions significantly promoted the VEGF production from cardiomyocytes, which up-regulated the HIF-1α pathway, which then mediated the behaviour of endothelial cells (Shi et al., 2021). This previous work therefore supports the evidence of synergy between Si, Ca and Co ions on VEGF production.
For the first time, a significant difference has been shown in the in vitro cellular response to conditioned media from a glass releasing pro-angiogenic ions compared to an equivalent amount of the ions in media, or conditioned media from an un-doped glass supplemented with the therapeutic ions. The effect of 5Co-F on HIF-1α and VEGF expression could not be replicated by the equivalent amount of CoCl2 in DMEM or by the addition of CoCl2 to 0Co-F conditioned media. To gain a further understanding of this effect, the pH was investigated. When bioactive glasses are incubated in aqueous buffer, the pH rises as modifying cations are exchanged with H+ from the solution (Hench, 1991). Although the pH of the glass conditioned media was left to equilibrate overnight immediately prior to incubation with the cells, the pH was still slightly more basic for all glass conditioned media compared to the DMEM and CoCl2 controls (Supplementary Figure S4). Measurement of HIF-1α after exposing keratinocytes to media adjusted to high pH and left to equilibrate overnight, with or without CoCl2, showed no effect of incubating keratinocytes in CoCl2 at a slightly basic pH, showing that the higher HIF-1α and VEGF measured with 5Co-F was not due to the combination of cobalt and an increased pH.
Finally, HIF-1α was measured after incubating keratinocytes with conditioned media from the glass particles, to understand if the synergistic effect observed was due to the glass composition or processing into glass fibres. Figure 7A shows a significantly higher amount of HIF-1α was measured when incubating cells 5Co-P conditioned media compared to DMEM with only CoCl2, indicating that the synergistic effect seen is due to the composition of the glass. Surprisingly, for the glass particle conditioned media, a significantly higher amount of HIF-1α is also measured for 0Co-P + CoCl2 compared to the CoCl2 control, and for 0Co-P compared to DMEM. When comparing the ionic concentration of the conditioned media, it was found that the cobalt concentration was similar between the glass fibre and glass particle conditioned media, suggesting that the differences seen between the fibre and particle conditioned media were not due to differences in cobalt concentration. The presence of nanoparticles in the particle conditioned media may have played a role, especially as cells can internalise silicate nanoparticles (Naruphontjiralkul et al., 2018; Naruphontjirakul et al., 2019).
The effect of the silicate nanoparticles on the HIF response of keratinocytes must be fully characterised, but here they appeared to increase HIF-1α measured for both the 0Co-P and 0Co-P + CoCl2 conditioned media compared to the DMEM and CoCl2 controls, but little additional effect of these nanoparticles was seen when the glass contained Co (5Co-P). These results are in contrast to those obtained when incubating the glass fibres and particles in DMEM at a ratio of 1.5 mg ml−1, which showed no difference between the ionic concentrations of DMEM incubated with the glass fibres or particles. The presence of such nanoparticles has not been previously reported.
The most important result is that Co species released from the bioactive glass produces a synergistic angiogenic effect with the other glass dissolution products that is different to the Co made available when CoCl2 is used as the Co source in the media. The laser melting method is a promising method for producing amorphous bioactive glasses with highly degradable (low silica) compositions.
## Conclusion
Glass compositions with and without cobalt were made into fibres, with diameters from tens of nanometers to tens of microns, by laser spinning. Although the glass fibres generally behaved in a similar manner to glass particles, when incubated in DMEM for 7 days at a ratio of 1.5 mg ml−1, there was a clear difference in the degradation properties of the glass fibres and particles when incubated in a ratio of 6.67 mg ml−1. Keratinocyte cell cultures exposed to conditioned media from the fibres showed that the glasses activated the HIF pathway and promoted the expression of VEGF. Cobalt containing glass fibres caused significantly increased HIF-1α measured when compared not only to the DMEM control, but also when compared to DMEM containing the equivalent amount of cobalt chloride. The most important result was that the increase in HIF-1α was not replicated by adding cobalt chloride to conditioned media from the glass that did not originally contain cobalt. This trend was also seen in the VEGF expression, although the glass without cobalt also promoted VEGF expression compared to the DMEM control. This suggested that the Co species released from the glass produces a synergistic effect with the other glass dissolution products that is different to the Co made available when CoCl2 is used as the Co source in the media. The morphology of the fibres is expected to produce improved results in vivo as they mimic the morphology of extracellular matrix fibrils. Further characterisation of the glass particle conditioned media revealed levels of silica that are expected to be above the supersaturation point. Nanoparticles were found to form in the conditioned media in the glass particle conditioned media, but not in the fibre conditioned media, and these appeared to affect the in vitro cell response. Taken together, this suggests that the glass fibres developed in this study have the potential to reduce the healing time of chronic wounds, but the mechanism by which they activate the HIF pathway requires further investigation.
## Data availability statement
The raw data supporting the conclusion of this article will be made available by the authors, without undue reservation. Raw data can be obtained on request from rdm-enquiries@imperial.ac.uk.
## Author contributions
AS, HA, FQ, JP, MS, and JJ developed experimental protocols and designed the experiments. AR, JP, MM, and AS produced all the samples used in the experiments. AS, HA, AR, SA, JP, MM, TW, AN-N, and EL conducted experiments, analysed and interpreted the data. AM, JP, MS, and JJ obtained the funding for the work and provided supervision. DP and HA provided supervision and analytical input. AS and JJ wrote the manuscript, all authors revised and approved the manuscript.
## Conflict of interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher’s note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## Supplementary material
The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fbioe.2023.1125060/full#supplementary-material
## References
1. Azevedo M. M., Tsigkou O., Nair R., Jones J. R., Jell G., Stevens M. M.. **Hypoxia inducible factor-stabilizing bioactive glasses for directing mesenchymal stem cell behavior**. *Tissue Eng. Part A* (2015) **21** 382-389. DOI: 10.1089/ten.tea.2014.0083
2. Baino F., Novajra G., Miguez-Pacheco V., Boccaccini A. R., Vitale-Brovarone C.. **Bioactive glasses: Special applications outside the skeletal system**. *J. Non-Cryst Solids* (2016) **432** 15-30. DOI: 10.1016/j.jnoncrysol.2015.02.015
3. Barr T. L.. **An XPS study of Si as it occurs in adsorbents, catalysts, and thin films**. *Appl. Surf. Sci.* (1983) **15** 1-35. DOI: 10.1016/0378-5963(83)90003-x
4. Barrioni B. R., de Laia A. G. S., Valverde T. M., Martins T. M. D., Caliari M. V., de Sa M. A.. **Evaluation of**. *Ceram. Int.* (2018) **44** 20337-20347. DOI: 10.1016/j.ceramint.2018.08.022
5. Barrioni B. R., Norris E., Jones J. R., Pereira M. D.. **The influence of cobalt incorporation and cobalt precursor selection on the structure and bioactivity of sol-gel-derived bioactive glass**. *J. Sol-Gel Sci. Technol.* (2018) **88** 309-321. DOI: 10.1007/s10971-018-4823-7
6. Boateng J. S., Matthews K. H., Stevens H. N., Eccleston G. M.. **Wound healing dressings and drug delivery systems: A review**. *J. Pharm. Sci.* (2008) **97** 2892-2923. DOI: 10.1002/jps.21210
7. Botusan I. R., Sunkari V. G., Savu O., Catrina A. I., Grunler J., Lindberg S.. **Stabilization of HIF-1 is critical to improve wound healing in diabetic mice**. *Proc. Natl. Acad. Sci.* (2008) **105** 19426-19431. DOI: 10.1073/pnas.0805230105
8. Boukamp P., Petrussevska R. T., Breitkreutz D., Hornung J., Markham A., Fusenig N. E.. **Normal keratinization in a spontaneously immortalized aneuploid human keratinocyte cell-line**. *J. Cell Biol.* (1988) **106** 761-771. DOI: 10.1083/jcb.106.3.761
9. Broderick N.. **Understanding chronic wound healing**. *Nurse Pract.* (2009) **34** 16-22. DOI: 10.1097/01.npr.0000361299.88402.f6
10. Chen S., Michalek M., Galuskova D., Michalkova M., Svancarek P., Talimian A.. **Multi-targeted B and Co co-doped 45S5 bioactive glasses with angiogenic potential for bone regeneration**. *Mater Sci. Eng. C* (2020) **112** 110909
11. Day R. M.. **Bioactive glass stimulates the secretion of angiogenic growth factors and angiogenesis**. *Tissue Eng.* (2005) **11** 768-777. DOI: 10.1089/ten.2005.11.768
12. Day R. M., Boccaccini A. R., Shurey S., Roether J. A., Forbes A., Hench L. L.. **Assessment of polyglycolic acid mesh and bioactive glass for soft-tissue engineering scaffolds**. *Biomaterials* (2004) **25** 5857-5866. DOI: 10.1016/j.biomaterials.2004.01.043
13. de Laia A. G. S., Barrioni B. R., Valverde T. M., de Goes A. M., de Sa M. A., Pereira M. D.. **Therapeutic cobalt ion incorporated in poly(vinyl alcohol)/bioactive glass scaffolds for tissue engineering**. *J. Mater Sci.* (2020) **55** 8710-8727. DOI: 10.1007/s10853-020-04644-0
14. de Laia A. G. S., Valverde T. M., Barrioni B. R., Cunha P. S., de Goes A. M., de Miranda M. C.. **Cobalt-containing bioactive glass mimics vascular endothelial growth factor A and hypoxia inducible factor 1 function**. *J. Biomed. Mater Res. Part A* (2021) **109** 1051-1064. DOI: 10.1002/jbm.a.37095
15. Deng Z. W., Lin B. C., Jiang Z. H., Huang W. H., Li J. S., Zeng X. Q.. **Hypoxia-mimicking cobalt-doped borosilicate bioactive glass scaffolds with enhanced angiogenic and osteogenic capacity for bone regeneration**. *Int. J. Biol. Sci.* (2019) **15** 1113-1124. DOI: 10.7150/ijbs.32358
16. Detmar M.. **The role of VEGF and thrombospondins in skin angiogenesis**. *J. Dermatol. Sci.* (2000) **24**. DOI: 10.1016/S0923-1811(00)00145-6
17. Dery M. A., Michaud M. D., Richard D. E.. **Hypoxia-inducible factor 1: Regulation by hypoxic and non-hypoxic activators**. *Int. J. Biochem. cell Biol.* (2005) **37** 535-540. DOI: 10.1016/j.biocel.2004.08.012
18. Dietrich E., Oudadesse H., Lucas-Girot A., Mami M.. *J. Biomed. Mater. Res. Part A* (2009) **88** 1087-1096. DOI: 10.1002/jbm.a.31901
19. Duscher D., Maan Z. N., Whittam A. J., Sorkin M., Hu M. S., Walmsley G. G.. **Fibroblast-specific deletion of hypoxia inducible factor-1 critically impairs murine cutaneous neovascularization and wound healing**. *Plastic Reconstr. Surg.* (2015) **136** 1004-1013. DOI: 10.1097/prs.0000000000001699
20. Dziadek M., Zagrajczuk B., Menaszek E., Dziadek K., Cholewa-Kowalska K.. **A simple way of modulating**. *Mater Lett.* (2018) **215** 87-90. DOI: 10.1016/j.matlet.2017.12.075
21. Esakkimuthukumar M., Swaroop A. K., Patnaik S. K., Kumar R. R., Praveen T. K., Naik M. R.. **A novel family of small molecule HIF-1 alpha stabilizers for the treatment of diabetic wounds; an integrated**. *Rsc Adv.* (2022) **12** 31293-31302. DOI: 10.1039/d2ra05364k
22. Elias P. M., Arbiser J., Brown B. E., Rossiter H., Man M. Q., Cerimele F.. **Epidermal vascular endothelial growth factor production is required for permeability barrier homeostasis, dermal angiogenesis, and the development of epidermal hyperplasia - Implications for the pathogenesis of psoriasis**. *Am. J. Pathol.* (2008) **173** 689-699. DOI: 10.2353/ajpath.2008.080088
23. Filgueiras M. R., La Torre G., Hench L. L.. **Solution effects on the surface reactions of a bioactive glass**. *J. Biomed. Mater. Res.* (1993) **27** 445-453. DOI: 10.1002/jbm.820270405
24. Fonder M. A., Lazarus G. S., Cowan D. A., Aronson-Cook B., Kohli A. R., Mamelak A. J.. **Treating the chronic wound: A practical approach to the care of nonhealing wounds and wound care dressings**. *J. Am. Acad. Dermatology* (2008) **58** 185-206. DOI: 10.1016/j.jaad.2007.08.048
25. Franks P. J., Morgan P. A.. **Health-related quality of life with chronic leg ulceration**. *Expert Rev. pharmacoeconomics outcomes Res.* (2003) **3** 611-622. DOI: 10.1586/14737167.3.5.611
26. Frykberg R. G., Banks J.. **Challenges in the treatment of chronic wounds**. *Adv. Wound Care* (2015) **4** 560-582. DOI: 10.1089/wound.2015.0635
27. Gerhardt L. C., Widdows K. L., Erol M. M., Burch C. W., Sanz-Herrera J. A., Ochoa I.. **The pro-angiogenic properties of multi-functional bioactive glass composite scaffolds**. *Biomaterials* (2011) **32** 4096-4108. DOI: 10.1016/j.biomaterials.2011.02.032
28. Gleadle J. M., Ebert B. L., Firth J. D., Ratcliffe P. J.. **Regulation of angiogenic growth factor expression by hypoxia, transition metals, and chelating agents**. *Am. J. Physiology - Cell Physiology* (1995) **268** C1362-C1368. DOI: 10.1152/ajpcell.1995.268.6.c1362
29. Gorustovich A., Perio C., Roether J. A., Boccaccini A. R.. **Effect of bioactive glasses on angiogenesis: A review of**. *Tissue Eng. Part B* (2010) **16** 199-207. DOI: 10.1089/ten.teb.2009.0416
30. Green J., Jester R., McKinley R., Pooler A.. **Patient perspectives of their leg ulcer journey**. *J. Wound Care* (2013) **22** 58-66. DOI: 10.12968/jowc.2013.22.2.58
31. Greenspan D. C.. **Bioglass at 50 – a look at Larry Hench’s legacy and bioactive materials**. *Biomed. Glas.* (2019) **5** 178-184. DOI: 10.1515/bglass-2019-0014
32. Gunnarsson I., Arnorsson S.. **Amorphous silica solubility and the thermodynamic properties of H4SiO°4 in the range of 0° to 350°C at Psat**. *Geochimica Cosmochimica Acta* (2000) **64** 2295-2307. DOI: 10.1016/s0016-7037(99)00426-3
33. Han G., Ceilley R.. **Chronic wound healing: A review of current management and treatments**. *Adv. Ther.* (2017) **34** 599-610. DOI: 10.1007/s12325-017-0478-y
34. Hench L. L.. **Bioceramics: From concept to clinic**. *J. Am. Ceram. Soc.* (1991) **74** 1487-1510. DOI: 10.1111/j.1151-2916.1991.tb07132.x
35. Hench L. L., Splinter R. J., Allen W. C., Greenlee T. K.. **Bonding mechanisms at the interface of ceramic prosthetic materials**. *J. Biomed. Mater. Res.* (1971) **5** 117-141. DOI: 10.1002/jbm.820050611
36. Ho S-W., Houlla M., Hercules D. M.. **Effect of particle size on carbon monoxide hydrogenation activity of silica supported cobalt catalysts**. *J. Phys. Chem.* (1990) **94** 6396-6399. DOI: 10.1021/j100379a045
37. Hoppe A., Jokic B., Janackovic D., Fey T., Greil P., Romeis S.. **Cobalt-releasing 1393 bioactive glass-derived scaffolds for bone tissue engineering applications**. *ACS Appl. Mater. interfaces* (2014) **6** 2865-2877. DOI: 10.1021/am405354y
38. Jones J. R.. **Review of bioactive glass: From hench to hybrids**. *Acta biomater.* (2013) **9** 4457-4486. DOI: 10.1016/j.actbio.2012.08.023
39. Jung S. B., Jones J. R., Clare A. G.. **Bioactive borate glasses**. *Bio-Glasses: An introduction* (2012) 75-94
40. Kargozar S., Baino F., Lotfibakhshaiesh N., Hill R. G., Milan P. B., Hamzehlou S.. **When size matters: Biological response to strontium- and cobalt-substituted bioactive glass particles**. *Mater. Today-Proceedings* (2018) **5** 15768-15775. DOI: 10.1016/j.matpr.2018.04.190
41. Kargozar S., Hamzehlou S., Baino F.. **Can bioactive glasses be useful to accelerate the healing of epithelial tissues?**. *Mater Sci. Eng. C* (2019) **97** 1009-1020. DOI: 10.1016/j.msec.2019.01.028
42. Kargozar S., Lotfibakhshaiesh N., Ai J., Mozafari M., Milan P. B., Hamzehlou S.. **Strontium- and cobalt-substituted bioactive glasses seeded with human umbilical cord perivascular cells to promote bone regeneration via enhanced osteogenic and angiogenic activities**. *Acta Biomater.* (2017) **58** 502-514. DOI: 10.1016/j.actbio.2017.06.021
43. Kermani F., Beidokhti S. M., Baino F., Gholamzadeh-Virany Z., Mozafari M., Kargozar S.. **Strontium- and cobalt-doped multicomponent mesoporous bioactive glasses (MBGs) for potential use in bone tissue engineering applications**. *Materials* (2020) **13** 1348. DOI: 10.3390/ma13061348
44. Khodakov A. Y., Griboval-Constant A., Bechara R., Zholobenko V. L.. **Pore size effects in fischer tropsch synthesis over cobalt-supported mesoporous silicas**. *J. Catal.* (2002) **206** 230-241. DOI: 10.1006/jcat.2001.3496
45. Leire E., Olson J., Isaacs H., Nizet V., Hollands A.. **Role of hypoxia inducible factor-1 in keratinocyte inflammatory response and neutrophil recruitment**. *J Inflamm. Lon.* (2013) **10**. DOI: 10.1186/1476-9255-10-28
46. Leu A., Leach J. K.. **Proangiogenic potential of a collagen/bioactive glass substrate**. *Pharm. Res.* (2008) **25** 1222-1229. DOI: 10.1007/s11095-007-9508-9
47. Lin C., Mao C., Zhang J., Li Y., Chen X.. **Healing effect of bioactive glass ointment on full-thickness skin wounds**. *Biomed. Mater.* (2012) **7** 045017. DOI: 10.1088/1748-6041/7/4/045017
48. Lin Y., Brown R. F., Jung S. B., Day D. E.. **Angiogenic effects of borate glass microfibers in a rodent model**. *J. Biomed. Mater. Res. Part A* (2014) **102** 4491-4499. DOI: 10.1002/jbm.a.35120
49. Littmann E., Autefage H., Solanki A. K., Kallepitis C., Jones J. R., Alini M.. **Cobalt-containing bioactive glasses reduce human mesenchymal stem cell chondrogenic differentiation despite HIF-1 alpha stabilisation**. *J. Eur. Ceram. Soc.* (2018) **38** 877-886. DOI: 10.1016/j.jeurceramsoc.2017.08.001
50. Liu X., Rahaman M. N., Day D. E.. **Conversion of melt-derived microfibrous borate (13-93B3) and silicate (45S5) bioactive glass in a simulated body fluid**. *J. Mater. Sci. Mater. Med.* (2013) **24** 583-595. DOI: 10.1007/s10856-012-4831-z
51. Macon A. L., Kim T. B., Valliant E. M., Goetschius K., Brow R. K., Day D. E.. **A unified**. *J. Mater. Sci. Mater. Med.* (2015) **26** 115. DOI: 10.1007/s10856-015-5403-9
52. Mao C., Chen X. F., Miao G. H., Lin C.. **Angiogenesis stimulated by novel nanoscale bioactive glasses**. *Biomed. Mater* (2015) **10** 025005. DOI: 10.1088/1748-6041/10/2/025005
53. Miguez-Pacheco V., Hench L. L., Boccaccini A. R.. **Bioactive glasses beyond bone and teeth: Emerging applications in contact with soft tissues**. *Acta biomater.* (2015) **13C** 1-15. DOI: 10.1016/j.actbio.2014.11.004
54. Naruphontjirakul P., Tsigkou O., Li S. W., Porter A. E., Jones J. R.. **Human mesenchymal stem cells differentiate into an osteogenic lineage in presence of strontium containing bioactive glass nanoparticles**. *Acta Biomater.* (2019) **90** 373-392. DOI: 10.1016/j.actbio.2019.03.038
55. Naruphontjiralkul P., Porter A. E., Jones J. R.. *Acta Biomater.* (2018) **66** 67-80. DOI: 10.1016/j.actbio.2017.11.008
56. Naseri S., Griffanti G., Lepry W. C., Maisuria V. B., Tufenkji N., Nazhat S. N.. **Silver-doped sol-gel borate glasses: Dose-dependent effect on**. *J. Am. Ceram. Soc.* (2022) **105** 1711-1722. DOI: 10.1111/jace.17802
57. Naseri S., Lepry W. C., Nazhat S. N.. **Bioactive glasses in wound healing: Hope or hype?**. *J. Mat. Chem. B* (2017) **5** 6167-6174. DOI: 10.1039/c7tb01221g
58. Norris E., Ramos-Rivera C., Poologasundarampillai G., Clark J. P., Ju Q., Obata A.. **Electrospinning 3D bioactive glasses for wound healing**. *Biomed. Mater* (2020) **15** 015014. DOI: 10.1088/1748-605x/ab591d
59. Olsson M., Jarbrink K., Divakar U., Bajpai R., Upton Z., Schmidtchen A.. **The humanistic and economic burden of chronic wounds: A systematic review**. *Wound Repair Regen.* (2019) **27** 114-125. DOI: 10.1111/wrr.12683
60. Posnett J., Franks P. J.. **The burden of chronic wounds in the UK**. *Nurs. Times* (2008) **104** 44-45
61. Quinlan E., Partap S., Azevedo M. M., Jell G., Stevens M. M., O'Brien F. J.. **Hypoxia-mimicking bioactive glass/collagen glycosaminoglycan composite scaffolds to enhance angiogenesis and bone repair**. *Biomaterials* (2015) **52** 358-366. DOI: 10.1016/j.biomaterials.2015.02.006
62. Quintero F., Dieste O., Penide J., Lusquinos F., Riveiro A., Pou J.. **Nonconventional production of glass nanofibers by laser spinning**. *J. Am. Ceram. Soc.* (2014) **97** 3116-3121. DOI: 10.1111/jace.13127
63. Quintero F., Pou J., Comesaña R., Lusquiños F., Riveiro A., Mann A. B.. **Laser spinning of bioactive glass nanofibers**. *Adv. Funct. Mater.* (2009) **19** 3084-3090. DOI: 10.1002/adfm.200801922
64. Rahaman M. N., Day D. E., Bal B. S., Fu Q., Jung S. B., Bonewald L. F.. **Bioactive glass in tissue engineering**. *Acta biomater.* (2011) **7** 2355-2373. DOI: 10.1016/j.actbio.2011.03.016
65. Ranmuthu C. D. S., Ranmuthu C. K. I., Russell J. C., Singhania D., Khan W. S.. **Evaluating the effect of non-cellular bioactive glass-containing scaffolds on osteogenesis and angiogenesis in**. *Front. Bioeng. Biotechnol.* (2020) **8** 430. DOI: 10.3389/fbioe.2020.00430
66. Rezvani H. R., Ali N., Nissen L. J., Harfouche G., de Verneuil H.. **HIF-1α in epidermis: oxygen sensing, cutaneous angiogenesis, cancer, and non-cancer disorders**. *J. Invest. Dermatol.* (2011) **131** 1793-1805. DOI: 10.1038/jid.2011.141
67. Sanders D. M., Person W. B., Hench L. L.. **Quantitative analysis of glass structure with the use of infrared reflection spectra**. *Appl. Spectrosc.* (1974) **28** 247-255. DOI: 10.1366/000370274774332623
68. Semenza G. L.. **Hydroxylation of HIF-1: Oxygen sensing at the molecular level**. *Physiology* (2004) **19** 176-182. DOI: 10.1152/physiol.00001.2004
69. Sen C. K., Gordillo G. M., Roy S., Kirsner R., Lambert L., Hunt T. K.. **Human skin wounds: A major and snowballing threat to public health and the economy**. *Wound Repair. Regen.* (2009) **17** 763-771. DOI: 10.1111/j.1524-475x.2009.00543.x
70. Serra J., González P., Liste S., Serra C., Chiussi S., León B.. **FTIR and XPS studies of bioactive silica based glasses**. *J. Non-Crystalline Solids* (2003) **332** 20-27. DOI: 10.1016/j.jnoncrysol.2003.09.013
71. Shi M., Zhao F. J., Sun L. Y., Tang F. L., Gao W. D., Xie W. H.. **Bioactive glass activates VEGF paracrine signaling of cardiomyocytes to promote cardiac angiogenesis**. *Mater Sci. Eng. C* (2021) **124** 112077
72. Shi Q. Y., Luo X., Huang Z. Q., Midgley A. C., Wang B., Liu R. H.. **Cobalt-mediated multi-functional dressings promote bacteria-infected wound healing**. *Acta Biomater.* (2019) **86** 465-479. DOI: 10.1016/j.actbio.2018.12.048
73. Singh A., Halder S., Menon G. R., Chumber S., Misra M. C., Sharma L. K.. **Meta-analysis of randomized controlled trials on hydrocolloid occlusive dressing versus conventional gauze dressing in the healing of chronic wounds**. *Asian J. Surg.* (2004) **27** 326-332. DOI: 10.1016/s1015-9584(09)60061-0
74. Solanki A. K., Lali F. V., Autefage H., Agarwal S., Nommeots-Nomm A., Metcalfe A. D.. **Bioactive glasses and electrospun composites that release cobalt to stimulate the HIF pathway for wound healing applications**. *Biomaterials Res.* (2021) **25** 1. DOI: 10.1186/s40824-020-00202-6
75. Tan B. J., Klabunde K. J., Sherwood P. M. A.. **XPS studies of solvated metal atom dispersed (SMAD) catalysts. Evidence for layered cobalt-manganese particles on alumina and silica**. *J. Am. Chem. Soc.* (1991) **113** 855-861. DOI: 10.1021/ja00003a019
76. Tobler D. J., Shaw S., Benning L. G.. **Quantification of initial steps of nucleation and growth of silica nanoparticles: An**. *Geochimica Cosmochimica Acta* (2009) **73** 5377-5393. DOI: 10.1016/j.gca.2009.06.002
77. Tokoro S., Satoh T., Okubo Y., Igawa K., Yokozeki H.. **Latent dystrophic subcutaneous calcification in patients with chronic venous insufficiency**. *Acta dermato-venereologica.* (2009) **89** 505-508. DOI: 10.2340/00015555-0680
78. Vyas V. K., Kumar A. S., Prasad S., Singh S. P., Pyare R.. **Bioactivity and mechanical behaviour of cobalt oxide-doped bioactive glass**. *Bull. Mater. Sci.* (2015) **38** 957-964. DOI: 10.1007/s12034-015-0936-6
79. Wu C., Zhou Y., Fan W., Han P., Chang J., Yuen J.. **Hypoxia-mimicking mesoporous bioactive glass scaffolds with controllable cobalt ion release for bone tissue engineering**. *Biomaterials* (2012) **33** 2076-2085. DOI: 10.1016/j.biomaterials.2011.11.042
80. Xynos I. D., Edgar A. J., Buttery L. D. K., Hench L. L., Polak J. M.. **Gene-expression profiling of human osteoblasts following treatment with the ionic products of Bioglass (R) 45S5 dissolution**. *J. Biomed. Mater Res.* (2001) **55** 151-157. DOI: 10.1002/1097-4636(200105)55:2<151::aid-jbm1001>3.0.co;2-d
81. Yang Q., Chen S., Shi H., Xiao H., Ma Y.. *Mater. Sci. Eng. C, Mater. Biol. Appl.* (2015) **55** 105-117. DOI: 10.1016/j.msec.2015.05.049
82. Yu H., Peng J., Xu Y., Chang J., Li H.. **Bioglass activated skin tissue engineering constructs for wound healing**. *ACS Appl. Mater. interfaces* (2016) **8** 703-715. DOI: 10.1021/acsami.5b09853
83. Yuan Y., Hilliard G., Ferguson T., Millhorn D. E.. **Cobalt inhibits the interaction between hypoxia-inducible factor-α and von Hippel-lindau protein by direct binding to hypoxia-inducible factor-α**. *J. Biol. Chem.* (2003) **278** 15911-15916. DOI: 10.1074/jbc.m300463200
84. Zhang H., Lancelot C., Chu W., Hong J., Khodakov A. Y., Chernavskii P. A.. **The nature of cobalt species in carbon nanotubes and their catalytic performance in Fischer–Tropsch reaction**. *J. Mater. Chem.* (2009) **19** 9241. DOI: 10.1039/b911355j
85. Zhang W. Y., Wang L. T., Guo H. Y., Chen L. L., Huang X.. **Dapagliflozin‐loaded exosome mimetics facilitate diabetic wound healing by HIF‐1**. *Adv. Healthc. Mater.* (2022a) 2202751. DOI: 10.1002/adhm.202202751
86. Zhang Y., Zhang Y. Y., Pan Z. W., Li Q. Q., Sun L. H., Li X.. **GDF11 promotes wound healing in diabetic mice via stimulating HIF-1 alpha-VEGF/SDF-1 alpha-mediated endothelial progenitor cell mobilization and neovascularization**. *Acta Pharmacol. Sin* (2022b). DOI: 10.1038/s41401-022-01013-2
87. Zhao S., Li L., Wang H., Zhang Y., Cheng X., Zhou N.. **Wound dressings composed of copper-doped borate bioactive glass microfibers stimulate angiogenesis and heal full-thickness skin defects in a rodent model**. *Biomaterials* (2015) **53** 379-391. DOI: 10.1016/j.biomaterials.2015.02.112
88. Zhou J., Wang H., Zhao S., Zhou N., Li L., Huang W.. *Mater. Sci. Eng. C, Mater. Biol. Appl.* (2016) **60** 437-445. DOI: 10.1016/j.msec.2015.11.068
|
---
title: Intrinsic brain abnormalities in chronic rhinosinusitis associated with mood
and cognitive function
authors:
- Simin Lin
- Miaomiao Nie
- Bingshan Wang
- Shaoyin Duan
- Qianwen Huang
- Naiming Wu
- Zhishang Chen
- Hengyu Zhao
- Yi Han
journal: Frontiers in Neuroscience
year: 2023
pmcid: PMC10036396
doi: 10.3389/fnins.2023.1131114
license: CC BY 4.0
---
# Intrinsic brain abnormalities in chronic rhinosinusitis associated with mood and cognitive function
## Abstract
### Background
Chronic rhinosinusitis (CRS) poses a risk for developing emotional and cognitive disorders. However, the neural evidence for this association is largely unclear. Resting-state functional magnetic resonance imaging (rs-fMRI) analysis can demonstrate abnormal brain activity and functional connectivity and contribute to explaining the potential pathophysiology of CRS-related mood and cognitive alterations.
### Methods
Chronic rhinosinusitis patients (CRS, $$n = 26$$) and gender- and age-matched healthy control subjects (HCs, $$n = 38$$) underwent resting-state functional MRI scanning. The amplitude of low-frequency fluctuations (ALFF) was calculated to observe the intrinsic brain activity. The brain region with altered ALFF was further selected as the seed for functional connectivity (FC) analysis. Correlation analysis was performed between the ALFF/FC and clinical parameters in CRS patients.
### Results
Compared with HCs, CRS patients exhibited significantly increased ALFF in the left orbital superior frontal cortex and reduced connectivity in the right precuneus using the orbital superior frontal cortex as the seed region. The magnitude of the orbital superior frontal cortex increased with inflammation severity. In addition, ALFF values in the orbital superior frontal cortex were positively correlated with the hospital anxiety and depression scale (HADS) scores. The ROC curves of altered brain regions indicated great accuracy in distinguishing between CRS patients and HCs.
### Conclusion
In this study, patients with CRS showed increased neural activity in the orbital superior frontal cortex, a critical region in emotional regulation, and this region also indicated hypoconnectivity to the precuneus with a central role in modulating cognition. This study provided preliminary insights into the potential neural mechanism related to mood and cognitive dysfunctions in CRS patients.
## Introduction
Chronic rhinosinusitis (CRS), one of the most common chronic diseases, is a set of disorders characterized by paranasal sinus and nose inflammation that lasts longer than 12 weeks (Gao et al., 2016). The prevalence of CRS is estimated to be more than $10\%$ around the world based on the presence of symptoms or objective findings (Sedaghat et al., 2022) and reaches $13\%$ of China’s population (Albu, 2020). CRS is a multi-factorial disease influenced by chronic inflammation, aberrant epithelial barrier and immunity, and imbalance of the nasal microflora (Sedaghat, 2017). There has been great advancement in the understanding of the pathophysiology of CRS: from the altered epithelial barrier and the epithelial-mesenchymal transition to the adaptive and innate immune system and, lastly, the effect of bacteria, including eosinophils and Staphylococcus aureus, on the persistence of disease (Bachert et al., 2020).
Among patients with CRS, the challenge is not only to constantly deal with CRS-related symptoms, such as nasal congestion and facial pain, but also a variety of potential complications, such as attention difficulties, possible depression and anxiety as recent reports (Ghadami, 2019; Kim et al., 2020; Jung et al., 2021). These symptoms may damage the patient’s quality of life, decrease the efficiency of work and incur serious medical costs (Gao et al., 2016). How CRS affects brain function is unclear. Previously proposed mechanisms mentioned that CRS-related brain function alterations is due to the effects of inflammatory cytokines on the brain (McAfoose and Baune, 2009; Jafari et al., 2021). Inflammatory cytokines may play a role in neurogenesis, neuromodulation and synaptic plasticity (McAfoose and Baune, 2009; Kim et al., 2016; Dehdar et al., 2019). The previous research demonstrated that inflammation might affect neural activity and neurogenesis (Fukushima et al., 2015; Salimi et al., 2021). The nasal epithelium is connected to the olfactory bulb via the olfactory nerve (Bohmwald et al., 2018). The mouse model studies of allergic rhinitis found neuroinflammation of the olfactory bulb (Lv et al., 2021). A recent study further discovered alterations in spontaneous brain activity in allergic rhinitis (Gao et al., 2021). We speculate that inflammatory cytokines in the upper respiratory tract, including CRS, enter the central nervous system via the olfactory nerve, which may lead to abnormal neural activity. Some studies also suggested that increased chronic inflammatory response in the nasal cavity may change the homeostasis of the local microbiome (Harrass et al., 2021). This microbial imbalance may affect the neuronal integrity of the central nervous system through mechanisms such as bacterial translocation to the central nervous system and/or modulation of the immune response in the central nervous system, leading to brain dysfunction (Hoggard et al., 2018; Harrass et al., 2021). Nevertheless, the pathophysiology of brain function alteration in CRS is still largely unknown.
Neuroimaging has proven to be a non-invasive tool for exploring the in vivo human brain, helping to detect the functional abnormalities in the brain at an early stage of various diseases (Lui et al., 2016). Resting-state functional magnetic resonance imaging (rs-fMRI) is a powerful and safe neuroimaging approach that inspects the changes in intrinsic neural activity via measurement of blood-oxygen level-dependent (BOLD) signal without task performance during the examination (Guo et al., 2022; Lin et al., 2022). Both amplitude of low-frequency fluctuation (ALFF) and regional homogeneity (ReHo) are crucial techniques to explore aberrant brain activity (Zang et al., 2004; Guo et al., 2022). The ALFF measures the low-frequency oscillation intensity of BOLD time courses in rs-fMRI (Zang et al., 2007; Guo et al., 2022). Although the exact biologic mechanisms of ALFF remain unclear, plenty of research has indicated that the ALFF changes are associated with local neuronal activity (Cui et al., 2017; Wang et al., 2019; Sun et al., 2022). ReHo measures the similarity in the time series of a given voxel to its nearest neighbors, which indicates that the ReHo alterations are associated with the coherence of spontaneous neuronal activity (Zang et al., 2004). Compared with ReHo, ALFF has the advantage of directly reflecting the intensity or amplitude of spontaneous brain activity (Qi et al., 2012). Resting-state functional connectivity (FC) measures interregional temporal correlation to reflect the intensity of interregional functional connectivity (Lu et al., 2017). Seed-based FC measures correlations of time series between selected seeds and the remaining voxels in the brain. This method is easy to measure and interpret. More to the point, it allows direct localization of the effect regions that show FC with the seed region at the whole-brain level (Gao et al., 2019; Liu et al., 2022). Some recent studies of chronic immune and inflammatory diseases, such as allergic rhinitis (Gao et al., 2021) and Crohn’s disease (Hou et al., 2019), used brain activity and functional connectivity to indicate abnormal brain regions associated with cognitive deficits or mood changes. Jafari et al. [ 2021] studied the functional connectivity of brain networks by independent component analysis and found altered functional connectivity in frontal medial cortices that regulated cognition function. However, to our knowledge, there are no reports on resting-state intrinsic neural activity in patients with CRS.
In this study, ALFF and seed-based FC methods are involved here for the first time to comprehensively explore the intraregional brain activity and interregional functional connectivity, attempting to provide potential information about the brain functional changes of CRS. We hypothesized that brain regions involved in emotional and cognitive functions might show abnormal brain activity and functional connectivity in patients with CRS, and altered brain regions may be associated with anxiety and depression.
## Participants
Twenty-six patients with CRS were recruited from the Zhongshan Hospital of Xiamen University from February to August 2022. The diagnostic criteria for CRS refer to the European Position Paper on Rhinosinusitis and Nasal Polyps (EP3OS) document (Fokkens et al., 2020). Inclusion criteria for patients of CRS included [1] the Lund-Mackay scoring (LMS) ≥ 8; [2] the age range was between 20 and 50. Thirty-eight gender- and age-matched healthy volunteers were also recruited from the local community as a healthy control group. The exclusion criteria for all subjects were [1] serious anxiety or depression (HADS-A or HADS-D scores more than 14; [2] history of brain surgery, tumor, or neuropsychiatric disease; [3] history of alcohol or drug abuse; [4] left-handed; [5] contraindications to MR examinations. The experiment was authorized by the Medical Ethics Committee of Zhongshan Hospital of Xiamen University. Written informed consent was obtained from participants enrolled in this study.
Demographic and clinical parameters such as sex, age, education level, disease course, clinical symptoms, medical and surgical history, and complications were collected. CRS severity was also calculated using objective and subjective scores, including the Lund-Mackay score (LMS) and the visual analog scale (VAS) score. In addition, we collected the hospital anxiety and depression scale (HADS) scores of each subject.
## MRI parameters
Images were obtained using a Philips Ingenia 3.0 T CX (Philips Healthcare, Best, Netherlands) at the Zhongshan Hospital of Xiamen University. Each subject was instructed to close their eyes, keep awake, try not to think and remain breathing calmly during the entire scanning process. Earplugs and foam pads were applied to minimize head motion and noise during scanning.
Resting-state functional MR imaging, including 200 volumes, was acquired by a gradient-echo-planar imaging sequence (EPI) in the axial plane with the following parameters. echo time (TE) = 25 ms; repetition time (TR) = 2,000 ms; flip angle (FA) = 65°; thickness = 2.5 mm; field of view (FOV) = 219 mm × 219 mm; matrix = 88 × 88; gap = 0 mm; voxel size = 2.5 mm × 2.5 mm × 2.5 mm; number of slices = 57.
Structural images including high-resolution T1-weighted and T2-weighted brain images. T1-weighted scans were obtained using a 3D magnetization-prepared rapid gradient-echo (MP-RAGE). The parameters are as follows: TE = 3.0 ms; TR = 6.6 ms; FA = 8°; thickness = 1.0 mm; FOV = 240 mm × 240 mm; matrix = 240 × 240; voxel size = 1.0 mm × 1.0 mm × 1.0 mm; gap = 0 mm; number of slices = 180. T2-weighted structural imaging was used to calculate all sinonasal Lund-Mackay scores of each patient by two experienced radiologists. Prior studies have already shown a good correlation between computed tomography and magnetic resonance imaging-based scores ($R = 0.837$; $P \leq 0.001$) (Lin and Bhattacharyya, 2009; Bhattacharyya, 2010). The parameters of the SPIR T2-weighted images: TE = 280 ms; TR = 3,000 ms; FA = 90°; thickness = 1.0 mm; matrix = 240 × 240; number of slices = 320.
## Data preprocessing
Resting-state BOLD data were preprocessed using Data Processing and Analysis for Brain Imaging (DPABI). This toolbox is based on Statistical Parametric Mapping 12 (SPM12). The preprocessing included the following steps: the first 10 volumes for each participant were abandoned to minimize the instability of initial imaging signals. The remaining volumes were corrected for slice timing and then realigned to correct head motions. Each subject who exhibited a rotational or translational motion parameter more than 2° or 2 mm was discarded. Each functional volume was spatially normalized to the Montreal Neurological Institute template using the DARTEL technique and resampled to 3-mm voxels. Subsequently, the images were smoothed using a 6 mm full-width-at half maximum (FWHM) Gaussian kernel. After smoothing, nuisance variables were regressed out from the data, including linear drift, cerebrospinal fluid signal, white matter signal, and Friston 24-parameter head motion parameters. Finally, the band-pass filtering (0.01–0.08 Hz) was performed to remove the influence of physiological noise.
## ALFF and seed-based FC analyses
Amplitude of low-frequency fluctuations and FC analyses were calculated via REST software. For ALFF analysis, after preprocessing, the fast Fourier transform was applied to transform the time series to the frequency domain (FFT). Then the square root of the power spectrum was computed, and the averaged square root was obtained across 0.01-0.08 Hz to acquire the ALFF. Eventually, to minimize the global influence, the ALFF values were divided by the global mean ALFF values for standardization. Based on the findings of ALFF, the abnormal brain region was selected as the seed. A radius of 5 mm around the peak MNI was used as the region of interest (ROI) for the seed-based FC analysis. Pearson correlation coefficients were conducted between the time courses of seed regions and the time series of all voxels in the entire brain. Fisher’s z-transform was performed on the Pearson correlation coefficients to generate an approximately normal distribution for further statistical analysis.
In comparison with the HCs, patients with CRS showed significantly increased ALFF values in the left orbital superior frontal cortex extending to the left rectus (cluster size of 130 voxels, $t = 5.0765$, $P \leq 0.05$, FDR corrected) (Figure 1). FC in the right precuneus was found to be decreased when the left orbital superior frontal cortex was used as the seed point (with a cluster size of 68 voxels, t = −4.0840, $P \leq 0.05$, FDR corrected) (Figure 2). Details are shown in Table 2.
**FIGURE 1:** *Amplitude of low-frequency fluctuations (ALFF) differences between CRS patients and HCs ($P \leq 0.05$, FDR corrected). Compared with HCs, patients with CRS showed significantly increased ALFF in the left orbital superior frontal cortex, extending to the left rectus. L, left; R, right.* **FIGURE 2:** *Seed-based FC differences between CRS patients and HCs ($P \leq 0.05$, FDR corrected). Compared with HCs, patients with CRS showed significantly decreased FC between the seed region in the left orbital superior frontal cortex and the right precuneus. SFC, superior frontal cortex; L, left; R, right.* TABLE_PLACEHOLDER:TABLE 2
## Statistical analysis
Demographics and clinical parameters between the two groups were calculated using SPSS 26.0. An independent two-sample t-test was performed for continuous variables, and a chi-squared test was performed for proportions.
Independent two-sample t-tests were performed for differences in ALFF and FC of each voxel between the CRS and HCs using SPM12 software. Sex, age, and levels of education were imported as covariates to minimize the impacts of confounding covariates. The cluster-level False Discovery Rate (FDR) method was used for multiple comparison correction, with a cluster-defined threshold of $$P \leq 0.001$$ and a corrected cluster significance of $P \leq 0.05.$ The HADS-A and VAS scores do not conform to the normal distribution, so Spearman correlation analysis was used to analyze the relationship between ALFF/FC and HADS-A and VAS. Pearson correlation analysis was performed to investigate the relationship between ALFF/FC and HADS-D scores and LMS.
To evaluate the diagnostic potential of ALFF and FC values in abnormal brain regions in separating CRS patients from HCs, we also used the receiver operating characteristic (ROC) curve for analysis. Besides, the level of precision was quantified by calculating the area under the curve.
## Demographics and clinical characteristics
A total of 26 patients with CRS and 38 HCs aged 20–50 years were included in the analysis, with no significant differences in gender (χ2 = 1.546, $$P \leq 0.214$$), age (t = −1.250, $$P \leq 0.216$$), HADS total scores (t = −0.966, $$P \leq 0.338$$), HADS-A (t = −1.112, $$P \leq 0.270$$) and HADS-D (t = −0.489, $$P \leq 0.626$$) scores between the two groups. There is a statistical difference between the two in the level of education ($t = 3.231$, $$P \leq 0.002$$). The disease duration of the patients varied widely, from a minimum of 6 months to a maximum of more than 20 years, resulting in a relatively large standard deviation. Details are summarized in Table 1.
**TABLE 1**
| Unnamed: 0 | CRS patients | HCs | Statistics | P-value |
| --- | --- | --- | --- | --- |
| | (n = 26) | (n = 38) | | |
| Demographic characteristics | Demographic characteristics | Demographic characteristics | Demographic characteristics | Demographic characteristics |
| Gender: Men (%) | 19 (73.1%) | 22 (57.9%) | χ2 = 1.546 | 0.214b |
| Age (year) | 36.88 ± 7.63 | 34.45 ± 7.69 | t = −1.250 | 0.216a |
| Education (year) | 14.00 ± 3.15 | 16.16 ± 2.20 | t = 3.231 | 0.002a |
| Clinical characteristics | Clinical characteristics | Clinical characteristics | Clinical characteristics | Clinical characteristics |
| Disease duration (year) | 8.08 ± 7.63 | | | |
| Nasal obstruction/congestion | 24 (92.3%) | | | |
| Nasal discharge | 19 (73.1%) | | | |
| Hyposmia/anosmia | 8 (30.8%) | | | |
| Facial pain/pressure | 7 (26.9%) | | | |
| Asthma | 2 (7.7%) | | | |
| Nasal polyp | 11 (42.3%) | | | |
| Previous sinonasal surgery | 7 (21.9%) | | | |
| LMS | 11.38 ± 2.53 | | | |
| Questionnaires | Questionnaires | Questionnaires | Questionnaires | Questionnaires |
| HADS total scores | 9.46 ± 5.85 | 8.16 ± 4.90 | t = −0.966 | 0.338a |
| HADS-A scores | 5.12 ± 3.50 | 4.26 ± 2.63 | t = −1.112 | 0.270a |
| HADS-D scores | 4.35 ± 3.12 | 3.97 ± 2.90 | t = −0.489 | 0.626a |
| VAS | 5.63 ± 1.81 | | | |
| Medicine in three months | Medicine in three months | Medicine in three months | Medicine in three months | Medicine in three months |
| Nasal glucocorticoid | 15 (57.7%) | | | |
| Antibiotics | 4 (15.4%) | | | |
## Correlation analysis
In patients with CRS, ALFF in the orbital superior frontal cortex was found to be positively correlated with the HADS-A scores ($R = 0.3457$, $$P \leq 0.0418$$, Figure 3A), HADS-D scores ($R = 0.3698$, $$P \leq 0.0315$$, Figure 3B) and LMS ($R = 0.4121$, $$P \leq 0.0182$$, Figure 3C), as well as a significant correlation was found between VAS and HADS-A scores ($R = 0.5125$, $$P \leq 0.0037$$, Figure 3D). However, there was no significant correlation between FC alterations in the right precuneus and clinical index, including HADS scores, LMS and VAS. Moreover, the correlation analysis between disease duration and both ALFF and FC values showed no statistical difference.
**FIGURE 3:** *Correlations between the ALFF value in the left orbital SFC and clinical assessment. Panels (A,B) revealed positive correlations between ALFF in the left orbital SFC and HADS-A and HADS-D scores. Panel (C) established a “dose-dependent” association between the ALFF value in the left orbital SFC and the objective severity of inflammation. Panel (D) showed a positive correlation between the patient’s subjective severity of inflammation and anxiety scores. SFC, superior frontal cortex; HADS-A, hospital anxiety and depression scale-anxiety; HADS-D, hospital anxiety and depression scale-depression; LMS, Lund-Mackay scoring.*
## Receiver operating characteristic curve
We also performed the ROC curve analysis of mean ALFF and FC values in these changed brain regions to find potential imaging biomarkers to separate CRS patients from healthy controls. The area under the ROC curve (AUC) of the ALFF values of the left orbital superior frontal cortex and the FC values of the right precuneus were 0.8229 (Figure 4A) and 0.7895 (Figure 4B), respectively. It showed good accuracy in distinguishing between CRS patients and HCs.
**FIGURE 4:** *ROC curve analysis for altered brain region. (A) The AUC of the ALFF values of the left orbital SFC was 0.8229 (P < 0.001; 95% CI: 0.7211–0.9247); (B) the AUC of the FC values of the right precuneus was 0.7895 (P < 0.001; 95% CI: 0.6762–0.9028). ROC, receiver operating characteristic; AUC, area under the curve; CI, confidence interval; SFC, superior frontal cortex.*
## Discussion
This study presents preliminary neurological evidence for ALFF and FC abnormalities as a potential basis for emotional and cognitive alterations in patients with CRS. We found that compared with healthy controls, patients with CRS presented enhanced ALFF values in the left orbital superior frontal cortex extending to the left rectus, both belonging to the orbitofrontal cortex (OFC), which is involved in emotion regulation and decision-making (Schoenbaum et al., 2006; Subramaniam et al., 2016). This brain region demonstrated decreased functional connectivity to the precuneus, which is prominent for its unique role in cognitive modulation (Cavanna and Trimble, 2006). The severity of inflammation as well as anxiety and depression problems are significantly positively correlated with spontaneous neural activity in the OFC. Taken together, these findings provide a neuropathophysiological basis for understanding the relationship between CRS and the increased risk of emotional and cognition dysfunctions, which contribute to clinical prevention and treatment for improving the life quality of patients.
The orbitofrontal cortex, as a part of the prefrontal cortex, is implicated in emotion regulation and decision-making (Schoenbaum et al., 2006; Subramaniam et al., 2016; Burks et al., 2018). The OFC receives environmental irritation, affective information, and emotional or social responses (Cheng et al., 2016). Banerjee et al. [ 2020] indicated that the orbitofrontal cortex of the cerebral cortex reprograms neurons located in sensory areas, allowing humans to make flexible decisions. Wikenheiser et al. [ 2017] found that OFC neurons encode the expected outcome of a decision task by recording neuronal activity in the OFC during the execution of the decision task in mice. A theory of depression has been developed that depression may be related to the over-responsiveness of the OFC to the non-reward or punishment system (Milad and Rauch, 2007). Neuroimaging has shown evidence of altered brain activity or functional connectivity in the OFC of patients with depression and anxiety disorders. For instance, rs-fMRI in healthy male participants revealed that trait anxiety is associated with fALFF in the medial OFC and the functional connectivity between the medial OFC and precuneus (Xue et al., 2018). Depression patients showed increased regional cerebral blood flow, higher ALFF, lower functional connectivity, reduced cortical thickness and gray matter volume in the OFC (Lyoo et al., 2004; Price and Drevets, 2010; Xue et al., 2018; Gong et al., 2020). A cohort study indicated the overall incidence of depression and anxiety disorders was increased in CRS patients than in healthy people during the 11-year follow-up (Kim et al., 2019). We speculate that increased brain activity in the OFC involved in emotion regulation in patients with CRS may be associated with anxiety and depression. Although HADS scores were higher in patients with CRS, we did not find a statistical difference between groups in scores of anxiety and depression. Thus, given the brain’s ability to adapt and compensate, especially among young individuals excluding severe anxiety and depression, our results may present early and subclinical intrinsic brain abnormalities (including ALFF and FC) that may precede or be more sensitive than clinical symptoms. To further support our speculation, we collected several clinical indexes for CRS patients to investigate the association between CRS and mood disorders. As expected, The ALFF values of OFC increase with the severity of inflammation, which indicates a dose-dependent association between neuronal activity in OFC and the degree of inflammation. This study also showed a significantly positive correlation between neuronal activity in OFC of CRS patients and scores of anxiety and depression, and we guess that sinusitis patients have overreaction of OFC to non-rewards, which may increase the incidence of mood disorders. Relating brain activity abnormalities in OFC to HADS scores may help to provide new insights into the relationship between CRS and anxiety and depression.
The precuneus, as a functional core of the default mode network, plays an important role in cognitive processing (Li et al., 2019). The precuneus is implicated in the representation of self-related processing, autobiographical memory, and visuospatial processing (Cavanna and Trimble, 2006; Zhang and Li, 2012). Besides, the neuroimaging studies had consistently identified that the precuneus was activated when subjects were triggered with a higher action identification level, which is powerful evidence of the connection of the precuneus to higher-order cognitive functions such as abstract mental imagination and attention shifting (Lou et al., 2004; Simon et al., 2004). Jafari et al. [ 2021] used independent component analysis to indicate altered brain connectivity within the frontoparietal and default mode networks, both of which play a critical role in cognition modulation. In contrast to us, their findings are based on a cohort of young participants identified from public databases, lacking relevant clinical history and not suitable to represent a clinical CRS population. A large sample functional connectivity study found that the precuneus is strongly associated with the effects of depressive problems (Cheng et al., 2018). It is supposed that the precuneus is involved in the sense of self, and depression is associated with an impaired representation of self. These findings together support our points that hypoconnectivity between OFC and precuneus may contribute to explaining the increased risk of mood and cognitive impairments of patients with CRS.
Given the evidence of chronic rhinosinusitis associated with mood and cognitive function, we make the following assumptions about the pathogenesis: [1] the physiological effect of nasal congestion and its influence on sleep quality may subsequently affect psychiatric symptoms negatively, such as anxiety and depression (Fang et al., 2010); [2] pro-inflammatory cytokines can enter the central nervous system and interact with a cytokine network in the brain to affect brain function (Capuron and Miller, 2011); [3] an imbalance of local microbiome homeostasis may affect the neuronal integrity of the central nervous system and cause brain-related symptoms (Harrass et al., 2021). Further study should focus on the specific regulatory mechanisms of molecules between the sinonasal inflammation and abnormal brain function aiming to find a way to block the vicious circle.
## Limitation
Several limitations are mentioned in this study. The sample size of this research was relatively small, which may reduce statistical power and limit the further investigation of the relationship between the abnormal activity in the brain regions and CRS. Furthermore, we deduced that CRS was associated with cognitive dysfunction. However, the cognitive analysis using professional scales or questionnaires did not conduct here, which may influence the credibility of the results. And the brain functional abnormal changes in CRS is a crucial project that deserves much further exploration.
## Conclusion
This study demonstrated the intrinsic abnormal brain activity in the orbital superior frontal cortex in chronic rhinosinusitis patients, which is associated with mood disorders, including depression and anxiety. Besides, the functional connectivity alterations between OFC and the precuneus in CRS patients. This rs-fMRI study provides preliminary evidence for alterations in brain activity and functional connectivity as a potential basis for mood and cognitive dysfunction in patients with CRS.
## Data availability statement
The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.
## Ethics statement
Written informed consent was obtained from the individual(s) for the publication of any potentially identifiable images or data included in this article.
## Author contributions
SL searched the literature, analyzed the data, and wrote the first version of the manuscript. MN and SD collected the clinical information, questionnaire scales, and imaging data of patients. BW and QH collected the clinical information, questionnaire scales, and imaging data of healthy controls. NW and ZC created tables and figures. YH and HZ revised the manuscript for critical intellectual content. All authors read and approved the final manuscript.
## Conflict of interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher’s note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## References
1. Albu S.. **Chronic rhinosinusitis-an update on epidemiology, pathogenesis and management.**. (2020) **9**. DOI: 10.3390/jcm9072285
2. Bachert C., Marple B., Schlosser R. J., Hopkins C., Schleimer R. P., Lambrecht B. N.. **Adult chronic rhinosinusitis.**. (2020) **6**
3. Banerjee A., Parente G., Teutsch J., Lewis C., Voigt F. F., Helmchen F.. **Value-guided remapping of sensory cortex by lateral orbitofrontal cortex.**. (2020) **585** 245-250. DOI: 10.1038/s41586-020-2704-z
4. Bhattacharyya N.. **The role of CT and MRI in the diagnosis of chronic rhinosinusitis.**. (2010) **10** 171-174. PMID: 20425010
5. Bohmwald K., Galvez N. M. S., Rios M., Kalergis A. M.. **Neurologic alterations due to respiratory virus infections.**. (2018) **12**. DOI: 10.3389/fncel.2018.00386
6. Burks J. D., Conner A. K., Bonney P. A., Glenn C. A., Baker C. M., Boettcher L. B.. **Anatomy and white matter connections of the orbitofrontal gyrus.**. (2018) **128** 1865-1872. PMID: 28862541
7. Capuron L., Miller A. H.. **Immune system to brain signaling: Neuropsychopharmacological implications.**. (2011) **130** 226-238. DOI: 10.1016/j.pharmthera.2011.01.014
8. Cavanna A. E., Trimble M. R.. **The precuneus: A review of its functional anatomy and behavioural correlates.**. (2006) **129** 564-583. PMID: 16399806
9. Cheng W., Rolls E. T., Qiu J., Liu W., Tang Y. Q., Huang C. C.. **Medial reward and lateral non-reward orbitofrontal cortex circuits change in opposite directions in depression.**. (2016) **139** 3296-3309. DOI: 10.1093/brain/aww255
10. Cheng W., Rolls E. T., Ruan H. T., Feng J. F.. **Functional connectivities in the brain that mediate the association between depressive problems and sleep quality.**. (2018) **75** 1052-1061. DOI: 10.1001/jamapsychiatry.2018.1941
11. Cui L. B. A., Liu L., Guo F., Chen Y. C., Chen G., Xi M.. **Disturbed brain activity in resting-state networks of patients with first-episode schizophrenia with auditory verbal hallucinations: A cross-sectional functional MR imaging study.**. (2017) **283** 809-818. DOI: 10.1148/radiol.2016160938
12. Dehdar K., Mahdidoust S., Salimi M., Gholami-Mahtaj L., Nazari M., Mohammadi S.. **Allergen-induced anxiety-like behavior is associated with disruption of medial prefrontal cortex–amygdala circuit.**. (2019) **9**. DOI: 10.1038/s41598-019-55539-3
13. Fang B. J., Tonelli L. H., Soriano J. J., Postolache T. T.. **Disturbed sleep: Linking allergic rhinitis, mood and suicidal behavior.**. (2010) **2** 30-46. DOI: 10.2741/s44
14. Fokkens W. J., Lund V. J., Hopkins C., Hellings P. W., Kern R., Reitsma S.. **European position paper on rhinosinusitis and nasal polyps 2020.**. (2020) **5** 1-464
15. Fukushima S., Nishikawa K., Furube E., Muneoka S., Ono K., Takebayashi H.. **Oligodendrogenesis in the fornix of adult mouse brain; the effect of LPS-induced inflammatory stimulation.**. (2015) **1627** 52-69. DOI: 10.1016/j.brainres.2015.09.011
16. Gao J., Zhou Y., Yang X. Y., Luo J., Meng F. Q., Zheng D.. **Abnormalities within and beyond the cortico-striato-thalamo-cortical circuitry in medication-free patients with OCD revealed by the fractional amplitude of low-frequency fluctuations and resting-state functional connectivity.**. (2019) **712**. DOI: 10.1016/j.neulet.2019.134449
17. Gao W. X., Ou C. Q., Fang S. B., Sun Y. Q., Zhang H., Cheng L.. **Occupational and environmental risk factors for chronic rhinosinusitis in China: A multicentre cross-sectional study.**. (2016) **17**. DOI: 10.1186/s12931-016-0366-z
18. Gao Z., Chen X. X., Xiang R., Zhang W., Tan L., Fan W. J.. **Changes in resting-state spontaneous brain activity in patients with allergic rhinitis: A pilot neuroimaging study.**. (2021) **15**. DOI: 10.3389/fnins.2021.697299
19. Ghadami M. R.. **Chronic rhinosinusitis and increased risk of depression and anxiety: The role of impaired sleep quality.**. (2019) **145** 689-690. DOI: 10.1001/jamaoto.2019.0916
20. Gong J. Y., Wang J. J., Qiu S. J., Chen P., Luo Z. Y., Wang J. R.. **Common and distinct patterns of intrinsic brain activity alterations in major depression and bipolar disorder: Voxel-based meta-analysis.**. (2020) **10**. DOI: 10.1038/s41398-020-01036-5
21. Guo P. F., Hu S. W., Jiang X. L., Zheng H. Y., Mo D. M., Cao X. M.. **Associations of neurocognition and social cognition with brain structure and function in early-onset schizophrenia.**. (2022) **13**. DOI: 10.3389/fpsyt.2022.798105
22. Harrass S., Yi C. J., Chen H.. **Chronic rhinosinusitis and Alzheimer’s disease-a possible role for the nasal microbiome in causing neurodegeneration in the elderly.**. (2021) **22**. DOI: 10.3390/ijms222011207
23. Hoggard M., Nocera A., Biswas K., Taylor M. W., Douglas R. G., Bleier B. S.. **The sinonasal microbiota, neural signaling, and depression in chronic rhinosinusitis.**. (2018) **8** 394-405. DOI: 10.1002/alr.22074
24. Hou J. C., Mohanty R., Nair V. A., Dodd K., Beniwal-Patel P., Saha S.. **Alterations in resting-state functional connectivity in patients with Crohn’s disease in remission.**. (2019) **9**
25. Jafari A., Xavier L. D., Bernstein J. D., Simonyan K., Bleier B. S.. **Association of sinonasal inflammation with functional brain connectivity.**. (2021) **147** 534-543. DOI: 10.1001/jamaoto.2021.0204
26. Jung H. J., Lee J. Y., Choi Y. S., Choi H. G., Wee J. H.. **Chronic rhinosinusitis and progression of cognitive impairment in dementia.**. (2021) **138** 147-151. DOI: 10.1016/j.anorl.2020.05.017
27. Kim J. Y., Ko I., Kim M. S., Kim D. W., Cho B. J., Kim D. K.. **Relationship of chronic rhinosinusitis with asthma, myocardial infarction, stroke, anxiety, and depression.**. (2020) **8** 721-727.e3. DOI: 10.1016/j.jaip.2019.09.001
28. Kim J. Y., Ko I., Kim M. S., Yu M. S., Cho B. J., Kim D. K.. **Association of chronic rhinosinusitis with depression and anxiety in a nationwide insurance population.**. (2019) **145** 313-319. DOI: 10.1001/jamaoto.2018.4103
29. Kim Y. K., Na K. S., Myint A. M., Leonard B. E.. **The role of pro-inflammatory cytokines in neuroinflammation, neurogenesis and the neuroendocrine system in major depression.**. (2016) **64** 277-284. PMID: 26111720
30. Li R., Utevsky A. V., Huettel S. A., Braams B. R., Peters S., Crone E. A.. **Developmental maturation of the precuneus as a functional core of the default mode network.**. (2019) **31** 1506-1519. PMID: 31112473
31. Lin H. W., Bhattacharyya N.. **Diagnostic and staging accuracy of magnetic resonance imaging for the assessment of sinonasal disease.**. (2009) **23** 36-39. DOI: 10.2500/ajra.2009.23.3260
32. Lin S., Han Y., Duan S., Wu P., Wu N., Xie T.. **Aberrant spontaneous brain activity in coronary heart disease using fractional amplitude of low-frequency fluctuations: A preliminary resting-state functional MRI study.**. (2022) **2022**. DOI: 10.1155/2022/2501886
33. Liu J., Cao L. X., Li H. L., Gao Y. X., Bu X., Liang K. L.. **Abnormal resting-state functional connectivity in patients with obsessive-compulsive disorder: A systematic review and meta-analysis.**. (2022) **135**
34. Lou H. C., Luber B., Crupain M., Keenan J. P., Nowak M., Kjaer T. W.. **Parietal cortex and representation of the mental self.**. (2004) **101** 6827-6832. PMID: 15096584
35. Lu S. J., Gao W. J., Wei Z. G., Wang D. D., Hu S. H., Huang M. L.. **Intrinsic brain abnormalities in young healthy adults with childhood trauma: A resting-state functional magnetic resonance imaging study of regional homogeneity and functional connectivity.**. (2017) **51** 614-623. DOI: 10.1177/0004867416671415
36. Lui S., Zhou X. H. J., Sweeney J. A., Gong Q. Y.. **Psychoradiology: The Frontier of neuroimaging in psychiatry.**. (2016) **281** 357-372. PMID: 27755933
37. Lv H., Liu P. Q., Zhou F. W., Gao Z., Fan W. J., Xu Y.. **TAK-242 ameliorates olfactory dysfunction in a mouse model of allergic rhinitis by inhibiting neuroinflammation in the olfactory bulb.**. (2021) **92**. DOI: 10.1016/j.intimp.2021.107368
38. Lyoo I. K., Kim M. J., Stoll A. L., Demopulos C. M., Parow A. M., Dager S. R.. **Frontal lobe gray matter density decreases in bipolar I disorder.**. (2004) **55** 648-651. PMID: 15013835
39. McAfoose J., Baune B. T.. **Evidence for a cytokine model of cognitive function.**. (2009) **33** 355-366. PMID: 18996146
40. Milad M. R., Rauch S. L.. **The role of the orbitofrontal cortex in anxiety disorders.**. (2007) **1121** 546-561. PMID: 17698998
41. Price J. L., Drevets W. C.. **Neurocircuitry of mood disorders.**. (2010) **35** 192-216. PMID: 19693001
42. Qi R. F., Zhang L. J., Wu S. Y., Zhong J. H., Zhang Z. Q., Zhong Y.. **Altered resting-state brain activity at functional MR imaging during the progression of hepatic encephalopathy.**. (2012) **264** 187-195. PMID: 22509052
43. Salimi M., Ghazvineh S., Nazari M., Dehdar K., Garousi M., Zare M.. **Allergic rhinitis impairs working memory in association with drop of hippocampal–prefrontal coupling.**. (2021) **1758**. DOI: 10.1016/j.brainres.2021.147368
44. Schoenbaum G., Roesch M. R., Stalnaker T. A.. **Orbitofrontal cortex, decision-making and drug addiction.**. (2006) **29** 116-124. PMID: 16406092
45. Sedaghat A. R.. **Chronic rhinosinusitis.**. (2017) **96** 500-506. PMID: 29094889
46. Sedaghat A. R., Kuan E. C., Scadding G. K.. **Epidemiology of chronic rhinosinusitis: Prevalence and risk factors.**. (2022) **10** 1395-1403. PMID: 35092822
47. Simon O., Kherif F., Flandin G., Poline J. B., Riviere D., Mangin J. F.. **Automatized clustering and functional geometry of human parietofrontal networks for language, space, and number.**. (2004) **23** 1192-1202. DOI: 10.1016/j.neuroimage.2004.09.023
48. Subramaniam P., DiMuzio J., McGlade E., Yurgelun-Todd D.. **Orbitofrontal connectivity is associated with depression and anxiety in marijuana-using adolescents.**. (2016) **79**. DOI: 10.1016/j.jad.2018.07.002
49. Sun T., Shu H. Y., Wu J. L., Su T., Liu Y. J., Zhang L. J.. **Investigation of changes in the activity and function of dry eye-associated brain regions using the amplitude of low-frequency fluctuations method.**. (2022) **42**. DOI: 10.1042/BSR20210941
50. Wang Y., Shao Y., Shi W. Q., Jiang L., Wang X. Y., Zhu P. W.. **The predictive potential of altered spontaneous brain activity patterns in diabetic retinopathy and nephropathy.**. (2019) **10** 249-259. DOI: 10.1007/s13167-019-00171-4
51. Wikenheiser A. M., Marrero-Garcia Y., Schoenbaum G.. **Suppression of ventral hippocampal output impairs integrated orbitofrontal encoding of task structure.**. (2017) **95** 1197-1207.e3. DOI: 10.1016/j.neuron.2017.08.003
52. Xue S. W., Lee T. W., Guo Y. H.. **Spontaneous activity in medial orbitofrontal cortex correlates with trait anxiety in healthy male adults.**. (2018) **19** 643-653. DOI: 10.1631/jzus.B1700481
53. Zang Y. F., He Y., Zhu C. Z., Cao Q. J., Sui M. Q., Liang M.. **Altered baseline brain activity in children with ADHD revealed by resting-state functional MRI.**. (2007) **29** 83-91. PMID: 16919409
54. Zang Y. F., Jiang T. Z., Lu Y. L., He Y., Tian L. X.. **Regional homogeneity approach to fMRI data analysis.**. (2004) **22** 394-400. PMID: 15110032
55. Zhang S., Li C. S. R.. **Functional connectivity mapping of the human precuneus by resting state fMRI.**. (2012) **59** 3548-3562. PMID: 22116037
|
---
title: Development and validation of a genomic nomogram based on a ceRNA network for
comprehensive analysis of obstructive sleep apnea
authors:
- Wang Liu
- Xishi Sun
- Jiewen Huang
- Jinjian Zhang
- Zhengshi Liang
- Jinru Zhu
- Tao Chen
- Yu Zeng
- Min Peng
- Xiongbin Li
- Lijuan Zeng
- Wei Lei
- Junfen Cheng
journal: Frontiers in Genetics
year: 2023
pmcid: PMC10036397
doi: 10.3389/fgene.2023.1084552
license: CC BY 4.0
---
# Development and validation of a genomic nomogram based on a ceRNA network for comprehensive analysis of obstructive sleep apnea
## Abstract
Objectives: Some ceRNA associated with lncRNA have been considered as possible diagnostic and therapeutic biomarkers for obstructive sleep apnea (OSA). We intend to identify the potential hub genes for the development of OSA, which will provide a foundation for the study of the molecular mechanism underlying OSA and for the diagnosis and treatment of OSA.
Methods: We collected plasma samples from OSA patients and healthy controls for the detection of ceRNA using a chip. Based on the differential expression of lncRNA, we identified the target genes of miRNA that bind to lncRNAs. We then constructed lncRNA-related ceRNA networks, performed functional enrichment analysis and protein-protein interaction analysis, and performed internal and external validation of the expression levels of stable hub genes. Then, we conducted LASSO regression analysis on the stable hub genes, selected relatively significant genes to construct a simple and easy-to-use nomogram, validated the nomogram, and constructed the core ceRNA sub-network of key genes.
Results: We successfully identified 282 DElncRNAs and 380 DEmRNAs through differential analysis, and we constructed an OSA-related ceRNA network consisting of 292 miRNA-lncRNAs and 41 miRNA-mRNAs. Through PPI and hub gene selection, we obtained 7 additional robust hub genes, CCND2, WT1, E2F2, IRF1, BAZ2A, LAMC1, and DAB2. Using LASSO regression analysis, we created a nomogram with four predictors (CCND2, WT1, E2F2, and IRF1), and its area under the curve (AUC) is 1. Finally, we constructed a core ceRNA sub-network composed of 74 miRNA-lncRNA and 7 miRNA-mRNA nodes.
Conclusion: Our study provides a new foundation for elucidating the molecular mechanism of lncRNA in OSA and for diagnosing and treating OSA.
## 1 Introduction
Obstructive sleep apnea (OSA) is the repeated partial or complete collapse of the upper airway during sleep, resulting in intermittent hypoxemia, which influences the onset and progression of the disease (Duarte et al., 2020; Zhou et al., 2021). OSA may be associated with multiple system diseases over time, including hypertension, coronary heart disease, type 2 diabetes, cerebral infarction, Alzheimer’s disease, Parkinson’s disease, and non-alcoholic fatty liver disease (Floras, 2015; Salman et al., 2020; Zhou et al., 2021). OSA affects at least $2\%$–$4\%$ of the adult population, and the prevalence of OSA in patients aged 65 and older exceeds $30\%$ (Ekin et al., 2021). In clinical practice, weight loss therapy, positive airway pressure ventilation therapy, surgical therapy, oral appliance therapy, and drug therapy are frequently used, but the results are not satisfactory (Kendzerska et al., 2021). High prevalence and ineffective treatment will have a negative impact on the quality of life of patients. Therefore, there is an urgent need to identify new biomarkers and treatment targets for OSA that are more effective.
Non-coding RNA (ncRNA) is a functional RNA molecule that is not translated into protein, such as microRNA (miRNA), long non-coding RNA (lncRNA), circular RNA (circRNA), intronic RNA, small interfering RNA (siRNA), small nucleolar RNA (snoRNA), and piwi-interacting RNA (piRNA), among others (Matsui and Corey, 2017). Long non-coding RNAs (lncRNAs) are a class of transcripts with a length of >200 nucleotides that are incapable of encoding proteins but play a crucial role in gene regulation, biological processes, and a variety of diseases (Zhou et al., 2021). Recent studies have found that lncRNA XIST promotes the occurrence and development of OSAHS by downregulating the expression of GRα in the adenoids of OSAHS children, which may provide a potential therapeutic target for OSAHS (Zhou et al., 2021). MicroRNA (miRNA) is a non-coding, single-stranded molecule of approximately 22 nucleotides that fine-tunes the expression of its target genes after transcription by interfering with the 3′-UTR region of mRNA (Haenisch et al., 2015; Li and Yuan, 2020). This interference results in mRNA degradation or inhibition of protein translation. Therefore, misregulation of miRNA will result in changes in protein expression, leading to disease development (Li and Yuan, 2020). Overexpression of miR-107 inhibits the expression of hypoxia-inducible factor 1 (HIF-1b) and hypoxia signaling [10] (Li et al., 2017); conversely, overexpression of miR-107 increases the expression of hypoxia-inducible factor 1 (HIF-1a) and hypoxia signaling.
There is evidence that the activities of lncRNA and miRNA are intertwined through a variety of complex mechanisms (Yamamura et al., 2018), as a result of the advancement of research. Among the mechanisms is the function of lncRNA as competing endogenous RNA (ceRNA), which compete with mRNAs for binding to miRNA binding sites, thereby negatively regulating miRNA and its target genes (Xu et al., 2021). In mice, cardiac apoptosis-related lncRNA CARL can acquire miR-539, which indirectly upregulates its target PHB2 and regulates apoptosis and mitochondrial fission (Rey et al., 2021). The highly expressed lncRNA-Adi in rat adipocytes interacts with miR-449a, which enhances the expression of the miRNA target CDK6, and then participates in the regulation of the formation of beige cellular tissue (Hou et al., 2018; Chen et al., 2020). The lncRNA MALAT1 can play a regulatory role by acting on miR-224-5p, thereby regulating the hippocampal NLRP3/IL-1β pathway and inhibiting the hippocampus inflammatory response in type 2 diabetic patients with OSA (Du et al., 2020).
In this study, plasma samples were collected from OSA patients and a normal control group, plasma RNA was extracted, and ceRNA chip detection was performed. Based on the differential expression of lncRNA, we identified the target genes of miRNA that bind to lncRNAs. We then constructed lncRNA-related ceRNA networks, performed functional enrichment analysis and protein-protein interaction analysis, and performed internal and external validation of the expression levels of stable hub genes. Then, we performed LASSO regression analysis on the stable hub genes, selected relatively significant genes to construct a simple and easy-to-use nomogram, validated the nomogram, and constructed the core ceRNA sub-network of key genes. This study identified potential target genes of miRNA that may be involved in the combination of lncRNA in OSA, providing a foundation for the study of the pathogenesis of OSA and the diagnosis and treatment of OSA.
## 2.1 Study subjects
In this study, 54 participants were recruited between December 2020 and May 2021. Patients who met the inclusion and exclusion criteria were selected, and the final sample consisted of 9 volunteers; 6 OSA patients and 3 healthy volunteers who served as the control group. All participants were subjected to a Watch-PAT examination (the specification model is Watch-PAT 200, and the manufacturer is Israel Ita), in addition to anthropometric measurements, blood pressure measurements, and blood biochemical tests. According to the inclusion and exclusion criteria, eligible patients were selected. Inclusion criteria: 1) The experimental group was comprised of volunteers whose Watch-PAT test result indicated OSA; 2) Patients aged between 28 and 36 years old (including 28 and 36 years old); 3) Patients with the capacity to act independently and consent to sign the informed consent form; and 4) Patients with a total sleep time >4 h5) Healthy volunteers were recruited for the control group. Exclusion criteria: 1) Patients with coronary heart disease, hypertension, diabetes, kidney disease, chronic pulmonary disease, or cerebrovascular disease; 2) patients with severe organ failure; 3) a history of brain tumors or epilepsy; 4) patients with various mental and psychological diseases who were taking sedatives and sleeping drugs; 5) OSA patients who had previously received treatment: The Medical Ethics Committee of the Second Affiliated Hospital of Guangdong Medical University approved this study (ethics number: GDEFEY2020LS030).
## 2.2 Basic data collection
In our study, we collected the patient’s name, age, gender, neck circumference, waist circumference, weight, height, and blood pressure. We then calculated the patient’s body mass index (BMI) using the formula: weight (kg)/height (m)2 (BMI = kg/m2). The researchers then modify the NoSAS score using general information. NoSAS (Marti-Soler et al., 2016) includes 5 questions: The first issue is that a neck circumference ≥40 cm is worth 4 points; the second issue is the BMI value range: 25 ≤ BMI <30 kg/m2 is worth 3 points, BMI ≥30 kg/m2 is worth 5 points, and snoring is worth 2 points. The answer to the fourth question is that age ≥55 is worth 4 points; the score for the fifth question for male patients is 2. If the NoSAS score is ≥ 8, it indicates that OSA patients are at high risk.
## 2.3 Watch-PAT detection
The primary function of the Watch-PAT sleep monitoring device is to detect sleep-disordered breathing. On the day of the examination, participants were instructed to abstain from alcohol, caffeine, and sleep aids. The Watch-PAT sleep monitoring device primarily monitors PAT, heart rate, blood oxygen saturation, snoring, body position, and additional sleep or waking stage parameters. The software analyzes the changes in the PAT signal throughout the entire sleep process. The sleep time of all patients should be monitored for at least 7 h. The diagnostic criteria for OSA were defined as apnea hypopnea index (AHI) ≥5 times/h, but the criteria were further subdivided into mild OSA (5 ≤ AHI <15 times/h), moderate OSA (15 ≤ AHI <30 times/h), and severe OSA (AHI ≥30 times/h).
## 2.4 Blood sample collection
Our study participants were divided into three groups: the normal group, the training cohort (obese OSA), and the internal validation cohort (non-obese OSA). There were three members in each group. All participants provided two blood samples at 8:00 a.m., following a full night of Watch-PAT sleep monitoring and overnight fasting. Then, we collected the blood into EDTA purple anticoagulant tubes; one was sent for blood glucose and blood lipid detection, while the other was used for ceRNA chip detection. Within 60 min of blood collection, the blood was centrifuged for 10 min at 3,000 g to separate plasma. The supernatant was transferred to an RNase-free Eppendorf tube and stored at −80°C until RNA extraction.
## 2.5 ceRNA expression profile
Total RNA was isolated using RNeasy Total RNA Isolation Kit (Qiagen, GmBH, Germany)/TRIzol reagent (Life technologies, Carlsbad, CA, United States) per the manufacturer’s instructions, purified using an RNeasy Mini Kit (Qiagen, GmBH, Germany), and quantified using Nanodrop. Using the Agilent Bioanalyzer 2,100 (Agilent technologies, Santa Clara, CA, United States), the fragment distribution of total RNA was analyzed. The RNA from each group was then used to generate biotinylated cRNA targets for the Sino Human ceRNA array V3.0. cRNA targets that were biotinylated were then hybridized with the slides. The Agilent Microarray Scanner was used to scan the slides after hybridization (Agilent technologies, Santa Clara, CA, United States). Using Feature Extraction software 10.7, data was extracted (Agilent technologies, Santa Clara, CA, United States). Quantile algorithm, R package “Limma” were used to normalize the raw data (Ritchie et al., 2015). At Sinotech Genomics Corporation, the microarray experiments were conducted according to the protocol developed by Agilent technologies Inc. Genes exhibiting a fold change of at least 1 were chosen for further examination.
## 2.6 Identification of differentially expressed lncRNA and mRNA between OSA group and normal group
The “limma” package in R is used to identify differentially expressed lncRNA and mRNA between OSA and normal groups, an efficient bioinformatics analysis technique. The statistical significance thresholds for differentially expressed lncRNA (DElncRNA) and mRNA (DEmRNA) samples were determined to be $p \leq 0.05$ and |log2FC| >1. Using these screening conditions, we identified the differential expression of lncRNAs and mRNAs between OSA patients and healthy controls. To reveal the sample specificity of differentially expressed lncRNA and mRNA, we utilized volcano plots and the “Pheatmap” package in R software (Khomtchouk et al., 2014) to conduct supervised hierarchical clustering based on the Euclide distance of the lncRNA and mRNA in the samples (Mielke and Berry, 2003; Bien and Tibshirani, 2011). Herein, P.Adjustp < adjustP & logFC > logFoldChange is an up-regulating gene, and P.Adjustp < logFC <(-log Fold Change) is a down-regulating gene.
## 2.7 Target gene prediction of differentially expressed miRNA
Identification of target genes is crucial for defining the function of miRNA. Due to the lack of miRNA information, the miRcode database (Jeggari et al., 2012) was utilized to predict the DElncRNA-targeted miRNA. Then, we predicted the target genes of differentially expressed miRNA using the mirtarBase (Chou et al., 2018), miRDB (Chen and Wang, 2020), and TargetScan (Lewis et al., 2003) databases. To improve the accuracy of miRNA prediction, we chose miRNA with common target genes across three databases.
## 2.8 Construction of OSA-related lncRNA-miRNA-mRNA network
The lncRNA that compete with miRNA for binding, miRNA of common target genes, and differential mRNA were incorporated into the ceRNA network, which was then visualized using Cytoscape software (Shannon et al., 2003) (version 3.8.2; http://cytoscape.org), resulting in the lncRNA-miRNA-mRNA ceRNA network diagram.
## 2.9 GO and KEGG functional enrichment analysis
For further analysis of the three domains of potential cell component (CC), molecular function (MF), and biological process (BP) of gene modules, the “ClusterProfiler” package (Yu et al., 2012) in R software was utilized for GO and KEGG pathway enrichment analysis of target genes. Each category describes the biological function of genes at varying depths. Utilizing KEGG pathway enrichment analysis, the enrichment degree of differential genes in pathways was analyzed. When $p \leq 0.05$, the GO term and KEGG pathway were designated as being enriched.
## 2.10 Construction of protein-protein interaction network and identification of hub genes
To further investigate the interactions between the corresponding genes in the ceRNA network, we constructed a PPI network using the Interaction Gene Retrieval Search Tool (STRING) (Szklarczyk et al., 2015) 11.0 (http://string-db.org/). It was assigned a confidence score greater than 0.15. Nodes in the PPI network results represent proteins, while lines represent protein interactions. We installed the Hubba plugin for Cytoscape (Chin et al., 2014) after identifying the hub genes among the common genes (http://hub.iis.sinica.edu.tw/cytohubba/). CytoHubba is a visualization program that generates dense relationships using degree, tight centrality, and moderate centrality algorithms. Using CytoHubba, the central gene in the ceRNA network was identified. Then, the top 10 genes were extracted using the five hub gene screening methods of MCC, degree, EPC, closeness, and betweenness. The “venndiagram” package (Lam et al., 2016) of R software was used to create a Venn diagram, and the final hub gene was determined by the intersection of the Venn diagrams.
## 2.11 Expression level and correlation analysis of hub genes
To understand the expression levels of the final hub genes, we used a t-test to compare the differences between the normal group and the OSA group for the final hub genes. Then, to gain a better understanding of the relationship between hub genes, Pearson’s correlation analysis was utilized, and the “Corrplot” package was used to visualize the results (Zhang et al., 2021a).
## 2.12 Verification of hub gene expression level
To validate the differential expression of hub genes between the OSA group and the normal group, we used the other three OSA patients as the internal validation cohort and the GSE135917 data set downloaded from the GEO data frame as the external validation cohort and then extracted the expression data from both sets of data. First, enter the search term “obstructive sleep apnea” on the homepage of the gene expression database (GEO) (http://www.ncbi.nlm.nih.gov/geo) for retrieval; the only allowed species is “Homo sapiens”; the data type is “expression profiling by array”. The dataset (GSE135, 917) was selected and queried from the GEO database, and then the platform file (GPL6244-17, 930) and matrix file (GSE135, 917) were downloaded. The GSE135917 dataset contains 8 normal patients and 10 OSA patients. This dataset also includes gene expression samples from 48 OSA patients receiving treatment. The t-test was utilized to compare the differences between the two groups, and the R packages “Ggplot2” (Zhang et al., 2021b) and “RColorBrewer” (Jędroszka et al., 2017) were utilized to visualize the results. $p \leq 0.05$ was considered statistically significant.
## 2.13 Construction of a genomic model based on predictor selection
To further screen the hub genes associated with a high risk of OSA, we used the “Glmnet” package in the R software to conduct the least absolute shrinkage and selection operator (LASSO) logistic regression to reduce the dimension of the data and determine the best prediction characteristics of the training cohort (Friedman et al., 2010). Then, the genes with non-zero coefficient characteristics in the LASSO regression model were chosen, and the “Rms” package in the R software was used to develop nomograms for them to identify patients at risk for OSA (Zheng et al., 2021).
## 2.14 Verification of nomogram
The nomogram is bootstrapped (1,000 bootstrap samples) in order to calculate the relative corrected C-index, which is used to evaluate the nomogram’s discrimination (Wolbers et al., 2009). The C-index ranges from 0.5 to 1.0, with 0.5 representing random chance and 1.0 representing complete discrimination (Wolbers et al., 2009). Medcalc software was used to evaluate the diagnostic value of the OSA nomogram using receiver operating characteristic (ROC) curves, and internal and external validation cohort ROC curves were used for further validation.
## 2.15 Construction of core ceRNA subnetworks for key genes
We remapped the validated key genes and their associated lncRNA and miRNA into ceRNA networks, which were visualized using the Cytoscape software.
## 2.16 Statistical methods
For correlation analysis, SPSS 26.0 statistical software, R 4.0.5 software, and Medcalc software were utilized. Counting data were expressed as frequency, while measurement data were expressed as mean ± standard deviation. t-test was used to analyze measurement data, while the chi-square test was used to analyze counting data. $p \leq 0.05$ was considered statistically significant.
## 3.1 Basic characteristics of patients
In our study, we included 6 OSA patients and 3 healthy subjects, with 3 of them serving as the normal group (BMI: 22.77 ± 2.15), 3 obese OSA patients serving as the training cohort (BMI: 22.77 ± 2.15 vs. 31.40 ± 1.18, $$p \leq 0.004$$), and 3 non-obese OSA patients serving as the internal validation cohort (BMI: 22.77 ± 2.15 vs. 25.00 ± 1.06, $$p \leq 0.181$$). The AHI of the normal group was 1.53 ± 1.20 (times/hour), the obese group AHI was 89.93 ± 44.04 (times/hour), and the non-obese group AHI was 47.13 ± 22.01 (times/hour). In addition, the NoSAS score for the normal group was (4.00 ± 0.00), for the obese OSA group it was (13.00 ± 0.00), and for the non-obese OSA group it was (9.00 ± 1.73). The experimental group (obese OSA and non-obese OSA) consisted of patients with a high risk of OSA. There were almost no statistically significant differences between the three groups in terms of age, blood glucose, total cholesterol, triglyceride, high-density lipoprotein, and low-density lipoprotein ($p \leq 0.05$), and all patients were male, so they were well-matched. OSA patients had a larger neck circumference and waist circumference, and lower minimum oxygen saturation (Min-NOX) and mean oxygen saturation (Mean-NOX), and the difference between them was statistically significant ($p \leq 0.05$) (Table 1).
**TABLE 1**
| Unnamed: 0 | Training cohort | Training cohort.1 | Unnamed: 3 | Internal validation cohort | Internal validation cohort.1 | Unnamed: 6 |
| --- | --- | --- | --- | --- | --- | --- |
| | Normal group | Obesity OSA | P | Normal group | Non-obese OSA | P |
| Number | 3 | 3 | — | 3 | 3 | — |
| Male | 3 | 3 | — | 3 | 3 | — |
| Age (years) | 33.00 ± 4.36 | 32.00 ± 1.00 | 0.718 | 33.00 ± 4.36 | 31.00 ± 3.61 | 0.573 |
| BMI(Kg/m2) | 22.77 ± 2.15 | 31.40 ± 1.18 | 0.004 | 22.77 ± 2.15 | 25.00 ± 1.06 | 0.181 |
| NC(cm) | 37.67 ± 0.58 | 44.33 ± 3.51 | 0.032 | 37.67 ± 0.58 | 42.67 ± 0.58 | <0.001 |
| WC(cm) | 84.67 ± 5.03 | 104.67 ± 8.50 | 0.025 | 84.67 ± 5.03 | 93.33 ± 3.79 | 0.076 |
| SBP(mmHg) | 120.33 ± 9.24 | 134.67 ± 4.93 | 0.077 | 120.33 ± 9.24 | 128.33 ± 10.97 | 0.389 |
| DBP(mmHg) | 72.67 ± 4.04 | 85.67 ± 4.16 | 0.018 | 72.67 ± 4.04 | 82.67 ± 8.50 | 0.140 |
| HR(times/min) | 68.00 ± 5.29 | 80.33 ± 16.62 | 0.288 | 68.00 ± 5.29 | 85.33 ± 5.51 | 0.017 |
| NoSAS(points) | 4.00 ± 0.00 | 13.00 ± 0.00 | — | 4.00 ± 0.00 | 9.00 ± 1.73 | 0.007 |
| Blood sugar(mmol/L) | 5.08 ± 0.68 | 5.21 ± 0.37 | 0.795 | 5.08 ± 0.68 | 5.37 ± 0.51 | 0.584 |
| CHO(mmol/L) | 3.72 ± 0.67 | 4.16 ± 0.64 | 0.450 | 3.72 ± 0.67 | 5.18 ± 0.55 | 0.043 |
| TG(mmol/L) | 0.85 ± 0.09 | 1.49 ± 0.62 | 0.151 | 0.85 ± 0.09 | 3.02 ± 1.94 | 0.126 |
| HDL-C(mmol/L) | 0.89 ± 0.19 | 0.95 ± 0.18 | 0.720 | 0.89 ± 0.19 | 1.20 ± 0.13 | 0.074 |
| LDL-C(mmol/L) | 2.73 ± 0.57 | 3.11 ± 0.66 | 0.492 | 2.73 ± 0.57 | 3.75 ± 0.78 | 0.141 |
| AHI(times/hour) | 1.53 ± 1.20 | 89.93 ± 44.04 | 0.025 | 1.53 ± 1.20 | 47.13 ± 22.01 | 0.023 |
| Mean-NOX(%) | 97.33 ± 0.58 | 91.67 ± 1.15 | 0.002 | 97.33 ± 0.58 | 94.67 ± 0.58 | 0.005 |
| Min-NOX(%) | 89.00 ± 6.08 | 59.33 ± 15.95 | 0.040 | 89.00 ± 6.08 | 73.67 ± 5.51 | 0.032 |
## 3.2 Differential expression of lncRNA and mRNA between the OSA group and normal group
To identify the differentially expressed lncRNA and mRNA between the OSA group and the normal group, the total RNA of 3 normal groups and 3 OSA groups was analyzed using microarrays. The microarray analysis revealed that lncRNA and mRNA were altered in the OSA group in comparison to the normal group. According to the screening criteria, 282 differential lncRNA, including 166 upregulated and 116 downregulated lncRNA, and 380 differential mRNA, including 225 upregulated and 155 downregulated mRNA, were screened. *All* gene expressions in the dataset were represented as volcano maps and cluster heat maps (Figure 1). Each point on the volcanic map represents a gene, with blue points representing genes with low expression and red points representing genes with high expression. The cluster heatmap displayed the differentially expressed lncRNA and mRNA between the OSA group and the normal group.
**FIGURE 1:** *Expression analysis of DElncRNA and DEmRNA in the training cohort (volcano map and cluster heat map). (A)and (C) show the volcano plots of DElncRNA and DEmRNA, respectively. Red dots represent upregulated expression, and blue dots represent downregulated expression; (B) and (D) represent the clustering heatmaps of DElncRNA and DEmRNA, respectively. Red stripes show upregulated expression, while blue stripes show downregulated expression. DElncRNA: differentially expressed lncRNA; DEmRNA: differentially expressed mRNA.*
## 3.3 Construction of OSA-related lncRNA-miRNA-mRNA network
We used the miRcode database to predict the 2,468 targeted miRNA of DElncRNA, followed by the mirtarBase, miRDB, and TargetScan databases to eliminate the 1935 miRNA of the common target genes of these three databases. Subsequently, lncRNAs [2,468] competing with miRNAs for binding, miRNAs of common target genes [1935], and differential mRNAs [380] were incorporated into the ceRNA network (Figure 2), and an OSA-related ceRNA network consisting of 292 miRNA-lncRNA and 41 miRNA-mRNA was developed,. The ceRNA network contained 23 predicted miRNA, 40 DElncRNA, and 28 DEmRNA.
**FIGURE 2:** *Network diagram of lncRNA-miRNA-mRNA. The construction of the ceRNA network includes 40 DElncRNA, 23 predicted miRNA and 28 DEmRNA. Blue circles, green triangles and red diamonds represent DEmRNA, miRNA and DElncRNA, respectively. DElncRNA: differentially expressed lncRNA; DEmRNA: differentially expressed mRNA; miRNA: microRNA.*
## 3.4 Functional enrichment analysis
To gain a deeper understanding of the cellular processes mediated by target genes, a GO functional enrichment analysis was conducted to investigate the functional roles of their target genes in the fields of biological processes, cellular components, and molecular functions. Taking pvalueCutoff = 0.05 and qvalueCutoff = 0.05 as the criteria, we screened the enrichment analysis of the top 10 of p-value, which was mainly enriched in protein serine/threonine kinase (PKB, also known as Akt) activity, transcriptional co-regulatory activity, DNA-binding transcription factor binding, ubiquitin-like protein ligase binding, ubiquitin protein ligase binding, RNA polymerase II-specific DNA-binding transcription factor binding, phosphatase binding, protein phosphatase binding, SMAD binding and nuclear receptor activity (Figure 3).
**FIGURE 3:** *GO enrichment analysis diagram. GO enrichment analysis of target genes (p < 0.05 and q < 0.05), GO: Gene Ontology.*
Then, a KEGG pathway enrichment analysis was performed to determine which pathways were significantly enriched in target genes. Using pvalueCutoff = 0.05 and qvalueCutoff = 0.05 as the criteria, we screened the pathway analysis of the top 10 of p-value, which was predominantly enriched in mitogen activated protein kinase (MAPK) signaling pathway, miRNA in cancer, human cytomegalovirus infection, Hepatitis B, Kaposi Sarcoma-associated herpesvirus infection, cellular senescence, breast cancer, Yersinia infection, neurotrophin signaling pathway, and EGFR tyrosine kinase inhibitor resistance (Figure 4).
**FIGURE 4:** *KEGG pathway enrichment analysis diagram. KEGG pathway enrichment analysis of target genes (p < 0.05 and q < 0.05), KEGG: Kyoto Encyclopedia of Genes and Genomes.*
## 3.5 PPI network analysis and hub gene selection
To distinguish hub genes from common genes, we inserted the corresponding genes in ceRNA into the STRING database to build a PPI network (Figure 5). Subsequently, we uploaded the aforementioned PPI network relationship to Cytoscape and utilized its cytohubba plugin to identify hub genes. The MCC, EPC, Degree, Closeness, and Betweenness algorithms in Cytohubba were utilized to determine the top 10 hub genes. The scores of the five algorithms that screened the top 10 hub genes are shown in Table 2. To obtain a more robust hub gene, the top 10 hub genes identified by these five algorithms were intersected (Figure 6), resulting in the identification of 7 more robust hub genes, namely, CCND2, WT1, E2F2, IRF1, BAZ2A, LAMC1, and DAB2.
**FIGURE 5:** *Key gene protein interaction (PPI) network.* TABLE_PLACEHOLDER:TABLE 2 **FIGURE 6:** *Venn diagram of five methods to screen the top ten hub genes.*
## 3.6 Expression level and correlation analysis of hub genes
In order to comprehend the expression levels and correlations between these seven hub genes, we utilized the t-test to compare the variances of these seven hub genes between the normal and OSA groups. According to Table 3, the difference between the normal group and the OSA group was nearly statistically significant ($p \leq 0.05$). Then, we continued with the correlation matrix analysis of the seven hub genes. According to the classification of the Pearson correlation coefficient (r) (Hazra and Gogtay, 2016), the absolute values of 0–0.30, 0.30–0.50, 0.50–0.70, and 0.70–1.00 indicate “weak” correlation, “general” or “moderate” correlation, “good” correlation, and “strong” correlation, respectively. In addition, “$r = 0$” indicates “no correlation whatsoever” and “$r = 1.00$” indicates “complete correlation”. As shown in Figure 7, there was a good or strong correlation between the seven hub genes, with IRF1 being strongly positively correlated with E2F2 ($r = 0.97$) and strongly negatively correlated with DAB2 (r = −0.97), respectively.
## 3.7 Validation of hub gene expression levels
The downloaded GSE135917 dataset was preprocessed, and relevant information regarding the CCND2, WT1, E2F2, IRF1, BAZ2A, LAMC1, and DAB2 genes was then searched for. According to the expression profiling analysis of the GSE135917 data set, the expressions of CCND2, WT1, E2F2, and IRF1 in the OSA group were significantly decreased ($p \leq 0.05$), while there was no significant difference in BAZ2A, LAMC1, and DAB2 ($p \leq 0.05$) (Table 3). Figure 8 displays the results of the comparison between the two groups of CCND2, WT1, E2F2 and IRF1. According to the expression data of the internal validation cohort, although there was no statistically significant difference in the expression of CCND2, WT1, E2F2, and IRF1 ($p \leq 0.05$), this may be due to the difference caused by non-obese patients or the small sample size that did not achieve statistical significance. Nevertheless, according to the expression data from the external validation cohort, their expression differences were statistically significant ($p \leq 0.05$). The expression analysis of the four hub genes in the internal and external validation datasets was generally consistent with the training data set.
**FIGURE 8:** *Boxplots of differential expression of key genes in the external validation cohort. External validation of four core genes in the GSE135917 dataset. “0”indicates the normal group; “1″indicates the OSA group. OSA stands for Obstructive Sleep Apnea. (A) The relative expression level of CCND2 between OSA and normal groups; (B) The relative expression level of WT1 between OSA and normal groups; (C) The relative expression level of E2F2 between OSA and normal groups; (D) The relative expression level of IRF1 between OSA and normal group. Data are presented as medians with interquartile ranges. t-test was used to compare relative expression levels between the two groups.*
## 3.8 Construction of a genomic model based on predictor selection
LASSO regression is appropriate for high-dimensional data regression. The compression coefficient is obtained by constructing a penalty function, and some compression coefficients are set to zero so that the most significant predictors can be extracted from the main data set and a more precise linear regression model can be developed (Friedman et al., 2010). In this study, a coefficient distribution curve was generated by calculating each subject’s risk score using a linear combination of factors weighted by the subject coefficient (Figure 9A). Figure 9B depicts the error plot for the cross-validation of the lasso regression model. The cross-validation error for the most regularized and parsimonious model was within 1 standard error of the minimum for 3 of the 7 variables. Four predictors (CCND2, WT1, E2F2, and IRF1) were ultimately chosen to develop an easy-to-use nomogram based on the expression level and correlation analysis of hub genes, the PPI network diagram, and its significance (Figure 10).
**FIGURE 9:** *Factor selection using the Least Absolute Shrinkage and Selection Operator (LASSO) logistic regression model. (A): The best parameter (lambda) selection in LASSO regression was selected using 10-fold cross-validation (through the minimum standard). Black vertical lines are drawn at the best values by using the minimum standard and one standard error of the minimum standard (1-SE standard). (B): Three features of the lasso coefficient profile. Coefficient profiles are plotted according to the logarithmic (λ) series.* **FIGURE 10:** *Nomogram for predicting obstructive sleep apnea syndrome. OSA: obstructive sleep apnea.*
## 3.9 Verification of nomogram
Using 1,000 bootstrap analyses, the validity of the nomogram was determined. In predicting OSA, the C-index of the nomogram for both the training cohort and the internal and external validation cohorts was 1, indicating that the model was sufficiently accurate; consequently, the model is appropriate for predicting OSA patients. Since there were fewer than 10 cases in the training and internal validation groups, no calibration plot could be generated. However, the calibration plot for external validation (Figure 11) revealed a relatively strong correlation between observed and predicted OSA. In addition, ROC curve analysis was used to evaluate OSA when the AHI cut-off value was 5 times/h based on the current nomogram. When the AUC of the nomogram was at the optimal cutoff point, regardless of whether the cohort was the training cohort, the internal validation cohort, or the external validation cohort, the ROC curve indicated that the diagnostic performance of the nomogram was improved (AUC = 1, AUC = 1, and AUC = 1), and their specificity and sensitivity were both $100\%$ (Figure 12).
**FIGURE 11:** *Calibration curve of nomogram in external validation cohort.* **FIGURE 12:** *ROC curve of nomogram. (A): Nomogram ROC of the training cohort; (B) Nomogram ROC of the internal validation cohort; (C) Nomogram ROC of the external validation cohort.*
## 3.10 Construction of core ceRNA subnetworks for key genes
Based on the ceRNA network, we remapped the four key genes of CCND2, WT1, E2F2, and IRF1, and related lncRNA and miRNA into the ceRNA network, thereby establishing a core ceRNA sub-network (Figure 13). It contained 74 miRNA-lncRNA and 7 miRNA-mRNA edges and 40 nodes [29 lncRNA (NSMCE4A, ZBTB16, TCERG1, UCHL5, ESRRB, DNAH8, FAM189A2, GARNL3, SLC35F4, ZNF890P, CCDC112, STRBP, IRX2, BPTF, TLR1, CTAGE10P, FANCB, NRBP2, LDB1, MIR659, MIR524, VSTM5, MIR300, CCDC26, TPRG1-AS1, EIF3IP1, OSTCP1, OPCML-IT1, and OR7E2P), 7 miRNA(hsa-miR-1297, hsa-miR- 33a-3p, hsa-miR-17–5p, hsa-miR-20b-5p, hsa-miR-125b-5p, hsa-miR-301b-3p, and hsa-miR-212–3p) and 4 mRNA(CCND2, WT1, E2F2 and IRF1)]. Figure 13 demonstrates that STRBP was the core lncRNA, capable of binding with hsa-miR-1297, hsa-miR-17–5p, hsa-miR-20b-5p, hsa-miR-125b-5p, hsa-miR-301b -3p, and hsa-miR-212–3p, which in turn affected four genes: CCND2, WT1, E2F2, and IRF1.
**FIGURE 13:** *Core ceRNA sub-network diagram of key genes. The construction of the ceRNA network included 29 DElncRNA, 7 predicted miRNA and 4 DEmRNA. Blue circles, green triangles and red diamonds represent DEmRNA, miRNA and DElncRNA, respectively. DElncRNA: differentially expressed lncRNA; DEmRNA: differentially expressed mRNA; miRNA: microRNA.*
## 4 Discussion
Obstructive sleep apnea (OSA) is the most prevalent respiratory sleep disorder, affecting up to 1 billion individuals worldwide (Kendzerska et al., 2021). In addition, the patient’s compliance with the diagnosis and treatment of OSA is poor, which can easily lead to the disease’s progression, which will impose a heavy burden on society and present treatment challenges for physicians. In order to prevent the progression of OSA, effective screening technologies, accurate diagnosis, and treatment remain crucial. Recent studies have highlighted the regulatory role of lncRNA as ceRNA in the development and occurrence of chronic intermittent hypoxia (Ge et al., 2019; Zhang et al., 2020a; Hu et al., 2021). We constructed an lncRNA-related ceRNA network based on the results of the OSA ceRNA chip to identify new targets with potential diagnostic or therapeutic value for OSA, and then validated the new targets.
In this study, we successfully identified 282 DElncRNA and 380 DEmRNA through the differential expression of lncRNA and mRNA in order to reduce the error interference between the OSA group and the normal group. Initially, we matched the key information of the two groups in order to reduce the error interference between the OSA group and the normal group. Combining lncRNA that compete with miRNA for binding, miRNA of common target genes, and differential mRNA yielded a ceRNA network. The target genes in the ceRNA network were then analyzed for enrichment in GO terms and KEGG pathways. GO enrichment analysis revealed that target genes were primarily enriched in protein serine/threonine kinase activity, transcriptional co-regulatory activity, DNA-binding transcription factor binding, ubiquitin-like protein ligase binding, ubiquitin protein ligase binding, and so on. Protein serine/threonine kinase activity was, without a doubt, the most important GO pathway. PKB consists of three widely expressed isoforms (PKBα, PKBβ, and PKBγ; also known as Akt1, Akt2, and Akt3, respectively), and PKBβ may be an important mediator in the insulin signaling transduction pathway (Lawlor and Alessi, 2001). Analysis of KEGG pathway annotations revealed that target genes were predominantly involved in the MAPK signaling pathway, miRNA in cancer, human cytomegalovirus infection, Hepatitis B and Kaposi sarcoma-associated herpesvirus infection, etc. However, the MAPK signaling pathway was the pathway with the greatest enrichment. MAPK is a ubiquitous family of proline-directed protein serine/threonine kinases that are required for the sequential transduction of biological signals from the cell membrane to the nucleus (Broom et al., 2009). OSA-induced intermittent hypoxia has been reported to excessively and persistently activate the MAPK signaling pathway (Zhao et al., 2016). Some studies have also demonstrated that at the cellular level, chronic intermittent hypoxia alters the equilibrium between the phosphatidylinositol 3-kinase (PI3K)-dependent insulin signaling pathway, which regulates the production of endothelial nitric oxide (NO), and the activation of the mitogen-activated protein kinase (MAPK)-dependent insulin signaling pathway, which regulates the secretion of vasoconstrictor endothelin-1 (ET-1), thus affecting vascular endothelial dysfunction (Sharma et al., 2018). Clearly, protein serine/threonine kinase activity plays a pivotal role in the MAPK signaling pathway during the progression of OSA disease.
Then, we constructed a network of lncRNA-related ceRNAs and identified 28 hub genes. Then, a PPI network was created, and the cytohubba plug-in was utilized to identify 7 stable hub genes (CCND2, WT1, E2F2, IRF1, BAZ2A, LAMC1, and DAB2). IRF1 was strongly positively correlated with E2F2 ($r = 0.97$), and IRF1 was strongly negatively correlated with DAB2 (r = −0.97), according to the results of the correlation analysis. In addition, the levels of expression of these seven hub genes were validated using both internal and external validation datasets. LASSO regression was then applied to the seven hub genes. In conjunction with the expression level and correlation analysis of hub genes, as well as the PPI network diagram and its significance, the number of candidate variables was reduced to 4 potential predictors (CCND2, WT1, E2F2, and IRF1). It has been reported that CCND2 regulates cell proliferation by binding to cyclin-dependent kinase 4 (CDK4) or cyclin-dependent kinase 6 (CDK6) to form a complex required for the G1/S cell cycle (Chermuła et al., 2019). Furthermore, it has been reported that CCND2 is one of the most important biomarkers of endothelial dysfunction (Zhu et al., 2021). WT1 is a transcription factor that is unique among transcription factors because it functions as both a tumor suppressor and an embryonic development regulator (Krueger et al., 2019). It has been demonstrated that the expression of WT1, which is upregulated by hypoxia in endothelial cells, and the proliferation of endothelial cells are regulated by WT1 (Duim et al., 2015). E2F2, a member of the E2F family, regulates the cell cycle by inhibiting or activating cell cycle regulators, such as cyclins, cyclin-dependent kinases (CDKs), and checkpoint regulators (Li et al., 2014). Experiments have demonstrated that E2F1 inhibits angiogenesis and endothelial cell proliferation following ischemic injury by suppressing the expression of pro-angiogenic cytokines, vascular endothelial growth factor, and placental growth factor (Zhou et al., 2013). Interferon regulatory factor-1 (IRF1), a member of the IRF family of transcription factors, regulates gene expression during inflammation, immune response, cell proliferation, cell cycle progression, T cell differentiation, and DNA damage (Huang et al., 2019). Studies indicate that hypoxia can regulate the transcription of KPNA2 by simultaneously increasing the expression of its positive regulator E2F1 and inhibiting the expression of its negative regulator IRF1 (Huang et al., 2019). Clearly, these four predictors are associated with hypoxia, and intermittent hypoxia is one of the underlying causes of OSA. Consequently, CCND2, WT1, E2F2, and IRF1 target genes are all associated with OSA, and their inclusion in the model is reasonable.
Nomogram is a risk prediction tool that has been used for decades in medicine. By combining important predictors to predict clinical events and outcomes, it has been widely used to predict the risk and prognosis of various diseases (Chen et al., 2021; Song et al., 2021). Zhu et al. [ 2020] constructed a nomogram using seven hub RNA (HMMR, RNF24, RAP2A, S100A10, ARL2, has-miR-326, and hsa-miR-421). Consistently, the calibration curve demonstrated that the risk prediction model for hepatocellular carcinoma based on seven hub RNA had an adequate predictive effect. Song and Fu, 2019 developed a nomogram that included the target gene CXCR5, age, and stage. In the training set, the AUC values for the nomogram’s ability to predict the 3-year and 5-year overall survival of colorectal cancer were 0.749 and 0.805, respectively, whereas, the corresponding values in the validation set were 0.706 and 0.779, respectively. Shi et al. [ 2020] established a nomogram that included waist-to-hip ratio, smoking status, BMI, uric acid, Homeostasis Model Assessment 2 Insulin Resistance Index (HOMA2-IR), and history of fatty liver, and the AUC for distinguishing non-OSA patients from OSA patients was 0.855. Luo et al. [ 2015] established a nomogram that incorporated numerous subjective and objective variables (disease duration, smoking status, sleep difficulties, lack of energy, and waist circumference), and its discrimination accuracy for non-OSA, moderate-to-severe OSA, and severe OSA was $83.8\%$, $79.9\%$, and $80.5\%$, respectively. Based on these prediction models, it is evident that, regardless of whether the research is fundamental or clinical, nomograms are generally effective at predicting disease, which provides a foundation for identifying reliable targets. Verification of the nomogram is crucial for avoiding overfitting and determining generality (Iasonos et al., 2008). In our study, after proper calibration, the validation cohort’s calibration curve revealed that the actual occurrence probability was relatively close to the predicted occurrence probability. In addition, the training cohort, internal validation cohort, and external validation cohort all have C-index and AUC values of 1, indicating that the model is sufficiently accurate and diagnostically efficient. Thus, the validity of our nomogram has been established.
The nomogram has been recognized as a reliable tool for quantifying disease risk based on multivariate modeling procedures (Luo et al., 2015), and the nomogram constructed by target genes such as CCND2, WT1, E2F2, and IRF1 has been demonstrated to be robust; consequently, CCND2, WT1, E2F2, and IRF1 may be reliable OSA targets. In order to gain a deeper understanding of the regulatory mechanisms of these four target genes, we remapped them back into the ceRNA network, establishing a core ceRNA sub-network to search for important lncRNA or miRNA, or even lncRNA-miRNA-mRNA regulatory axes. The results demonstrated that STRBP was a core lncRNA that could bind competitively with hsa-miR-1297, hsa-miR-17-5p, hsa-miR-20b-5p, hsa-miR-125b-5p, hsa-miR-301b-3p, and hsa-miR-212-3p, thereby regulating the four genes CCND2, WT1, E2F2, and IRF1.
The ceRNA mechanism is a critical mode of regulation for cellular active metabolism and disease. STRBP is a sperm perinuclear RNA-binding protein that resides on chromosome 9q33, is widely expressed in lymph nodes, testis, and other tissues, and plays a crucial role in mammalian spermatogenesis (Zhang et al., 2020b). According to reports, STRBP can be detected in lung adenocarcinoma, breast cancer, and hematological malignancies (Zhang et al., 2020b). STRBP may be associated with body weight, according to studies (Wang et al., 2020). miRNA are commonly used in bioinformatics target prediction algorithms, and seed matching, sequence conservation, and thermodynamics of miRNA-mRNA interactions are commonly used to predict potential targets (Angerstein et al., 2012). miR-1297 inhibits KPNA2 in glioblastoma to negatively regulate metabolic reprogramming (Li and Yuan, 2020). According to previous studies, the CCND2 gene is a potential target of miR-1297, which inhibits the progression of colorectal cancer by inhibiting the transcription of CCND2 in colorectal cancer cells (Wang et al., 2017). It has been reported that miR-17-5p plays a role in the proliferation of pulmonary vascular smooth muscle cells, making it a potential new therapeutic target for the control of pulmonary hypertension (Yao et al., 2021).hsa-miR-17-5p may play a significant role in hypertrophic cardiomyopathy and is anticipated to serve as a diagnostic biomarker for this condition (Shi et al., 2019). Drobna et al. [ 2020] discovered that hsa-miR-20b-5p affected the expression of the tumor suppressor genes PTEN and BIM and regulated the survival of T-cell acute lymphoblastic leukemia cells in vitro. In addition, miR-20b-5p is predicted to regulate the TNFα signaling pathway, which supports the notion that diabetic retinopathy progression is primarily driven by retinal inflammation (Trotta et al., 2021). It has been reported that miR-125b-5p, a member of the miR-125 family, regulates the proliferation of differentiated tumor cells and may be a diagnostic biomarker for early cervical cancer and rheumatoid arthritis (Deping et al., 2021). Wu et al. [ 2020] discovered that the regulatory axis of hsa_circ_0000069/hsa-miR-125b-5p/CDKN2A may play a role in the occurrence and progression of cervical squamous cell carcinoma. miR-301b expression was induced by hypoxia in PrCa cell lines (DU145, PC-3, LNCaP) in vitro, resulting in increased autophagy and loss of radiosensitivity, thereby influencing the occurrence and progression of prostate cancer (Fort et al., 2018). Validation of cell lines and cell line-derived exosomes demonstrated that exosome-specific hsa-miR-301b-3p was upregulated in both eye cancer cell lines and their exosomes (Ravishankar et al., 2020). Previous research has demonstrated that the entire genome of hsa-miR-212-3p is downregulated in Alzheimer’s disease, with a more pronounced decrease in Alzheimer’s disease samples containing gray matter (Pichler et al., 2017). In the study by Cheng and Wang, 2020 lncRNA XIST regulates the expression of ASF1A and BRWD1 via miR-212–3p, influencing the occurrence and development of acute kidney injury. As far as we know, hypoxia is a condition of insufficient tissue oxygenation that plays an important role in numerous pathophysiologies, including embryonic development, high-altitude adaptation, inflammation, tissue repair, and tumor growth (Krueger et al., 2019), whereas chronic intermittent hypoxia can cause OSA (Zhou et al., 2021). In addition, OSA has been linked to cardiovascular disease, type 2 diabetes, Alzheimer’s disease, pulmonary hypertension, and kidney damage (Daulatzai, 2013; Abuyassin et al., 2019; Zhou et al., 2021). In conclusion, we hypothesize that lncRNA STRBP may compete with miRNA (hsa-miR-1297, hsa-miR-17-5p, hsa-miR-20b-5p, hsa-miR-125b-5p, hsa-miR-301b-3p, and hsa-miR-212-3p) for binding, thereby regulating the target genes of CCND2, WT1, E2F2, and IRF1, affecting the occurrence and development of OSA; however, the specific pathogenesis still warrants further investigation. Although these four key hub genes and related mechanism networks may not be specific and require further validation, they can still provide a new direction for the diagnosis and treatment of OSA in patients.
## 5 Advantages and limitations
There are advantages and limitations to this study. First, to the best of our knowledge, this may be the first study to construct a human plasma lncRNA-related ceRNA network, followed by the development of a predictive model for key hub genes and internal and external validation. The findings in this study provide a new perspective on the functional mechanism of OSA and theoretical support for the potential diagnostic and therapeutic targets. Nevertheless, our study has many limitations. First, we only compared ceRNA between OSA and normal plasma; however, it may differ between OSA severity levels and must be identified further. Second, the sample size used for analysis and validation is smaller than the sample size typically required for biomarker analysis, which may result in errors. Thirdly, the external validation is based solely on public databases, and our results require additional in vivo and in vitro validation. Therefore, we must conduct a prospective cohort study with a larger sample size to further confirm our position.
## 6 Conclusion
In conclusion, our findings indicate that protein serine/threonine kinase activity plays a crucial role in the MAPK signaling pathway during the progression of OSA disease. CCND2, WT1, E2F2, and IRF1 could be new OSA targets for diagnosis and treatment. Using these four key hub genes, we designed and validated a new nomogram to predict the risk of OSA patients that has sufficient performance and discrimination ability to serve as a basis for clinical decision-making. LncRNA STRBP may compete with miRNA (hsa-miR-1297, hsa-miR-17-5p, hsa-miR-20b-5p, hsa-miR-125b-5p, hsa-miR-301b-3p and hsa-miR-212-3p) for binding, thereby regulating the target genes of CCND2, WT1, E2F2 and IRF1, affecting the occurrence and development of OSA. In conclusion, despite the fact that our results are preliminary, these analyses provide a new direction for the pathogenesis of OSA; consequently, they may aid in the future translation of this study into clinical work.
## Data availability statement
The original contributions presented in the study are publicly available. This data can be found here: https://www.ncbi.nlm.nih.gov/geo/. Accession number: GSE226379.
## Ethics statement
The studies involving human participants were reviewed and approved by the Medical Ethics Committee of the Second Affiliated Hospital of Guangdong Medical University (ethics number: GDEFEY2020LS030). The patients/participants provided their written informed consent to participate in this study.
## Author contributions
JC and WeL conceived and designed the study. WaL and XS conducted data analysis and wrote manuscripts. JH, JjZ, ZL, JrZ, TC and YZ conducted patient recruitment, examination, blood sample processing and most of the experiments. MP, XL and LZ have given great support in medical facilities, patient examination and funding. All authors participated in the editing of the manuscript.
## Conflict of interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher’s note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## References
1. Abuyassin B., Badran M., Ayas N. T., Laher I.. **The antioxidant α-lipoic acid attenuates intermittent hypoxia-related renal injury in a mouse model of sleep apnea**. *Sleep* (2019) **42** zsz066. DOI: 10.1093/sleep/zsz066
2. Angerstein C., Hecker M., Paap B. K., Koczan D., Thamilarasan M., Thiesen H. J.. **Integration of MicroRNA databases to study MicroRNAs associated with multiple sclerosis**. *Mol. Neurobiol.* (2012) **45** 520-535. DOI: 10.1007/s12035-012-8270-0
3. Bien J., Tibshirani R.. **Hierarchical clustering with prototypes via minimax linkage**. *J. Am. Stat. Assoc.* (2011) **106** 1075-1084. DOI: 10.1198/jasa.2011.tm10183
4. Broom O. J., Widjaya B., Troelsen J., Olsen J., Nielsen O. H.. **Mitogen activated protein kinases: A role in inflammatory bowel disease?**. *Clin. Exp. Immunol.* (2009) **158** 272-280. DOI: 10.1111/j.1365-2249.2009.04033.x
5. Chen W., Feng J., Wang Y., Wang C., Dong Z.. **Development and validation of a nomogram for predicting obstructive sleep apnea in bariatric surgery candidates**. *Nat. Sci. Sleep.* (2021) **13** 1013-1023. DOI: 10.2147/NSS.S316674
6. Chen Y., Li K., Zhang X., Chen J., Li M., Liu L.. **The novel long noncoding RNA lncRNA-Adi regulates adipogenesis**. *Stem Cells Transl. Med.* (2020) **9** 1053-1067. DOI: 10.1002/sctm.19-0438
7. Chen Y., Wang X.. **miRDB: an online database for prediction of functional microRNA targets**. *Nucleic Acids Res.* (2020) **48** D127-D131. DOI: 10.1093/nar/gkz757
8. Cheng Q., Wang L.. **LncRNA XIST serves as a ceRNA to regulate the expression of ASF1A, BRWD1M, and PFKFB2 in kidney transplant acute kidney injury via sponging hsa-miR-212-3p and hsa-miR-122-5p**. *Cell Cycle* (2020) **19** 290-299. DOI: 10.1080/15384101.2019.1707454
9. Chermuła B., Brązert M., Jeseta M., Ożegowska K., Kocherova I., Jankowski M.. **Transcriptomic pattern of genes regulating protein response and status of mitochondrial activity are related to oocyte maturational competence-A transcriptomic study**. *Int. J. Mol. Sci.* (2019) **20** 2238. DOI: 10.3390/ijms20092238
10. Chin C. H., Chen S. H., Wu H. H., Ho C. W., Ko M. T., Lin C. Y.. **cytoHubba: identifying hub objects and sub-networks from complex interactome**. *BMC Syst. Biol.* (2014) **8** S11. DOI: 10.1186/1752-0509-8-S4-S11
11. Chou C. H., Shrestha S., Yang C. D., Chang N. W., Lin Y. L., Liao K. W.. **miRTarBase update 2018: a resource for experimentally validated microRNA-target interactions**. *Nucleic Acids Res.* (2018) **46** D296-D302. DOI: 10.1093/nar/gkx1067
12. Daulatzai M. A.. **Death by a thousand cuts in Alzheimer's disease: hypoxia--the prodrome**. *Neurotox. Res.* (2013) **24** 216-243. DOI: 10.1007/s12640-013-9379-2
13. Deping C., Bofan J., Yaogang Z., Mingquan P.. **microRNA-125b-5p is a promising novel plasma biomarker for alveolar echinococcosis in patients from the southern province of Qinghai**. *BMC Infect. Dis.* (2021) **21** 246. DOI: 10.1186/s12879-021-05940-z
14. Drobna M., Szarzyńska B., Jaksik R., Sędek Ł., Kuchmiy A., Taghon T.. **hsa-miR-20b-5p and hsa-miR-363-3p affect expression of PTEN and BIM tumor suppressor genes and modulate survival of T-ALL cells**. *Cells* (2020) **9** 1137. DOI: 10.3390/cells9051137
15. Du P., Wang J., Han Y., Feng J.. **Blocking the LncRNA MALAT1/miR-224-5p/NLRP3 Axis inhibits the hippocampal inflammatory response in T2DM with OSA**. *Front. Cell Neurosci.* (2020) **14** 97. DOI: 10.3389/fncel.2020.00097
16. Duarte R. L., Magalhães-da-Silveira F. J., Oliveira-E-Sá T. S., Silva J. A., Mello F. C., Gozal D.. **Obstructive sleep apnea screening with a 4-item instrument, named GOAL questionnaire: Development, validation and comparative study with No-apnea, STOP-bang, and NoSAS**. *Nat. Sci. Sleep.* (2020) **12** 57-67. DOI: 10.2147/NSS.S238255
17. Duim S. N., Kurakula K., Goumans M. J., Kruithof B. P. T.. **Cardiac endothelial cells express Wilms' tumor-1: Wt1 expression in the developing, adult and infarcted heart**. *J. Mol. Cell Cardiol.* (2015) **81** 127-135. DOI: 10.1016/j.yjmcc.2015.02.007
18. Ekin S., Yildiz H., Alp H. H.. **IMA and oxidative DNA damage: Can these parameters be used to estimate the presence and severity of OSA?**. *Sleep. Breath.* (2021) **25** 529-536. DOI: 10.1007/s11325-020-02093-2
19. Floras J. S.. **Hypertension and sleep apnea**. *Can. J. Cardiol.* (2015) **31** 889-897. DOI: 10.1016/j.cjca.2015.05.003
20. Fort R. S., Mathó C., Oliveira-Rizzo C., Garat B., Sotelo-Silveira J. R., Duhagon M. A.. **An integrated view of the role of miR-130b/301b miRNA cluster in prostate cancer**. *Exp. Hematol. Oncol.* (2018) **7** 10. DOI: 10.1186/s40164-018-0102-0
21. Friedman J., Hastie T., Tibshirani R.. **Regularization paths for generalized linear models via coordinate descent**. *J. Stat. Softw.* (2010) **33** 1-22. DOI: 10.18637/jss.v033.i01
22. Ge H., Liu J., Liu F., Sun Y., Yang R.. **Long non-coding RNA ROR mitigates cobalt chloride-induced hypoxia injury through regulation of miR-145**. *Artif. Cells Nanomed Biotechnol.* (2019) **47** 2221-2229. DOI: 10.1080/21691401.2019.1620759
23. Haenisch S., Zhao Y., Chhibber A., Kaiboriboon K., Do L. V., Vogelgesang S.. **SOX11 identified by target gene evaluation of miRNAs differentially expressed in focal and non-focal brain tissue of therapy-resistant epilepsy patients**. *Neurobiol. Dis.* (2015) **77** 127-140. DOI: 10.1016/j.nbd.2015.02.025
24. Hazra A., Gogtay N.. **Biostatistics series module 6: Correlation and linear regression**. *Indian J. Dermatol* (2016) **61** 593-601. DOI: 10.4103/0019-5154.193662
25. Hou X., Zhang Y., Li W., Hu A. J., Luo C., Zhou W.. **CDK6 inhibits white to beige fat transition by suppressing RUNX1**. *Nat. Commun.* (2018) **9** 1023. DOI: 10.1038/s41467-018-03451-1
26. Hu C., Li J., Du Y., Yang Y., Jia Y., Li J.. **Impact of chronic intermittent hypoxia on the long non-coding RNA and mRNA expression profiles in myocardial infarction**. *J. Cell Mol. Med.* (2021) **25** 421-433. DOI: 10.1111/jcmm.16097
27. Huang J. X., Wu Y. C., Cheng Y. Y., Wang C. L., Yu C. J.. **IRF1 negatively regulates oncogenic KPNA2 expression under growth stimulation and hypoxia in lung cancer cells**. *Onco Targets Ther.* (2019) **12** 11475-11486. DOI: 10.2147/OTT.S221832
28. Iasonos A., Schrag D., Raj G. V., Panageas K. S.. **How to build and interpret a nomogram for cancer prognosis**. *J. Clin. Oncol.* (2008) **26** 1364-1370. DOI: 10.1200/JCO.2007.12.9791
29. Jędroszka D., Orzechowska M., Hamouz R., Gorniak K., Bednarek A. K.. **Markers of epithelial-to-mesenchymal transition reflect tumor biology according to patient age and Gleason score in prostate cancer**. *PLoS One* (2017) **12** e0188842. DOI: 10.1371/journal.pone.0188842
30. Jeggari A., Marks D. S., Larsson E.. **miRcode: a map of putative microRNA target sites in the long non-coding transcriptome**. *Bioinformatics* (2012) **28** 2062-2063. DOI: 10.1093/bioinformatics/bts344
31. Kendzerska T., van Walraven C., McIsaac D. I., Povitz M., Mulpuru S., Lima I.. **Case-ascertainment models to identify adults with obstructive sleep apnea using health administrative data: Internal and external validation**. *Clin. Epidemiol.* (2021) **13** 453-467. DOI: 10.2147/CLEP.S308852
32. Khomtchouk B. B., Van Booven D. J., HeatmapGenerator W. C.. **HeatmapGenerator: High performance RNAseq and microarray visualization software suite to examine differential gene expression levels using an R and C++ hybrid computational pipeline**. *Source Code Biol. Med.* (2014) **9** 30. DOI: 10.1186/s13029-014-0030-2
33. Krueger K., Catanese L., Sciesielski L. K., Kirschner K. M., Scholz H.. **Deletion of an intronic HIF-2α binding site suppresses hypoxia-induced WT1 expression**. *Biochim. Biophys. Acta Gene Regul. Mech.* (2019) **1862** 71-83. DOI: 10.1016/j.bbagrm.2018.11.003
34. Lam F., Lalansingh C. M., Babaran H. E., Wang Z., Prokopec S. D., Fox N. S.. **VennDiagramWeb: A web application for the generation of highly customizable Venn and euler diagrams**. *BMC Bioinforma.* (2016) **17** 401. DOI: 10.1186/s12859-016-1281-5
35. Lawlor M. A., Alessi D. R.. **PKB/Akt: A key mediator of cell proliferation, survival and insulin responses?**. *J. Cell Sci.* (2001) **114** 2903-2910. DOI: 10.1242/jcs.114.16.2903
36. Lewis B. P., Shih I. H., Jones-Rhoades M. W., Bartel D. P., Burge C. B.. **Prediction of mammalian microRNA targets**. *. Cell* (2003) **115** 787-798. DOI: 10.1016/s0092-8674(03)01018-3
37. Li H., Yuan H.. **MiR-1297 negatively regulates metabolic reprogramming in glioblastoma via repressing KPNA2**. *Hum. Cell* (2020) **33** 619-629. DOI: 10.1007/s13577-019-00316-7
38. Li K., Wei P., Qin Y., Wei Y.. **MicroRNA expression profiling and bioinformatics analysis of dysregulated microRNAs in obstructive sleep apnea patients**. *Med. Baltim.* (2017) **96** e7917. DOI: 10.1097/MD.0000000000007917
39. Li T., Yang J., Lv X., Liu K., Gao C., Xing Y.. **miR-155 regulates the proliferation and cell cycle of colorectal carcinoma cells by targeting E2F2**. *Biotechnol. Lett.* (2014) **36** 1743-1752. DOI: 10.1007/s10529-014-1540-3
40. Luo M., Zheng H. Y., Zhang Y., Feng Y., Li D. Q., Li X. L.. **A nomogram for predicting the likelihood of obstructive sleep apnea to reduce the unnecessary polysomnography examinations**. *Chin. Med. J. Engl.* (2015) **128** 2134-2140. DOI: 10.4103/0366-6999.162514
41. Marti-Soler H., Hirotsu C., Marques-Vidal P., Vollenweider P., Waeber G., Preisig M.. **The NoSAS score for screening of sleep-disordered breathing: A derivation and validation study**. *Lancet Respir. Med.* (2016) **4** 742-748. DOI: 10.1016/S2213-2600(16)30075-3
42. Matsui M., Corey D. R.. **Non-coding RNAs as drug targets**. *Nat. Rev. Drug Discov.* (2017) **16** 167-179. DOI: 10.1038/nrd.2016.117
43. Mielke P. J., Berry K. J.. **Multivariate multiple regression prediction models: A euclidean distance approach**. *Psychol. Rep.* (2003) **92** 763-769. DOI: 10.2466/pr0.2003.92.3.763
44. Pichler S., Gu W., Hartl D., Gasparoni G., Leidinger P., Keller A.. **The miRNome of Alzheimer's disease: Consistent downregulation of the miR-132/212 cluster**. *Neurobiol. Aging* (2017) **50** 167.e1-167167.e10. DOI: 10.1016/j.neurobiolaging.2016.09.019
45. Ravishankar H., Mangani A. S., Phoebe M. G., Mani S. P., Parameswaran S., Khetan V.. **Serum exosomal miRNA as biomarkers for Retinoblastoma**. *Exp. Eye Res.* (2020) **199** 108184. DOI: 10.1016/j.exer.2020.108184
46. Rey F., Urrata V., Gilardini L., Bertoli S., Calcaterra V., Zuccotti G. V.. **Role of long non-coding RNAs in adipogenesis: State of the art and implications in obesity and obesity-associated diseases**. *Obes. Rev.* (2021) **22** e13203. DOI: 10.1111/obr.13203
47. Ritchie M. E., Phipson B., Wu D., Hu Y., Law C. W., Shi W.. **Limma powers differential expression analyses for RNA-sequencing and microarray studies**. *Nucleic Acids Res.* (2015) **43** e47. DOI: 10.1093/nar/gkv007
48. Salman L. A., Shulman R., Cohen J. B.. **Obstructive sleep apnea, hypertension, and cardiovascular risk: Epidemiology, pathophysiology, and management**. *Curr. Cardiol. Rep.* (2020) **22** 6. DOI: 10.1007/s11886-020-1257-y
49. Shannon P., Markiel A., Ozier O., Baliga N. S., Wang J. T., Ramage D.. **Cytoscape: A software environment for integrated models of biomolecular interaction networks**. *Genome Res.* (2003) **13** 2498-2504. DOI: 10.1101/gr.1239303
50. Sharma P., Dong Y., Somers V. K., Peterson T. E., Zhang Y., Wang S.. **Intermittent hypoxia regulates vasoactive molecules and alters insulin-signaling in vascular endothelial cells**. *Sci. Rep.* (2018) **8** 14110. DOI: 10.1038/s41598-018-32490-3
51. Shi H., Li J., Song Q., Cheng L., Sun H., Fan W.. **Systematic identification and analysis of dysregulated miRNA and transcription factor feed-forward loops in hypertrophic cardiomyopathy**. *J. Cell Mol. Med.* (2019) **23** 306-316. DOI: 10.1111/jcmm.13928
52. Shi H., Xiang S., Huang X., Wang L., Hua F., Jiang X.. **Development and validation of a nomogram for predicting the risk of obstructive sleep apnea in patients with type 2 diabetes**. *Ann. Transl. Med.* (2020) **8** 1675. DOI: 10.21037/atm-20-6890
53. Song G. Q., Wang D. L., Wu H. M., Wang Q. J., Han F., Hu G. Q.. **Two effective clinical prediction models to screen for obstructive sleep apnoea based on body mass index and other parameters**. *Sleep Breath.* (2021) **26** 923-932. DOI: 10.1007/s11325-021-02347-7
54. Song W., Fu T.. **Circular RNA-associated competing endogenous RNA network and prognostic nomogram for patients with colorectal cancer**. *Front. Oncol.* (2019) **9** 1181. DOI: 10.3389/fonc.2019.01181
55. Szklarczyk D., Franceschini A., Wyder S., Forslund K., Heller D., Huerta-Cepas J.. **STRING v10: Protein-protein interaction networks, integrated over the tree of life**. *Nucleic Acids Res.* (2015) **43** D447-D452. DOI: 10.1093/nar/gku1003
56. Trotta M. C., Gesualdo C., Platania C., De Robertis D., Giordano M., Simonelli F.. **Circulating miRNAs in diabetic retinopathy patients: Prognostic markers or pharmacological targets?**. *Biochem. Pharmacol.* (2021) **186** 114473. DOI: 10.1016/j.bcp.2021.114473
57. Wang Y., Xue J., Kuang H., Zhou X., Liao L., Yin F.. **microRNA-1297 inhibits the growth and metastasis of colorectal cancer by suppressing cyclin D2 expression**. *DNA Cell Biol.* (2017) **36** 991-999. DOI: 10.1089/dna.2017.3829
58. Wang Z., Guo J., Guo Y., Yang Y., Teng T., Yu Q.. **Genome-wide detection of cnvs and association with body weight in sheep based on 600k snp arrays**. *Front. Genet.* (2020) **11** 558. DOI: 10.3389/fgene.2020.00558
59. Wolbers M., Koller M. T., Witteman J. C., Steyerberg E. W.. **Prognostic models with competing risks: Methods and application to coronary risk prediction**. *Epidemiology* (2009) **20** 555-561. DOI: 10.1097/EDE.0b013e3181a39056
60. Wu Q., Liu P., Lao G., Liu Y., Zhang W., Ma C.. **Comprehensive analysis of circRNA-miRNA-mRNA network in cervical squamous cell carcinoma by integrated analysis**. *Onco Targets Ther.* (2020) **13** 8641-8650. DOI: 10.2147/OTT.S254323
61. Xu Y., Dong Y., Deng Y., Qi Q., Wu M., Liang H.. **Identifying an lncRNA-related ceRNA network to reveal novel targets for a cutaneous squamous cell carcinoma**. *. Biol. (Basel)* (2021) **10** 432. DOI: 10.3390/biology10050432
62. Yamamura S., Imai-Sumida M., Tanaka Y., Dahiya R.. **Interaction and cross-talk between non-coding RNAs**. *Cell Mol. Life Sci.* (2018) **75** 467-484. DOI: 10.1007/s00018-017-2626-6
63. Yao X., Jing T., Wang T., Gu C., Chen X., Chen F.. **Molecular characterization and elucidation of pathways to identify novel therapeutic targets in pulmonary arterial hypertension**. *Front. Physiol.* (2021) **12** 694702. DOI: 10.3389/fphys.2021.694702
64. Yu G., Wang L. G., Han Y., He Q. Y.. **clusterProfiler: an R package for comparing biological themes among gene clusters**. *OMICS* (2012) **16** 284-287. DOI: 10.1089/omi.2011.0118
65. Zhang H., Liu R., Sun L., Guo W., Ji X., Hu X.. **Comprehensive analysis of gene expression changes and validation in hepatocellular carcinoma**. *Onco Targets Ther.* (2021) **14** 1021-1031. DOI: 10.2147/OTT.S294500
66. Zhang J., Hu C., Jiao X., Yang Y., Li J., Yu H.. **Potential role of mRNAs and LncRNAs in chronic intermittent hypoxia exposure-aggravated atherosclerosis**. *Front. Genet.* (2020) **11** 290. DOI: 10.3389/fgene.2020.00290
67. Zhang X. Y., Dai H. P., Li Z., Yin J., Lang X. P., Yang C. X.. **Identification of STRBP as a novel JAK2 fusion partner gene in a young adult with philadelphia chromosome-like B-lymphoblastic leukemia**. *Front. Oncol.* (2020) **10** 611467. DOI: 10.3389/fonc.2020.611467
68. Zhang Y., Xu Y., Li S.. **Bibliometrics analysis of diagnostic test accuracy studies of bladder cancer**. *. Dis. Markers* (2021) **2021** 8870353. DOI: 10.1155/2021/8870353
69. Zhao Y. N., Wang H. Y., Li J. M., Chen B. Y., Xia G., Zhang P. P.. **Hippocampal mitogen-activated protein kinase activation is associated with intermittent hypoxia in a rat model of obstructive sleep apnea syndrome**. *Mol. Med. Rep.* (2016) **13** 137-145. DOI: 10.3892/mmr.2015.4505
70. Zheng H., Li Z., Li J., Zheng S., Zhao E.. **Construction, validation, and visualization of two web-based nomograms to predict overall and cancer-specific survival in patients with gastric cancer and lung metastases**. *J. Oncol.* (2021) **2021** 5495267. DOI: 10.1155/2021/5495267
71. Zhou J., Cheng M., Wu M., Boriboun C., Jujo K., Xu S.. **Contrasting roles of E2F2 and E2F3 in endothelial cell growth and ischemic angiogenesis**. *J. Mol. Cell Cardiol.* (2013) **60** 68-71. DOI: 10.1016/j.yjmcc.2013.04.009
72. Zhou Z., Ni H., Li Y., Jiang B.. **LncRNA XIST promotes inflammation by downregulating GRα expression in the adenoids of children with OSAHS**. *Exp. Ther. Med.* (2021) **21** 500. DOI: 10.3892/etm.2021.9931
73. Zhu J., Wang L., Zhou Y., Hao J., Wang S., Liu L.. **Comprehensive analysis of the relationship between competitive endogenous RNA (ceRNA) networks and tumor infiltrating-cells in hepatocellular carcinoma**. *J. Gastrointest. Oncol.* (2020) **11** 1381-1398. DOI: 10.21037/jgo-20-555
74. Zhu Q. Q., Pu X. B., Chen T. C., Qiu C. Y., Wu Z. H., Tian L.. **Hsa_circ_0008360 sponges miR-186-5p to target CCND2 to modulate high glucose-induced vascular endothelial dysfunction**. *Cell Cycle* (2021) **20** 1389-1401. DOI: 10.1080/15384101.2021.1918877
|
---
title: Voluntary exercise modulates pathways associated with amelioration of retinal
degenerative diseases
authors:
- Joshua A. Chu-Tan
- Adrian V. Cioanca
- Yvette Wooff
- Max Kirkby
- Marissa Ellis
- Pranay Gulati
- Tim Karl
- Jeffrey H. Boatright
- Katie Bales
- John Nickerson
- Riccardo Natoli
journal: Frontiers in Physiology
year: 2023
pmcid: PMC10036398
doi: 10.3389/fphys.2023.1116898
license: CC BY 4.0
---
# Voluntary exercise modulates pathways associated with amelioration of retinal degenerative diseases
## Abstract
Background: Exercise has been shown to promote a healthier and longer life and linked to a reduced risk of developing neurodegenerative diseases including retinal degenerations. However, the molecular pathways underpinning exercise-induced cellular protection are not well understood. In this work we aim to profile the molecular changes underlying exercise-induced retinal protection and investigate how exercise-induced inflammatory pathway modulation may slow the progression of retinal degenerations.
Methods: Female C57Bl/6J mice at 6 weeks old were given free access to open voluntary running wheels for a period of 28 days and then subjected to 5 days of photo-oxidative damage (PD)-induced retinal degeneration. Following, retinal function (electroretinography; ERG), morphology (optical coherence tomography; OCT) and measures of cell death (TUNEL) and inflammation (IBA1) were analysed and compared to sedentary controls. To decipher global gene expression changes as a result of voluntary exercise, RNA sequencing and pathway and modular gene co-expression analyses were performed on retinal lysates of exercised and sedentary mice that were subjected to PD, as well as healthy dim-reared controls.
Results: Following 5 days of PD, exercised mice had significantly preserved retinal function, integrity and reduced levels of retinal cell death and inflammation, compared to sedentary controls. In response to voluntary exercise, inflammatory and extracellular matrix integrity pathways were significantly modulated, with the gene expression profile of exercised mice more closely trending towards that of a healthy dim-reared retina.
Conclusion: We suggest that voluntary exercise may mediate retinal protection by influencing key pathways involved in regulating retinal health and shifting the transcriptomic profile to a healthy phenotype.
## Background
Physical activity and exercise are synonymous with health and longevity. The benefits of exercise to the human body are well-established, with overwhelming evidence demonstrating amelioration of both physical and mental aspects of human physiology (Pedersen and Saltin, 2015). Exercise has been shown to improve the pathology of several chronic diseases including cardiovascular disease, Type-2 diabetes, obesity, and cancer (Pedersen and Saltin, 2015). In addition, as a modifiable risk factor alongside diet, physical activity has more recently been considered to be an important alternative therapeutic intervention to slow cognitive decline and pathophysiological processes in age-related neurological conditions and neurodegenerative disorders, including in Parkinson’s (Chen et al., 2005; Xu et al., 2010) and Alzheimer’s diseases (Buchman et al., 2012), as well as improving memory function in adults over the age of 65 (Colcombe and Kramer, 2003; Larson et al., 2006).
The broad beneficial effects of exercise observed across disorders of the central nervous system (CNS), have more recently been demonstrated in the retina, with strong evidence emerging that exercise offers protection against retinal degenerations, including in glaucoma (Williams, 2009a; Chrysostomou et al., 2014; Meier et al., 2018), diabetic retinopathy (Allen et al., 2018), retinitis pigmentosa (Hanif et al., 2015) and in age-related macular degeneration (AMD) (McGuinness et al., 2017; Mees et al., 2019).
AMD, the leading cause of blindness in the developed world (Deloitte Access Economics and Mitchell, 2011), is a chronic and progressive disease characterised by overburdening levels of inflammation, consequently resulting in the death of the light-sensing photoreceptor cells and irreversible blindness (Kauppinen et al., 2016). Exercise has been shown to reduce the severity of AMD, with higher levels of vigorous exercise in middle-aged adults (Williams, 2009b) and moderate exercise in people over 75 associated with a lower incidence rate of AMD (Gopinath et al., 2014). Conversely, decreased physical activity was associated with an increase in early pathogenic features, or “precursors” of AMD (Munch et al., 2013).
It has been suggested that one of the main mechanisms of action for the health benefits of exercise in neurodegenerative diseases is through modulating the activation of glia leading to reduced pro-inflammatory cytokines and neuroinflammation (Seo et al., 2019). As neuroinflammation is a hallmark feature of neuro- and retinal degenerative diseases such as AMD, targeting, or influencing endogenous anti-inflammatory mechanisms are of considerable interest in therapeutic development (Kauppinen et al., 2016). Given that AMD is projected to affect 1.7 million Australians by 2030, and 288 million worldwide by 2040 (Deloitte Access Economics and Mitchell, 2011; Wong et al., 2014), any treatments which could slow the progression of this disease, and others where underlying neuroinflammation is a key pathological feature, would greatly reduce the significant economic impacts and have major impacts on an individual’s quality of life.
Research has shown that exercise is able to regulate systemic inflammation, by modulating key inflammatory pathways and pro-inflammatory cytokine profiles that lead to downstream dampening of innate immune responses (Nieman and Wentz, 2019; Chu‐Tan et al., 2021). In the retina, inflammatory modulation in response to exercise has been correlated with protection against degeneration (Chrysostomou et al., 2016; Mees et al., 2019; Dantis Pereira de Campos et al., 2020; Bales et al., 2022). In rodent models of retinitis pigmentosa, voluntary exercise has been shown to be effective against photoreceptor cell loss and inflammation as well as increase visual acuity (Barone et al., 2012; Barone et al., 2014; Zhang et al., 2019), while swimming used in a model of glaucoma was protected against astrocytic gliosis, macrophage activation and age-related vulnerability (Chrysostomou et al., 2014). Furthermore, treadmill exercise was shown to protect visual function, and decrease inflammation and apoptotic neuronal cell death in a model of diabetic retinopathy (Ji et al., 2013; Kim et al., 2013; Allen et al., 2018; Dantis Pereira de Campos et al., 2020). Finally, in a light-induced model of degeneration that recapitulates aspects of atrophic AMD (Natoli et al., 2016), forced treadmill exercise in albino mice was shown to preserve both retinal morphology and function (Lawson et al., 2014; Mees et al., 2019) with low-intensity exercise proving the most beneficial (Mees et al., 2019). Additionally, this model also revealed retinal astrocytes have altered morphology and increased expression of brain-derived neurotrophic factor (BDNF) and a specific isoform of its high-affinity receptor TrkB (Bales et al., 2022).
Although these studies collectively shed light on the therapeutic potential of exercise to reduce inflammation and protect against, or slow the progression of, retinal degenerations, knowledge on the precise molecular events occurring within the retina during exercise is still lacking. Clinical research has largely been confined to correlative epidemiological studies, whereas animal studies have been limited to implicative processes and/or based on single, select molecular targets (Chu‐Tan et al., 2021). Further, as animal studies to date have predominantly employed a forced model of exercise, it is unclear what potentially confounding effects stress induced shock-forcing methods may have on the results of these studies, with known stress and anxiety-like behaviours exhibited in forced vs. voluntary-exercise animals (Leasure and Jones, 2008) as well as stress having a known impact on inflammation in the retina (Malan et al., 2020).
Therefore, to address these highlighted problems and help decipher the molecular effect of physical activity on retinal protection, we have employed a voluntary rodent model of aerobic exercise using open voluntary running wheels, and a well-established photo-oxidative damage (PD) model of retinal degeneration that recapitulates key inflammatory aspects of AMD (Natoli et al., 2016). We determined that voluntary exercise provides protection to the retina against photo-oxidative damage with both functional, morphological and histological improvements, in particular in reducing photoreceptor cell death and inflammation. Further, RNA sequencing and gene module co-expression analyses identified that inflammatory and extracellular matrix integrity pathways associated with AMD pathogenesis were both modulated by exercise, with the gene expression profiles of these modules trending towards those of healthy control retinas. Taken together this study indicates that voluntary exercise may confer retinal protection against degeneration and drives the transcriptomic profile of the retina more towards what is observed in a healthy phenotype.
## Animal handling
All experiments were conducted in accordance with the ARVO Statement for the Use of Animals in Ophthalmic and Vision Research and with approval from the Australian National University’s (ANU) Animal Experimentation Ethics Committee (AEEC) (Ethics ID: A$\frac{2020}{41}$; Rodent models and treatments for retinal degenerations). Adult male and female C57BL/6J wild-type (WT) mice (aged 50 postnatal days; (P50) at experimental onset) were purchased from Australian BioResources (ABR), (Moss Vale, New South Wales, Australia). Mice were bred, reared and housed under 12 h light/dark cycle conditions (5 lux) with free access to food and water.
## Voluntary exercise model
Mice were housed individually in standard type III rodent cages with, or without TSE Systems running wheels (drum diameter of 115 mm and width of 40 mm, Cat# E−303400-RW-V-M-D1-S) for 14 or 28 days. Mice were single-housed and acclimatised to their boxes/running wheels for 3 days prior to experimental onset. Mice housed with running wheels (exercise) had unlimited and free access to wheels, while control mice (sedentary) were housed in cages of the same dimensions with no running wheels but chew block and half-tunnel enrichment. It should be noted that sedentary mice are simply those that did not have access to a running wheel and there was nothing preventing them from moving within their cages. All running wheel drums were connected to a basic unit (TSE Systems, Cat# 303400-RW-V-M-BU) for each cage. All units were connected to a central control unit (TSE Systems, Cat# E−303400-C/16) for running measurements on the PhemoMaster Control System software. For exercised mice, data on total distance (km), total time (mins), total number of runs, max speed (m/s) and average speed (m/s) were collected every hour and collated at the end of each day and on experiment completion. Data on spontaneous in-cage physical activity performed by mice housed without running wheels was obtained from previous work published by Pernold et al. [ 2019]. This study employed both capacitance sensing technology (CST) and video tracking to examine the movement of C57BL/6J mice housed in similarly sized cages (Pernold et al., 2019). All mice were housed under 12 h light/dark cycle conditions (5 lux), with free access to food, water and environmental enrichment in the form of chew blocks.
## Photo-oxidative damage
Exercised and sedentary mice were subjected to photo-oxidative damage (PD) for 5 days as described previously (Natoli et al., 2016). Briefly, animals were placed into Perspex boxes coated with a reflective interior surface and exposed to 100 K lux white light from light-emitting diodes (LED, High CRI LED, Yuji, Beijing). The LED is a 100-W 65,000 k natural white LED with an emission spectrum more closely resembling daylight than halogen or incandescent bulbs. No exercise wheels were provided in these PD boxes. Animals were administered pupil dilator (Minims® atropine sulphate $1\%$ w/v; Bausch and Lomb) to both eyes twice a day during the course of the damage paradigm. Following degeneration paradigms, retinal function and morphology were assessed and compared between exercised and sedentary mice.
## Retinal function via electroretinography (ERG)
To assess retinal function full-field scotopic ERG was performed as previously described (Natoli et al., 2016). Briefly, mice were dark-adapted overnight before being anesthetized with an intraperitoneal injection of Ketamine (100 mg/kg; Troy Laboratories, NSW, Australia) and Xylazil (10 mg/kg; Troy Laboratories, NSW, Australia). Both pupils were dilated with one drop each of $2.5\%$ w/v Phenylephrine hydrochloride and $1\%$ w/v Tropicamide (Bausch and Lomb, NY, United States).
Anesthetized and pupil-dilated mice were placed on the thermally regulated stage of the Celeris ERG system (Diagnosys LLC, MA, United States). The Celeris ERG system has combined Ag/AgCl electrode-stimulator eye probes that measure the response from both eyes simultaneously and uses 32-bit ultra-low noise amplifiers fitted with impedance testing. Eye probes were cleaned with $70\%$ ethanol and then a $0.3\%$ Hypromellose eye drop solution (GenTeal; Novartis, NSW, Australia) was applied to both probes. The probes were then placed covering and just touching the surface of each eye. A single- or twin-flash paradigm was used to elicit a mixed response from rods and cones. Flash stimuli for mixed responses were provided using 6500K white flash luminance range over stimulus intensities from −0.01 – 40 log.cd.s.m–2. Responses were recorded and analysed using Espion V6 Software (Diagnosys LLC, MA, United States). Statistics were performed in Prism V7.0 using a two-way analysis of variance (ANOVA) to test for differences in a-wave and b-wave responses. Data was expressed as the mean wave amplitude ± SEM (μV).
## Optical coherence tomography (OCT)
Cross-sectional and fundus images of live mouse retinas were taken using a MICRON® IV device (Phoenix-Micron, Inc., OR, United States). Cross-sectional images were taken at 1 mm increments from the optic nerve. Eye gel (GenTeal; Novartis, NSW, Australia) was administered to both eyes for recovery. Using OCT cross-sectional retinal images, and ImageJ V2.0 software (National Institutes of Health, Bethesda, MD, United States), the thickness of the outer nuclear layer (ONL), was calculated as the ratio to the whole retinal thickness (outer limiting membrane to the inner limiting membrane).
## Tissue collection and preparation
Animals were ethically euthanized with CO2 following ERG/OCT analysis. The superior surface of the left eye was marked and enucleated, then immersed in $4\%$ paraformaldehyde for 3 h. Eyes were then cryopreserved in $15\%$ sucrose solution overnight, embedded in OCT medium (Tissue Tek, Sakura, Japan) and cryosectioned at 12 μm in a parasagittal plane (superior to inferior) using a CM 1850 Cryostat (Leica Biosystems, Germany). To ensure accurate comparisons were made for histological analysis, only sections containing the optic nerve head were used for analysis. The retina from the right eye was excised through a corneal incision and placed into RNAlater solution (Thermo Fisher Scientific, MA, United States) at 4°C overnight and then stored at −80°C until further use.
## Immunolabelling
Immunohistochemical analysis of retinal cryosections was performed as previously described (Rutar et al., 2015). Fluorescence was visualized and images taken using a laser-scanning A1+ confocal microscope at ×20 and ×40 magnification (Nikon, Tokyo, Japan). Images panels were analyzed using ImageJ V2.0 software and assembled using Illustrator software (Adobe Systems, CA, United States).
## IBA-1 immunohistochemistry
Immunolabeling for IBA-1 (1:500, 019-19741, Wako, Osaka, Japan) and quantification was performed as previously described (Rutar et al., 2015). The number of IBA-1+ cells (a marker of retinal microglia and macrophages) was counted across the superior and inferior retina using two retinal sections per mouse and then averaged. Retinal cryosections were stained with the DNA-specific dye bisbenzimide (1:10,000, Sigma-Aldrich, MO, United States) to visualise the cellular layers.
## TUNEL assay
Terminal deoxynucleotidyl transferase (TdT) dUTP nick end labeling (TUNEL), was used as a measure of photoreceptor cell death. TUNEL in situ labeling was performed on retinal cryosections using a Tdt enzyme (Cat# 3333566001, Sigma-Aldrich, MO, United States) and biotinylated deoxyuridine triphosphate (dUTP) (Cat# 11093070910, Sigma-Aldrich, MO, United States) as previously described (Natoli et al., 2010). Images of TUNEL staining were captured with the A1+Nikon confocal microscope at ×20 and ×40 magnification. The total number of TUNEL+ cells were counted including both the superior and inferior retina using two retinal sections along the entire arc of the retina per animal and is represented as the average number of TUNEL+ cells per retinal section.
To further quantify photoreceptor survival, the thickness of the ONL on retinal cryosections was determined by counting the number of nuclei rows (photoreceptor cell bodies) in the area of retinal lesion development (1 mm superior to the optic nerve head). Photoreceptor cell row quantification was performed five times per retina using two retinal cryosections at comparable locations per mouse.
## RNA extraction
RNA extraction was performed using RNAqueous micro total RNA isolation kit (Thermo Fisher Scientific, MA, United States) according to the manufacturer’s instructions. The concentration and purity of each RNA sample was assessed using the ND-1000 spectrophotometer (Nanodrop Technologies, DE, United States).
## High-throughput sequencing (HTS) and bioinformatics
RNA was placed in RNA stabilisation tubes (Azenta Life Sciences, Suzhou, China) and shipped to Azenta Life Sciences (Suzhou, China) for bulk RNA sequencing.
RNA libraries were constructed with the Illumina TruSeq RNA unstranded library preparation kit using polyA selection for mRNA enrichment, then sequenced on the Illumina NovaSeq6000 platform acquiring ∼20 million, 150 base-pair, paired-end reads per sample. Read phred scores, adapter/index contamination were checked with FastQC (Babraham Bioinformatics), then aligned to mouse genome (mm39) using HISAT2 aligner with default parameters. Alignments were summarised with featureCounts, genes with low expression (<1 count per million) were filtered out, and normalisation factors for remaining genes were calculated using trimmed means of m-values (TMM) (Robinson and Oshlack, 2010) method.
## Statistical and clustering analyses
RNA sequencing normalised counts were prepared for linear modelling using voomWithQualityWeights transformation (Law et al., 2014) and then a statistical model was fitted using lmFit (Ritchie et al., 2015) and moderated t-statistics were computed using ebayes function. Tables of fold change and p-value estimates were generated with top Table function and genes with p-value<0.05 and absolute fold changes >0.5 were deemed differentially expressed. Principal component analysis (PCA) and hierarchical clustering (HC) were used as methods for sample clustering and computations were performed using prcomp and hclust R functions (Team RC, 2019). Gene-wise scaling was implemented prior to sample clustering and Euclidean distance (calculated with dist R function) and complete linkage were used as clustering distance and agglomeration method respectively. Statistical testing of sample grouping on PCA was performed using the envfit function from the vegan R package using 10^6 sample permutations. *Modular* gene co-expression analysis and gene set enrichment analysis were carried out using CEMiTool R package and MSigDB C2 canonical pathway as reference set (Subramanian et al., 2005; Russo et al., 2018). Unpaired Student’s t-tests, one-way analysis of variance (ANOVA), or two-way ANOVA were performed using Prism V7.0. p-values < 0.05 were deemed statistically significant. All data was expressed as the mean ± SEM. Normality and homoscedasticity assumptions for parametric tests were visually verified by inspecting distributions of standardised residuals and the relationship between the predictor variables and the square root of standardised residuals. Kruskal–Wallis and Wilcoxon rank sum tests were employed where parametric tests were deemed unsuitable.
## Voluntary exercise paradigm and daily running data outputs
Exercise has well known and documented systemic benefits, including to the CNS and in protection against neurodegenerative diseases (reviewed in Chu-Tan et al. [ 2022]. Therefore, to investigate the effects of exercise on retinal protection against degeneration, mice either with or without (sedentary) free-access to running wheels for 14 or 28 days (14 days in supplementary data), were subjected to PD-induced retinal degeneration. Note that for the 14-day experiments, mixtures of male and female mice were used due to experimental constraints. Therefore, this may not be directly comparable to the 28-day data, where only females were used to control for sex-associated effects. We thus focus on the 28-day cohorts, with the 14-day experiment placed in the supplementary data and alluded to. Following experimental paradigms, retinas were analysed and compared between groups for functional, morphological and molecular changes (Figure 1A). To verify exercise outputs, and correlate running data to measures of retinal protection; recorded measures of distance (km), time (min), number of runs, max speed (m/s) and, average speed (m/s) in hour bins over a period of 28 days were analysed.
**FIGURE 1:** *Voluntary exercise paradigm and daily running data outputs. (A) Schematic depicting voluntary exercise and degeneration paradigms, with output measures of retinal function, morphology and RNA sequencing. (B) Cumulative distance, time and runs made chronologically per animal. (C) Box-and-whisker graphs showing daily and average running wheel outputs: distance (km), time (mins), number of runs, max speed (m/s) and, average speed (m/s) across 28 days for each animal. Two days were omitted due to incomplete data recordings from required cage maintenance. (D) Utilisation of running wheel during the 12 h light/dark housing conditions. Blue lines show the trend in physical activity over 24 h summarised using a generalised additive model. Grey shaded area indicates the 95% confidence interval. (E) Daily distance and mean speed resulting from access to the running wheels or spontaneous in-cage movement. n = 7.*
Mice initiated physical activity immediately after being provided with access to the running wheel and continued to utilise the running wheel until the cessation of the 28-day exercise period (Figure 1B). After deriving daily totals for distance, time, and number of runs and deriving cumulative sums over the 28 days period we report that, mice on average ran a total distance of 316 km (166—428, SEM 32.3), utilised the running wheel for an average of 128 h (70—128, SEM 677) and performed on average a total of 4,670 individual runs (4,048—5,799, SEM 220) (Figure 1B). All three exercise parameters were accrued over the 28 days in a consistent linear fashion with an almost invariant slope (Figure 1B). To understand the exercise behaviour of mice in further detail, daily averages were next computed for distance, time, number of runs as well as mean and maximum speed. Mice on average, ran 10.2 km (5.54—14.2, SE 1.84), 257 min (137—323, SE 24.3), and 155 runs (134—192, SE7.9) runs per day, with a max speed of 0.598 m/s (0.417—0.679, SE 0.035), and average speed of 0.060 m/s (0.033—0.083, SE 0.0068) (Figure 1C). Intraday exercise activity as measured by running distance, time, number of runs and maximum speed occurred cyclically and coincided with the 12 h light/dark housing conditions with most of the exercise activity being performed during the nocturnal phase (Figure 1D).
Next, we compared the daily exercise output of mice with access to running wheels to that of mice performing spontaneous in-cage activity without running wheels as reported by Pernold et al. [ 2019]. The metrics chosen for this comparison were daily distance and daily mean speed measured using both CST and video tracking by Pernold et al. [ 2019]. In the absence of a running wheel, mice covered a distance of 0.2 km per day at an average speed of 0.0024 m/s (Figure 1E) which represents approximately 50-fold decrease in activity level when compared to the 10.2 km daily distance and 0.060 m/s daily mean speed registered on the running wheel system (Figure 1E).
Taken together, tracking of the exercise output over a 28-day period demonstrates that the use of a running wheel setup is a reliable method for increasing physical activity in mice, orders of magnitude over the level of activity performed as a result of spontaneous in-cage movement.
## Voluntary exercise protects retinal function and morphology against degeneration
To investigate the effect of exercise on retinal protection against degeneration, mice ($$n = 8$$–9) were subjected to 5 days of photo-oxidative damage immediately after 30 days of exercise then retinal function and morphology were assessed using ERG, OCT and fundoscopy. Relative to sedentary controls, exercised mice had significantly improved retinal function for both a-wave (photoreceptor function) and b-wave (second order neuron) ($p \leq 0.05$, Figures 2A–C). Fundus/OCT imaging (Figures 2D–G), further demonstrate the protective effects of exercise on the retina during degeneration. Areas of free of degenerative lesions (defines as retinal regions presenting with of hyperreflective puncta, hyperpigmentation and severe ONL loss shown by dashed areas in Figure 2D) were larger in exercised mice ($39.7\%$ ± $6.84\%$) compared to sedentary controls ($17.4\%$ ± $2.42\%$) (Figure 2E, $p \leq 0.05$). Thickness measurements carried out two-disc diameters superior to the optic nerve head indicated that both the ONL (Figure 2E) and whole retinal thickness (WR) (Figure 2F) were significantly ($p \leq 0.05$) larger in the exercise animals across five distinct eccentricities. Exercised mice had a WR of 116.2 ± 2.08 μm and ONL thickness of 26.7 ± 1.73 μm while the same two measurements were 104.8 ± 1.34 μm and 12.4 ± 0.99 μm for sedentary mice respectively. In the 14-day cohort, no significant differences were observed in either ERG or ONL thickness (Supplementary Figure S1).
**FIGURE 2:** *Voluntary exercise protects retinal function and morphology against degeneration. (A–C) Mice with access to voluntary exercise had significantly improved retinal function in both (A) a-wave, and (B) b-wave measures. (C) Representative ERG trace of exercised and sedentary mice using a flash intensity of 1.6 log cd.s.m2. (D) Representative fundi from exercise and sedentary mice. White dashed lines show retinal areas free of severe degeneration marked by the presence of hyper-reflective puncta and hyperpigmentation. Blue arrows show the corresponding location of the OCT retinas on the right. (E) Quantification of lesion free areas expressed as a percent of the total retinal area superior to the optic nerve head. (F) Quantification of total retinal thickness and (G) ONL thickness. OCT images used for thickness measurements were captured at two-disc diameters superior to the optic nerve head measurements were taken directly above the optic nerve head and at 0.3 and 0.6 mm nasally and temporally. *Significance using two-way ANOVA with Sidak’s post hoc test for multiple comparisons, or Student's t-test, p < 0.05 and error bars indicate SEM, n = 9.*
## Voluntary exercise protects the retina against photoreceptor cell death and inflammation
Given the partial protection in retinal function and preservation of retinal morphology observed in exercised mice, (Figure 2), key features of retinal degeneration (photoreceptor cell death, and presence and activation of microglia/macrophage immune cells) were measured and compared between sedentary and exercised mice following PD. Compared to sedentary controls, mice with access to voluntary exercise had significantly reduced levels of photoreceptor cell death ($p \leq 0.05$, Figures 3A,B), as measured by a decreased number of TUNEL+ cells across the ONL, and in the superior ONL ($p \leq 0.05$, Figure 3C), and an increased number of photoreceptor rows ($p \leq 0.05$, Figure 3D). Further, as a measure of retinal inflammation, IBA1+ microglial/macrophage cells were counted and morphologically characterised in the ONL. In exercised mice, IBA1+ cells were significantly reduced in the ONL compared to respective sedentary controls ($p \leq 0.05$, Figures 3E,F), notably with significant reductions in the superior retina ($p \leq 0.05$, Figure 3G), and in both ramified and amoeboid IBA1+ morphologies ($p \leq 0.05$, Figure 3H). This was true also for the 14-day cohort where photoreceptor cell death and inflammation as measured by IBA1 was significantly reduced in the exercised mice (Supplementary Figure S2). Collectively these results demonstrate the protective effect of voluntary exercise against retinal degeneration within the 28-day time frame, with functional preservation attributed to reduced levels of photoreceptor cell death and retinal inflammation in exercised mice.
**FIGURE 3:** *Voluntary exercise protects the retina against photoreceptor cell death and inflammation. Mice with access to voluntary exercise had decreased levels of photoreceptor cell death, as shown in (A–B) representative confocal images and quantified by (C) reduced numbers of total and superior TUNEL+ cells (red) in the ONL, and (D) increased numbers of photoreceptor nuclei rows, compared to sedentary controls. Further, exercised mice had reduced levels of IBA1+ cells (green) in the ONL, as shown by (E–F) representative confocal images, and quantified by (G–H) decreased numbers of total, superior, ramified and amoeboid IBA1+ cells in the ONL. *Significance using a Student's t-test, p < 0.05 and error bars indicate SEM. Scale bar = 50 μM. n = 9.*
## Voluntary exercise induces gene expression profile similarities to healthy control mice
To investigate the molecular pathways underpinning the observed retinal protection, bulk RNA sequencing was performed on retinal lysates from exercised and sedentary mice exposed to PD and compared to dim sedentary mice. Following read count normalisation (Figure 4A), a two-dimensional principal component analysis was employed to visualise global differences between the expression of retinal mRNA profiles from exercised and sedentary mice (Figure 4B). Using envfit analysis and permutation-based statistical testing, we determined that exercise and sedentary mice are arranged in two distinct clusters on the PCA space ($p \leq 0.05$), thus suggesting that when compared to sedentary mice, exercise induces consistent gene expression changes in mice exposed to PD (Figure 4B). Next, we sought to determine if global gene expression patterns in exercised mice shifts the RNA profile of a heathy (non-PD) mouse retina. To achieve, a subset of genes (differentially expressed between exercised and sedentary mice, log2fold change (FC) > 0.5, $p \leq 0.05$) were identified (Figure 4C; orange dots), then using the expression values of this subset of genes, the Euclidean distances (ED) of both exercise and sedentary mice to dim controls were calculated. Hierarchical clustering of EDs computed based on differentially expressed genes, indicated that $\frac{7}{9}$ mice had a gene expression profile more similar to a heathy retina than sedentary controls (Figure 4D). Besides clustering, the magnitude of EDs was statistically tested (Figure 4E) and revealed that the exercise shortens the separation between heathy controls and PD-retinas (EDdim-sedentary = 24.5 ± 0.47, EDdim-active = 20.2 ± 0.21, $p \leq 0.05$). Taken together these results suggest that exercise partially shifts the transcriptome of the PD retina towards an expression profile that aligns more closely to a dim-reared, “healthy” retina.
**FIGURE 4:** *Voluntary exercise induces gene expression to a profile more similar to healthy control mice. (A) Relative Log Expression plots showing effective normalisation. (B) PCA plot showing significant clustering of the exercise and control groups. p-values were derived by performing envfit analysis on the PCA object with 106 permutations. Shaded area represents the probability ellipse of each group calculated by using the group standard deviation and a confidence limit of 0.95. (C) Volcano plot showing differentially expressed genes between exercised and sedentary retinas (absolute Log2FC>0.5, p < 0.05). (D) Clustering of Euclidean distances calculated using expression values of differentially expressed genes. Distances for both exercise and sedentary groups were calculated relative to a reference matrix containing expression values of differentially expressed genes (exercise vs. sedentary) in heathy dim-reared retinas. This heatmap indicates that the expression values of the 284 differentially expressed genes trend more strongly towards heathy controls (dim-reared mice) values in exercised mice than in sedentary mice. (E) Summary of Euclidean distances between exercise and sedentary mice relative to dim-reared mice based on the 284 differentially expressed genes in (C). Significance tested using Wilcoxon signed-rank test. n = 4-9.*
## Modular gene co-expression analysis identifies inflammatory and extracellular matrix integrity pathway modulation in response to voluntary exercise
As the retinal gene expression profile from exercised mice was found to be trending towards that of healthy control (dim) retinas, to identify the specific gene and gene pathways modulated in response to voluntary exercise and responsible for this profile shift, modular gene co-expression analysis was performed in combination with gene set enrichment analysis. Following modularisation of normalised gene counts, six gene expression modules were identified, with modules (M) M2, M3, and M4 showing significantly positive enrichment, and modules M1, M5 and M6 showing significantly decreased enrichment in exercised mice compared to sedentary controls ($p \leq 0.05$, Figure 5A). Of the six modules (Supplementary Table S2), M5 and M6 (Figure 5B) were further explored because they contained signalling pathways and immune mediators central to the pathogenesis of retinal degeneration. Pathways associated M5 included microglial and leukocyte cell migration and phagocytosis, complement activation and inflammasome activation ($p \leq 0.05$, Figure 5C). Consistent with a downregulation of in the enrichment of pathways in M5, key mediators of retinal inflammation such as C3, C1q, Cx3CR1 and Cd74 were observed to be downregulated in exercised animals (Figure 5C) relative to sedentary controls. Further, pathways significantly associated with M6 genes, and known to play key roles in retinal degeneration included extracellular matrix modulation, integrity and organisation, and oxidative stress responses ($p \leq 0.05$, Figure 5D).
**FIGURE 5:** *Modular gene co-expression analysis identifies inflammatory and extracellular matrix integrity pathway modulation in response to voluntary exercise. (A) Gene module enrichment identified six modules (M) which were significantly enriched (M2-M4, positively enriched, M1, M5 and M6, decreased enrichment) compared to sedentary controls. (B) Profile plots of genes within modules M5 and M6 to show representative similar gene expression trends within modules. (C,D) Pathway analyses showing inflammatory and extracellular matrix integrity pathways were associated with genes in M5 and M6 respectively.*
Taken together, these results support a role for exercise-mediated retinal protection against degeneration through a combination of influencing inflammation and ECM modulation and show a transcriptomic shift towards a healthy phenotype.
## Discussion
The benefit exercise provides to the CNS are only just starting to be discovered, with insight into the molecular changes that occur, in particular in the retina, still largely unknown. With accumulating evidence demonstrating the ability of physical activity to surmount neuro- and retinal degenerative pathophysiology (reviewed in Chu-Tan et al. [ 2022], there is a critical need to unlock the complete molecular events underpinning exercise-induced neuroprotection. Therefore, in this work, using a voluntary model of aerobic exercise that allows for free access to open running wheels and a well-characterised degeneration model that recapitulates key facets of AMD pathogenesis, we provide an overview of the molecular changes attributed to voluntary exercise in the retina, and how they may confer protection against retinal degeneration. Our results demonstrated that voluntary exercise in mice for 28 days prior to retinal degeneration provided significant protection to retinal function and integrity, and reduced levels of photoreceptor cell death and microglial/macrophage activation/infiltration. Further, that this observed protection may be attributable to decreased activation of inflammatory pathways and preservation of extracellular matrix integrity as hyper-inflammatory responses and loss of ECM are implicated in retinal degeneration and AMD progression. Overall, our results indicated that exercise was able to induce a transcriptomic gene expression profile shift towards that of a healthy retina, ultimately supporting the potential use of exercise as a non-pharmacological intervention to slow the progression of retinal degenerations and AMD, and identification of a novel therapeutic factors that can target key pathways such as complement.
## Exercise induces molecular changes that drive towards a healthy retinal phenotype
We have provided for the first time a foundational picture of the major molecular pathways underlying exercise-mediated retinal protection and identified two major themes that had overall transcriptomic shifts towards a healthy phenotype: inflammation, and extracellular matrix integrity.
Over-activation of the immune response within the CNS is a hallmark feature of neurodegeneration and has been established to influence the progression of retinal degenerations, including AMD (Ambati et al., 2013; Kauppinen et al., 2016). Chronic and progressive inflammation in retinal degenerations is primarily mediated by the innate immune system, and largely controlled by both the resident retinal microglia as well as recruited peripheral macrophages (Amor et al., 2010; Ambati et al., 2013; Kauppinen et al., 2016). Reducing pathological levels of inflammation may therefore be a key therapeutic strategy in slowing the progression of retinal degenerations, making the anti-inflammatory properties of exercise a considerable therapeutic interest. There is a strong link between exercise and the inflammatory system in the human body including in the CNS (Nieman and Wentz, 2019). Exercise is known to modulate key inflammatory mechanisms, mitigating and regulating systemic inflammation in neurological diseases such as multiple sclerosis (Dalgas et al., 2012), systemic lupus erythematous (Perandini et al., 2012), Alzheimer’s and Parkinson’s diseases (Fuller et al., 2020), and preclinical models of retinal degeneration (Crowston et al., 2017; Sellers et al., 2019; Zhang et al., 2019; Dantis Pereira de Campos et al., 2020). However, these anti-neuroinflammatory effects, whilst observed, have not been well-substantiated or often investigated past single target identification.
Results from this study support the anti-inflammatory effects of exercise, with reduced infiltration and activation of IBA1+ microglia/macrophages in the retina of exercised mice. in particular, reduced level of microglia/macrophages were detected in the ONL of the central/superior retina—an area well-characterised to be the primary site of damage during light-induced degeneration (Natoli et al., 2016). In support of these observations, pathway analysis of RNA sequencing results indicated decreased expression of microglial/macrophage markers, including Cd68, Cd74 and Cx3Cr1—all known to have increased expression and drive pathological inflammatory processes in AMD (Ambati et al., 2013). Further, bioinformatic analyses indicated that the complement system, which has been previously identified as a major instigator of inflammatory dysregulation in AMD (Edwards et al., 2005; Hageman et al., 2005; Haines et al., 2005; Klein et al., 2005), was less active in exercised animals. Notably, complement pathway genes C1qa, C1qb, C1qc, C3 and Serping1 were all significantly downregulated in the retinas of exercised mice. In addition, the Nlrp3 pathway also appeared to be less active in exercised animals. In AMD, NLRP3-dependent cleavage and activation of pro-inflammatory cytokine IL-1β (Halle et al., 2008) has been shown to drastically increase excessive inflammation in the retina, (Wooff et al., 2019), and is identified to play a key role in disease pathogenesis causing progressive photoreceptor cell death. Taken together, these results highlight potential pathways modulated in response to exercise that could be key to dampening the observed retinal immune response and providing protection against degeneration.
In addition to inflammatory modulation, key findings from this work identified that extracellular matrix integrity pathways were also significantly modulated in response to exercise. The extracellular matrix, acellular protein-rich scaffold, consisting of proteoglycans and glycoproteins, plays an important homeostatic role in maintaining retinal health and integrity functioning in both structural and mechanical support (Al-Ubaidi et al., 2013). In retinal degenerations including AMD, extracellular matrix degradation and remodelling have been strongly associated with disease pathogenesis (Al-Ubaidi et al., 2013). Results from our study show that in exercised mice extracellular matrix genes such as Itgam (Cd11b), a key mediator of monocyte adhesion (Solovjov et al., 2005), were significantly downregulated compared to sedentary mice. As integrin-mediated adhesion to the ECM plays a key role in infiltration of immune cells during damage or degeneration (Baiula et al., 2021; Mrugacz et al., 2021), modulations in this pathway may lead to reduced immune cell presence in the retina hereby haltering damaging and pathological immune cascades. Our results strongly support this hypothesis, with an observed reduction in recruited IBA1+ cells and microglial/macrophage markers, as well as a downregulation in several notable immune pathways known to be propagated by these immune cells and involved in AMD disease progression (Rutar et al., 2015; Jiao et al., 2018). Additionally, Tissue Inhibitor of Metalloproteinases 3 (Timp3) which was also found to be downregulated in exercised mice, has been heavily implicated as a potential candidate in AMD pathogenesis, with elevated levels seen to contribute to the thickening of Bruch’s membrane commonly seen in AMD patients (Kamei and Hollyfield, 1999). The diverse roles of Timp3 in inflammation, angiogenesis, and extracellular matrix turnover/degradation, have made it a closely looked at molecule for complex degenerative diseases including Alzheimer’s disease and AMD (Dewing et al., 2019).
These novel results provide further insight into the molecular pathways that are involved in providing neuronal protection to the retina following exercise. It must be noted that unlike in previous studies (Poo, 2001; Binder and Scharfman, 2004), we did not observe a significant change in brain-derived neurotrophic factor (BDNF) which has been a heavily postulated to be a crucial molecule in exercise-mediated neuronal protection (Poo, 2001; Binder and Scharfman, 2004). However, the sensitive nature of Bdnf deems it difficult to detect changes without conducting high-sensitivity experiments on its effect on specific pathways such as TrkB (Allen et al., 2018) or detecting their protein rather than RNA levels. Further, the protective effects of exercise are pleiotropic with the same gene influencing different phenotypic traits and expression patterns differing based on various forms of exercise (Côté et al., 2011; Keeler et al., 2012). Therefore, we cannot rule out a possibility of Bdnf playing a role in the protection that we observed despite no significant changes to it or its primary receptor.
## Voluntary exercise is protective against retinal degeneration
To date, the majority of animal studies investigating the effect of exercise in the retina have used forced exercise models where an often unpleasant or painful stimulus is used to ensure the animals complete a defined amount of physical activity (Ji et al., 2013; Kim et al., 2013; Chrysostomou et al., 2014; Lawson et al., 2014; Chrysostomou et al., 2016; Allen et al., 2018; Mees et al., 2019; Dantis Pereira de Campos et al., 2020). While it is understandable that studies using forced-exercise models use this method in an attempt to ensure equal and consistent levels of desired exercise, output measures from our study revealed that mice would voluntarily and consistently utilize the running wheels, running large distances of 10.2 km/day on average, mitigating any need to force this seemingly natural behavior onto them. Further, given the stress and anxiety-like behaviours known to be inflicted in these forced exercise models (Leasure and Jones, 2008), forced-exercise methods may potential confound the true effects of exercise. Studies have in fact demonstrated not only significant differences in brain behaviours, but also differences in the intensity and duration of activity under forced and voluntary exercise animal paradigms, despite attempts at standardization (Leasure and Jones, 2008). Even when distance was held constant, voluntary exercisers reached much higher speeds but for shorter time periods (43.7 m/min in voluntary exercise animals and 15.5 m/min in forced exercise animals) (Leasure and Jones, 2008). Due to the much higher speeds in the voluntary exercise groups, to effectively standardise the distance; forced exercise animals had to run for far-longer periods of time, creating inherent differences within the exercise paradigms (Leasure and Jones, 2008), and making the results difficult to compare and decipher. Further, stress caused by such models may disturb the inflammatory balance of the retina and impact on output measures as a result (Malan et al., 2020). As the use and output measures of voluntary models of exercise more closely mimic what may be prescribed in a clinical setting to human patients, it appears more intuitive to use voluntary models such as the one in this study to truly understand the molecular mechanisms and pathways underlying exercise-induced neuroprotection.
It should also be noted that our supplementary results demonstrated that there at 14 days of exercise, the protective effect seen at 28 days is not present in the functional results and only in histology did we observe protection against photo-oxidative damage. Whilst interesting and potentially indicative of a certain amount of exercise needed to provide the full suite of protection, these results cannot be directly compared due to differences in sex and age of animals for the experiments.
## Future directions
It remains unclear precisely how systemic effects from exercise can target the retina and confer protection against degeneration. It is likely however that tissue crosstalk is occurring to communicate the message of exercise to the rest of the body. Landmark studies have in fact identified that exercise-mediated protection may be facilitated, in part, by cellular communication via extracellular vesicles (EV) (Frühbeis et al., 2015; Whitham et al., 2018). EV (including exosomes and microvesicles) are nanosized cell-to-cell delivery vesicles which are released by nearly all cell types in the body and function to induce a biological response in recipient cells (Hessvik and Llorente, 2018) via the transfer of molecular cargo, including proteins, lipids and non-coding RNA, such as miRNA (Hessvik and Llorente, 2018). Studies have shown that following exhaustive cycle and treadmill exercise, a significant increase in EV numbers was found in the circulation immediately post exercise (Frühbeis et al., 2015; Whitham et al., 2018; Brahmer et al., 2019). Further, beneficial molecular cargo including proteins and miRNA carried within exercise-EV, have been shown to be responsible for exerting systemic neuroprotective effects, including promoting advantageous metabolic changes (Castaño et al., 2020), increasing skeletal muscle fibre growth and recovery (Wu et al., 2022), and modulating immune responses in the CNS (Frühbeis et al., 2015; Whitham et al., 2018; Vechetti et al., 2021). These findings largely support a role for EV/EV cargo in mediating exercise-induced systemic protection, suggesting that investigations into the role of EV in mediating retinal protection during exercise warrant further investigations, and should be avenues for future exploration.
## Conclusion
The benefits of exercise are far-ranging with increasing evidence demonstrating the therapeutic effects of exercise in the CNS and retina. Here we have demonstrated that voluntary exercise is protective to the retina both in function and retinal integrity, likely due to inflammatory pathway modulation and maintenance of the extracellular matrix. These results will help lay the foundation for deeper investigations into the therapeutic molecular mechanisms that exercise provides in protecting against retinal degenerations such as AMD.
## Data availability statement
The datasets presented in this study can be found in the article and Supplementary Material and is accessible for download from the BioProject repository under project ID PRJNA934406.
## Ethics statement
The animal study was reviewed and approved by Australian National University Animal Ethics Committee.
## Author contributions
Project conceptualisation—JC and RN; Methodology—JC, AC, MK, ME, and PG; Investigation—JC, AC, MK, ME, and PG; Data analysis—JC, AC, MK, ME, and PG; Manuscript writing—JC, AC, YW, and RN; Manuscript editing and reviewing—JC, AC, YW, TK, JB, KB, JN, and RN. All authors read and approved the final manuscript.
## Conflict of interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher’s note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## Supplementary material
The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fphys.2023.1116898/full#supplementary-material
## References
1. Allen R. S., Hanif A. M., Gogniat M. A., Prall B. C., Haider R., Aung M. H.. **TrkB signalling pathway mediates the protective effects of exercise in the diabetic rat retina**. *Eur. J. Neurosci.* (2018) **47** 1254-1265. DOI: 10.1111/ejn.13909
2. Al-Ubaidi M. R., Naash M. I., Conley S. M.. **A perspective on the role of the extracellular matrix in progressive retinal degenerative disorders**. *Invest. Ophthalmol. Vis. Sci.* (2013) **54** 8119-8124. DOI: 10.1167/iovs.13-13536
3. Ambati J., Atkinson J. P., Gelfand B. D.. **Immunology of age-related macular degeneration**. *Nat. Rev. Immunol.* (2013) **13** 438-451. DOI: 10.1038/nri3459
4. Amor S., Puentes F., Baker D., van der Valk P.. **Inflammation in neurodegenerative diseases**. *Immunology* (2010) **129** 154-169. DOI: 10.1111/j.1365-2567.2009.03225.x
5. Baiula M., Caligiana A., Bedini A., Zhao J., Santino F., Cirillo M.. **Leukocyte integrin antagonists as a novel option to treat dry age-related macular degeneration**. *Front. Pharmacol.* (2021) **11** 617836. DOI: 10.3389/fphar.2020.617836
6. Bales K. L., Chacko A. S., Nickerson J. M., Boatright J. H., Pardue M. T.. **Treadmill exercise promotes retinal astrocyte plasticity and protects against retinal degeneration in a mouse model of light-induced retinal degeneration**. *J. Neurosci. Res.* (2022) **100** 1695-1706. DOI: 10.1002/jnr.25063
7. Barone I., Novelli E., Piano I., Gargini C., Strettoi E.. **Environmental enrichment extends photoreceptor survival and visual function in a mouse model of retinitis pigmentosa**. *PLoS One* (2012) **7** e50726. DOI: 10.1371/journal.pone.0050726
8. Barone I., Novelli E., Strettoi E.. **Long-term preservation of cone photoreceptors and visual acuity in rd10 mutant mice exposed to continuous environmental enrichment**. *Mol. Vis.* (2014) **20** 1545-1556. PMID: 25489227
9. Binder D. K., Scharfman H. E.. **Brain-derived neurotrophic factor**. *Growth Factors.* (2004) **22** 123-131. DOI: 10.1080/08977190410001723308
10. Brahmer A., Neuberger E., Esch-Heisser L., Haller N., Jorgensen M. M., Baek R.. **Platelets, endothelial cells and leukocytes contribute to the exercise-triggered release of extracellular vesicles into the circulation**. *J. Extracell. Vesicles* (2019) **8** 1615820. DOI: 10.1080/20013078.2019.1615820
11. Buchman A. S., Boyle P. A., Yu L., Shah R. C., Wilson R. S., Bennett D. A.. **Total daily physical activity and the risk of AD and cognitive decline in older adults**. *Neurology* (2012) **78** 1323-1329. DOI: 10.1212/WNL.0b013e3182535d35
12. Castaño C., Mirasierra M., Vallejo M., Novials A., Párrizas M.. **Delivery of muscle-derived exosomal miRNAs induced by HIIT improves insulin sensitivity through down-regulation of hepatic FoxO1 in mice**. *Proc. Natl. Acad. Sci.* (2020) **117** 30335-30343. DOI: 10.1073/pnas.2016112117
13. Chen H., Zhang S. M., Schwarzschild M. A., Hernan M. A., Ascherio A.. **Physical activity and the risk of Parkinson disease**. *Neurology* (2005) **64** 664-669. DOI: 10.1212/01.WNL.0000151960.28687.93
14. Chrysostomou V., Kezic J. M., Trounce I. A., Crowston J. G.. **Forced exercise protects the aged optic nerve against intraocular pressure injury**. *Neurobiol. Aging* (2014) **35** 1722-1725. DOI: 10.1016/j.neurobiolaging.2014.01.019
15. Chrysostomou V., Galic S., van Wijngaarden P., Trounce I. A., Steinberg G. R., Crowston J. G.. **Exercise reverses age-related vulnerability of the retina to injury by preventing complement-mediated synapse elimination via a BDNF-dependent pathway**. *Aging Cell* (2016) **15** 1082-1091. DOI: 10.1111/acel.12512
16. Chu‐Tan J. A., Kirkby M., Natoli R.. **Running to save sight: The effects of exercise on retinal health and function**. *Clin. Exp. Ophthalmol.* (2021)
17. Chu-Tan J. A., Kirkby M., Natoli R.. **Running to save sight: The effects of exercise on retinal health and function**. *Clin. Exp. Ophthalmol.* (2022) **50** 74-90. DOI: 10.1111/ceo.14023
18. Colcombe S., Kramer A. F.. **Fitness effects on the cognitive function of older adults: A meta-analytic study**. *Psychol. Sci.* (2003) **14** 125-130. DOI: 10.1111/1467-9280.t01-1-01430
19. Côté M. P., Azzam G. A., Lemay M. A., Zhukareva V., Houlé J. D.. **Activity-dependent increase in neurotrophic factors is associated with an enhanced modulation of spinal reflexes after spinal cord injury**. *J. Neurotrauma* (2011) **28** 299-309. DOI: 10.1089/neu.2010.1594
20. Crowston J. G., Fahy E. T., Fry L., Trounce I. A., van Wijngaarden P., Petrou S.. **Targeting retinal ganglion cell recovery**. *Eye (Lond).* (2017) **31** 196-198. DOI: 10.1038/eye.2016.281
21. Dalgas U., Severinsen K., Overgaard K.. **Relations between 6 minute walking distance and 10 meter walking speed in patients with multiple sclerosis and stroke**. *Arch. Phys. Med. Rehabil.* (2012) **93** 1167-1172. DOI: 10.1016/j.apmr.2012.02.026
22. Dantis Pereira de Campos T., da Cruz Rodrigues K. C., Martins Pereira R., Morelli A. P., da Rocha A. L., Dos Santos Canciglieri R.. **Short-Term combined exercise improves inflammatory profile in the retina of obese mice**. *Int. J. Mol. Sci.* (2020) **21** 6099. DOI: 10.3390/ijms21176099
23. Deloitte Access Economics P.. *Eyes on the future: A clear outlook on age-related macular degeneration.* (2011)
24. Dewing J. M., Carare R. O., Lotery A. J., Ratnayaka J. A.. **The diverse roles of TIMP-3: Insights into degenerative diseases of the senescent retina and brain**. *Cells* (2019) **9** 39. DOI: 10.3390/cells9010039
25. Edwards A. O., Ritter R., Abel K. J., Manning A., Panhuysen C., Farrer L. A.. **Complement factor H polymorphism and age-related macular degeneration**. *Science* (2005) **308** 421-424. DOI: 10.1126/science.1110189
26. Frühbeis C., Helmig S., Tug S., Simon P., Krämer-Albers E-M.. **Physical exercise induces rapid release of small extracellular vesicles into the circulation**. *J. Extracell. Vesicles* (2015) **4** 28239. DOI: 10.3402/jev.v4.28239
27. Fuller O. K., Whitham M., Mathivanan S., Febbraio M. A.. **The protective effect of exercise in neurodegenerative diseases: The potential role of extracellular vesicles**. *Cells* (2020) **9** 2182. DOI: 10.3390/cells9102182
28. Gopinath B., Liew G., Burlutsky G., Mitchell P.. **Physical activity and the 15-year incidence of age-related macular degeneration**. *Invest. Ophthalmol. Vis. Sci.* (2014) **55** 7799-7803. DOI: 10.1167/iovs.14-15575
29. Hageman G. S., Anderson D. H., Johnson L. V., Hancox L. S., Taiber A. J., Hardisty L. I.. **A common haplotype in the complement regulatory gene factor H (HF1/CFH) predisposes individuals to age-related macular degeneration**. *Proc. Natl. Acad. Sci. U. S. A.* (2005) **102** 7227-7232. DOI: 10.1073/pnas.0501536102
30. Haines J. L., Hauser M. A., Schmidt S., Scott W. K., Olson L. M., Gallins P.. **Complement factor H variant increases the risk of age-related macular degeneration**. *Science* (2005) **308** 419-421. DOI: 10.1126/science.1110359
31. Halle A., Hornung V., Petzold G. C., Stewart C. R., Monks B. G., Reinheckel T.. **The NALP3 inflammasome is involved in the innate immune response to amyloid-beta**. *Nat. Immunol.* (2008) **9** 857-865. DOI: 10.1038/ni.1636
32. Hanif A. M., Lawson E. C., Prunty M., Gogniat M., Aung M. H., Chakraborty R.. **Neuroprotective effects of voluntary exercise in an inherited retinal degeneration mouse model**. *Invest. Ophthalmol. Vis. Sci.* (2015) **56** 6839-6846. DOI: 10.1167/iovs.15-16792
33. Hessvik N. P., Llorente A.. **Current knowledge on exosome biogenesis and release**. *Cell. Mol. Life Sci.* (2018) **75** 193-208. DOI: 10.1007/s00018-017-2595-9
34. Ji E. S., Ko I. G., Cho J. W., Davis R. W., Hwang G. Y., Jee Y. S.. **Treadmill exercise inhibits apoptotic neuronal cell death with suppressed vascular endothelial growth factor expression in the retinas of the diabetic rats**. *J. Exerc Rehabil.* (2013) **9** 348-353. DOI: 10.12965/jer.130043
35. Jiao H., Rutar M., Fernando N., Yednock T., Sankaranarayanan S., Aggio-Bruce R.. **Subretinal macrophages produce classical complement activator C1q leading to the progression of focal retinal degeneration**. *Mol. Neurodegener.* (2018) **13** 45. DOI: 10.1186/s13024-018-0278-0
36. Kamei M., Hollyfield J. G.. **TIMP-3 in bruch's membrane: Changes during aging and in age-related macular degeneration**. *Invest. Ophthalmol. Vis. Sci.* (1999) **40** 2367-2375. PMID: 10476804
37. Kauppinen A., Paterno J. J., Blasiak J., Salminen A., Kaarniranta K.. **Inflammation and its role in age-related macular degeneration**. *Cell. Mol. life Sci.* (2016) **73** 1765-1786. DOI: 10.1007/s00018-016-2147-8
38. Keeler B. E., Liu G., Siegfried R. N., Zhukareva V., Murray M., Houlé J. D.. **Acute and prolonged hindlimb exercise elicits different gene expression in motoneurons than sensory neurons after spinal cord injury**. *Brain Res.* (2012) **1438** 8-21. DOI: 10.1016/j.brainres.2011.12.015
39. Kim D. Y., Jung S. Y., Kim C. J., Sung Y. H., Kim J. D.. **Treadmill exercise ameliorates apoptotic cell death in the retinas of diabetic rats**. *Mol. Med. Rep.* (2013) **7** 1745-1750. DOI: 10.3892/mmr.2013.1439
40. Klein R. J., Zeiss C., Chew E. Y., Tsai J. Y., Sackler R. S., Haynes C.. **Complement factor H polymorphism in age-related macular degeneration**. *Science* (2005) **308** 385-389. DOI: 10.1126/science.1109557
41. Larson E. B., Wang L., Bowen J. D., McCormick W. C., Teri L., Crane P.. **Exercise is associated with reduced risk for incident dementia among persons 65 years of age and older**. *Ann. Intern Med.* (2006) **144** 73-81. DOI: 10.7326/0003-4819-144-2-200601170-00004
42. Law C. W., Chen Y., Shi W., Smyth G. K.. **voom: precision weights unlock linear model analysis tools for RNA-seq read counts**. *Genome Biol.* (2014) **15** R29. DOI: 10.1186/gb-2014-15-2-r29
43. Lawson E. C., Han M. K., Sellers J. T., Chrenek M. A., Hanif A., Gogniat M. A.. **Aerobic exercise protects retinal function and structure from light-induced retinal degeneration**. *J. Neurosci.* (2014) **34** 2406-2412. DOI: 10.1523/JNEUROSCI.2062-13.2014
44. Leasure J. L., Jones M.. **Forced and voluntary exercise differentially affect brain and behavior**. *Neuroscience* (2008) **156** 456-465. DOI: 10.1016/j.neuroscience.2008.07.041
45. Malan L., Hamer M., von Känel R., van Wyk R. D., Wentzel A., Steyn H. S.. **Retinal-glia ischemia and inflammation induced by chronic stress: The SABPA study**. *Brain, Behav. Immun. - Health* (2020) **2** 100027. DOI: 10.1016/j.bbih.2019.100027
46. McGuinness M. B., Le J., Mitchell P., Gopinath B., Cerin E., Saksens N. T. M.. **Physical activity and age-related macular degeneration: A systematic literature review and meta-analysis**. *Am. J. Ophthalmol.* (2017) **180** 29-38. DOI: 10.1016/j.ajo.2017.05.016
47. Mees L. M., Coulter M. M., Chrenek M. A., Motz C. T., Landis E. G., Boatright J. H.. **Low-intensity exercise in mice is sufficient to protect retinal function during light-induced retinal degeneration**. *Invest. Ophthalmol. Vis. Sci.* (2019) **60** 1328-1335. DOI: 10.1167/iovs.18-25883
48. Meier N. F., Lee D. C., Sui X., Blair S. N.. **Physical activity, cardiorespiratory fitness, and incident glaucoma**. *Med. Sci. Sports Exerc.* (2018) **50** 2253-2258. DOI: 10.1249/MSS.0000000000001692
49. Mrugacz M., Bryl A., Falkowski M., Zorena K.. **Integrins: An important link between angiogenesis, inflammation and eye diseases**. *Cells* (2021) **10** 1703. DOI: 10.3390/cells10071703
50. Munch I. C., Linneberg A., Larsen M.. **Precursors of age-related macular degeneration: Associations with physical activity, obesity, and serum lipids in the inter99 eye study**. *Invest. Ophthalmol. Vis. Sci.* (2013) **54** 3932-3940. DOI: 10.1167/iovs.12-10785
51. Natoli R., Zhu Y., Valter K., Bisti S., Eells J., Stone J.. **Gene and noncoding RNA regulation underlying photoreceptor protection: Microarray study of dietary antioxidant saffron and photobiomodulation in rat retina**. *Mol. Vis.* (2010) **16** 1801-1822. PMID: 20844572
52. Natoli R., Jiao H., Barnett N. L., Fernando N., Valter K., Provis J. M.. **A model of progressive photo-oxidative degeneration and inflammation in the pigmented C57BL/6J mouse retina**. *Exp. Eye Res.* (2016) **147** 114-127. DOI: 10.1016/j.exer.2016.04.015
53. Nieman D. C., Wentz L. M.. **The compelling link between physical activity and the body's defense system**. *J. Sport Health Sci.* (2019) **8** 201-217. DOI: 10.1016/j.jshs.2018.09.009
54. Pedersen B. K., Saltin B.. **Exercise as medicine - evidence for prescribing exercise as therapy in 26 different chronic diseases**. *Scand. J. Med. Sci. Sports* (2015) **25** 1-72. DOI: 10.1111/sms.12581
55. Perandini L. A., de Sa-Pinto A. L., Roschel H., Benatti F. B., Lima F. R., Bonfa E.. **Exercise as a therapeutic tool to counteract inflammation and clinical symptoms in autoimmune rheumatic diseases**. *Autoimmun. Rev.* (2012) **12** 218-224. DOI: 10.1016/j.autrev.2012.06.007
56. Pernold K., Iannello F., Low B. E., Rigamonti M., Rosati G., Scavizzi F.. **Towards large scale automated cage monitoring – diurnal rhythm and impact of interventions on in-cage activity of C57BL/6J mice recorded 24/7 with a non-disrupting capacitive-based technique**. *PLOS ONE* (2019) **14** e0211063. DOI: 10.1371/journal.pone.0211063
57. Poo M. M.. **Neurotrophins as synaptic modulators**. *Nat. Rev. Neurosci.* (2001) **2** 24-32. DOI: 10.1038/35049004
58. Ritchie M. E., Phipson B., Wu D., Hu Y., Law C. W., Shi W.. **Limma powers differential expression analyses for RNA-sequencing and microarray studies**. *Nucleic Acids Res.* (2015) **43** e47. DOI: 10.1093/nar/gkv007
59. Robinson M. D., Oshlack A.. **A scaling normalization method for differential expression analysis of RNA-seq data**. *Genome Biol.* (2010) **11** R25. DOI: 10.1186/gb-2010-11-3-r25
60. Russo P. S. T., Ferreira G. R., Cardozo L. E., Bürger M. C., Arias-Carrasco R., Maruyama S. R.. **CEMiTool: A bioconductor package for performing comprehensive modular co-expression analyses**. *BMC Bioinforma.* (2018) **19** 56. DOI: 10.1186/s12859-018-2053-1
61. Rutar M., Natoli R., Chia R. X., Valter K., Provis J. M.. **Chemokine-mediated inflammation in the degenerating retina is coordinated by Müller cells, activated microglia, and retinal pigment epithelium**. *J. Neuroinflamm.* (2015) **12** 8. DOI: 10.1186/s12974-014-0224-1
62. Sellers J. T., Chrenek M. A., Girardot P. E., Nickerson J. M., Pardue M. T., Boatright J. H.. **Initial assessment of lactate as mediator of exercise-induced retinal protection**. *Adv. Exp. Med. Biol.* (2019) **1185** 451-455. DOI: 10.1007/978-3-030-27378-1_74
63. Seo D-Y., Heo J-W., Ko J. R., Kwak H-B.. **Exercise and neuroinflammation in health and disease**. *Int. Neurourol. J.* (2019) **23** S82-S92. DOI: 10.5213/inj.1938214.107
64. Solovjov D. A., Pluskota E., Plow E. F.. **Distinct roles for the alpha and beta subunits in the functions of integrin alphaMbeta2**. *J. Biol. Chem.* (2005) **280** 1336-1345. DOI: 10.1074/jbc.M406968200
65. Subramanian A., Tamayo P., Mootha Vamsi K., Mukherjee S., Ebert Benjamin L., Gillette Michael A.. **Gene set enrichment analysis: A knowledge-based approach for interpreting genome-wide expression profiles**. *Proc. Natl. Acad. Sci.* (2005) **102** 15545-15550. DOI: 10.1073/pnas.0506580102
66. Team RC (2019). R: A language and environment for statistical computing. Available from: http://www.R-project.org/
.. *R: A language and environment for statistical computing* (2019)
67. Vechetti I. J., Valentino T., Mobley C. B., McCarthy J. J.. **The role of extracellular vesicles in skeletal muscle and systematic adaptation to exercise**. *J. Physiol.* (2021) **599** 845-861. DOI: 10.1113/JP278929
68. Whitham M., Parker B. L., Friedrichsen M., Hingst J. R., Hjorth M., Hughes W. E.. **Extracellular vesicles provide a means for tissue crosstalk during exercise**. *Cell Metab.* (2018) **27** 237-251. DOI: 10.1016/j.cmet.2017.12.001
69. Williams P. T.. **Relationship of incident glaucoma versus physical activity and fitness in male runners**. *Med. Sci. Sports Exerc.* (2009a) **41** 1566-1572. DOI: 10.1249/MSS.0b013e31819e420f
70. Williams P. T.. **Prospective study of incident age-related macular degeneration in relation to vigorous physical activity during a 7-year follow-up**. *Invest. Ophthalmol. Vis. Sci.* (2009b) **50** 101-106. DOI: 10.1167/iovs.08-2165
71. Wong W. L., Su X., Li X., Cheung C. M., Klein R., Cheng C. Y.. **Global prevalence of age-related macular degeneration and disease burden projection for 2020 and 2040: A systematic review and meta-analysis**. *Lancet Glob. health* (2014) **2** e106-e116. DOI: 10.1016/S2214-109X(13)70145-1
72. Wooff Y., Man S. M., Aggio-Bruce R., Natoli R., Fernando N.. **IL-1 family members mediate cell death, inflammation and angiogenesis in retinal degenerative diseases**. *Front. Immunol.* (2019) **10** 1618. DOI: 10.3389/fimmu.2019.01618
73. Wu Y. F., De La Toba E. A., Dvoretskiy S., Jung R., Kim N., Daniels L.. **Development of a cell-free strategy to recover aged skeletal muscle after disuse**. *J. Physiol.* (2022). DOI: 10.1113/JP282867
74. Xu Q., Park Y., Huang X., Hollenbeck A., Blair A., Schatzkin A.. **Physical activities and future risk of Parkinson disease**. *Neurology* (2010) **75** 341-348. DOI: 10.1212/WNL.0b013e3181ea1597
75. Zhang X., Girardot P. E., Sellers J. T., Li Y., Wang J., Chrenek M. A.. **Wheel running exercise protects against retinal degeneration in the I307N rhodopsin mouse model of inducible autosomal dominant retinitis pigmentosa**. *Mol. Vis.* (2019) **25** 462-476. PMID: 31523123
|
---
title: Prediction model for milk transfer of drugs by primarily evaluating the area
under the curve using QSAR/QSPR
authors:
- Tae Maeshima
- Shin Yoshida
- Machiko Watanabe
- Fumio Itagaki
journal: Pharmaceutical Research
year: 2023
pmcid: PMC10036427
doi: 10.1007/s11095-023-03477-1
license: CC BY 4.0
---
# Prediction model for milk transfer of drugs by primarily evaluating the area under the curve using QSAR/QSPR
## Abstract
### Purpose
Information on milk transferability of drugs is important for patients who wish to breastfeed. The purpose of this study is to develop a prediction model for milk-to-plasma drug concentration ratio based on area under the curve (M/PAUC). The quantitative structure–activity/property relationship (QSAR/QSPR) approach was used to predict compounds involved in active transport during milk transfer.
### Methods
We collected M/P ratio data from literature, which were curated and divided into M/PAUC ≥ 1 and M/PAUC < 1. Using the ADMET Predictor® and ADMET Modeler™, we constructed two types of binary classification models: an artificial neural network (ANN) and a support vector machine (SVM).
### Results
M/P ratios of 403 compounds were collected, M/PAUC data were obtained for 173 compounds, while 230 compounds only had M/Pnon-AUC values reported. The models were constructed using 129 of the 173 compounds, excluding colostrum data. The sensitivity of the ANN model was 0.969 for the training set and 0.833 for the test set, while the sensitivity of the SVM model was 0.971 for the training set and 0.667 for the test set. The contribution of the charge-based descriptor was high in both models.
### Conclusions
We built a M/PAUC prediction model using QSAR/QSPR. These predictive models can play an auxiliary role in evaluating the milk transferability of drugs.
### Supplementary Information
The online version contains supplementary material available at 10.1007/s11095-023-03477-1.
## Introduction
Breastfeeding is known to have various advantages, such as improving the immune [1] and cognitive functions of the infant [2], reducing the prevalence of certain diseases in the future [3], reducing the risk of various diseases in the mother [4, 5], and building a good mother-infant relationship [6]. Due to these reasons, the World Health Organization recommends exclusive breastfeeding [7]. According to a 2015 survey in Japan, over $90\%$ of pregnant women chose to breastfeed over other alternatives [8]. However, with advancement in mothers’ age at the time of childbirth and improved treatment for pregnancy complications, the number of women receiving drug treatment during their pregnancy and delivery is increasing [9]. Even if a pregnant woman on medication wishes to breastfeed, there is insufficient information available regarding the milk transferability of drugs. Moreover, it is ethically challenging to conduct human clinical trials to assess the milk transferability of drugs. Several factors affecting this transferability are known, including molecular weight, pH, lipophilicity, and plasma protein-binding [10]. Milk transfer of drugs involves passive diffusion. However, some drugs have also been shown to be actively carried by transporters, such as breast cancer resistance protein (BCRP) [11].
The milk-to-plasma drug concentration (M/P) ratio is an indicator of the transferability of drugs to milk. The safety of drug therapy for nursing mothers cannot be evaluated based on the M/P ratio alone; maternal and infant clearance and other factors must be considered. The M/P ratio is also used to calculate relative infant dose (RID) and exposure index (EI), which are pharmacokinetic measures [12]. These indicators can help determine the precise amount of drug ingested by the child. This is important to evaluate whether breastfeeding can be combined with drug therapy, making these indicators helpful in drug therapy for lactating women [13]. Recently, the M/P ratio has been used as a parameter in the in silico Physiologically based pharmacokinetic model, contributing to the development of drug therapy for lactating women [14].
Prediction of M/P ratios has been attempted since the 1980s, when a phase distribution model was reported to predict M/P ratios from the physicochemical properties of compounds [15, 16]. These were noted for their independence from clinical data and the fact that they did not take into account the effects of active transport [14]. Subsequently, an M/P prediction model using quantitative structure-activity relationships (QSAR/QSPR) was reported. The QSAR/QSPR approach aims to find correlations between structural features or physicochemical constants of a drug and its biological activity, and can be applied to predict physical and chemical properties by means of descriptors that explain changes in the physical or chemical properties of that drug group. A number of linear regression models have been reported for QSAR models predicting M/P ratios [17–23], but because M/P ratio data are collected from individual reports, uncertainties in subjects, measurement methods, and variations in the number of cases may affect the models. A classification model was also constructed based on the idea that prediction by linear regression is not realistic [24, 25]. Even with the establishment of highly accurate models, predicting milk transferability of actively transported drugs remains a challenge [14].
It is important to have organized and curated data for QSAR/QSPR model building. Datasets from previously reported models included inconsistent M/P ratios for animals and inconsistent sampling times for milk and plasma. Colostrum and mature milk also differ in pH, fat content, and secretion, but were not necessarily separated in the dataset. In addition, it is more appropriate to evaluate the transfer of a drug to milk using the area under the curve (AUC) (M/PAUC) rather than its concentration at a specific time point. This is because drug concentrations in maternal plasma and breast milk are not always in equilibrium. Furthermore, some drugs have been reported to take a longer time to reach equilibrium [26, 27]. However, frequent sampling for AUC calculation in clinical practice is not easy, and M/PAUC is not often reported. M/PAUC should be used to evaluate milk transfer.
The purpose of this study was to curate M/P ratio data and build a binomial classification model based on M/PAUC for screening drugs involved in active transport in human mature milk.
## Datasets
Human M/P ratio data were obtained from original papers with reference to various sources such as books [28, 29] and the LactMed database [30]. Only data from mature milk were extracted; data from colostrum up to 7 days postpartum were excluded. Only data evaluated by AUC were used for M/P ratios for model building, and M/P ratios calculated by other methods such as single point evaluation were excluded as M/Pnon-AUC. Reports of misplaced timing between breast milk and maternal blood samples were also excluded. When using data for which the M/P ratio was measured at multiple sampling times but the AUC was not calculated, the trapezoidal method was used to calculate the AUC. When only blood and milk drug concentration graphs were reported, these graphs were reproduced and AUCs were calculated. Protein preparations and metal-containing compounds whose physical properties could not be predicted by ADMET predictor® were excluded. A comprehensive list of selected original articles for each compound can be found in Online Resource.
## Descriptors
The simplified molecular input line entry system (SMILES) format of the compounds was obtained from the public database PubChem [31]. These were incorporated into the ADMET predictor® and 254 descriptors were generated. Descriptors belonging to the following categories were generated: simple constitutional descriptors, topological indices, atom-type electrophysiological state indices, charge-based descriptors, hydrogen bonding descriptors, molecular ionization descriptors, functional groups, Moriguchi descriptors, pattern-recognition flags, and Meylan flags. In contrast, the following categories were excluded: textual description, indicators, and 3D descriptors. The ADMET ModelerTM allows the user to add any descriptor from among the 48 descriptors for the calculated or simulated physical properties of the compounds in ADMET Predictor®, in addition to the 254 molecular descriptors during model construction. The choice of descriptors used to build the model was made in two steps. First, descriptors were reduced in an unsupervised process based on their characteristics and relevance to other descriptors. Descriptors were eliminated based on three conditions: a coefficient of variation lower than the Minimum Coefficient variation setting; a non-zero value lower than the Minimum representation; and a Maximum absolute with the least amount of information when a pair of descriptors shows a higher absolute correlation coefficient than the value set in "Max absolute correlation".
This was followed by a supervised prioritization based on sensitivity. This was done using the functions 'Input Gradient', 'Truncated Linear analysis', 'Iterative truncated linear analysis', and 'Genetic algorithm'. Descriptor selection settings are described in the following sections for each model.
## Artificial neural network (ANN) model settings
Multilayer perceptron was used as the architecture model in the ANN. The Kohonen self-organizing map method was used for the test set selection [32]. That is, using the Kohonen map method, the compounds in the data set are divided into three sets: training set, test set, and validation set. This is a method of selecting test sets from cells in a toroidal two-dimensional Kohonen map that clusters compounds by chemical similarity in a descriptor space. The Kohonen size was set automatically and the Kohonen map was not reused. The minimum test set size was $10\%$. The settings for descriptor number reduction were minimal coefficient of variation: 1, minimum representation: 4, and maximum absolute correlation: 0.98. Sensitivity analysis was performed using truncated linear analysis. One Monte Carlo attempt was applied, and the maximum weight of data was $75\%$. In each ensemble, 33 individual networks were used and the network multiplier was 5. This number multiplied by the ensemble size represents the total number of networks trained per architecture. The Autofill function was used to set the minimum, maximum, and step values of hidden neurons and network inputs.
## Support vector machine (SVM) model setting
The support vector machine was used as the architecture model. The Kohonen map method was used for the test set selection. The Kohonen size was set automatically and the Kohonen map was not reused. The minimum test set size was $10\%$. The settings for descriptor number reduction were minimal coefficient of variation: 1, minimum representation: 4, and maximum absolute correlation: 0.98. In the SVM model, the descriptor "S+log D,” which is the value of log D predicted by the ADMET predictor®, was added manually. When calculating log D, the pH of the maternal blood was set at 7.4. Sensitivity analysis was performed using a genetic algorithm (GA) [33]. The target on which the genetic algorithm is executed can be selected from the grid, row, or cell. Because the cell level is recommended for small datasets, GA was run by the cell. The GA’s max steps was 30,000, and its max training was 1,000. The number of individual SVM models used in each ensemble was 33. The total number of SVM models used to train the ensembles was set to 40. These are both default settings in the software.
## Evaluation of the model using a confusion matrix
To evaluate the model, each index was calculated using a confusion matrix (Fig. 1).Fig. 1Confusion matrix and formulas for each indicator.
## Software
DigitizeIt version 2.3.3 (I. Bormann, Braunschweig, Germany) was used to plot blood and milk concentration curves. ADMET Predictor® version 10.3 (Simulations Plus Inc., Lancaster, CA, USA) was used to generate the molecular descriptors. Furthermore, ADMET ModelerTM version 10.3 (Simulations Plus Inc., Lancaster, CA, USA) was used to build the model.
## Results
The M/P ratios of 403 compounds were obtained from literature. Of these, M/PAUC was reported for 173 compounds, and data for 21 compounds were calculated by replotting the graph. Data for the remaining 230 compounds were obtained using the M/Pnon-AUC. Of the 173 compounds with M/PAUC data, 129 compounds, excluding colostrum, were used in the dataset.
## ANN model performance
Of the 254 descriptors, 37 under-represented descriptors and 40 highly correlated descriptors were excluded. The remaining 177 descriptors were used to build the model as candidate inputs. Of the 129 compounds, 76 were assigned to the training set, 39 to the verification set, and the remaining 14 to the test set. The best model was made using 33 ANNs, 3 neuron, 24 inputs, and 79 weights. Table I lists the effect indicators for model evaluation. The sensitivity was 0.969 for the training set and 0.833 for the test set. The specificity was 0.940 for the training set and 1.000 for the test set. Furthermore, the Matthews correlation coefficient (MCC) was 0. 878 for the training set and 0.861 for the test set. The statistical performance results are shown in Fig. 2a. As shown in Table II, the most contributing descriptor in the ANN model was EEM_XFC. This descriptor is classified as a Charge-based Descriptor. Next was T-RDmtr. The third was Pi_AQc, which is also a Charge-based Descriptor. As an example, acetaminophen was predicted by the ANN model to have M/PAUC ≥ 1 with $95\%$ confidence; the sensitivity of EEM_XFc was -0.53, that of T_RDmtr was -0.77, and that of Pi_AQc was 0.71. On the other hand, for morphine, which is predicted to have M/PAUC ≥ 1 with $65\%$ confidence, EEM_XFc was -0.30, T_RDmtr was -0.64, and Pi_AQc was 0.70, with acetaminophen having the greater absolute value. The maximum uncertainty was $57.5\%$.Table IModel Output Statistics of ANN Model and SVM ModelsModelsData setAccuracyPrecisionSensitivitySpecificityYoudenMCCFalse rateF-measureANNTraining0.9480.8570.9690.9400.9090.8780.0520.910Test0.9291.0000.8331.0000.8330.8610.0710.909SVMTraining0.9911.0000.9711.0000.9710.9790.0090.985Test0.9381.0000.6671.0000.6670.7870.0630.800ANN, artificial neural network; SVM, support vector machine; Youden, Youden index; MCC, Matthews correlation coefficientFig. 2Statistical performance results for ANN model (a) and SVM model (b) using the ADMET Modeler™. Blue circles: training set data; red squares: test set data. FP, False positive; TP, True positive; TN, True negative; FN, False negative. Table IIDescriptors with High Contributions to Binary Classification ModelANN modelSVM modelIndexNameDescriptionSensitivityRelative SensitivityIndexNameDescriptionSensitivityRelative Sensitivity1EEM_XFcMaximum sigma Fukui index on C0.5721.0001SaaaCAtom-type E-state index for aCaa groups2.2681.0002T_RDmtrRelative topological diameter: maximal topological distance divided by the number of atoms0.4680.8172EEM_NFplMinimum sigma Fukui index on polar atoms2.2610.9973Pi_AQcSum of absolute values of Hückel pi atomic charges, but only on C atoms0.4500.7863EEM_XFonMaximum sigma Fukui index on N and O2.2500.9924Pi_FPl4Fourth component of the autocorrelation vector of pi Fukui(+) indices0.4290.7504T_RDmtrRelative topological diameter: maximal topological distance divided by the number of atoms2.2500.9925MinQMinimal PEOE Partial Atomic Charge0.4040.7065S + logDoctanol–water distribution coefficient (log D) calculated from S + pKa and S + logP2.2430.9896F_DbleBDouble bonds as fraction of total bonds0.3970.6946NPA_Q3Third component of the autocorrelation vector of estimated NPA partial atomic charges2.2390.9877T_RadbTopological equivalent of Radb__3D0.3810.6667EEM_XFcMaximum sigma Fukui index on C2.2360.9868ArHdrxl_-OHNumber of aromatic hydroxyl groups0.3230.5658SecAmine_ > NHNumber of primary and aliphatic N secondary amines2.2250.9819EEM_XFplMaximum sigma Fukui index on polar atoms0.3150.5519EEM_NFnpMinimum sigma Fukui index on nonpolar atoms2.2180.97810T_RadaTopological equivalent of Rada__3D0.3110.54310NPA_MinQMinimal Estimated NPA Partial Atomic Charge2.1860.964
## SVM model performance
Of the 255 descriptors, 37 under-represented and 40 highly correlated descriptors were excluded. The remaining 178 descriptors were used to build the model as candidate inputs. Of the 129 compounds, 72 were assigned to the training set, 41 to the verification set, and the remaining 16 to the test set. The best SVM ensemble model uses 72 inputs. Table I lists the effect indicators for model evaluation. The sensitivity for the training set was 0.971 and 0.667 for the test set. The specificity was 1.000 for the training set and 1.000 for the test set. Moreover, the MCC was 0.979 for the training set and 0.787 for the test set. The statistical performance results are shown in Fig. 2b. Table II lists the top 10 descriptors with the highest contribution. The descriptor with the highest contribution in the SVM model was SaaaC, which is an Atom-type Electropological State Index. This is followed by EEM_NFpl and EEM_XFon, which are Charge-based Descriptors. EEM_XFc and T_RDmtr, the first and second highest contributors in the ANN model, were in the fourth and seventh positions in the SVM model.
## Discussion
In this study, we constructed a binary classification model to predict M/PAUC using two models, ANN and SVM. Descriptors calculated specifically from the molecular structure of the compounds were used in the predictions. The performance of the models was within the acceptable range for both ANN and SVM. The ADMET Predictor® is a software package that can rapidly predict ADMET (absorption, distribution, metabolism, excretion, and toxicity) properties based on molecular structures [34]. The predictive model is built using carefully selected descriptors and optimal learning algorithms based on actual measurements collected from public sources and academic literature. It shows excellent performance in predicting the aqueous solubility of chemicals [35], organic compounds [36], and the plasma protein binding of compounds in humans [37]. The ADMET Modeler™ module allows users to build QSAR/QSPR prediction models from chemical structures and measured datasets. These were also used to develop in silico models to identify androgen-active chemicals [38] and to build machine-learning models to predict chemotherapy-induced peripheral neuropathy [39]. The ADMET ModelerTM is equipped with an early learning stop system to prevent overlearning, compensating for the shortcomings of ANNs, which are known to be prone to overlearning. The performance of the ANN model was checked using confidence analysis. A maximum uncertainty of 40–$60\%$ is recommended as the statistical stability of the model [40]. Therefore, the model we constructed with $57.5\%$ uncertainty meets this requirement.
We focused on the AUC to evaluate the M/P ratio. In the data collected, M/PAUC was obtained for only $43\%$ of the 403 compounds with M/P ratio data. The problems with evaluating non-AUC M/P ratio with have been discussed in previous studies [41]. Furthermore, we excluded colostrum data and ultimately built our model with 129 compounds. While some of the previously reported models covered more than 300 compounds, curation allowed us to use a carefully selected dataset of compounds in this study. This is the first report of curation using these criteria for over 400 compounds.
In this study, the models were constructed to identify compounds with M/PAUC greater than 1. Of the compounds with M/PAUC greater than 1 collected from the original articles, none showed false negative results in either model. Specifically, the M/PAUC ratios of 1.43 and 1.77 for mirtazapine and moxidectin [42, 43], respectively, were less than 1 by the ANN model and were false negative, while the SVM model correctly predicted their values greater than 1. The M/PAUC ratios of 1.32 and 5.99 for fluvoxamine and N-monodesalkyldisopyramide [44, 45], respectively, were predicted to be less than 1 by the SVM model and were false negative, while the ANN model correctly predicted these values greater than 1. Therefore, we believe that a combined analysis of various models will also contribute to model reconstruction in the future and improvement in accuracy, although the results of prediction and reliability analyses for compounds near the M/PAUC=1 boundary value need to be carefully considered.
Although the ANN and SVM models had different contributing descriptors, both models were characterized by a high number of charge-based descriptors among the top 10 contributing descriptors. Charge-based descriptors with high contributions in the ANN model included EEM_XFc, Pi_AQc, Pi_FPl4, MinQ, and EEM_XFpl. Charge-based descriptors with high contribution rates in the SVM model included EEM_NFpl, EEM_Xfon, NPA_Q3, EEM_XFc, EEM_NFnp, and NPA_MinQ. Among them, EEM_XFc, EEM_XFp, EEM_NFpl, EEM_Xfon, and EEM_NFnp are the sigma Fukui indices, derivatives of the atomic partial charge relative to the total number of electrons. The sigma charges were provided as input to parameterize the two-dimensional deformation of the EEM based on Chaves' formalism [46] for the EEM kernel. Fukui (+) indicators like Pi_FPl4 are related to the Pi electron density in the lowest unoccupied molecular orbital (LUMO) [47]. Previously reported milk transfer prediction models suggested that electronic properties play an important role [18, 24, 48], consistent with the high contribution of electronic properties in this study. The "SaaaC" with the highest contribution in the SVM model was Atom-type Electropological State Indices. It is dominated by the E-state descriptor, an electrophysiological state indicator for the atom type, which is incrementally perturbed by the eigenstates of the atoms to which they are connected and weighted by the topological distance to each other [49]. In addition, "S+logD" simulated by ADMET Predictor also has a high contribution in the SVM model, confirming that fat solubility is also an important factor. S+logD indicates values related to pKa and log P. pKa and logP are important factors in the evaluation of milk transferability according to previously published models for predicting milk transfer [16, 17]. We performed a preliminary study using descriptors that can be added arbitrarily to the ADMET ModelarTM. The results confirmed that adding S+logD to the descriptors in the SVM model improved its accuracy. A similar trend was not observed in the ANN model. We plan to further examine how known factors related to milk transfer affect the construction of the QSAR/QSPR model.
The M/PAUC prediction models developed in this study can be used to help evaluate the milk transfer potential of drugs, metabolites, impurities, and even enantiomers that have never been administered to lactating women. A case of child death due to morphine intoxication caused by breastfeeding from a mother who took codeine has been reported [50]. Although one factor in this case was that the mother was an ultra-rapid metabolizer of CYP2D6, an enzyme that metabolizes codeine, it is important to recognize that exposure to metabolites can affect the infant. Compounds like fluoxetine need to be evaluated for metabolites and racemates [51].
Since transporters such as BCRP have been shown to be involved in the milk transfer of drugs [52], it is important to know whether a compound is a substrate drug. If significant milk transfer is observed, contrary to the results predicted by pH partitioning theory, active transport may be involved. The ADMET Predictor® has a module that predicts whether a compound can be a substrate for BCRP. This predictive model showed $85.9\%$ concordance for the training set and $85.6\%$ for the test set [40]. Thus, the QSAR model can predict the characteristics of drug milk transfer, including the involvement of transporters, which are difficult to predict based on the physicochemical properties alone. However, BCRP is not the only transporter whose expression increases in the mammary gland during lactation; there are also the sodium/iodide symporter and Organic Cation Transporter 1, and there may be other transporters that are not yet known [52]. If predictive modeling indicates that active transport may be involved in milk transfer of a target compound, more detailed clinical data is needed.
This study has several limitations, including data consistency for model building. We curated the reported data in order to build an accurate model. However, the M/PAUC data were collected from various papers and thus the target patients and measurement methods were not standardized. In addition, M/P cannot be quantitatively evaluated in the classification model constructed in this study. Due to the ethical challenges in conducting large clinical trials in lactating women, much of the data must rely on case reports. We hope that further clinical data will be available in the future, as well as improvements in the in silico prediction accuracy. Furthermore, the compounds to which the ADMET predictor® can be applied only include organic compounds consisting of boron, carbon, nitrogen, oxygen, sulfur, phosphorus, fluorine, chlorine, bromine, and iodine. Therefore, it is unsuitable for the prediction of compounds containing metals or polymers. However, the constructed predictive model can be applied to many drugs, and polymeric compounds are less likely to migrate into milk. For example, the milk: serum ratios of abatacept and tocilizumab have been reported to be 0.003–0.005 and 0.001–0.002, respectively [53, 54].
It is not appropriate to predict the advisability of breastfeeding during drug therapy solely on the basis of the M/P ratio; the child's drug intake and metabolic capacity should also be considered. A physiologically based pharmacokinetic model for predicting milk transfer of drugs has been constructed and is anticipated to be applied to some drugs [55, 56]. As new drugs are launched, medical professionals evaluate milk transfer based on the results of animal experiments, physicochemical properties of compounds, or past case reports, and consider drug therapy for lactating women as a time-consuming process. If in silico prediction can be used to evaluate the milk transfer properties of drugs quickly and easily, the possibility of achieving compatibility between drug therapy and breastfeeding can be increased. It is also necessary to analyze the prediction results obtained by machine learning not only with one model but a combination of models to avoid deterioration of the prediction accuracy [57]. We believe that the model developed in this study can assist in the evaluation of the milk transfer characteristics of compounds, especially those involving active transport by BCRP.
## Conclusions
The purpose of this study was to develop a model to predict whether the human M/PAUC would exceed 1 for screening drugs transported actively in milk transfer. We built a milk transfer prediction model based on QSAR/QSPR using two methods, ANN and SVM. These two models showed satisfactory performance. Subsequently, in the process of building the predictive model, we confirmed the high contribution of the charge-based descriptor. The specific charge-based descriptor types and contribution rates, their relationship to active transport, and their effects on M/P ratios require further investigation but the charge properties indicated that these may be added to molecular weight, plasma protein binding, lipid solubility, and acid–base properties as factors to evaluate milk transfer of drugs.
Although further study is needed for the descriptors, we believe that the model constructed in this study can play an auxiliary role in evaluating the milk transferability of drugs.
## Supplementary Information
Below is the link to the electronic supplementary material. Supplementary file1 (PDF 1213 KB)
## References
1. Ladomenou F, Moschandreas J, Kafatos A, Tselentis Y, Galanakis E. **Protective effect of exclusive breastfeeding against infections during infancy: a prospective study**. *Arch Dis Child* (2010.0) **95** 1004-1008. DOI: 10.1136/adc.2009.169912
2. Kramer MS, Aboud F, Mironova E, Vanilovich I, Platt RW, Matush L. **Breastfeeding and child cognitive development: new evidence from a large randomized trial**. *Arch Gen Psychiatry* (2008.0) **65** 578-584. DOI: 10.1001/archpsyc.65.5.578
3. Horta BL, de Lima NP. **Breastfeeding and Type 2 diabetes: systematic review and meta-analysis**. *Curr Diab Rep* (2019.0) **19** 1. DOI: 10.1007/s11892-019-1121-x
4. Gunderson EP, Hurston SR, Ning X, Lo JC, Crites Y, Walton D. **Lactation and progression to type 2 diabetes mellitus after gestational diabetes mellitus: A prospective cohort study**. *Ann Intern Med* (2015.0) **163** 889-898. DOI: 10.7326/M15-0807
5. Schwarz EB, Ray RM, Stuebe AM, Allison MA, Ness RB, Freiberg MS. **Duration of lactation and risk factors for maternal cardiovascular disease**. *Obstet Gynecol* (2009.0) **113** 974-982. DOI: 10.1097/01.AOG.0000346884.67796.ca
6. Weaver JM, Schofield TJ, Papp LM. **Breastfeeding duration predicts greater maternal sensitivity over the next decade**. *Dev Psychol* (2018.0) **54** 220-227. DOI: 10.1037/dev0000425
7. 7.World Health Organization. Global strategy for infant and young child feeding; 2003. https://www.who.int/publications/i/item/9241562218. Accessed 18 Oct 2021.
8. 8.Ministry of Health, Labour and Welfare. Heisei. 27 nendo nyuyoji eiyo chosa kekka no gaiyo; 2015. https://www.mhlw.go.jp/file/06-Seisakujouhou-11900000-Koyoukintoujidoukateikyoku/0000134207.pdf. Accessed 18 Oct 2021 [Results of the 2015 infant nutrition survey].
9. 9.Ministry of Health, Labour and Welfare. The 2nd Study Group on the Insurance and Medical System for Expectant and Nursing Mothers. Current Status and Issues in the Medical Care of Pregnant and Nursing Women [Ninsanpu no shinryo no genjyo to kadai (in Japanese)]; 2019. https://www.mhlw.go.jp/content/12401000/000488877.pdf. Accessed 12 May 2022.
10. Verstegen RHJ, Ito S. **Drugs in lactation**. *J Obstet Gynaecol Res* (2019.0) **45** 522-531. DOI: 10.1111/jog.13899
11. Ito N, Ito K, Ikebuchi Y, Toyoda Y, Takada T, Hisaka A. **Prediction of drug transfer into milk considering breast cancer resistance protein (BCRP)-mediated transport**. *Pharm Res* (2015.0) **32** 2527-2537. DOI: 10.1007/s11095-015-1641-2
12. Verstegen RHJ, Anderson PO, Ito S. **Infant drug exposure via breast milk**. *Br J Clin Pharmacol* (2020.0). DOI: 10.1111/bcp.14538
13. Ito S, Koren G. **A novel index for expressing exposure of the infant to drugs in breast milk**. *Br J Clin Pharmacol* (1994.0) **38** 99-102. DOI: 10.1111/j.1365-2125.1994.tb04331.x
14. Anderson PO, Momper JD. **Clinical lactation studies and the role of pharmacokinetic modeling and simulation in predicting drug exposures in breastfed infants**. *J Pharmacokinet Pharmacodyn* (2020.0) **47** 295-304. DOI: 10.1007/s10928-020-09676-2
15. Wilson JT, Brown RD, Cherek DR, Dailey JW, Hilman B, Jobe PC. **Drug excretion in human breast milk: principles, pharmacokinetics and projected consequences**. *Clin Pharmacokinet* (1980.0) **5** 1-66. DOI: 10.2165/00003088-198005010-00001
16. Begg EJ, Atkinson HC. **Modelling of the passage of drugs into milk**. *Pharmacol Ther* (1993.0) **59** 301-310. DOI: 10.1016/0163-7258(93)90072-l
17. Meskin MS, Lien EJ. **QSAR analysis of drug excretion into human breast milk**. *J Clin Hosp Pharm* (1985.0) **10** 269-278. DOI: 10.1111/j.1365-2710.1985.tb00924.x
18. Agatonovic-Kustrin S, Tucker IG, Zecevic M, Zivanovic LJ. **Prediction of drug transfer into human milk from theoretically derived descriptors**. *Anal Chim Acta* (2000.0) **418** 181-195. DOI: 10.1016/S0003-2670(00)00963-6
19. Katritzky AR, Dobchev DA, Hür E, Fara DC, Karelson M. **QSAR treatment of drugs transfer into human breast milk**. *Bioorg Med Chem* (2005.0) **13** 1623-1632. DOI: 10.1016/j.bmc.2004.12.015
20. Abraham MH, Gil-Lostes J, Fatemi M. **Prediction of milk/plasma concentration ratios of drugs and environmental pollutants**. *Eur J Med Chem* (2009.0) **44** 2452-2458. DOI: 10.1016/j.ejmech.2009.01.009
21. Agatonovic-Kustrin S, Morton DW, Celebic D. **QSAR: an**. *Comb Chem High Throughput Screen* (2013.0) **16** 223-232. DOI: 10.2174/1386207311316030007
22. Kar S, Roy K. **Prediction of Milk/Plasma Concentration Ratios of Drugs and Environmental Pollutants Using**. *Mol Inform* (2013.0) **32** 693-705. DOI: 10.1002/minf.201300018
23. Wanat K, Khakimov B, Brzezińska E. **Comparison of statistical methods for predicting penetration capacity of drugs into human breast milk using physicochemical, pharmacokinetic and chromatographic descriptors**. *SAR QSAR Environ Res* (2020.0) **31** 457-475. DOI: 10.1080/1062936x.2020.1772365
24. Zhao C, Zhang H, Zhang X, Zhang R, Luan F, Liu M. **Prediction of milk/plasma drug concentration (M/P) ratio using support vector machine (SVM) method**. *Pharm Res* (2006.0) **23** 41-48. DOI: 10.1007/s11095-005-8716-4
25. Fatemi MH, Ghorbanzad'e M. **Classification of drugs according to their milk/plasma concentration ratio**. *Eur J Med Chem* (2010.0) **45** 5051-5055. DOI: 10.1016/j.ejmech.2010.08.013
26. Somogyi A, Gugler R. **Cimetidine excretion into breast milk**. *Br J Clin Pharmacol* (1979.0) **7** 627-629. DOI: 10.1111/j.1365-2125.1979.tb04655.x
27. Shyu WC, Shah VR, Campbell DA, Venitz J, Jaganathan V, Pittman KA. **Excretion of cefprozil into human breast milk**. *Antimicrob Agents Chemother* (1992.0) **36** 938-941. DOI: 10.1128/aac.36.5.938
28. 28.Hale TW. Medications & Mothers' Milk. Springer Publishing company; 2019.
29. 29.Briggs GG. Drugs in pregnancy and lactation : a reference guide to fetal and neonatal risk. 11th ed. Lippincott Williams & Wilkins; 2017.
30. 30.Drugs and Lactation Database (LactMed); 2006. https://www.ncbi.nlm.nih.gov/books/NBK501922/. Accessed 17 Sep 2019.
31. 31.PubChem. https://pubchem.ncbi.nlm.nih.gov/. Accessed 13 Jul 2020.
32. Kohonen T. **Self-organized formation of topologically correct feature maps**. *Biological Cybernetics volume* (1982.0) **43** 59-69. DOI: 10.1007/BF00337288
33. Leardi R, Boggia R, Terrile M. **Genetic algorithms as a strategy for feature selection**. *J Chemometrics* (1992.0) **6** 267-281. DOI: 10.1002/cem.1180060506
34. 34.Simulation Plus Website. https://www.simulations-plus.com/ Accessed 19 Nov 2022.
35. Hewitt M, Cronin MT, Enoch SJ, Madden JC, Roberts DW, Dearden JC. *J Chem Inf Model* (2009.0) **49** 2572-2587. DOI: 10.1021/ci900286s
36. Cappelli CI, Manganelli S, Lombardo A, Gissi A, Benfenati E. **Validation of quantitative structure-activity relationship models to predict water-solubility of organic compounds**. *Sci Total Environ* (2013.0) **463–464** 781-789. DOI: 10.1016/j.scitotenv.2013.06.081
37. Yun YE, Tornero-Velez R, Purucker ST, Chang DT, Edginton AN. **Evaluation of Quantitative Structure Property Relationship Algorithms for Predicting Plasma Protein Binding in Humans**. *Comput Toxicol.* (2021.0) **17** 100142. DOI: 10.1016/j.comtox.2020.100142
38. Manganelli S, Roncaglioni A, Mansouri K, Judson RS, Benfenati E, Manganaro A. **Development, validation and integration of**. *Chemosphere* (2019.0) **220** 204-215. DOI: 10.1016/j.chemosphere.2018.12.131
39. Bloomingdale P, Mager DE. **Machine Learning Models for the Prediction of Chemotherapy-Induced Peripheral Neuropathy**. *Pharm Res* (2019.0) **36** 35. DOI: 10.1007/s11095-018-2562-7
40. 40.Simulation Plus I. ADMET Predictor X.3 Manual. 2021.
41. Wilson JT, Brown RD, Hinson JL, Dailey JW. **Pharmacokinetic pitfalls in the estimation of the breast milk/plasma ratio for drugs**. *Annu Rev Pharmacol Toxicol* (1985.0) **25** 667-689. DOI: 10.1146/annurev.pa.25.040185.003315
42. Kristensen JH, Ilett KF, Rampono J, Kohan R, Hackett LP. **Transfer of the antidepressant mirtazapine into breast milk**. *Br J Clin Pharmacol* (2007.0) **63** 322-327. DOI: 10.1111/j.1365-2125.2006.02773.x
43. Korth-Bradley JM, Parks V, Chalon S, Gourley I, Matschke K, Gossart S. **Excretion of moxidectin into breast milk and pharmacokinetics in healthy lactating women**. *Antimicrob Agents Chemother* (2011.0) **55** 5200-5204. DOI: 10.1128/aac.00311-11
44. Hägg S, Granberg K, Carleborg L. **Excretion of fluvoxamine into breast milk**. *Br J Clin Pharmacol* (2000.0) **49** 286-288. DOI: 10.1046/j.1365-2125.2000.00142-3.x
45. Ellsworth AJ, Horn JR, Raisys VA, Miyagawa LA, Bell JL. **Disopyramide and N-monodesalkyl disopyramide in serum and breast milk**. *DICP* (1989.0) **23** 56-57. DOI: 10.1177/106002808902300112
46. Chaves J, Barroso JM, Bultinck P, Carbó-Dorca R. **Toward an alternative hardness kernel matrix structure in the Electronegativity Equalization Method (EEM)**. *J Chem Inf Model* (2006.0) **46** 1657-1665. DOI: 10.1021/ci050505e
47. Anderson JSM, Melin J, Ayers PW. **Conceptual Density-Functional Theory for General Chemical Reactions, Including Those That Are Neither Charge- nor Frontier-Orbital-Controlled. 1. Theory and Derivation of a General-Purpose Reactivity Indicator**. *J Chem Theor Comput.* (2007.0) **3** 358-74. DOI: 10.1021/ct600164j
48. Agatonovic-Kustrin S, Ling LH, Tham SY, Alany RG. **Molecular descriptors that influence the amount of drugs transfer into human breast milk**. *J Pharm Biomed Anal* (2002.0) **29** 103-119. DOI: 10.1016/s0731-7085(02)00037-7
49. Hall LH, Mohney B, Kier LB. **The electrotopological state: structure information at the atomic level for molecular graphs**. *J Chem Inf Comput Sci* (1991.0) **31** 76-82. DOI: 10.1021/ci00001a012
50. Koren G, Cairns J, Chitayat D, Gaedigk A, Leeder SJ. **Pharmacogenetics of morphine poisoning in a breastfed neonate of a codeine-prescribed mother**. *Lancet* (2006.0) **368** 704. DOI: 10.1016/s0140-6736(06)69255-6
51. Anderson PO, Sauberan JB. **Modeling drug passage into human milk**. *Clin Pharmacol Ther* (2016.0) **100** 42-52. DOI: 10.1002/cpt.377
52. 52.García-Lino AM, Álvarez-Fernández I, Blanco-Paniagua E, Merino G, Álvarez AI. Transporters in the Mammary Gland-Contribution to Presence of Nutrients and Drugs into Milk. Nutrients. 2019;11. 10.3390/nu11102372.
53. Saito J, Yakuwa N, Takai C, Kaneko K, Goto M, Nakajima K. **Abatacept concentrations in maternal serum and breast milk during breastfeeding and an infant safety assessment: a case study**. *Rheumatol (Oxf Engl)* (2019.0) **58** 1692-1694. DOI: 10.1093/rheumatology/kez135
54. Saito J, Yakuwa N, Takai C, Nakajima K, Kaneko K, Goto M. **Tocilizumab concentrations in maternal serum and breast milk during breastfeeding and a safety assessment in infants: a case study**. *Rheumatol (Oxf Engl)* (2018.0) **57** 1499-1501. DOI: 10.1093/rheumatology/key091
55. Job KM, Dallmann A, Parry S, Saade G, Haas DM, Hughes B. **Development of a Generic Physiologically-Based Pharmacokinetic Model for Lactation and Prediction of Maternal and Infant Exposure to Ondansetron via Breast Milk**. *Clin Pharmacol Ther* (2022.0) **111** 1111-1120. DOI: 10.1002/cpt.2530
56. Abduljalil K, Pansari A, Ning J, Jamei M. **Prediction of drug concentrations in milk during breastfeeding, integrating predictive algorithms within a physiologically-based pharmacokinetic model**. *CPT Pharmacometrics Syst Pharmacol* (2021.0) **10** 878-889. DOI: 10.1002/psp4.12662
57. Kiryu Y. **Medical Big Data Analysis Using Machine Learning Algorithms in the Field of Clinical Pharmacy**. *Yakugaku Zasshi* (2022.0) **142** 319-326. DOI: 10.1248/yakushi.21-00178-1
|
---
title: 'Sex differences in body composition in people with prediabetes and type 2
diabetes as compared with people with normal glucose metabolism: the Maastricht
Study'
authors:
- Rianneke de Ritter
- Simone J. S. Sep
- Marleen M. J. van Greevenbroek
- Yvo H. A. M. Kusters
- Rimke C. Vos
- Michiel L. Bots
- M. Eline Kooi
- Pieter C. Dagnelie
- Simone J. P. M. Eussen
- Miranda T. Schram
- Annemarie Koster
- Martijn C. G. Brouwers
- Niels M. R. van der Sangen
- Sanne A. E. Peters
- Carla J. H. van der Kallen
- Coen D. A. Stehouwer
journal: Diabetologia
year: 2023
pmcid: PMC10036428
doi: 10.1007/s00125-023-05880-0
license: CC BY 4.0
---
# Sex differences in body composition in people with prediabetes and type 2 diabetes as compared with people with normal glucose metabolism: the Maastricht Study
## Abstract
### Aims/hypothesis
Obesity is a major risk factor for type 2 diabetes. However, body composition differs between women and men. In this study we investigate the association between diabetes status and body composition and whether this association is moderated by sex.
### Methods
In a population-based cohort study ($$n = 7639$$; age 40–75 years, $50\%$ women, $25\%$ type 2 diabetes), we estimated the sex-specific associations, and differences therein, of prediabetes (i.e. impaired fasting glucose and/or impaired glucose tolerance) and type 2 diabetes (reference: normal glucose metabolism [NGM]) with dual-energy x-ray absorptiometry (DEXA)- and MRI-derived measures of body composition and with hip circumference. Sex differences were analysed using adjusted regression models with interaction terms of sex-by-diabetes status.
### Results
Compared with their NGM counterparts, both women and men with prediabetes and type 2 diabetes had more fat and lean mass and a greater hip circumference. The differences in subcutaneous adipose tissue, hip circumference and total and peripheral lean mass between type 2 diabetes and NGM were greater in women than men (women minus men [W–M] mean difference [$95\%$ CI]: 15.0 cm2 [1.5, 28.5], 3.2 cm [2.2, 4.1], 690 g [8, 1372] and 443 g [142, 744], respectively). The difference in visceral adipose tissue between type 2 diabetes and NGM was greater in men than women (W–M mean difference [$95\%$ CI]: −14.8 cm2 [−26.4, −3.1]). There was no sex difference in the percentage of liver fat between type 2 diabetes and NGM. The differences in measures of body composition between prediabetes and NGM were generally in the same direction, but were not significantly different between women and men.
### Conclusions/interpretation
This study indicates that there are sex differences in body composition associated with type 2 diabetes. The pathophysiological significance of these sex-associated differences requires further study.
### Supplementary Information
The online version contains peer-reviewed but unedited supplementary material available at 10.1007/s00125-023-05880-0.
## Introduction
Obesity is associated with a proinflammatory state and dyslipidaemia and is a major risk factor for type 2 diabetes [1]. The amount and distribution of fat and lean mass (i.e. body composition) differ between women and men, with women having proportionally more fat mass and men more muscle mass [2].
Sex differences have been reported in the association of excess body fat with type 2 diabetes [3]. *In* general, women have a higher BMI at diagnosis of type 2 diabetes [3]. *Women* generally have a greater amount of total body fat than men, but an increase in body fat appears to have a smaller effect on their insulin sensitivity [4]. The transition from normoglycaemia to type 2 diabetes may be associated with more fat accumulation in women, because they tend to store excess fat first in less metabolically harmful regions (i.e. subcutaneously and on their lower extremities) and subsequently in more harmful regions (e.g. as visceral adipose tissue [VAT] in the abdominal region) [4, 5]. In contrast, men predominately store fat more rapidly as VAT, which is associated with metabolic disturbances and higher risks of type 2 diabetes and CVD [4].
Meanwhile, the role of lean mass is ambiguous. One study reported that a higher lean mass was significantly associated with a lower risk of diabetes in women, and this association was directionally similar in men [6]. Additionally, in both sexes, hyperglycaemia has been associated with a lower lean mass [7, 8]. While lean mass may be beneficial for glucose metabolism, it has also been suggested that greater lean mass may not protect against insulin resistance [9, 10]. More specifically, among men, greater lean mass accompanied by greater fat mass may be detrimental for glucose regulation, whereas, among women, greater fat mass is the major determinant of glucose intolerance [9]. Additionally, people with both a high fat and a high lean mass were shown to have the most unfavourable cardiometabolic risk profile [10]. Therefore, it is important to take both fat and lean mass into account with regard to diabetes development. However, no large studies are available that have analysed sex differences in body composition associated with prediabetes (i.e. impaired fasting glucose and/or impaired glucose tolerance) and/or type 2 diabetes.
In view of these considerations, more insight into sex differences in the amount of fat and lean mass between people with (pre)diabetes and people with normal glucose metabolism (NGM) could contribute to a greater understanding of the sex-specific role of body composition in the development of type 2 diabetes. Therefore, we investigated sex-specific associations, and differences therein (i.e. interactions), of (pre)diabetes with dual-energy x-ray absorptiometry (DEXA)- and MRI-derived measures of body composition and with hip circumference.
## Study design and population
Data from the Maastricht Study, an observational prospective population-based cohort study, were used in this study. The rationale and methodology have been described previously [11]. In brief, the study focuses on the aetiology, pathophysiology, complications and comorbidities of type 2 diabetes and is characterised by an extensive phenotyping approach. All individuals aged between 40 and 75 years and living in the southern part of the Netherlands were eligible for participation. Participants were recruited through mass media campaigns and from the municipal registries and the regional Diabetes Patient Registry through mail-outs. Recruitment was stratified according to known diabetes status, with an oversampling of individuals with type 2 diabetes, for reasons of efficiency. This study includes cross-sectional data from the first 7689 participants, who completed the baseline survey between November 2010 and December 2017. All examinations of participants were performed within a time frame of 3 months, except for the DEXA and MRI scans. DEXA measurements were implemented from January 2015 onwards, with a mean lag time of 2.6 years. MRI measurements were implemented from December 2013 onwards and had a mean lag time of 1.2 years. The study was approved by the institutional medical ethics committee (NL31329.068.10) and the Minister of Health, Welfare and Sport of the Netherlands (permit 131088-105234-PG). All participants gave written informed consent. For the current study, individuals with other types of diabetes than type 2 diabetes were excluded ($$n = 50$$).
## Assessment of body composition
A DEXA scanner was used to assess participants’ fat and lean mass (electronic supplementary material [ESM] Fig. 1) as described in ESM Methods. MRI was performed to determine the amount of abdominal subcutaneous adipose tissue (SAT) and VAT and liver fat percentage (ESM Fig. 1, ESM Methods).
Because of a technical error in the case of 250 participants, measurements of the amount of SAT ($$n = 250$$) and VAT ($$n = 28$$) were incomplete. We estimated these values as described in ESM Methods.
We used hip circumference, determined as described elsewhere [11], as a proxy for thigh and buttock fat [12] (ESM Fig. 1).
## Assessment of glucose metabolism status
To determine glucose metabolism status (GMS), all participants underwent a standardised 2 h 75 g OGTT after fasting overnight. Further details on the assessment of GMS, as well as the assessments of covariates and population characteristics, are described in ESM Methods.
## Statistical analyses
SSPS version 27.0 (IBM, USA) was used for the statistical analyses. Population characteristics were described as mean ± SD and median (IQR), for normally and non-normally distributed variables, respectively, or n (%) for discrete variables. Variables were log-transformed if residuals were skewed.
Sex and the interaction of sex-by-(pre)diabetes need to be distinguished as potential determinants, as described in more detail elsewhere [13]. We used generalised linear models to estimate adjusted (model 3 as described below) sex-specific amounts of DEXA-estimated fat and lean mass, MRI-estimated amounts of VAT and SAT and liver fat percentage, and hip circumference in participants with NGM, prediabetes and type 2 diabetes. We used linear regression analyses to test whether sex was a determinant in these associations. Our main goal was to investigate sex-by-(pre)diabetes interactions; therefore, we used linear regression analyses (based on two-sided tests) to estimate sex-specific associations, and differences therein (i.e. interactions), of prediabetes and type 2 diabetes (reference category: NGM) with DEXA-estimated fat and lean mass, with MRI-estimated VAT, SAT and liver fat percentage and with hip circumference. To test for sex differences, interaction terms of sex-by-dummy-coded (pre)diabetes status (i.e. sex-by-prediabetes and sex-by-type 2 diabetes) were incorporated into the models. Several sets of adjustments were made. Model 1 was adjusted for age and height. We adjusted for height as a measure of body size in the associations with fat and lean mass, expressed as an amount (g) or area (cm2). Thus, only liver fat percentage was not adjusted for height, as we considered height not to be a potential confounder. Model 2 was additionally adjusted for other potential confounders, that is, physical activity, healthy diet score, educational level, alcohol consumption and smoking status. If total or peripheral lean mass was the outcome, model 2 was additionally adjusted for total fat mass. Model 3 (main model) was additionally adjusted for the use of medication that may cause weight gain and/or weight loss as a side effect. For each potential confounder included, an interaction term (sex-by-potential confounder) was also incorporated in the same models to ensure that the adjustments made in the interaction models would vary by sex as they do in the sex-specific models [14]. For the interactions of sex with (pre)diabetes, $p \leq 0.05$ was considered statistically significant and the results are presented with $95\%$ CIs. Multiple imputation was performed for both potential confounders and outcomes (i.e. measures of body composition). The percentage of missing values was a maximum of $11.7\%$ for potential confounders and $34.7\%$ for outcomes (Table 1, Fig. 1). We imputed data using multiple imputation by chained equations under the assumption that data were missing at random. We used predictive mean matching to impute 20 datasets with ten iterations for each dataset. For the main analysis, we pooled the results across all imputed datasets with the use of Rubin’s rule [15]. Table 1Study population characteristics according to sex and GMSCharacteristicWomen ($$n = 3788$$)Men ($$n = 3851$$)NGM ($$n = 2632$$)Prediabetes ($$n = 525$$)T2D ($$n = 631$$)NGM ($$n = 1973$$)Prediabetes ($$n = 616$$)T2D ($$n = 1262$$)Demographic characteristics and health measures Age (years)57.4 ± 8.561.6 ± 8.562.3 ± 8.258.7 ± 8.762.8 ± 7.963.4 ± 7.6 Height (cm)a165 ± 6163 ± 6163 ± 6178 ± 7177 ± 6176 ± 7 Education level (n)a Low842 (32.4)229 (44.6)364 (59.1)463 (23.6)200 (32.9)509 (41.3) Medium751 (28.9)136 (26.5)147 (23.9)532 (27.2)166 (27.3)346 (28.1) High1005 (38.7)149 (29.0)105 (17.0)964 (49.2)241 (39.7)377 (30.6) *Smoking status* (n)a Never1105 (42.3)195 (37.4)248 (39.9)781 (39.7)183 (29.9)309 (24.9) Former1203 (46.0)265 (50.9)278 (44.8)909 (46.2)351 (57.4)733 (59.1) Current307 (11.7)61 (11.7)95 (15.3)276 (14.0)78 (12.7)199 (16.0) Alcohol use (g/day)b8.5 ± 9.38.3 ± 10.45.1 ± 8.216.3 ± 14.718.7 ± 19.413.8 ± 16.0 Healthy diet score (score)b81 ± 1379 ± 1377 ± 1474 ± 1573 ± 1470 ± 14 Systolic blood pressure (mmHg)126 ± 17134 ± 18139 ± 18134 ± 15140 ± 17143 ± 18 Diastolic blood pressure (mmHg)72 ± 975 ± 1076 ± 1077 ± 979 ± 1078 ± 10 Use of blood pressure-lowering medication (n)a520 (19.8)222 (42.4)437 (69.3)516 (26.2)304 (49.4)909 (72.1) Total cholesterol (mmol/l)5.6 ± 1.05.6 ± 1.14.8 ± 1.05.3 ± 1.05.2 ± 1.14.3 ± 1.0 HDL-cholesterol (mmol/l)1.8 ± 0.51.6 ± 0.41.5 ± 0.41.4 ± 0.41.3 ± 0.41.2 ± 0.3 LDL-cholesterol (mmol/l)3.3 ± 0.93.3 ± 1.02.6 ± 0.93.3 ± 0.93.1 ± 1.02.3 ± 0.9 Triglycerides (mmol/l)1.1 ± 0.61.6 ± 0.91.7 ± 0.91.3 ± 0.71.7 ± 1.21.8 ± 1.1 Use of lipid-modifying medication (n)a321 (12.2)151 (28.8)425 (67.4)394 (20.0)240 (39.0)938 (74.4) Use of medication that affects weight (n)a,c421 (16.0)142 (27.1)285 (45.2)284 (14.4)178 (28.9)526 (41.7) Moderate/vigorous physical activity (h/week)b6.1 ± 4.55.0 ± 3.84.5 ± 4.25.9 ± 4.55.1 ± 4.54.2 ± 4.2 *Postmenopausal status* (n)a1941 (74.9)432 (83.9)537 (87.9)n/an/an/a Use of postmenopausal hormone replacement medicationa55 (2.8)10 (2.3)14 (2.6)n/an/an/aMeasures of body composition Total body fat mass (kg)b27.1 ± 7.930.7 ± 8.734.3 ± 10.223.4 ± 6.726.9 ± 7.529.2 ± 8.6 Total body lean mass (kg)b41.4 ± 5.142.6 ± 5.844.9 ± 6.558.3 ± 6.559.6 ± 7.260.1 ± 7.5 Peripheral fat mass (kg)b13.5 ± 3.914.3 ± 4.415.2 ± 4.89.8 ± 2.810.8 ± 3.311.3 ± 3.5 Peripheral lean mass (kg)b17.4 ± 2.517.7 ± 2.718.4 ± 3.125.9 ± 3.226.1 ± 3.525.8 ± 3.6 Trunk adipose tissue (kg)b12.6 ± 4.515.4 ± 4.818.1 ± 6.012.4 ± 4.114.9 ± 4.516.7 ± 5.3 Gynoid fat mass (kg)b3.5 ± 1.03.9 ± 1.14.1 ± 1.24.8 ± 1.35.0 ± 1.45.2 ± 1.6 SAT (cm2)b220.9 ± 92.1256.0 ± 99.7283.7 ± 107.8184.7 ± 68.7206.1 ± 72.8214.8 ± 79.4 VAT (cm2)b99.0 ± 56.1143.1 ± 64.4181.4 ± 78.8176.8 ± 88.0229.7 ± 88.5279.2 ± 107.9 Liver fat percentage (%)b2.4 (1.6–3.9)4.2 (2.4–8.4)6.3 (3.2–11.9)3.0 (2.1–5.0)4.7 (2.9–8.1)6.4 (3.6–11.5) BMI (kg/m2)25.2 ± 4.027.7 ± 4.730.7 ± 5.626.1 ± 3.228.0 ± 3.729.3 ± 4.6 Waist circumference (cm)86 ± 1193 ± 12102 ± 1496 ± 10103 ± 10108 ± 12 Hip circumference (cm)101 ± 9104 ± 10109 ± 1299 ± 6102 ± 7104 ± 8Measures of glucose metabolism Fasting glucose (mmol/l)5.0 ± 0.45.7 ± 0.67.5 ± 1.85.2 ± 0.46.0 ± 0.58.0 ± 2.0 Post-load glucose (mmol/l)d5.4 ± 1.18.6 ± 1.214.5 ± 3.95.3 ± 1.28.0 ± 1.814.2 ± 3.7 HbA1c (mmol/mol)35.3 ± 3.838.0 ± 4.450.2 ± 11.235.3 ± 3.938.1 ± 4.551.6 ± 11.7 HbA1c (%)5.4 ± 0.45.6 ± 0.46.7 ± 1.05.4 ± 0.45.6 ± 0.46.9 ± 1.1 Use of oral drugs for T2D (n)n/an/a409 (64.8)n/an/a897 (71.1) Use of insulin for T2D (n)n/an/a100 (15.8)n/an/a259 (20.5)Data are presented as mean ± SD, median (25th–75th percentile) in case of a skewed distribution or n (%)aMissing data <$5\%$ per variable: height, $$n = 7636$$; education level, $$n = 7526$$; smoking status, $$n = 7576$$; blood pressure-lowering medication, $$n = 7633$$; lipid-modifying medication, $$n = 7633$$; medication that affects weight, $$n = 7633$$; postmenopausal status (women only), $$n = 3719$$; hormone replacement medication (postmenopausal women only), $$n = 2910$$bMissing data >$5\%$ per variable: alcohol use, $$n = 7150$$; healthy diet score, $$n = 7150$$; moderate/vigorous physical activity, $$n = 6748$$; DEXA-derived measures of body composition, $$n = 6413$$; SAT and VAT, $$n = 5139$$; liver fat percentage, $$n = 4990$$cUse of medication that affects weight was defined as using one or more of the following medications: hormonal contraceptives, antidepressants, antipsychotic drugs, lithium, medicinal cannabis, β-blockers, anti-epileptics (i.e. valproic acid, gabapentin, carbamazepine and topiramate), mineralocorticoids (i.e. fludrocortisone), glucocorticoids (i.e. betamethasone, dexamethasone, methylprednisolone, prednisolone, prednisone, triamcinolone [acetonide], hydrocortisone and cortisone)dMissing data for $23\%$ of individuals with type 2 diabetes per protocoln/a, not applicable; T2D, type 2 diabetesFig. 1Flowchart of study participants From an aetiological perspective, we were interested in the potential effect of body composition on the development of (pre)diabetes in men and women. Although it may seem counterintuitive, we specifically chose to analyse the data with (pre)diabetes as the determinant and measures of body composition as outcomes, and not the other way around, because results of analyses with body composition measures as determinants are difficult to interpret. For example, men have more VAT than women and therefore a 1 cm2 increase in VAT is a relatively smaller increase for men than for women. Hence, the results, for example the odds of having (pre)diabetes compared with NGM per 1 cm2 increase in VAT, are difficult to compare between women and men. Similarly, men also have a higher SD of VAT than women, because of their greater amount of VAT, so comparing SDs between men and women would also give results that are difficult to interpret.
In the current analyses, the results are expressed as linear regression coefficients, which represent mean differences (βs) or geometric mean ratios (GMRs) for measures of body composition (g or cm2) according to (pre)diabetes status (reference category: NGM). The results are presented for men and women separately and can be seen as a snapshot indicating the amount of fat and lean mass in (pre)diabetes compared with NGM.
To investigate the robustness of the results obtained by the above analyses we performed several sensitivity analyses as described in ESM Methods.
## Results
The study population consisted of 3788 women (age 58.8 ± 8.7 years) and 3851 men (age 60.9 ± 8.5 years). Of these individuals, 4605 ($57.2\%$ women) had NGM, 1141 ($46.0\%$ women) had prediabetes and 1893 ($33.3\%$ women) had type 2 diabetes (Table 1).
## Sex as determinant
Compared with men, independent of GMS, women had significantly higher levels of total fat, peripheral fat, trunk fat and gynoid fat mass (Fig. 2a,e,i,k). In contrast, men, independent of GMS, had significantly higher levels of total and peripheral lean mass than women (Fig. 2c,g). Fig. 2(a, c, e, g, i, k, m, o, q, s) Sex as a determinant of body composition in participants with NGM, prediabetes and type 2 diabetes: total body fat mass (a), total body lean mass (c), peripheral fat mass (e), peripheral lean mass (g), trunk fat mass (i), gynoid fat mass (k), SAT (m), VAT (o), liver fat percentage (q) and hip circumference (s). The graphs shows adjusted (fully adjusted model) sex-specific means and corresponding $95\%$ CIs. Statistically significant adjusted (fully adjusted model) differences in body composition between women and men (sex differences) are indicated. * $p \leq 0.05$, **$p \leq 0.01.$ ( b, d, f, h, j, l, n, p, r, t) Sex-by-(pre)diabetes as a determinant of body composition in participants with NGM, prediabetes and type 2 diabetes: total body fat mass (b), total body lean mass (d), peripheral fat mass (f), peripheral lean mass (h), trunk fat mass (j), gynoid fat mass (l), SAT (n), VAT (p), liver fat percentage (r) and hip circumference (t). The graphs show adjusted (fully adjusted model) sex-specific mean differences (for all body composition variables except liver fat percentage) or GMRs (for liver fat percentage; r) between (pre)diabetes and NGM (reference category). Results are expressed as adjusted (fully adjusted model) linear regression coefficients and corresponding $95\%$ CIs. Statistically significant differences between women and men (sex differences) are indicated. * $p \leq 0.05$, **$p \leq 0.01.$ preD, prediabetes; ref, reference; T2D, type 2 diabetes Women, independent of GMS, had significantly higher levels of SAT than men (Fig. 2m). In contrast men, independent of GMS, had significantly higher levels of VAT than women (Fig. 2o). Men with NGM, but not with prediabetes or type 2 diabetes, had a significantly higher liver fat percentage than women (Fig. 2q).
Women, independent of GMS, had a significantly greater hip circumference than men (Fig. 2s).
## Sex-by-(pre)diabetes interaction
Compared with their NGM counterparts, both women and men with prediabetes and type 2 diabetes had significantly higher levels of total fat and total lean mass and peripheral fat, trunk fat and gynoid fat mass (Table 2; Fig. 2b,d,f,j,l). Women with prediabetes and type 2 diabetes had significantly higher levels of peripheral lean mass than women with NGM. In men, this association was statistically significant only for prediabetes, not type 2 diabetes (Table 2; Fig. 2h). The differences in total and peripheral lean mass between type 2 diabetes and NGM, but not between prediabetes and NGM, were significantly greater in women than in men (women minus men [W–M] mean difference [$95\%$ CI]: 690 g [8, 1372] and 443 g [142, 744], respectively) (Table 2, model 3; Fig. 2d,h). The differences in total fat, peripheral fat, trunk fat and gynoid fat mass between type 2 diabetes and NGM and between prediabetes and NGM were not significantly different for women and men (Table 2, model 3; Fig. 2b,f,j,l). Table 2Differences within and between sexes in mean differences in measures of body composition according to glucose metabolismVariablePrediabetes β or GMR ($95\%$ CI)Type 2 diabetes β or GMR ($95\%$ CI)Sex difference WM-β or WM-GMR ($95\%$ CI)WomenMenWomenMenPrediabetesT2DDEXA-derived measures of body compositionTotal body fat mass (g) ($$n = 7639$$) Model 13665 [2855, 4476]3497 [2703, 4291]6502 [5691, 7313]5689 [5025, 6354]169 (−985, 1322)813 (−240, 1865) Model 23042 [2242, 3841]2996 [2221, 3771]5045 [4226, 5865]4640 [3986, 5294]46 (−1087, 1179)405 (−655, 1465) Model 32853 [2059, 3647]2698 [1927, 3469]4413 [3570, 5257]4064 [3405, 4724]155 (−969, 1278)349 (−725, 1423)Total body lean mass (g) ($$n = 7639$$) Model 12147 [1643, 2651]2144 [1639, 2649]3773 [3270, 4275]3250 [2828, 3673]3 (−722, 728)523 (−144, 1189) Model 21010 [553, 1466]726 [259, 1193]1763 [1261, 2265]964 [558, 1370]283 (−382, 948)799 [120, 1478]* Model 31007 [550, 1464]773 [305, 1241]1751 [1250, 2251]1060 [643, 1478]234 (−431, 899)690 [8, 1372]*Peripheral fat mass (g) ($$n = 7639$$) Model 11020 [649, 1391]1118 [774, 1462]1808 [1439, 2176]1693 [1402, 1984]−98 (−618, 422)115 (−357, 586) Model 2753 [390, 1116]927 [591, 1262]1136 [774, 1499]1270 [980, 1559]−174 (−683, 336)−133 (−597, 330) Model 3689 [326, 1052]822 [487, 1157]922 [544, 1300]1068 [777, 1359]−133 (−643, 376)−146 (−624, 333)Peripheral lean mass (g) ($$n = 7639$$) Model 1794 [549, 1039]742 [497, 986]1349 [1111, 1588]853 [657, 1049]52 (−298, 403)496 [190, 803]* Model 2324 [103, 545]221 (−7, 449)515 [276, 754]22 (−176, 219)102 (−216, 421)493 [193, 793]* Model 3330 [109, 552]258 [29, 487]541 [298, 784]98 (−101, 298)72 (−248, 392)443 [142, 744]*Trunk fat mass (g) ($$n = 7639$$) Model 12641 [2144, 3138]2312 [1842, 2781]4685 [4199, 5171]3923 [3545, 4300]329 (−337, 996)763 [152, 1373]* Model 22275 [1787, 2763]2006 [1546, 2465]3880 [3389, 4371]3299 [2920, 3679]269 (−381, 919)581 (−32, 1193) Model 32149 [1665, 2633]1822 [1366, 2278]3459 [2973, 3945]2944 [2562, 3325]327 (−316, 971)515 (−94, 1125)Gynoid fat mass (g) ($$n = 7639$$) Model 1327 [202, 452]395 [279, 510]496 [378, 615]554 [463, 646]−67 (−238, 104)−58 (−204, 88) Model 2241 [118, 364]327 [213, 441]296 [175, 416]420 [328, 512]−86 (−256, 84)−124 (−272, 23) Model 3221 [99, 344]291 [177, 405]230 [108, 351]349 [258, 441]−69 (−239, 101)−120 (−269, 29)MRI-derived measures of body compositionSAT (cm2) ($$n = 7639$$) Model 130.6 (20.9, 40.4)20.5 (11.9, 29.1)52.3 (42.0, 62.5)29.1 (21.4, 36.9)10.2 (−2.5, 22.9)23.1 (10.3, 35.9)* Model 224.6 (15.0, 34.3)17.6 (9.1, 26.0)38.0 (27.6, 48.5)21.3 (13.5, 29.0)7.1 (−5.5, 19.6)16.7 (3.7, 29.8)* Model 322.7 (13.1, 32.4)15.2 (6.7, 23.6)31.6 (20.9, 42.3)16.7 (8.6, 24.7)7.5 (−5.1, 20.2)15.0 (1.5, 28.5)*VAT (cm2) ($$n = 7639$$) Model 130.0 (21.4, 38.6)37.8 (29.0, 46.5)53.9 (45.1, 62.7)72.1 (64.1, 80.0)−7.8 (−19.9, 4.4)−18.1 (−29.7, −6.6)* Model 226.1 (17.6, 34.6)31.4 (22.7, 40.0)46.0 (37.3, 54.7)61.7 (53.9, 69.5)−5.3 (−17.3, 6.7)−15.8 (−27.2, −4.3)* Model 324.4 (16.0, 32.9)28.0 (19.4, 36.6)40.4 (31.6, 49.2)55.2 (47.2, 63.2)−3.6 (−15.5, 8.3)−14.8 (−26.4, −3.1)*Liver fat percentage (%)a ($$n = 7639$$) Model 11.39 (1.29, 1.50)1.37 (1.26, 1.49)1.71 (1.60, 1.83)1.59 (1.49, 1.71)1.04 (0.93, 1.17)1.08 (0.97, 1.20) Model 21.37 (1.26, 1.49)1.28 (1.20, 1.38)1.59 (1.46, 1.74)1.55 (1.47, 1.64)1.04 (0.94, 1.15)1.03 (0.94, 1.13) Model 31.35 (1.24, 1.47)1.30 (1.20, 1.42)1.52 (1.38, 1.66)1.48 (1.38, 1.60)1.03 (0.91, 1.16)1.02 (0.91, 1.15)Other anthropometric variableHip circumference (cm) ($$n = 7639$$) Model 13.7 (2.9, 4.4)3.0 (2.3, 3.8)9.0 (8.3, 9.7)4.9 (4.3, 5.4)0.6 (−0.5, 1.7)4.1 (3.2, 5.1)* Model 23.0 (2.2, 3.8)2.7 (2.0, 3.4)7.5 (6.7, 8.2)4.1 (3.5, 4.7)0.3 (−0.7, 1.4)3.3 (2.4, 4.3)* Model 32.9 (2.1, 3.6)2.5 (1.8, 3.3)7.0 (6.2, 7.7)3.8 (3.2, 4.4)0.3 (−0.7, 1.4)3.2 (2.2, 4.1)*Sex-specific differences are expressed as linear regression coefficients ($95\%$ CIs) of the dependent variables, indicating mean differences (βs) in fat mass, lean mass, SAT, VAT or hip circumference, or GMRs for liver fat percentage, according to GMS. The reference category for prediabetes and type 2 diabetes is normal GMSDifferences between sexes are expressed as linear regression coefficients ($95\%$ CIs) of the interaction terms sex-by-prediabetes and sex-by-type 2 diabetes, indicating differences between women and men in mean differences (WM-βs) in fat mass, lean mass, SAT, VAT or hip circumference or in GMRs (WM-GMRs) for liver fat percentage, according to GMSModel 1: adjusted for age and height. Associations with liver fat percentage were not adjusted for heightModel 2: additionally adjusted for physical activity, healthy diet score, educational level, alcohol consumption and smoking status. Associations with total and peripheral lean mass were additionally adjusted for total fat massModel 3: additionally adjusted for the use of medication known to cause weight gain and/or loss as a possible side effectFor each potential confounder included, an interaction term (sex-by-potential confounder) was incorporated in the same modelaGMRs or WM-GMRs*$p \leq 0.05$ Both women and men with prediabetes and type 2 diabetes had significantly higher levels of SAT and VAT and a significantly higher liver fat percentage than their NGM counterparts (Table 2; Fig. 2n,p,r). The difference in SAT between type 2 diabetes and NGM, but not between prediabetes and NGM, was significantly greater in women than in men (W–M mean difference [$95\%$ CI]: 15.0 cm2 [1.5, 28.5]) (Table 2, model 3; Fig. 2n). The difference in VAT between type 2 diabetes and NGM, but not between prediabetes and NGM, was significantly greater in men than in women (W–M mean difference [$95\%$ CI]: −14.8 cm2 [−26.4, −3.1]) (Table 2, model 3; Fig. 2p). The differences in liver fat percentage between type 2 diabetes and NGM and between prediabetes and NGM were not significantly different between women and men (Table 2; Fig. 2r).
Both women and men with prediabetes and type 2 diabetes, had a significantly greater hip circumference than their NGM counterparts, (Table 2; Fig. 2t). The difference in hip circumference between type 2 diabetes and NGM, but not between prediabetes and NGM, was significantly greater in women than in men (W–M mean difference [$95\%$ CI]: 3.2 cm [2.2, 4.1]) (Table 2; Fig. 2t).
*In* general, for all sex-by-prediabetes interactions investigated, the results of the more basic models (models 1 and 2) were comparable to those of the main model (model 3).
## Sensitivity analyses
After additional adjustment for DEXA and MRI lag time, the results did not materially change (ESM Table 1). In separate analyses of participants with a lag time ≤6 months, the greater differences in total lean mass and in SAT between type 2 diabetes and NGM in women than in men were attenuated (W–M mean difference [$95\%$ CI]: total lean mass from 690 g [8, 1372] to 559 g [−648, 1766], SAT from 15.0 cm2 [1.5, 28.5] to 5 cm2 [−16, 27]; ESM Table 2). The results of other analyses in participants with a lag time ≤6 months or >6 months were not materially different (ESM Table 2). Exclusion of premenopausal women ($$n = 809$$) and women in whom menopausal status was unclear ($$n = 69$$; analysis population $$n = 6761$$) attenuated the greater difference in SAT between type 2 diabetes and NGM in women than in men (W–M mean difference [$95\%$ CI] from 15.0 cm2 [1.5, 28.5] to 11.2 cm2 [−3.0, 25.4]; ESM Table 3). Sex differences in hip circumference, VAT and lean mass did not materially change (ESM Table 3). Exclusion of participants with estimated values of SAT and VAT ($$n = 250$$; analysis population $$n = 4119$$) attenuated the greater difference in VAT between type 2 diabetes and NGM in men than in women (W–M mean difference [$95\%$ CI] from −11.2 cm2 [−24.2, 1.7] to −7.9 cm2 [−20.9, 5.1]; ESM Table 4). Sex differences in the results for SAT did not materially change (ESM Table 4). Complete case analysis (data not shown) gave similar sex differences to the multiple imputation approach. The statistical significance of the sex differences investigated differed for only two variables (i.e. trunk fat mass $$p \leq 0.02$$ in the original dataset and $$p \leq 0.10$$ in the imputed dataset; VAT $$p \leq 0.09$$ in the original dataset and $$p \leq 0.01$$ in the imputed dataset).
## Discussion
To our knowledge, this is the most comprehensive study to date that has investigated sex differences in body composition between people with prediabetes or type 2 diabetes and people with NGM. We showed that both women and men with prediabetes or type 2 diabetes had more fat mass, more lean mass and a greater hip circumference than their NGM counterparts. After adjustment for potential confounders, the differences in SAT and hip circumference between people with type 2 diabetes and people with NGM were greater in women than in men, whereas the difference in VAT was greater in men than in women. In addition, the difference in lean mass between those with type 2 diabetes and those with NGM was greater in women than in men. The differences in measures of body composition between those with prediabetes and those with NGM were generally in the same direction, but not statistically different for women and men.
In addition, women had more total, peripheral and gynoid fat mass than men, similar to previous findings [2, 16]. Women also had more trunk fat mass, for which previous findings have been inconsistent [17–19]. Regardless, we found that the larger amounts of total, peripheral, trunk and gynoid fat mass in people with type 2 diabetes than in those with NGM were not significantly different for women and men. These results suggest that, although women and men have different amounts of total, peripheral, trunk and gynoid fat mass, changes in the amounts that accompany the development of prediabetes and type 2 diabetes are similar among women and men.
Our results showed that women have more SAT and a greater hip circumference and men have more VAT, which is in line with previous studies [4, 20]. In addition, we observed that both women and men with prediabetes or type 2 diabetes had more SAT, a greater hip circumference and more VAT than their NGM counterparts, which is also in line with previous studies [21, 22].
The differences in SAT and hip circumference between people with type 2 diabetes and those with NGM were greater in women than in men. Excess body fat is associated with type 2 diabetes [23] and the observed sex differences in SAT and hip circumference could be explained by the preferential storage of excess body fat in subcutaneous and peripheral adipose tissue in women [4]. Excess body fat in these fat depots is considered less harmful than excess VAT [4, 24]. Moreover, subcutaneous thigh fat and gluteofemoral body fat have been associated with more favourable levels of glucose and lipids [25, 26] and a lower likelihood of the metabolic syndrome [26, 27]. Large hip and thigh circumferences have also been associated with a lower risk of type 2 diabetes [28]. However, the results regarding the sex difference in hip circumference should be interpreted carefully, as the difference in gynoid fat mass between people with type 2 diabetes and those with NGM was not significantly greater in women than in men.
In individuals with obesity, excessive amounts of VAT and related lipid accumulation in the liver and pancreas define the increased risk of type 2 diabetes and CVD [23, 29]. However, our results imply that this process differs between women and men. We observed that the difference in VAT between people with type 2 diabetes and those with NGM was greater in men than in women, but the difference in liver fat percentage was not statistically different between women and men. This may imply that women developing type 2 diabetes have a similar increase in liver fat despite a smaller increase in VAT than men. These differences might be explained by women’s greater increase in SAT during the transition to type 2 diabetes, as implied by our results. SAT can be divided into two layers: superficial and deep SAT [30]. Deep SAT is thought to be more metabolically harmful than superficial SAT [31], but whether any harmful effects of superficial and deep SAT differ between men and women is not clear [30, 32]. Taken together, with increasing weight, women may have a smaller increase in VAT but a higher increase in SAT than men. Because of the harmful effects of deep SAT, this might have a similar adverse effect on liver fat deposition and the development of type 2 diabetes. Alternatively, or additionally, VAT may also have sex-specific effects on diabetes development. VAT seems to have a stronger association with metabolic risk factors [33] and insulin resistance [34] in women than in men. Thus, although women may have a smaller increase in VAT, this does not necessarily indicate that it is less detrimental for diabetes development. In our data we did not distinguish between superficial and deep SAT. The sex-specific role of SAT and VAT in diabetes development requires further investigation. If the observed sex differences are important for the pathogenesis of type 2 diabetes, sex-specific prevention measures may be necessary.
*In* general, our results indicate that, in diabetes, women have more total fat mass than men, which is distributed differently, that is, more peripheral, trunk and gynoid fat mass, a larger hip circumference, more SAT and less VAT and a similar liver fat percentage. Furthermore, our results imply that, during the transition to type 2 diabetes, women have greater increases in SAT and hip circumference and a smaller increase in VAT, but a similar increase in liver fat.
We attribute the attenuated greater difference in lean mass and SAT between type 2 diabetes and NGM in women than in men in participants having a DEXA and MRI lag time ≤6 months to chance and loss of power. We measured weight at baseline and also at the time of the DEXA scan. We have no information on weight at the time of the MRI scan. For men, the difference in weight at the time of the DEXA scan compared with baseline was 0.5 kg (mean ± SD 87.0 ± 14.3 vs 86.5 ± 13.8). For women, the difference was 0.4 kg (mean ± SD 72.0 ± 13.7 vs 71.6 ± 13.4). This weight difference is probably comparable to that which would have been observed at the time of the MRI scan compared with baseline. The small differences in weight are unlikely to have affected the results and indeed most results in this sensitivity analysis were not materially different.
The exclusion of premenopausal women in the additional analyses attenuated the greater difference in SAT between type 2 diabetes and NGM in women than in men. This was a result of the smaller mean difference in SAT between postmenopausal women with type 2 diabetes and postmenopausal women with NGM, which can be explained by a decrease in oestrogen levels after the menopause. Oestrogen favours the deposition of SAT and decreased oestrogen levels lead to a smaller proportion of fat gain in SAT [4].
For men and women who had a body size that prohibited the direct determination of the complete amounts of VAT and SAT, estimated values were used. The exclusion of participants with estimated values of SAT and VAT attenuated the greater difference in VAT between type 2 diabetes and NGM in men than in women. As these participants generally had larger amounts of VAT and were not missing at random, range restriction could explain the observed attenuated sex difference [35].
We observed that men had more lean mass than women, which is in line with previous research [2]. We additionally observed that both women and men with prediabetes or type 2 diabetes had more lean mass than their NGM counterparts and that the difference in lean mass between type 2 diabetes and NGM was greater in women than in men. Previous data on diabetes-associated lean mass have been inconsistent [7–10, 36]. Hyperglycaemia and type 2 diabetes have been associated with lower levels of lean mass [7, 8, 36]; however, it has also been suggested that high levels of lean mass do not protect against insulin resistance [9, 10]. Although we adjusted for fat mass, residual effects of adiposity may underlie the positive associations of prediabetes and type 2 diabetes with lean mass. Increased adiposity has been suggested to act as a chronic overload stimulus on the muscles, increasing muscle size and strength [37]. A further possibility is that, in people with increased adiposity and type 2 diabetes, lean mass is less functional because of skeletal muscle lipid infiltration [10]. In turn, lean mass has to increase to compensate for malfunction. Whether women are more susceptible to these mechanisms is unclear. Nevertheless, women’s greater difference in lean mass between type 2 diabetes and NGM does not seem to be explained by differences in lifestyle factors, as we previously found that there were no sex differences in the association of type 2 diabetes with physical activity and healthy diet score [38].
The strengths of our study include its population-based design combined with oversampling of individuals with type 2 diabetes, which enabled an accurate comparison of individuals with and without type 2 diabetes. Additionally, this study benefits from the large sample size and the detailed phenotypic assessment. There are also some limitations. First, the data were cross-sectional; therefore, we cannot determine the causality and direction in the associations of measures of body composition and (pre)diabetes. However, we do not expect this to affect the sex differences investigated. Second, our population was generally relatively healthy; this may have resulted in an underestimation of the sex-specific associations found and the differences therein. Additionally, our study population consisted of middle-aged white individuals. Our results are generalisable to individuals with similar characteristics, but it should be kept in mind that the associations and sex differences found may differ in populations with a different distribution of determinants or in other ethnic groups. Third, some variables had missing data. After multiple imputation, the observed sex differences were comparable to those found in the complete case analysis. Fourth, the interplay of sex, body composition and type 2 diabetes is complex. We did not investigate sex differences in the associations of measures of body composition with insulin resistance or beta cell function, or sex differences in the association of pancreatic fat with (pre)diabetes, which may aid our understanding of sex differences in the role of body composition in the development of type 2 diabetes, as this was beyond the scope of this study. Finally, the percentage of women in the type 2 diabetes study population was about 14 percentage points lower than that in the source population [39]. If the apparent under-representation of women with type 2 diabetes was the result of health selection, this could have influenced the sex differences seen. However, the recruitment strategy was the same for women and men.
## Conclusion
In conclusion, we found that differences in SAT, hip circumference and lean mass between people with type 2 diabetes and people with NGM were greater in women than in men, and the difference in VAT was greater in men than in women. There was no difference in the percentage of liver fat between people with type 2 diabetes and people with NGM. These results suggest that there is a sex-specific role of body composition in the development of type 2 diabetes and that sex-specific prevention measures are necessary.
## Supplementary Information
ESM(PDF 695 kb)
## Authors’ relationships and activities
The authors declare that there are no relationships or activities that might bias, or be perceived to bias, their work.
## Contribution statement
RdR wrote this paper under the supervision of CS, SS and CvdK. All authors contributed to the conception and design of the study, acquistion of data and/or interpretation of data. All authors reviewed the draft paper and provided critical intellectual content and approved the final version of the manuscript and its submission to Diabetologia. CS takes responsibility for the contents of the article.
## References
1. Scherer PE, Hill JA. **Obesity, diabetes, and cardiovascular diseases: a compendium**. *Circ Res* (2016.0) **118** 1703-1705. DOI: 10.1161/CIRCRESAHA.116.308999
2. Schorr M, Dichtel LE, Gerweck AV. **Sex differences in body composition and association with cardiometabolic risk**. *Biol Sex Differ* (2018.0) **9** 28. DOI: 10.1186/s13293-018-0189-3
3. Paul S, Thomas G, Majeed A, Khunti K, Klein K. **Women develop type 2 diabetes at a higher body mass index than men**. *Diabetologia* (2012.0) **55** 1556-1557. DOI: 10.1007/s00125-012-2496-2
4. Power ML, Schulkin J. **Sex differences in fat storage, fat metabolism, and the health risks from obesity: possible evolutionary origins**. *Br J Nutr* (2008.0) **99** 931-940. DOI: 10.1017/S0007114507853347
5. Einstein FH, Atzmon G, Yang XM. **Differential responses of visceral and subcutaneous fat depots to nutrients**. *Diabetes* (2005.0) **54** 672-678. DOI: 10.2337/diabetes.54.3.672
6. Yeung CHC, Au Yeung SL, Fong SSM, Schooling CM. **Lean mass, grip strength and risk of type 2 diabetes: a bi-directional Mendelian randomisation study**. *Diabetologia* (2019.0) **62** 789-799. DOI: 10.1007/s00125-019-4826-0
7. Kalyani RR, Tra Y, Egan JM, Ferrucci L, Brancati F. **Hyperglycemia is associated with relatively lower lean body mass in older adults**. *J Nutr Health Aging* (2014.0) **18** 737-743. DOI: 10.1007/s12603-014-0445-0
8. Sugimoto K, Tabara Y, Ikegami H. **Hyperglycemia in non-obese patients with type 2 diabetes is associated with low muscle mass: the Multicenter Study for Clarifying Evidence for Sarcopenia in Patients with Diabetes Mellitus**. *J Diabetes Investig* (2019.0) **10** 1471-1479. DOI: 10.1111/jdi.13070
9. Rehunen SK, Kautiainen H, Korhonen PE, Eriksson JG. **Lean body mass is not beneficial, but may be detrimental for glucose tolerance: splitting body mass index according to body composition**. *Prim Care Diabetes* (2020.0) **14** 747-752. DOI: 10.1016/j.pcd.2020.05.003
10. Rehunen SKJ, Kautiainen H, Korhonen PE, Eriksson JG. **A high lean body mass is not protecting from type 2 diabetes in the presence of a high body fat mass**. *Diabetes Metab* (2021.0) **47** 101219. DOI: 10.1016/j.diabet.2020.101219
11. Schram MT, Sep SJ, van der Kallen CJ. **The Maastricht Study: an extensive phenotyping study on determinants of type 2 diabetes, its complications and its comorbidities**. *Eur J Epidemiol* (2014.0) **29** 439-451. DOI: 10.1007/s10654-014-9889-0
12. Katz EG, Stevens J, Truesdale KP, Cai J, Adair LS, North KE. **Hip circumference and incident metabolic risk factors in Chinese men and women: the People’s Republic of China study**. *Metab Syndr Relat Disord* (2011.0) **9** 55-62. DOI: 10.1089/met.2010.0045
13. de Ritter R, Sep SJS, van der Kallen CJH. **Sex differences in the association of prediabetes and type 2 diabetes with microvascular complications and function: the Maastricht Study**. *Cardiovasc Diabetol* (2021.0) **20** 102. DOI: 10.1186/s12933-021-01290-x
14. Woodward M. **Rationale and tutorial for analysing and reporting sex differences in cardiovascular associations**. *Heart* (2019.0) **105** 1701-1708. DOI: 10.1136/heartjnl-2019-315299
15. 15.Heymans MW, Eekhout I (2019) Applied missing data analysis with SPSS and (R)Studio. Available from https://bookdown.org/mwheymans/bookmi/. Accessed 5 Dec 2022.
16. Wiklund P, Toss F, Jansson JH. **Abdominal and gynoid adipose distribution and incident myocardial infarction in women and men**. *Int J Obes (Lond)* (2010.0) **34** 1752-1758. DOI: 10.1038/ijo.2010.102
17. Wijnhoven HA, Snijder MB, van Bokhorst-de van der Schueren MA, Deeg DJ, Visser M. **Region-specific fat mass and muscle mass and mortality in community-dwelling older men and women**. *Gerontology* (2012.0) **58** 32-40. DOI: 10.1159/000324027
18. Langeard A, Fakrahnak Z, Vrinceanu T. **Sex-moderated association between body composition and cognition in older adults**. *Exp Gerontol* (2020.0) **138** 111002. DOI: 10.1016/j.exger.2020.111002
19. Wu B, Huang J, Fukuo K, Suzuki K, Yoshino G, Kazumi T. **Different associations of trunk and lower-body fat mass distribution with cardiometabolic risk factors between healthy middle-aged men and women**. *Int J Endocrinol* (2018.0) **2018** 1289485. DOI: 10.1155/2018/1289485
20. Bredella MA. **Sex differences in body composition**. *Adv Exp Med Biol* (2017.0) **1043** 9-27. DOI: 10.1007/978-3-319-70178-3_2
21. Hanley AJ, Wagenknecht LE, Norris JM. **Insulin resistance, beta cell dysfunction and visceral adiposity as predictors of incident diabetes: the Insulin Resistance Atherosclerosis Study (IRAS) Family study**. *Diabetologia* (2009.0) **52** 2079-2086. DOI: 10.1007/s00125-009-1464-y
22. Conway B, Xiang YB, Villegas R. **Hip circumference and the risk of type 2 diabetes in middle-aged and elderly men and women: the Shanghai women and Shanghai men's health studies**. *Ann Epidemiol* (2011.0) **21** 358-366. DOI: 10.1016/j.annepidem.2011.02.005
23. Piche ME, Tchernof A, Despres JP. **Obesity phenotypes, diabetes, and cardiovascular diseases**. *Circ Res* (2020.0) **126** 1477-1500. DOI: 10.1161/CIRCRESAHA.120.316101
24. Cypess AM. **Reassessing human adipose tissue**. *N Engl J Med* (2022.0) **386** 768-779. DOI: 10.1056/NEJMra2032804
25. Snijder MB, Visser M, Dekker JM. **Low subcutaneous thigh fat is a risk factor for unfavourable glucose and lipid levels, independently of high abdominal fat. The Health ABC Study**. *Diabetologia* (2005.0) **48** 301-308. DOI: 10.1007/s00125-004-1637-7
26. Manolopoulos KN, Karpe F, Frayn KN. **Gluteofemoral body fat as a determinant of metabolic health**. *Int J Obes (Lond)* (2010.0) **34** 949-959. DOI: 10.1038/ijo.2009.286
27. Koster A, Stenholm S, Alley DE. **Body fat distribution and inflammation among obese older adults with and without metabolic syndrome**. *Obesity (Silver Spring)* (2010.0) **18** 2354-2361. DOI: 10.1038/oby.2010.86
28. Snijder MB, Dekker JM, Visser M. **Associations of hip and thigh circumferences independent of waist circumference with the incidence of type 2 diabetes: the Hoorn Study**. *Am J Clin Nutr* (2003.0) **77** 1192-1197. DOI: 10.1093/ajcn/77.5.1192
29. Lim EL, Hollingsworth KG, Aribisala BS, Chen MJ, Mathers JC, Taylor R. **Reversal of type 2 diabetes: normalisation of beta cell function in association with decreased pancreas and liver triacylglycerol**. *Diabetologia* (2011.0) **54** 2506-2514. DOI: 10.1007/s00125-011-2204-7
30. Marinou K, Hodson L, Vasan SK. **Structural and functional properties of deep abdominal subcutaneous adipose tissue explain its association with insulin resistance and cardiovascular risk in men**. *Diabetes Care* (2014.0) **37** 821-829. DOI: 10.2337/dc13-1353
31. Kim SH, Chung JH, Song SW, Jung WS, Lee YA, Kim HN. **Relationship between deep subcutaneous abdominal adipose tissue and metabolic syndrome: a case control study**. *Diabetol Metab Syndr* (2016.0) **8** 10. DOI: 10.1186/s13098-016-0127-7
32. Brand T, van den Munckhof ICL, van der Graaf M. **Superficial vs deep subcutaneous adipose tissue: sex-specific associations with hepatic steatosis and metabolic traits**. *J Clin Endocrinol Metab* (2021.0) **106** e3881-e3889. DOI: 10.1210/clinem/dgab426
33. Fox CS, Massaro JM, Hoffmann U. **Abdominal visceral and subcutaneous adipose tissue compartments: association with metabolic risk factors in the Framingham Heart Study**. *Circulation* (2007.0) **116** 39-48. DOI: 10.1161/CIRCULATIONAHA.106.675355
34. de Mutsert R, Gast K, Widya R. **Associations of abdominal subcutaneous and visceral fat with insulin resistance and secretion differ between men and women: the Netherlands Epidemiology of Obesity Study**. *Metab Syndr Relat Disord* (2018.0) **16** 54-63. DOI: 10.1089/met.2017.0128
35. Bland JM, Altman DG. **Correlation in restricted ranges of data**. *BMJ* (2011.0) **342** d556. DOI: 10.1136/bmj.d556
36. Al-Sofiani ME, Ganji SS, Kalyani RR. **Body composition changes in diabetes and aging**. *J Diabetes Complications* (2019.0) **33** 451-459. DOI: 10.1016/j.jdiacomp.2019.03.007
37. Tomlinson DJ, Erskine RM, Morse CI, Winwood K, Onambele-Pearson G. **The impact of obesity on skeletal muscle strength and structure through adolescence to old age**. *Biogerontology* (2016.0) **17** 467-483. DOI: 10.1007/s10522-015-9626-4
38. de Ritter R, Sep SJS, van der Kallen CJH. **Adverse differences in cardiometabolic risk factor levels between individuals with pre-diabetes and normal glucose metabolism are more pronounced in women than in men: the Maastricht Study**. *BMJ Open Diabetes Res Care* (2019.0) **7** e000787. DOI: 10.1136/bmjdrc-2019-000787
39. 39.Nielen M, Poos R, Korevaar J (2020) Diabetes mellitus in Nederland. Prevalentie en incidentie: heden, verleden en toekomst. Nivel, Utrecht, the Netherlands. Available from https://www.nivel.nl/sites/default/files/bestanden/1003898.pdf. Accessed 5 Dec 2022 [article in Dutch]
40. 40.de Ritter R (2022) Sex differences in causes and consequences of type 2 diabetes. PhD thesis. Maastricht University, the Netherlands. 10.26481/dis.20220908rr
|
---
title: The utility of a type 2 diabetes polygenic score in addition to clinical variables
for prediction of type 2 diabetes incidence in birth, youth and adult cohorts in
an Indigenous study population
authors:
- Lauren E. Wedekind
- Anubha Mahajan
- Wen-Chi Hsueh
- Peng Chen
- Muideen T. Olaiya
- Sayuko Kobes
- Madhumita Sinha
- Leslie J. Baier
- William C. Knowler
- Mark I. McCarthy
- Robert L. Hanson
journal: Diabetologia
year: 2023
pmcid: PMC10036431
doi: 10.1007/s00125-023-05870-2
license: CC BY 4.0
---
# The utility of a type 2 diabetes polygenic score in addition to clinical variables for prediction of type 2 diabetes incidence in birth, youth and adult cohorts in an Indigenous study population
## Abstract
### Aims/hypothesis
There is limited information on how polygenic scores (PSs), based on variants from genome-wide association studies (GWASs) of type 2 diabetes, add to clinical variables in predicting type 2 diabetes incidence, particularly in non-European-ancestry populations.
### Methods
For participants in a longitudinal study in an Indigenous population from the Southwestern USA with high type 2 diabetes prevalence, we analysed ten constructions of PS using publicly available GWAS summary statistics. Type 2 diabetes incidence was examined in three cohorts of individuals without diabetes at baseline. The adult cohort, 2333 participants followed from age ≥20 years, had 640 type 2 diabetes cases. The youth cohort included 2229 participants followed from age 5–19 years (228 cases). The birth cohort included 2894 participants followed from birth (438 cases). We assessed contributions of PSs and clinical variables in predicting type 2 diabetes incidence.
### Results
Of the ten PS constructions, a PS using 293 genome-wide significant variants from a large type 2 diabetes GWAS meta-analysis in European-ancestry populations performed best. In the adult cohort, the AUC of the receiver operating characteristic curve for clinical variables for prediction of incident type 2 diabetes was 0.728; with the PS, 0.735. The PS’s HR was 1.27 per SD ($$p \leq 1.6$$ × 10−8; $95\%$ CI 1.17, 1.38). In youth, corresponding AUCs were 0.805 and 0.812, with HR 1.49 ($$p \leq 4.3$$ × 10−8; $95\%$ CI 1.29, 1.72). In the birth cohort, AUCs were 0.614 and 0.685, with HR 1.48 ($$p \leq 2.8$$ × 10−16; $95\%$ CI 1.35, 1.63). To further assess the potential impact of including PS for assessing individual risk, net reclassification improvement (NRI) was calculated: NRI for the PS was 0.270, 0.268 and 0.362 for adult, youth and birth cohorts, respectively. For comparison, NRI for HbA1c was 0.267 and 0.173 for adult and youth cohorts, respectively. In decision curve analyses across all cohorts, the net benefit of including the PS in addition to clinical variables was most pronounced at moderately stringent threshold probability values for instituting a preventive intervention.
### Conclusions/interpretation
This study demonstrates that a European-derived PS contributes significantly to prediction of type 2 diabetes incidence in addition to information provided by clinical variables in this Indigenous study population. Discriminatory power of the PS was similar to that of other commonly measured clinical variables (e.g. HbA1c). Including type 2 diabetes PS in addition to clinical variables may be clinically beneficial for identifying individuals at higher risk for the disease, especially at younger ages.
### Supplementary Information
The online version of this article (10.1007/s00125-023-05870-2) contains peer-reviewed but unedited supplementary material.
## Introduction
Type 2 diabetes-associated genetic variants, derived from genome-wide association studies (GWASs), have largely been reproducible across populations. There is limited information on how polygenic scores (PSs) based on these variants add to clinical variables for predicting type 2 diabetes incidence. Such prediction could help identify individuals at increased risk of type 2 diabetes for targeted prevention efforts.
Previous studies assessing contributions of a type 2 diabetes PS for prediction of type 2 diabetes incidence have mostly been conducted in European-ancestry populations [1, 2, 3, 4, 5, 6]. These studies, using PSs constructed from 15 variants to over six million common variants, have generally found that PSs were significantly associated with type 2 diabetes incidence but contributed little beyond clinical variables to overall prediction of type 2 diabetes [1, 2, 3, 4, 5, 6].
Previous studies were largely conducted in adults, but the utility of PSs for prediction of subsequent type 2 diabetes may be greater earlier in life (in youth or even at birth). The present study employed a PS for prediction of type 2 diabetes incidence in an Indigenous population from the Southwestern USA with a high prevalence of type 2 diabetes and obesity, and in which long-term follow-up data are available. In this population, the age-adjusted prevalence of diabetes is approximately six times higher than in non-Hispanic white people in the USA [7]. We aimed to analyse how genetic and clinical variables could inform strategies for screening and prevention in three cohorts of individuals in different age groups (birth, youth and adulthood) at baseline.
## Study design and participants
A longitudinal study of diabetes (1965–2007) was conducted in an Indigenous study population from the Southwestern USA; methods for this study have been described previously [8]. Before participation, volunteers were fully informed of the nature and purpose of the study and adult participants provided written informed consent, including consent for genetic studies; minor participants provided written assent. Protocols were approved by the institutional review board of the National Institute of Diabetes and Digestive and Kidney Diseases, and research was conducted in accordance with the principles of the Declaration of Helsinki.
Briefly, individuals at least 5 years old were invited for health examinations every 2 years. At each exam, a 75 g oral glucose tolerance test was administered with measurement of HbA1c and fasting and 2 h plasma glucose (FPG, 2hPG). Diabetes was diagnosed using 1997 American Diabetes *Association criteria* (FPG ≥7.0 mmol/l, 2hPG ≥11.1 mmol/l or clinical diagnosis) [9]. Height and weight were measured to calculate BMI and birthweight was collected from clinical information and Arizona state birth certificates. Participants had not been directly asked to report parental diabetes; however, since many participants’ parents had also participated in the study, we were able to approximate the information that would be available in clinical encounters by using information from direct examination in the parents. We defined parental diabetes using three categories (yes, no or unknown) per parent. Characteristics of participants are summarised in electronic supplementary material (ESM) Tables 1 and 2.
## Genotypic data
Of the study participants, 7701 had genotypes available from previous GWASs, generated using a custom Axiom array designed to capture common variation in members of this community (minor allele frequency (MAF) ≥0.05, or ≥0.01 for coding variants), using methods described previously (Affymetrix, Santa Clara, CA, USA) [10]. Missing and ungenotyped variants were imputed with whole genome sequence data for 266 community members as a reference panel using Impute 2, resulting in 6.6 million variants with MAF >0.01 and imputation quality score >0.5 (median 0.95) [11]. Previous work in this population suggests that a population-specific reference panel is optimal for imputing common variants, with little value from including samples from outside populations [12]. Variants were excluded from analyses if they had an imputation quality score <0.5 or MAF <0.01 (ESM Method 1).
## Study cohorts
Of the 7701 individuals with genotypes available, we constructed three cohorts based on age at baseline examination for those who had data for at least two exams with availability of clinical variables. There were 2333 participants followed from first examination in adulthood (age ≥20 years); 640 cases of type 2 diabetes occurred over 16,686 person-years of follow-up. There were 2229 participants followed from first examination in youth (age 5–19 years); 228 cases of type 2 diabetes occurred over 17,803 person-years of follow-up. There were 2894 participants with birthweight data available who were considered to be followed from birth; 438 cases of type 2 diabetes occurred over 61,591 person-years of follow-up. Individuals were included in multiple cohorts if suitable data were available.
## Construction of type 2 diabetes PSs
We compared associations of ten different constructions of type 2 diabetes PS, derived from GWASs conducted for populations from various world regions. We used ‘pruning and thresholding’ methods to select variants for the PS, selecting independent genome-wide significant variants from the GWASs for other populations (ESM Tables 3–6). These PSs included the following, each named for the meta-analysis from which it was derived: Diabetes Genetics Replication And Meta-analysis consortium (DIAGRAM) 2018 (constituting 293 variants derived from European populations) [13], Asian Genetic Epidemiology *Network consortium* (AGEN) 2020 (125 variants derived from East Asian populations) [14] and Diabetes Meta-Analysis of Trans-Ethnic association studies consortium (DIAMANTE) 2022. The seven DIAMANTE PSs are constructions of multi-ancestry PSs (287 variants) with weights taken from meta-analyses of populations representing the following ancestry groups: multi-ancestry, African, East Asian, European, Hispanic/Latino and South Asian [15], in addition to a ‘population-specific weight’ PS from these same DIAMANTE 2022 variants with weights derived from the present population, using tenfold cross-validation to address overfitting. Finally, we also derived a ‘population-specific variant’ PS by selecting 287 type 2 diabetes-associated variants from the 515,692 variants typed in the type 2 diabetes GWAS in the study population, using twofold cross-validation (ESM Method 2). While PSs can be constructed using a larger number of variants, by using less stringent significance thresholds or accounting for linkage disequilibrium, applicability across populations with different linkage disequilibrium patterns is uncertain. We, thus, employed the widely used method of selecting significant variants.
We constructed each PS using imputed genotypes available in the present study population. The products of the number of risk alleles for each individual with the effect size (logarithm of the OR) from the corresponding GWAS were summed across variants. PSs were standardised across the entire study population to have mean of 0 and SD of 1: HRs for PSs were expressed in terms of SD of that PS.
## Statistical analyses
Analyses were completed in SAS 9.4 (SAS Institute, Cary, NC, USA). For each cohort, individuals were followed from inception (first examination with clinical data available for youth and adult cohorts; birth for birth cohort) until they developed type 2 diabetes or until their last examination, whichever came first. We evaluated the relative contributions of various combinations of clinical variables and/or the PS in the following analyses: cumulative incidence, survival, AUC of the receiver operating characteristic curve, net reclassification improvement (NRI) and decision curve. Cumulative incidence, survival, decision curve and NRI analyses required calculation of the predicted occurrence of type 2 diabetes at a specified follow-up time for all individuals to ensure comparability: a follow-up of 10 years was used for the adult and youth cohorts, and 30 years for the birth cohort.
The variables available for the adult cohort included: age, sex, parental diabetes, BMI, HbA1c, FPG, 2hPG and the type 2 diabetes PS [16]. Those for the youth cohort included: age, sex, parental diabetes, modified BMI z score [17], HbA1c, FPG, 2hPG and PS. Those for the birth cohort included: sex, parental diabetes, birthweight and PS. This specific set of clinical variables was chosen because the US Preventive Services Task Force focuses on measures of obesity, family history and hyperglycaemia in recommendations for screening and prevention of type 2 diabetes [18]. Most previous studies have assessed prediction with control for a similar set of clinical predictors as used here, but many studies have also included measurement of lipids [1, 2, 3, 4, 5, 6]. Measurements of serum HDL and triglycerides/triacylglycerols were available in a subset of the present cohort (measured since 1993) and adjustment for these variables yielded similar results to those observed in primary analyses (ESM Table 7).
Since HbA1c was only measured for examinations after 1989, we conducted additional analyses that did not require HbA1c to allow for longer follow-up and greater sample size: these analyses returned similar findings to analyses that included HbA1c (ESM Table 8). Given the U-shaped relationship between birthweight and type 2 diabetes in this population [19], we analysed birthweight using two binary variables, one denoting birthweight <3000 g and another denoting birthweight >4000 g. We also conducted analyses that included a continuous birthweight variable and its squared term to capture the quadratic relationship between birthweight and type 2 diabetes. While these analyses gave similar results, the dichotomised birthweight variables yielded a better fit according to Akaike’s information criterion.
We also conducted analyses including stated admixture as a covariate; its inclusion returned virtually the same results as without. To further control for population stratification, we conducted additional analyses after adjustment of the PS for the first ten genetic principal components derived from the GWAS, with separate estimation of principal components in each of the three target cohorts; results were similar to those of the primary analyses (ESM Table 9). Previous studies in this population demonstrated that genetic variants at KCNQ1 rs2237895 (risk allele frequency=0.49, OR 1.31; exhibits parent-of-origin effects) [20] and ABCC8 rs1272388614 (risk allele frequency=0.017, OR 2.02) [21] are significantly and strongly associated with type 2 diabetes. We conducted further analyses to assess the contributions of these genotypes in addition to the type 2 diabetes PS for prediction of type 2 diabetes incidence.
## Cumulative incidence and survival analyses
We used Cox proportional hazards regression to evaluate associations of clinical variables and PS with type 2 diabetes incidence. Cumulative incidence of type 2 diabetes was calculated as the proportion of individuals that developed type 2 diabetes over the specified follow-up time, using Breslow’s method (PROC PHREG in SAS). To assess separate contributions of PS and clinical risk, we calculated predicted cumulative incidence according to different levels of PS and of clinical risk, as determined by linear predictors from the clinical variables in the proportional hazards model.
## AUC analyses
We compared the Harrell’s C statistic [22] of models that included clinical variables alone with the C statistic of those that included clinical variables and the PS. The C statistic expresses the probability within a pair of individuals, one who developed type 2 diabetes and one who did not, that the individual who developed type 2 diabetes had a higher predicted probability of doing so [23]. In the context of survival analysis (e.g. in the proportional hazards models used here), the C statistic is equivalent to the AUC of the receiver operating characteristic curve [23], and we refer to it as ‘AUC’ throughout the manuscript.
We conducted AUC analyses to evaluate the predictive accuracy of models containing combinations of clinical variables and the PS. In the adult cohort, the AUC for the model with age and sex was 0.590 ($95\%$ CI 0.566, 0.615); with the PS, 0.619 ($95\%$ CI 0.596, 0.643); the difference in AUC (i.e. ∆AUC) was 0.029 ($$p \leq 0.003$$). In the youth cohort, corresponding AUCs were 0.625 ($95\%$ CI 0.587, 0.663) and 0.682 ($95\%$ CI 0.648, 0.716); the ∆AUC was 0.057 ($$p \leq 3.96$$ × 10−4). In the birth cohort, AUC for the model with sex was 0.537 ($95\%$ CI 0.512, 0.562); with the PS, 0.638 ($95\%$ CI 0.610, 0.666); the ∆AUC was 0.101 ($p \leq 10$−5).
Though the PS was strongly associated with incident type 2 diabetes, the improvement in AUC compared with clinical variables alone was modest. In the adult cohort, AUC for the full clinical model was 0.728 ($95\%$ CI 0.706, 0.750); with the PS, 0.735 ($95\%$ CI 0.714, 0.757); and the ∆AUC was 0.007 ($$p \leq 0.023$$) (Table 1). In the youth cohort, AUC for the full clinical model was 0.805 ($95\%$ CI 0.778, 0.832); with the PS, 0.812 ($95\%$ CI 0.785, 0.839); and the ∆AUC was 0.007 ($$p \leq 0.173$$) (Table 2). For the birth cohort, the increment in AUC with addition of the PS was greater: the AUC for the model including clinical variables was 0.613 ($95\%$ CI 0.582, 0.644); with the PS, 0.685 ($95\%$ CI 0.657, 0.713); the ∆AUC was 0.071 ($p \leq 10$−5) (Table 3).
## NRI analyses
Continuous-variable NRI quantifies the amount of correct reclassification introduced by using a model with an additional variable [24]. We analysed NRI by calculating the net proportion of events reclassified correctly (assigned a higher probability value) plus the net proportion of nonevents reclassified correctly (assigned a lower probability value) [25]. Confidence intervals for the NRI were calculated by a bootstrap method.
While AUC provides a measure of overall predictive accuracy, it does not fully capture the extent to which addition of a variable can affect individual risk estimates. To examine this, we calculated predicted cumulative incidence of type 2 diabetes according to PS for various levels of clinical risk. Across all cohorts, greater type 2 diabetes PS and greater percentiles of clinical linear predictor were both directly and separately associated with predicted cumulative incidence of type 2 diabetes (Fig. 2). Fig. 2Predicted cumulative incidence of type 2 diabetes for the scaled DIAGRAM 2018 PS and specified percentiles of the clinical linear predictor. Cumulative incidence was calculated for various combinations of the PS and clinical risk (based on the linear predictor derived from the clinical variables in the models). Across all cohorts, greater type 2 diabetes PS and greater percentiles of clinical linear predictor were both directly and separately associated with predicted cumulative incidence of type 2 diabetes. ( a) Predicted cumulative incidence of type 2 diabetes over 10 years of follow-up in the adult cohort: clinical variables include age, sex, parental diabetes, BMI, FPG, HbA1c. At 10 year follow-up, 504 individuals had developed type 2 diabetes and 635 remained at risk. ( b) Predicted cumulative incidence of type 2 diabetes over 10 years of follow-up in the youth cohort: clinical variables include age, sex, parental diabetes, modified BMI z score, FPG, HbA1c. At 10 year follow-up, 152 had developed type 2 diabetes and 745 remained at risk. ( c) Predicted cumulative incidence of type 2 diabetes over 30 years of follow-up in the birth cohort: clinical variables include sex, parental diabetes, birth weight. At 30 year follow-up, 340 had developed type 2 diabetes and 474 remained at risk To further quantify the contribution of each variable to the model’s risk classification, we calculated the NRI of each variable. NRI quantifies the extent to which type 2 diabetes cases and non-cases are consequently reclassified upon inclusion of an additional variable. The NRI for adding the PS to clinical variables was 0.270 ($95\%$ CI 0.149, 0.392; 0.092 for events, 0.178 for nonevents) in the adult cohort (Table 1); in the youth cohort, 0.268 ($95\%$ CI 0.073, 0.464; 0.085 for events, 0.183 for nonevents) (Table 2); in the birth cohort, 0.362 ($95\%$ CI 0.222, 0.502; 0.106 for events, 0.256 for nonevents) (Table 3). In comparison, the NRI for HbA1c was 0.267 in the adult cohort and 0.173 in the youth cohort.
## Decision curve analyses
We employed decision-analytic methods to assess consequences of clinical decisions and expected outcomes of alternative clinical management (i.e. including various combinations of clinical variables with and without the PS in prediction models). These analyses assume that the threshold probability (pt) of developing type 2 diabetes at which one would opt for an intervention is informative of how one weighs the relative benefits and harms of true-positive and false-positive predictions, and the net benefit of using a predictive model to select individuals above a given pt is calculated accordingly [26]. We used extensions to decision curve methods for survival analysis to plot net benefit across a range of pt values to evaluate for which pt ranges and what corresponding proportion of the population the PS had marginal net benefit [27].
We employed decision curve analyses to estimate the net benefit of including the PS at a range of threshold probabilities (i.e. minimum probabilities of disease that would warrant intervention). When the costs of false-positives are low (i.e. as pt approaches 0), population-wide interventions may be favoured; thus, screening by clinical or genetic means would have little net benefit. When false-positive costs are higher (i.e. at higher pt values), net clinical benefit can be increased by screening to target the intervention to higher-risk individuals.
In the adult cohort, the net benefit of including the PS in addition to clinical variables was most pronounced at pt values 0.3 to 0.5 (up to $18\%$ improvement); this corresponded to 15–$40\%$ of the highest-risk individuals selected for the intervention (Fig. 3). In the youth cohort, the net benefit of including the PS was most pronounced at pt values 0.05 to 0.35 (up to $21\%$ improvement) (Fig. 3). In the birth cohort, the net benefit of including the PS was most pronounced at pt values 0.15 to 0.35 (up to $56\%$ improvement) (Fig. 3). Fig. 3Net benefit of predictive models with or without the DIAGRAM 2018 PS for type 2 diabetes prediction. For each model in each cohort, the net benefit is plotted on the y-axis against the pt for implementing an intervention on the x-axis. The proportion of the population that would be selected at each pt is shown below the x-axis. ( a) In the adult cohort, the net benefit of including the PS in addition to clinical variables was most pronounced at pt values 0.3–0.5 (up to $18\%$ improvement); this corresponded to 15–$40\%$ of the highest-risk individuals selected for the intervention. ( b) In the youth cohort, the net benefit of including the PS was most pronounced at pt values 0.05–0.35 (up to $21\%$ improvement). ( c) In the birth cohort, the net benefit of including the PS was most pronounced at pt values 0.15–0.35 (up to $56\%$ improvement). Bwt, birthweight
## Comparisons of associations of PSs among cohorts
To compare the effects of the PSs (e.g. HRs) among the different age cohorts, a bootstrap analysis was conducted as previously described [28]. In brief, the 4770 individuals who were included in at least one cohort were resampled 2000 times, and the analyses were repeated for each iteration. The resulting differences in the logarithm of the HR between each pair of cohorts and their standard errors were calculated and used to test statistical significance of the differences. Since the availability and predictive power of different clinical covariates may affect the HR estimates, these analyses were conducted without any covariates.
## Ten constructions of type 2 diabetes PSs
All ten PSs for type 2 diabetes, constructed using the overlap of published type 2 diabetes GWAS summary statistics and genotypes available in this study population, had significant associations with type 2 diabetes incidence in the study population. HRs for the PSs in models adjusted for clinical variables (age, sex, BMI, FPG, HbA1c and parental diabetes for the adult cohort; age, sex, modified BMI z score, FPG, HbA1c and parental diabetes for the youth cohort; and sex, birthweight and parental diabetes for the birth cohort) ranged from 1.13 to 1.27 per SD for the adult cohort, from 1.19 to 1.49 for the youth cohort and from 1.27 to 1.48 for the birth cohort (ESM Table 10). The PS that consistently had the strongest associations with type 2 diabetes incidence (largest HRs) was constructed using the DIAGRAM 2018 GWAS. Thus, for the rest of this text, we present results for the DIAGRAM 2018 PS. Calibration plots for models for this PS for each of the three cohorts show that these models are well-calibrated (ESM Fig. 1). The DIAMANTE 2022 multi-ancestry PS and the population-specific variant PS also had strong associations with type 2 diabetes incidence, though not as strong (ESM Tables 10–12 and ESM Figs 2–7).
## Association of PS with incidence of type 2 diabetes
The best-performing PS was significantly associated with type 2 diabetes incidence in adult, youth and birth cohorts (Fig. 1). In the adult cohort, 10 year cumulative incidence of type 2 diabetes in the lowest decile of PS was $20.5\%$; in the highest, $42.5\%$ (unadjusted HR=1.31 per SD, $$p \leq 6.9$$ × 10−11). In the youth cohort, 10 year cumulative incidence of type 2 diabetes in the lowest decile of PS was $2.4\%$; in the highest, $21.5\%$ (HR=1.59 per SD, $$p \leq 6.8$$ × 10−12). In the birth cohort, 30 year cumulative incidence of type 2 diabetes in the lowest decile of PS was $15.1\%$; in the highest, $37.3\%$ (HR=1.47 per SD, $$p \leq 1.7$$ × 10−15). The clinical predictors were also strongly associated with incidence of type 2 diabetes (ESM Fig. 8). Fig. 1Cumulative incidence of type 2 diabetes by decile of the DIAGRAM 2018 PS. The PS was significantly associated with type 2 diabetes incidence in adult, youth and birth cohorts. ( a) Cumulative incidence in the adult cohort at 10 year follow-up; at 10 years, 504 individuals had developed type 2 diabetes and 635 remained at risk. ( b) Cumulative incidence in the youth cohort at 10 year follow-up; at 10 years, 152 individuals had developed type 2 diabetes and 745 remained at risk. ( c) Cumulative incidence in the birth cohort at 30 year follow-up; at 30 years, 340 individuals had developed type 2 diabetes and 474 remained at risk. T2D, type 2 diabetes
## Survival analyses with adjustment for clinical predictors
We conducted survival analyses to assess associations of individual clinical variables and the PS with type 2 diabetes incidence. In the adult cohort, in the model with clinical variables, the HR of the PS was 1.27 per SD ($$p \leq 1.6$$ × 10−8; $95\%$ CI 1.17, 1.38; Table 1). In the youth cohort, in the model with clinical variables, the HR of the PS was 1.49 ($$p \leq 4.3$$ × 10−8; $95\%$ CI 1.29, 1.72) (Table 2). In the birth cohort, in the model with clinical variables, the HR of the PS was 1.48 ($$p \leq 2.8$$ × 10−16; $95\%$ CI 1.35, 1.63) (Table 3). Adding 2hPG to adult and youth cohorts’ models did not substantially alter the HRs of the PS. *In* general, the HRs associated with clinical variables were only modestly affected with the addition of the PS. Table 1Results of survival, AUC and NRI analyses for the adult cohortVariableClinical variables (AUC=0.728)Clinical variables + PS (AUC=0.735)Clinical variables + 2hPG (AUC=0.760)Clinical variables + 2hPG + PS (AUC=0.765)HR ($95\%$ CI)p valueNRIHR ($95\%$ CI)p valueNRIHR ($95\%$ CI)p valueNRIHR ($95\%$ CI)p valueNRIAge (decades)1.01 (0.93, 1.10)0.8690.0111.04 (0.95, 1.13)0.3930.0671.00 (0.92, 1.09)0.992−0.0451.03 (0.94, 1.12)0.5400.112Sex (F/M)1.35 (1.14, 1.59)5.48 × 10−40.1151.34 (1.13, 1.58)7.70 × 10−40.1161.13 (0.953, 1.34)0.1570.1101.13 (0.954, 1.35)0.1540.106Mother diabetic/NonDb1.52 (1.21, 1.90)2.75 × 10−40.1611.46 (1.17, 1.83)2.08 × 10−30.1401.43 (1.14, 1.80)2.60 × 10−30.1781.38 (1.10, 1.73)0.01310.126Mother unknown/NonDb1.45 (1.15, 1.82)1.38 (1.09, 1.74)1.39 (1.10, 1.74)1.33 (1.06, 1.67)Father diabetic/NonDb1.45 (1.06, 1.96)1.38 (1.01, 1.87)1.38 (1.02, 1.88)1.33 (0.975, 1.81)Father unknown/NonDb1.20 (0.921, 1.56)1.17 (0.900, 1.52)1.20 (0.93, 1.56)1.17 (0.900, 1.52)BMI (kg/m2)1.02 (1.01, 1.03)1.22 × 10−40.1581.03 (1.01, 1.03)1.03 × 10−50.2011.02 (1.01, 1.03)2.11 × 10−50.2421.03 (1.02, 1.04)1.88 × 10−60.269FPG (mmol/l)1.99 (1.70, 2.33)2.78 × 10−170.3031.95 (1.66, 2.29)3.13 × 10−160.2951.44 (1.22, 1.71)2.51 × 10−40.1761.44 (1.21, 1.71)2.93 × 10−50.191HbA1c (mmol/mol)1.08 (106, 1.10)1.18 × 10−180.2721.08 (1.06, 1.10)1.14 × 10−170.2671.06 (1.05, 1.08)9.64 × 10−130.2151.06 (1.04, 1.08)5.67 × 10−120.2232hPG (mmol/l)––––––1.32 (1.25, 1.40)3.74 × 10−240.4201.31 (1.24, 1.38)6.71 × 10−230.426PS (SD)–––1.27 (1.17, 1.38)1.61 × 10−80.270–––1.24 (1.15, 1.35)2.90 × 10−70.252‘Clinical variables’ refers to: age, sex, parental diabetes, BMI and FPG for the adult cohortNRI is calculated for each predictor by comparing the model including the predictor with a model that does not include the predictorDb, diabetic with regard to parental diabetes; F, female; M, maleTable 2Results of survival, AUC and NRI analyses for the youth cohortVariableClinical variables (AUC=0.805)Clinical variables + PS (AUC=0.812)Clinical variables + 2hPG (AUC=0.820)Clinical variables + 2hPG + PS (AUC=0.825)HR ($95\%$ CI)p valueNRIHR ($95\%$ CI)p valueNRIHR ($95\%$ CI)p valueNRIHR ($95\%$ CI)p valueNRIAge (decades)2.11 (1.46, 3.04)6.50 × 10−50.3732.21 (1.53, 3.18)2.02 × 10−50.3722.15 (1.49, 3.11)4.96 × 10−50.3672.26 (1.57, 3.27)1.37 × 10−50.382Sex (F/M)1.40 (1.06, 1.84)0.01700.0461.32 (1.01, 1.74)0.04580.0461.09 (0.816, 1.45)0.5620.0201.04 (0.78, 1.39)0.7940.032Mother diabetic/NonDb2.46 (1.83, 3.29)1.21 × 10−110.5242.27 (1.70, 3.05)1.28 × 10−90.5142.26 (1.68, 3.04)2.38 × 10−90.4572.10 (1.56, 2.83)1.29 × 10−70.472Mother unknown/NonDb1.84 (1.27, 2.68)1.87 (1.29, 2.72)1.82 (1.26, 2.64)1.83 (1.26, 2.65)Father diabetic/NonDb1.88 (1.27, 2.78)1.72 (1.16, 2.55)1.74 (1.18, 2.57)1.61 (1.09, 2.39)Father unknown/NonDb1.01 (0.740, 1.37)0.972 (0.714, 1.32)1.01 (0.743, 1.37)0.977 (0.719, 1.33)Modified BMI z score1.50 (1.37, 1.66)1.43 × 10−160.6071.53 (1.39, 1.69)4.48 × 10−170.5871.44 (1.30, 1.59)6.63 × 10−130.5231.47 (1.33, 1.63)1.39 × 10−130.578FPG (mmol/l)2.32 (1.66, 3.23)7.87 × 10−70.2612.06 (1.48, 2.88)2.10 × 10−50.2201.46 (1.01, 2.10)0.04350.0041.31 (0.904, 1.88)0.155−0.031HbA1c (mmol/mol)1.06 (1.03, 1.10)2.77 × 10−40.1821.06 (1.03, 1.09)6.29 × 10−40.1731.05 (1.01, 1.08)6.19 × 10−30.0991.04 (1.01, 1.08)0.009100.0772hPG (mmol/l)––––––1.35 (1.21, 1.49)1.71 × 10−80.3031.34 (1.21, 1.48)4.20 × 10−80.310PS (SD)–––1.49 (1.29, 1.72)4.31 × 10−80.268–––1.48 (1.28, 1.71)1.06 × 10−70.277‘Clinical variables’ refers to: age, sex, parental diabetes, modified BMI z score and FPG for the youth cohortNRI is calculated for each predictor by comparing the model including the predictor with a model that does not include the predictorDb, diabetic with regard to parental diabetes; F, female; M, maleTable 3Results of survival, AUC and NRI analyses for the birth cohortVariableSex, parental diabetes (AUC=0.597)Sex, parental diabetes + PS (AUC=0.683)Sex, parental diabetes, birthweight (AUC=0.613)Sex, parental diabetes, birthweight + PS (AUC=0.685)HR ($95\%$ CI)p valueNRIHR ($95\%$ CI)p valueNRIHR ($95\%$ CI)p valueNRIHR ($95\%$ CI)p valueNRISex (F/M)1.18 (0.974, 1.43)0.09150.0901.18 (0.972, 1.43)0.09470.0901.15 (0.951, 1.40)0.1460.0901.14 (0.939, 1.38)0.1840.090Mother diabetic/NonDb7.28 (5.00, 10.6)6.79 × 10−300.2127.86 (5.39, 11.5)1.12 × 10−300.2097.40 (5.08, 10.77)4.00 × 10−300.3497.95 (5.44, 11.6)5.20 × 10−310.317Mother unknown/NonDb0.871 (0.711, 1.07)0.887 (0.724, 1.09)0.875 (0.715, 1.07)0.886 (0.723, 1.09)Father diabetic/NonDb3.93 (2.36, 6.54)3.98 (2.39, 6.64)3.98 (2.38, 6.65)4.05 (2.42, 6.78)Father unknown/NonDb1.01 (0.818, 1.26)1.07 (0.863, 1.33)0.977 (0.787, 1.21)1.03 (0.829, 1.28)Low birthweight––––––1.56 (1.24, 1.96)6.11 × 10−40.1641.16 (0.877, 1.53)5.99 × 10−40.164High birthweight––––––1.17 (0.887, 1.55)1.57 (1.25, 1.97)PS (SD)–––1.48 (1.35, 1.63)2.83 × 10−160.345–––1.48 (1.35, 1.63)2.77 × 10−160.362NRI is calculated for each predictor by comparing the model including the predictor with a model that does not include the predictorDb, diabetic with regard to parental diabetes; F, female; M, male
## Additional genotypic analyses
The effects of some variants strongly associated with type 2 diabetes in this Indigenous study population were not captured in the DIAGRAM 2018 PS. To address this, we assessed the contribution of genotypes for KCNQ1 rs2237895 (which exhibits parent-of-origin effects) and ABCC8 rs1272388614 in the adult cohort. For each genotype, associations were significant; however, they contributed modestly to the model of clinical variables and the PS, as assessed by AUC and NRI analyses (ESM Table 13).
## Discussion
PSs potentially have utility for identification of individuals with higher risk of type 2 diabetes. Previous studies generally reported significant associations between type 2 diabetes PS and diabetes incidence and modest prediction improvement as measured by AUC: ∆AUC from 0.005 to 0.02 [1, 2, 3, 4, 5, 6]. A limited number of studies include measures of reclassification: continuous NRIs ranged from 0.044 to 0.285 [4, 5, 6]. Most previous studies have been done in European-ancestry populations, but some have been done in non-European populations, including Korean [5], African American [29] and Iranian [30]. Findings in these populations have generally been similar to those in European-ancestry groups. In the present study, the DIAGRAM 2018 PS was strongly statistically significant in predicting type 2 diabetes incidence in adult, youth and birth cohorts in an Indigenous study population from the Southwestern USA.
Results of AUC analyses are consistent with findings of previous studies: improvement in prediction contributed by the type 2 diabetes PS was statistically significant but modest. However, ∆AUC does not fully capture the contribution of a single variable to individual risk [31]. We calculated NRIs for individual variables to address this limitation. NRIs for the PS across all cohorts ranged from 0.2 to 0.3, which is considered intermediate power for identifying type 2 diabetes risk [24], and were comparable to those of commonly measured clinical variables (e.g. HbA1c and FPG). Our findings are consistent with the evidence that for most chronic diseases PSs generally provide additional predictive information beyond that provided by traditional risk factors [32].
## Implications of decision curve analyses
Ultimately, clinical utility may depend on how the type 2 diabetes PS affects the decision to implement preventive interventions. Across adult, youth and birth cohorts, results of our decision curve analyses suggest modest increases in clinical benefit for using the PS at moderately stringent pt values. There are few data on optimal pt values for type 2 diabetes prevention: they depend upon preferences of individual patients and clinicians, healthcare system characteristics and the nature of the interventions considered. Many clinicians would recommend lifestyle prevention for individuals with impaired glucose regulation (e.g. FPG ≥5.5 mmol/l or HbA1c ≥39 mmol/mol ($5.7\%$)); in the adult cohort, the prevalence of impaired glucose regulation at baseline was $35\%$, and this would correspond to pt=0.32 (0.21–0.49 based on cumulative incidence). This is in the range in which our analyses suggest meaningful, albeit modest, improvement in clinical benefit from incorporating the type 2 diabetes PS.
Decision curve analysis assumes that the intervention will be equally effective regardless of how risk is determined. There are limited data on how type 2 diabetes PS affects response to preventive interventions. However, a study within the Diabetes Prevention Program Outcomes Study suggested that lifestyle and metformin interventions were both effective, even in those with greater type 2 diabetes PS [33].
## Construction of PS
While all type 2 diabetes PSs we examined were significantly associated with type 2 diabetes incidence across all cohorts, the DIAGRAM 2018 PS, derived from European-ancestry populations, performed slightly better than the others. While we have previously shown modest heterogeneity in effects of established type 2 diabetes variants between Europeans and this study population [7], the DIAGRAM 2018 PS even out-performed a population-specific variant PS with a comparable number of variants, derived by twofold cross-validation in the present population (n≈3850). The expectation is that a PS derived from a GWAS in a more closely matched ancestry group would perform better than one from a different ancestry group, if GWAS sample sizes are equal [34], but PSs derived in a large European-ancestry group can outperform ancestry-specific PSs when the sample size available for deriving the ancestry-specific PS is small [15]. In the present study, the DIAGRAM 2018 PS likely performed well due to the large sample size and extensive fine-mapping in the DIAGRAM type 2 diabetes meta-analysis. Achieving adequate sample sizes for GWASs to derive ancestry-specific PSs in Indigenous study populations is challenging, but many Indigenous populations have extensive linkage disequilibrium which may facilitate the ability of PSs to capture causal variants [35]. While further work is needed to optimise type 2 diabetes PSs across diverse populations, the present study suggests that PSs constructed using results of GWASs in larger populations may be suitable for translation across study populations in which well-powered GWASs are not available. Studies in additional populations are needed.
## Optimal age for preventive interventions
Genetic effects of the PSs with respect to type 2 diabetes incidence were greatest in the youth and birth cohorts. This is consistent with the hypothesis that genetic effects for many chronic diseases are strongest earlier in life [36], and consistent with the finding that familial recurrence risk of diabetes in this population is higher when it occurs at younger ages [37]. The present findings could also reflect the limited availability of phenotypic data for study participants at birth or young ages. However, when analysed without any clinical covariates, the HRs associated for the birth cohort (HR=1.47) and the youth cohort (HR=1.59) were significantly higher than that for the adult cohort (HR=1.31); tests for differences in the HRs between the adult and youth cohorts and adult and birth cohorts yielded $$p \leq 0.037$$ and $$p \leq 0.006$$, respectively, while differences between birth and youth cohorts were not significant ($$p \leq 0.15$$). The improvements in AUC and net benefit upon adding the PS to clinical variables were greatest in the birth cohort. The use of type 2 diabetes PS at birth could be particularly beneficial as phenotypic manifestations of risk (e.g. hyperglycaemia and obesity) are less apparent. However, some relevant clinical measures that may be readily obtained at birth (e.g. birth length for calculation of adiposity measures) were not available in the present study. In adults, there is strong evidence that type 2 diabetes can be prevented by lifestyle modification, pharmacologic treatment or bariatric surgery, but there are few data on the efficacy of preventive efforts initiated in youth or infancy [38]. Thus, while our analyses suggest that the type 2 diabetes PS has the strongest contribution to prediction of type 2 diabetes incidence in the birth cohort, adults may be a more appropriate target population for preventive interventions in the near term.
## Future research
This study shows that type 2 diabetes PSs, as currently constructed, can provide utility for assessing type 2 diabetes risk; as measured by NRI analyses, information from the PS for classifying type 2 diabetes risk is comparable to that from widely used clinical variables (e.g. HbA1c and BMI) in this study population. Further optimisation of the PS is expected to provide better prediction in the future [30]. Such investigations could assess whether differences in population genetic characteristics, obesity and incidence of type 2 diabetes are paralleled by differences in performance of PSs. Results from the present study were derived from an Indigenous population from the Southwestern USA with a relatively high prevalence of type 2 diabetes.
Beyond the scientific issues, however, technical, logistical and cultural issues need consideration before PSs can be incorporated into clinical practice. For example, advances in laboratory methods and informatics are required to make PSs and risk algorithms available to clinicians and patients. Health economics studies are needed to investigate which clinical settings and constructions of type 2 diabetes PS would maximise net benefit for prediction of type 2 diabetes incidence. With such knowledge, more informed decisions about the use of genetic information in prevention of type 2 diabetes could be made.
## Supplementary information
ESM 1(PDF 2.08 kb)
## Authors’ relationships and activities
MIM has served on advisory panels for Pfizer, Novo Nordisk and Zoe Global; has received honoraria from Merck, Pfizer, Novo Nordisk and Eli Lilly; and has received research funding from AbbVie, Astra Zeneca, Boehringer Ingelheim, Eli Lilly, Janssen, Merck, NovoNordisk, Pfizer, Roche, Sanofi Aventis, Servier and Takeda. As of June 2019, MIM and AM are employees of Genentech and holders of Roche stock.
## Contribution statement
LEW, AM, MIM and RLH contributed to the conceptualisation and design of the manuscript. AM, W-CH, PC, MTO and SK contributed to the acquisition of data. LEW, AM, LJB, MS, WCK, MIM and RLH contributed to the analysis and interpretation of data. All co-authors contributed to the drafting of this article and revised it critically for important intellectual content. All authors gave final approval of the version to be published. RLH is the guarantor of this work.
## References
1. Meigs JB, Shrader P, Sullivan LM. **Genotype score in addition to common risk factors for prediction of type 2 diabetes**. *N Engl J Med* (2008) **359** 2208-2219. DOI: 10.1056/NEJMoa0804742
2. Lyssenko V, Jonsson A, Almgren P. **Clinical risk factors, DNA variants, and the development of type 2 diabetes**. *N Engl J Med* (2008) **359** 2220-2232. DOI: 10.1056/NEJMoa0801869
3. Mars N, Koskela JT, Ripatti P. **Polygenic and clinical risk scores and their impact on age at onset and prediction of cardiometabolic diseases and common cancers**. *Nat Med* (2020) **26** 549-557. DOI: 10.1038/s41591-020-0800-0
4. Vaxillaire M, Yengo L, Lobbens S. **Type 2 diabetes-related genetic risk scores associated with variations in fasting plasma glucose and development of impaired glucose homeostasis in the prospective DESIR study**. *Diabetologia* (2014) **57** 1601-1610. DOI: 10.1007/s00125-014-3277-x
5. Park HY, Choi HJ, Hong YC. **Utilizing genetic predisposition score in predicting risk of type 2 diabetes mellitus incidence: a community-based cohort study on middle-aged Koreans**. *J Korean Med Sci* (2015) **30** 1101-1109. DOI: 10.3346/JKMS.2015.30.8.1101
6. He Y, Lakhani CM, Rasooly D, Manrai AK, Tzoulaki I, Patel CJ. **Comparisons of polyexposure, polygenic, and clinical risk scores in risk prediction of type 2 diabetes**. *Diabetes Care* (2021) **44** 935-943. DOI: 10.2337/dc20-2049
7. Hanson RL, Rong R, Kobes S. **Role of established type 2 diabetes-susceptibility genetic variants in a high prevalence American Indian population**. *Diabetes* (2015) **64** 2646-2657. DOI: 10.2337/db14-1715
8. Knowler WC, Bennett PH, Hamman RF, Miller M. **Diabetes incidence and prevalence in Pima Indians: a 19-fold greater incidence than in Rochester, Minnesota**. *Am J Epidemiol* (1978) **108** 497-505. DOI: 10.1093/oxfordjournals.aje.a112648
9. Kahn R. **Report of the expert committee on the diagnosis and classification of diabetes mellitus**. *Diabetes Care* (1997) **20** 1183-1197. DOI: 10.2337/diacare.20.7.1183
10. Piaggi P, Masindova I, Muller YL. **A genome-wide association study using a custom genotyping array identifies variants in GPR158 associated with reduced energy expenditure in American Indians**. *Diabetes* (2017) **66** 2284-2295. DOI: 10.2337/db16-1565
11. Howie BN, Donnelly P, Marchini J. **A flexible and accurate genotype imputation method for the next generation of genome-wide association studies**. *PLoS Genet* (2009) **5** e1000529. DOI: 10.1371/journal.pgen.1000529
12. Malhotra A, Kobes S, Bogardus C, Knowler WC, Baier LJ, Hanson RL. **Assessing accuracy of genotype imputation in American Indians**. *PLoS One* (2014) **9** e102544. DOI: 10.1371/JOURNAL.PONE.0102544
13. Mahajan A, Taliun D, Thurner M. **Fine-mapping type 2 diabetes loci to single-variant resolution using high-density imputation and islet-specific epigenome maps**. *Nat Genet* (2018) **50** 1505-1513. DOI: 10.1038/s41588-018-0241-6
14. Spracklen CN, Horikoshi M, Kim YJ. **Identification of type 2 diabetes loci in 433,540 East Asian individuals**. *Nature* (2020) **582** 240-245. DOI: 10.1038/s41586-020-2263-3
15. Mahajan A, Spracklen CN, Zhang W. **Multi-ancestry genetic study of type 2 diabetes highlights the power of diverse populations for discovery and translation**. *Nat Genet* (2022) **54** 560-572. DOI: 10.1038/s41588-022-01058-3
16. Vijayakumar P, Nelson RG, Hanson RL, Knowler WC, Sinha M. **HbA1c and the prediction of type 2 diabetes in children and adults**. *Diabetes Care* (2017) **40** 16-21. DOI: 10.2337/dc16-1358
17. Chambers M, Tanamas SK, Clark EJ. **Growth tracking in severely obese or underweight children**. *Pediatrics* (2017) **140** e20172248. DOI: 10.1542/peds.2017-2248
18. Davidson KW, Barry MJ, Mangione CM. **Screening for prediabetes and type 2 diabetes: US preventive services task force recommendation statement**. *JAMA* (2021) **326** 736-743. DOI: 10.1001/JAMA.2021.12531
19. Olaiya MT, Wedekind LE, Hanson RL. **Birthweight and early-onset type 2 diabetes in American Indians: differential effects in adolescents and young adults and additive effects of genotype, BMI and maternal diabetes**. *Diabetologia* (2019) **62** 1628-1637. DOI: 10.1007/s00125-019-4899-9
20. Hanson RL, Guo T, Muller YL. **Strong parent-of-origin effects in the association of KCNQ1 variants with type 2 diabetes in American Indians**. *Diabetes* (2013) **62** 2984-2991. DOI: 10.2337/db12-1767
21. Baier LJ, Muller YL, Remedi MS. **ABCC8 R1420H loss-of-function variant in a southwest American Indian community: Association with increased birth weight and doubled risk of type 2 diabetes**. *Diabetes* (2015) **64** 4322-4332. DOI: 10.2337/db15-0459
22. Harrell FE, Califf RM, Pryor DB, Lee KL, Rosati RA. **Evaluating the yield of medical tests**. *JAMA* (1982) **247** 2543-2546. DOI: 10.1001/JAMA.1982.03320430047030
23. Pencina MJ, D’Agostino RB. **Overall C as a measure of discrimination in survival analysis: Model specific population value and confidence interval estimation**. *Stat Med* (2004) **23** 2109-2123. DOI: 10.1002/sim.1802
24. Pencina MJ, D’Agostino RB, D’Agostino RB, Vasan RS. **Evaluating the added predictive ability of a new marker: from area under the ROC curve to reclassification and beyond**. *Stat Med* (2008) **27** 157-172. DOI: 10.1002/sim.2929
25. Chambless LE, Cummiskey CP, Cui G. **Several methods to assess improvement in risk prediction models: extension to survival analysis**. *Stat Med* (2011) **30** 22-38. DOI: 10.1002/sim.4026
26. Vickers AJ, Elkin EB. **Decision curve analysis: a novel method for evaluating prediction models**. *Med Decis Mak* (2006) **26** 565-574. DOI: 10.1177/0272989X06295361
27. Vickers AJ, Cronin AM, Elkin EB, Gonen M. **Extensions to decision curve analysis, a novel method for evaluating diagnostic tests, prediction models and molecular markers**. *BMC Med Inform Decis Mak* (2008) **8** 53. DOI: 10.1186/1472-6947-8-53
28. Olaiya MT, Knowler WC, Sinha M. **Weight tracking in childhood and adolescence and type 2 diabetes risk**. *Diabetologia* (2020) **63** 1753-1763. DOI: 10.1007/S00125-020-05165-W
29. Vassy JL, Hivert MF, Porneala B. **Polygenic type 2 diabetes prediction at the limit of common variant detection**. *Diabetes* (2014) **63** 2172-2182. DOI: 10.2337/db13-1663
30. Moazzam-Jazi M, Najd Hassan Bonab L, Zahedi AS, Daneshpour MS. **High genetic burden of type 2 diabetes can promote the high prevalence of disease: a longitudinal cohort study in Iran**. *Sci Rep* (2020) **10** 14006. DOI: 10.1038/s41598-020-70725-4
31. Pencina MJ, D’Agostino RB, Pencina KM, Janssens ACJW, Greenland P. **Interpreting incremental value of markers added to risk prediction models**. *Am J Epidemiol* (2012) **176** 473-481. DOI: 10.1093/aje/kws207
32. Lambert SA, Abraham G, Inouye M. **Towards clinical utility of polygenic risk scores**. *Hum Mol Genet* (2019) **28** R133-R142. DOI: 10.1093/HMG/DDZ187
33. Hivert MF, Jablonski KA, Perreault L. **Updated genetic score based on 34 confirmed type 2 diabetes loci is associated with diabetes incidence and regression to normoglycemia in the diabetes prevention program**. *Diabetes* (2011) **60** 1340-1348. DOI: 10.2337/db10-1119
34. Martin AR, Kanai M, Kamatani Y, Okada Y, Neale BM, Daly MJ. **Current clinical use of polygenic scores will risk exacerbating health disparities**. *Nat Genet* (2019) **51** 584. DOI: 10.1038/S41588-019-0379-X
35. Conrad DF, Jakobsson M, Coop G. **A worldwide survey of haplotype variation and linkage disequilibrium in the human genome**. *Nat Genet* (2006) **38** 1251-1260. DOI: 10.1038/ng1911
36. Jiang X, Holmes C, McVean G. **The impact of age on genetic risk for common diseases**. *PLOS Genet* (2021) **17** e1009723. DOI: 10.1371/JOURNAL.PGEN.1009723
37. Hanson RL, Knowler WC. **Analytic strategies to detect linkage to a common disorder with genetically determined age of onset: diabetes mellitus in Pima Indians**. *Genet Epidemiol* (1998) **15** 299-315. DOI: 10.1002/(SICI)1098-2272(1998)15:3
38. Crandall JP, Knowler WC, Kahn SE. **The prevention of type 2 diabetes**. *Nat Clin Pract Endocrinol Metab* (2008) **4** 382-393. DOI: 10.1038/ncpendmet0843
|
---
title: Skeletal muscle and intermuscular adipose tissue gene expression profiling
identifies new biomarkers with prognostic significance for insulin resistance progression
and intervention response
authors:
- Dominik Lutter
- Stephan Sachs
- Marc Walter
- Anna Kerege
- Leigh Perreault
- Darcy E. Kahn
- Amare D. Wolide
- Maximilian Kleinert
- Bryan C. Bergman
- Susanna M. Hofmann
journal: Diabetologia
year: 2023
pmcid: PMC10036433
doi: 10.1007/s00125-023-05874-y
license: CC BY 4.0
---
# Skeletal muscle and intermuscular adipose tissue gene expression profiling identifies new biomarkers with prognostic significance for insulin resistance progression and intervention response
## Abstract
### Aims/hypothesis
Although insulin resistance often leads to type 2 diabetes mellitus, its early stages are often unrecognised, thus reducing the probability of successful prevention and intervention. Moreover, treatment efficacy is affected by the genetics of the individual. We used gene expression profiles from a cross-sectional study to identify potential candidate genes for the prediction of diabetes risk and intervention response.
### Methods
Using a multivariate regression model, we linked gene expression profiles of human skeletal muscle and intermuscular adipose tissue (IMAT) to fasting glucose levels and glucose infusion rate. Based on the expression patterns of the top predictive genes, we characterised and compared individual gene expression with clinical classifications using k-nearest neighbour clustering. The predictive potential of the candidate genes identified was validated using muscle gene expression data from a longitudinal intervention study.
### Results
We found that genes with a strong association with clinical measures clustered into three distinct expression patterns. Their predictive values for insulin resistance varied substantially between skeletal muscle and IMAT. Moreover, we discovered that individual gene expression-based classifications may differ from classifications based predominantly on clinical variables, indicating that participant stratification may be imprecise if only clinical variables are used for classification. Of the 15 top candidate genes, ST3GAL2, AASS, ARF1 and the transcription factor SIN3A are novel candidates for predicting a refined diabetes risk and intervention response.
### Conclusion/interpretation
Our results confirm that disease progression and successful intervention depend on individual gene expression states. We anticipate that our findings may lead to a better understanding and prediction of individual diabetes risk and may help to develop individualised intervention strategies.
### Supplementary Information
The online version of this article (10.1007/s00125-023-05874-y) contains peer-reviewed but unedited supplementary material.
## Introduction
Obesity is a frequent precondition for the development of chronic metabolic diseases such as insulin resistance and type 2 diabetes. Based on the recently published results from the 2017–2018 National Health and Nutrition Examination Survey (NHANES), $42.5\%$ of US adults are currently obese and are thus at high risk for developing type 2 diabetes and its complications [1]. Moreover, the IDF predicts that there will be a $51\%$ increase in the number of individuals with diabetes worldwide by 2045, from 463 million to 700 million, and indicates that one in two adults with diabetes remain undiagnosed at presence [2]. Although the current assessment of diabetes and impaired glucose tolerance is based on purely glycaemic indicators, it is important to emphasise that the risk for developing diabetes is also dependent on age, sex, fat tissue distribution, genetics and gene expression, ethnicity and environmental characteristics. Depending on these individual risk factors and on the inclusion criteria for the cohorts studied, wide heterogeneity in the progression from impaired glucose tolerance to diabetes has been observed. Emerging evidence from a population-based study with 381,363 participants indicates that even people referred to as having ‘metabolically healthy obesity’ are at a substantially higher risk of developing diabetes and its complications [3]. Although medical interventions or changes in lifestyle (diet, exercise) reduce the risk of severe complications, evidence is emerging in population-based cohorts that treatment efficacy also depends on individual genetics [4–7]. This means that patients treated with glucose-lowering interventions will vary in their response, with some gaining a considerable benefit, others seeing no benefit and some experiencing limiting side effects. Taking all of the evidence together, it is becoming increasingly clear that the current clinical standards for defining the metabolic health status of an individual are not adequate and that new strategies for the effective prevention of diabetes are critically important to reduce the burden of this disease. A deeper understanding of the individual features and precise phenotyping of impaired glucose tolerance may improve stratification of disease risk and optimise the benefit/risk ratio and cost-effectiveness of therapeutic approaches for the prevention and treatment of type 2 diabetes.
Given that skeletal muscle is responsible for more than $85\%$ of insulin-stimulated whole-body glucose disposal [8], and that any dysfunction impairing glucose metabolism in this tissue will affect whole-body glucose homeostasis, ultimately contributing to the development of diabetes [9], mechanistic studies mostly focus on this tissue in attempts to elucidate mechanisms involved in metabolic adaptation and its regulation. More recently, evidence has pointed to intermuscular adipose tissue (IMAT) accumulation as another local regulator of muscular insulin resistance and the progression to diabetes [10, 11]. In this study we hypothesised that tissue-specific gene expression profiling of skeletal muscle and/or IMAT could achieve a more specific and detailed characterisation and classification of individual physiological states than circulating variables alone. We further presumed that the expression of individual genes might [1] allow the prediction of individual disease-related states; [2] identify individuals with a high or low risk for diabetes; and [3] enable the potential response of a given individual to a specific treatment strategy to be predicted.
To this end, we aimed to investigate dependencies between gene expression in skeletal muscle and/or IMAT and clinical diabetes markers from individuals with obesity, with and without type 2 diabetes. We used multivariate regression to model the tissue-specific gene expression impact on the two key insulin resistance markers, glucose infusion rate (GIR) during a hyperinsulinaemic–euglycaemic clamp and fasting glucose (FG). We used a clustering approach to compare states (obesity and type 2 diabetes) defined by metabolic-related gene expression patterns with binary clinical classifications. Finally, we tested selected genes for their potential to predict individual intervention response based on an independent lifestyle and exercise intervention study.
## Human transcriptional profiling dataset
Human muscle and IMAT transcriptional profiles were obtained from a cross-sectional study previously reported by Sachs et al [12] (Fig. 1). To identify features suitable for characterisation of individual prediabetic states and potentially predictive for disease progression we selected all 16 participants with obesity and type 2 diabetes for whom paired samples were available. All participants were clinically characterised by determining age, BMI, body weight, FG, fasting insulin, fat-free mass, glucagon, height, insulin sensitivity via GIR during a hyperinsulinaemic–euglycaemic clamp and relative fat mass (Table 1). In total, 13 participants were of white ethnicity and three were of Hispanic ethnicity. Fig. 1Study designs. Data for cross-sectional transcriptional profiling were obtained from a cohort of individuals with obesity and type 2 diabetes (T2D). Intervention study data were collected from individuals with obesity, with and without impaired glucose tolerance and impaired FG, pre and post exercise intervention. The design of the metabolic profiling, including the hyperinsulinaemic–euglycaemic clamp, was identical for both studiesTable 1Participant demographics: human transcriptional profiling ($$n = 16$$)Clinical variableObesityType 2 diabetesNo. of participants106Age (years)40.5 ± 2.445.7 ± 2.5BMI (kg/m2)36.7 ± 1.634.8 ± 1.7BW (kg)116.8 ± 7.4101.4 ± 6.4FFM (kg)71.9 ± 4.863.0 ± 4.5FG (mmol/l)4.9 ± 0.210.2 ± 0.9***Fasting insulin (pmol/l)122.2 ± 22.9173.6 ± 27.8Glucagon (ng/l)70.6 ± 9.787.7 ± 10.1Height (m)1.8 ± 0.041.7 ± 0.03Insulin sensitivity/GIR (mg kg–1 min–1)5.1 ± 1.01.7 ± 0.7*RelFat (%)38.2 ± 2.637.8 ± 2.7Data are mean ± SEMThe human transcriptional profiling dataset included eight participants of white ethnicity and two participants of Hispanic ethnicity with obesity, and five participants of white ethnicity and one of Hispanic ethnicity with type 2 diabetes. Insulin sensitivity data were normalised to kg of BW*$p \leq 0.05$ and ***$p \leq 0.001$ for difference between obesity and type 2 diabetes (one-way ANOVA)
## Longitudinal intervention dataset
Seventeen individuals with obesity (BMI 30–40 kg/m2), with and without impaired glucose tolerance and impaired FG, were recruited for this study from the local Denver area. Impaired FG was defined as FG between 5.6 and 7 mmol/l, with postprandial glucose <7.8 mmol/l 2 h after a 75g OGTT. Impaired glucose tolerance was defined as normal FG <5.6 mmol/l, with postprandial glucose >7.8 mmol/l 2 h after a 75g OGTT. A list of participant exclusion criteria is available in the electronic supplementary material (ESM; see ‘Longitudinal intervention study’). Participants were asked to refrain from planned physical activity for 48 h before the first and second metabolic studies and were given a standardised diet for 7 days prior to each study (Fig. 1). After overnight fasting a basal muscle biopsy was taken; this was followed by metabolic profiling including a 3 h hyperinsulinaemic–euglycaemic clamp. After the first metabolic study, participants entered a 12 week supervised weight loss and exercise training intervention. The weight loss intervention consisted of a medically supervised low energy diet comprising a meal replacement product that can be consumed as a liquid or made into a variety of food forms (Health & Nutrition Technology, Carmel, CA, USA). Participants were provided with powdered Health *One formula* and instructed to consume five portions per day, providing 3724 kJ/day (890 kcal/day), 75 g of protein, 15 g of fat and 110 g of carbohydrate and $100\%$ of the daily recommended intake of all vitamins, minerals and micronutrients. The exercise training consisted of four individually supervised sessions per week of whole-body aerobic activity. During the first 2–3 weeks of training, the exercise duration and intensity were gradually increased to 45 min at 80–$85\%$ of the maximal heart rate. After the 3 month intervention, participants transitioned to a 2 week weight maintenance diet. Participants continued to exercise during the weight stabilisation period. After completing the intervention and 2 week weight maintenance period, participants then repeated the metabolic study with the muscle biopsy and 3 h hyperinsulinaemic–euglycaemic clamp (Table 2). Metabolic studies consisted of measurements of age, BMI, BW, FG, fasting insulin, FFM, glucagon, insulin sensitivity via GIR during the hyperinsulinaemic–euglycaemic clamp, relative fat mass and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \dot{V}{\mathrm{O}}_{2\mathrm{peak}} $$\end{document}V˙O2peak (Table 2). See ESM, ‘Longitudinal intervention study’, for further details. Table 2Participant demographics: longitudinal intervention study ($$n = 17$$)Clinical variablePre interventionPost interventionAge (years)46.5 ± 2.2BMI (kg/m2)34.7 ± 1.030.7 ± 1.0***BW (kg)96.9 ± 2.785.9 ± 2.6***FFM (kg)56.7 ± 1.852.9 ± 1.5***FG (mmol/l)5.2 ± 0.15.0 ± 0.1Fasting insulin (pmol/l)110.4 ± 9.778.5 ± 9.7***Glucagon (ng/l)82.2 ± 4.071.4 ± 3.7***Insulin sensitivity/GIR (mg kg–1 min–1)3.5 ± 0.45.4 ± 0.5***RelFat (%)41.3 ± 1.538.0 ± 1.8***\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \dot{V}{\mathrm{O}}_{2\mathrm{peak}} $$\end{document}V˙O2peak (l/min)2.2 ± 0.12.5 ± 0.1*Relative \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \dot{V}{\mathrm{O}}_{2\mathrm{peak}} $$\end{document}V˙O2peak (ml/kg min–1)23.2 ± 1.029.6 ± 1.4*Data are mean ± SEMThe study included 12 individuals of white, three of Hispanic, one of East Indian and one of African American ethnicity. Ethnicity was not taken into account in statistical analyses. Insulin sensitivity data were normalised to kg of BW*$p \leq 0.05$ and ***$p \leq 0.001$ for difference from pre intervention (paired t test) Changes in metabolic variables pre to post intervention were estimated using a paired t test. Pre- and post-intervention biopsies were used for gene expression analysis. Because insufficient RNA was isolated from the IMAT samples or the RNA integrity number did not match quality requirements for gene expression analysis, we removed all IMAT samples and used only the remaining muscle samples for gene expression analysis. Pre- to post-differential gene expression was estimated using one-way ANOVA. See ESM, ‘Gene expression analysis’, for further information.
## Models and statistics
To estimate the impact of gene expression on FG and GIR we used linear multivariate regression models. Thus, we created one predictive model for each gene, simultaneously predicting clinical variables based on gene expression in skeletal muscle and IMAT. Models can be formalised in matrix notation as Y = βXj + ε, where Y is a matrix of the sampled response variables GIR (g) and FG (f) and X is a matrix of the predictor values, the expression of gene j in muscle (m) and IMAT (i). β forms the 2 × 2 matrix of the four estimated regression coefficients βmg, βmf, βig and βif describing the four relationships between tissue-specific gene expression and response variables. The residues or errors are formed in ε. Our approach can be interpreted as a mixture model that allows us to jointly estimate these four coefficients in one model to predict insulin sensitivity and glucose homeostasis from gene expression in muscle and IMAT. Genes that contributed the most to insulin sensitivity and glucose homeostasis were then scored based on the log-likelihood and negative log-likelihood of the single regression models. Subsequently, hierarchical clustering was performed to group the selected genes into distinct clusters with similar expression profiles. These clusters were then used to compare gene expression profiles with diagnosed disease states by generating participant k-nearest neighbour (kNN)-networks with $k = 3$ nearest neighbours, using the Euclidean distance metric. The predictive classification score was calculated from the maximum ratio of each participant’s direct neighbour’s clinical classification (percentage with obesity vs percentage with type 2 diabetes). Analysis was carried out using Matlab R2020a (https://www.mathworks.com). See ESM for further information.
## Transcriptomic profiling
Skeletal muscle and IMAT samples from the cross-sectional study were used for transcriptional profiling. See ESM for further information.
## Gene expression analysis
Quantitative reverse transcription PCR (qRT-PCR) was used to determine relative mRNA expression levels. See ESM for further information.
## Multivariate regression unravels tissue-specific gene expression patterns correlating with insulin resistance
We first compared participants’ demographic and metabolic variables. As expected, we found sex-specific differences in RelFat ($p \leq 0.05$), FFM ($p \leq 0.001$) and height ($p \leq 0.01$) (ESM Fig. 1a). We also found significant differences in GIR ($p \leq 0.05$) and FG ($p \leq 0.001$) between participants with obesity and those with type 2 diabetes (Table 1, Fig. 2c,d). Additionally, as expected, we found that low GIR values correlate with high FG levels (ESM Fig. 1b). As shown in Fig. 2c,d, individuals with obesity (BMI >30 kg/m2 and FG <7 mmol/l) exhibited a wide range of insulin sensitivities (GIR 0.8–11.1 mg kg–1 min–1). Some individuals with more severe insulin resistance (GIR <3 mg kg–1 min–1) were still able to maintain FG levels at <7 mmol/l. In contrast, some individuals with diabetes exhibited a better GIR, with levels up to 4 mg kg–1 min–1. To explore whether transcriptional changes in muscle and IMAT at the transition from obesity with compensated insulin resistance (normoglycaemia) to type 2 diabetes (hyperglycaemia) reflect the inconsistency between FG and GIR, we performed a multivariate regression analysis to identify genes whose expression had a strong link to insulin resistance and glucose homeostasis. Fig. 2Multivariate regression analysis unravels tissue-specific gene expression patterns correlating with insulin resistance. ( a–d) Boxplots comparing FG (mmol/l) and insulin sensitivity (GIR, mg kg–1 min–1) distributions between participants of different sexes (a, b) and classifications (c, d). F, female; M, male; OB, obesity; T2D, type 2 diabetes. Red circles indicate outliers. * $p \leq 0.05$ and ***$p \leq 0.001$ (one-way ANOVA). ( e) Heatmaps of muscle and IMAT genes correlating with insulin resistance identified by multivariate regression. Colours in the dendrogram refer to clusters 1 (blue), 2 (red) and 3 (yellow). The colours in the bars below the heatmaps indicate individual disease classification, GIR, FG levels and sex. Vertical colour bars show the four estimated regression coefficients for each gene, indicating the four relationships between tissues and response variables (GIR and FG). ( f, g) Scatterplots comparing gene-specific β coefficients for GIR and FG for muscle (f) and IMAT (g). The colours of the dots refer to gene cluster assignment After multivariate regression analysis we selected the top 59 genes (ESM Fig. 1c) contributing to GIR and FG and used k-means to cluster them into three clusters of 14 (cluster 1), 23 (cluster 2) and 22 (cluster 3) genes with distinct expression patterns in muscle and IMAT (Fig. 2e, ESM Fig. 1d, ESM Table 1). We compared the corresponding β values of the gene clusters and observed three distinct patterns (Fig. 2f,g, ESM Fig. 1e). Cluster 1 included genes whose expression was positively associated with GIR and negatively associated with FG in both muscle and IMAT, thus correlating with healthy glucose metabolism. Cluster 2 contained genes whose expression was positively correlated with GIR and negatively correlated with FG in muscle, similar to cluster 1, with opposing associations for most of the observed genes in IMAT. Cluster 3, on the contrary, contained genes whose expression was negatively correlated with GIR and positively correlated with FG in muscle, with no effect observed in IMAT.
Our results suggest that these largely different gene expression profiles in muscle and IMAT are associated with varying impacts on glucose metabolism. In particular, expression of PDK4, which has been linked to diabetes and glucose metabolism previously [13], shows a high correlation with GIR and FG in muscle but almost none in IMAT (Fig. 2f,g). We also identified genes with opposing effects on glucose metabolism in muscle compared with IMAT, such as UBTD1 and ST3GAL2. *Both* genes show strong positive β coefficients for FG and negative β coefficients for GIR in muscle whereas in IMAT we observed negative β coefficients for FG and positive β coefficients for GIR (Fig. 2f,g). *Both* genes were associated with cluster 3. In contrast, NAPB from cluster 2 shows the opposite associations. In a third observation we found genes, here represented by SIN3A, that seem to have a relatively high predictive value for FG but a low or no predictive value for GIR in muscle but completely opposing values in IMAT (high value for GIR, low value for FG).
Taken together, the genes identified in all three clusters show a striking association between expression in muscle and GIR and FG, while only genes in cluster 1 show an association between IMAT expression and glucose homeostasis and insulin sensitivity (Fig. 2f,g, ESM Fig. 1e). These results suggest that muscle gene expression profiles may allow for a more specific and detailed characterisation and classification of individual physiological states than serum-based physiological variables alone.
## Gene expression-based classification enables a refined view of the individual physiological state of individuals with obesity
To test our hypothesis that gene expression patterns are superior to conventional clinical markers for categorising individual insulin resistance states, we performed a kNN classification for each tissue and gene cluster based solely on expression profiles. Thus, we generated six nearest neighbour networks (NNNs) representing expression-based participant similarities for all 16 participants (Fig. 3a,b). Based on direct network neighbours we then calculated a predictive classification score for each individual (Fig. 3a,b, ESM Fig. 2). For muscle, we found a non-unique classification over all three NNNs for five of the 16 participants (two with obesity: Pb029, Pb043; three with type 2 diabetes: Pb034, PB053, Pb032). After averaging over the three clusters, one participant with obesity was classified as having type 2 diabetes (Pb043) and two participants with type 2 diabetes were classified with obesity (Pb034, Pb053). For the IMAT-derived NNNs we found a predicted classification that differed from the clinical classification for seven participants (three with obesity: Pb048, Pb028, Pb043; four with type 2 diabetes: Pb034, Pb053, Pb033, Pb032). Averaging over the three clusters resulted in a different classification for two participants compared with the clinical classification: Pb043 (obesity) and Pb033 (type 2 diabetes). Averaging over both tissues, we identified two participants with a divergent classification: Pb043 and Pb053. Overall, the classification of obesity was more consistent than the classification of type 2 diabetes, with only two participants consistently classified as having type 2 diabetes (Pb036, Pb052). Fig. 3Gene expression-based participant classification reveals a refined view of physiological state. ( a, b) kNN-networks for the three clusters in muscle (a) and IMAT (b). Nodes refer to individual participants. Node shape refers to the assigned clinical classification: obesity (OB; diamond) or type 2 diabetes (T2D; circle). Node colour refers to the estimated disease state based on connected individuals. ( c–e) Scatter plots displaying the clinical variables FG and GIR for all individuals. Node shape refers to the assigned clinical classification: OB (diamond) or T2D (circle). Node colour refers to the estimated disease state across all three gene clusters for muscle (c), IMAT (d) and both tissues combined (e) When comparing the NNN-based classification with metabolic variables, we found that an increase in the probability of developing type 2 diabetes correlated with decreasing GIR for participants with obesity in both muscle tissue and IMAT (Fig. 3c–e). In contrast, for hyperglycaemic participants clinically classified as having type 2 diabetes (FG >7 mmol/l), the NNN-based classification did not correlate with either GIR or FG in either tissue. These results suggest that, beyond a binary clinical classification of type 2 diabetes and normoglycaemia as FG >7 mmol/l and FG ≤7 mmol/l, respectively, there is a continuous development from insulin resistance to diabetes that follows an individual trajectory, which means that there is diagnostic potential to predict an individual’s risk for diabetes or their potential to respond to interventions.
## ST3GAL2, SIN3A, ARF1 and AASS mRNA levels in muscle tissue predict intervention response
To evaluate whether gene expression profiles within muscle and/or IMAT define individual health states with predictive potential for disease progression or modulation of insulin sensitivity, we analysed 17 individuals with obesity, with and without impaired glucose tolerance, undergoing a combined weight loss and exercise training intervention (Table 2). Clinical variables such as GIR, FG, BW, RelFat, FFM and BMI were measured pre and post intervention. Almost all individuals showed an increase in GIR ($$p \leq 2.2$$ × 10–5) after the intervention and a decrease in BMI ($$p \leq 2.3$$ × 10–8), BW ($$p \leq 6.1$$ × 10–8), RelFat ($$p \leq 2.7$$ × 10–6) and FFM ($$p \leq 1.4$$ × 10–6). A change in FG levels post intervention was not observed ($$p \leq 0.12$$) (Fig. 4a–f). However, when correlating the relative pre/post change (Δ%) between all clinical variables, we found that a change in GIR was significantly correlated only with a change in BW (ESM Fig. 3). A decrease in FG, in turn, was significantly correlated with a relative decrease in BMI, FFM and BW. Fig. 4ST3GAL2, SIN3A, ARF1 and AASS mRNA expression in muscle predicts intervention response. ( a–f) Changes in clinical variables from pre to post intervention: (a) BMI, (b) BW, (c) FFM, (d) FG, (e) GIR and (f) RelFat. *** $p \leq 0.001$ (paired t test). ( g) Relative mRNA levels (to the reference gene TBP) of selected genes pre and post intervention, shown as log2 (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ {2}^{-\Delta \Delta {\mathrm{C}}_{\mathrm{t}}} $$\end{document}2−ΔΔCt). * $p \leq 0.05$, **$p \leq 0.01$ and ***$p \leq 0.001$ (one-way ANOVA). Error bars denote SEMs. ( h–m) Correlation volcano plots of pre-intervention mRNA expression and relative change in clinical variables between pre and post intervention: (h) BMI (kg/km2), (i) BW (kg), (j) FFM (kg), (k) FG (mmol/l), (l) GIR (mg kg–1 min–1) and (m) RelFat (%). Significantly correlated mRNAs are shown in orange. FC, fold change To test if these physiological changes are linked to individual gene expression in muscle, biopsies taken before and after the intervention were used for RNA expression analysis. Six pre- and eight post-intervention muscle samples did not meet quality requirements and were removed from subsequent analyses. We combined various criteria to select 15 candidate genes for validation from the three clusters of 59 genes initially identified in our first patient cohort. Gene expression in muscle tissue was measured using qRT-PCR (ESM Table 2) in this second independent intervention trial. Among the genes selected, SIN3A, UBTD1, ST3GAL2 and NAPB showed notable β value profiles (Fig. 2f,g). AASS, DBNDD1, PDK4, PIGA, POLR3GL, SNAP23, SPCS2, SSU72 and UBTD1 could be linked to diabetes-associated SNPs identified in the Type 2 Diabetes Knowledge Portal [14] and ARF1, BCAT2 and LDHD could be linked to skeletal muscle lipid and glucose metabolism and insulin resistance [15–17]. PDK4 was included as a well-described marker of muscle insulin resistance and as a potential therapeutic target [18]. Of these 15 genes, five (LDHD, ARF1 NAPB, POLR3GL and SNAP23) showed a significant change in expression between pre and post intervention (Fig. 4g). As we hypothesised that distinct gene expression states may relate to individual disease states, we tested selected genes for their predictive potential for individual intervention response. To this end, we correlated individual pre-intervention gene expression with the relative change (pre to post) in the clinical variables BMI, BW, FFM, FG, GIR and RelFat (Fig. 4a–f). ΔGIR and ΔRelFat could not be significantly correlated with any of the genes tested. A change in the remaining variables could be significantly predicted by the genes ST3GAL2 (FG, BW, FFM and BMI), SIN3A (FG, FFM and BMI), ARF1 (FG) and AASS (FFM) (Fig. 4h–m, ESM Fig. 4, ESM Table 3). In contrast to the five genes that showed a change in expression post intervention (LDHD, ARF1, NAPB, POLL3GL and SNAP23), none of the four genes identified with predictive character appeared to be differentially expressed between pre and post intervention (Fig. 4g). Together, these findings indicate that individual susceptibility to exercise intervention for the improvement of glucose homeostasis is independent of the individual clinical variables, but correlates with individual gene expression profiles prior to intervention. We next compared these four identified genes with PDK4, a well-described muscle marker for insulin resistance. To our surprise, PDK4 was not significantly associated with any intervention-induced change in metabolic variables (ESM Fig. 4).
Next, we found that low expression levels of three of the four genes (AASS, ARF1 and SIN3A) was associated with a good health prognosis. In particular, ARF1 showed a significant decrease in expression on exercise intervention. In turn, ST3GAL2 was the only gene for which increased expression levels in muscle tissue increased the likelihood of an effective intervention. In summary, within this independent intervention trial we were able to validate our hypothesis that muscle gene expression profiles have predictive potential for individual insulin resistance states.
## Discussion
In this study we showed that human transcriptional profiles of skeletal muscle and IMAT from individuals with obesity, with and without type 2 diabetes, are differentially coupled to insulin resistance and glucose homeostasis. We identified predictive gene clusters that mirror gene expression states reflecting a continuous progression from early insulin resistance to type 2 diabetes according to individual traits. From a subset, the genes AASS, ARF1, SIN3A and ST3GAL2 predicted individual improvement of impaired glucose metabolism by means of an exercise and lifestyle intervention.
We started our analysis with the observation that there is overlap of GIR measurements between obesity and type 2 diabetes. We hypothesised that the binary clinical classification of type 2 diabetes does not reflect individual underlying gene expression states and that specific gene expression patterns in skeletal muscle and/or IMAT may have the potential to identify and predict individuals with a high or low risk of developing diabetes or to predict individual susceptibility to interventions.
Our multivariate regression analysis revealed a strong association of all three gene clusters in muscle with GIR and FG, while only the genes in cluster 1 in IMAT were associated with glucose homeostasis and insulin resistance. Together with the observation that β coefficients estimates in cluster 2 showed opposing associations with FG and GIR in muscle compared with IMAT, we concluded that IMAT and muscle contribute differentially to glucose metabolism. However, there is higher variance in IMAT gene expression [12], which may mean that any correlation is harder to detect. The increased variability in IMAT gene expression may arise from technical difficulties in dissecting IMAT from muscle, resulting in less material for RNA extraction, or the higher heterogeneity of IMAT itself, which is composed of multiple cell types such as pre-adipocytes, adipocytes, adipocyte-like cells, myoblasts and stromal and vascular cells.
Participant classification based on kNN-networks revealed that insulin sensitivity could be accurately predicted for individuals without diabetes from gene expression patterns, whereas gene expression patterns for participants with hyperglycaemia scored differently from the clinical classification in several cases. The latter observation is consistent with the idea that hyperglycaemia occurs in a late state of disease progression as a consequence of pancreatic beta cell failure. In addition, insulin sensitivity is associated with multiple organ malfunction and, in particular, skeletal muscle is the primary organ for glucose uptake [19]. Although GIR measured using a hyperinsulinaemic–euglycaemic clamp is still the gold standard for directly measuring insulin resistance, it is highly invasive and time-consuming and has very limited predictive potential. FG levels by themselves are unlikely to identify individuals with obesity and impaired glucose tolerance; rather, they identify individuals with severe insulin resistance with an increased risk for irreversible damage of tissues and organs [20]. We thus conclude that both GIR and FG levels are not suitable for a reliable early diagnosis and prognosis of disease progression. In contrast, the gene expression profiles identified here, which represent a muscle-specific state of individual insulin resistance, have predictive potential for the characterisation of individual insulin sensitivity.
This predictive potential was tested on muscle tissue from an additional independent cohort of 17 individuals with impaired glucose metabolism undergoing a 12 week combined weight loss and exercise intervention. By correlating the pre-intervention expression levels of our candidate genes with the relative change in clinical variables post intervention we identified four genes with significant predictive value: AASS, ARF1, SIN3A and ST3GAL2. Lower levels of expression of AASS, ARF1 and SIN3A indicated a positive prognosis. AASS encodes the enzyme aminoadipate-semialdehyde synthase, which is involved in mammalian lysine degradation and in hyperlysinaemia [21], but which has not yet been characterised in the context of impaired glucose metabolism, insulin resistance or diabetes. Beside its predictive potential we also found that expression of ARF1, which encodes ADP ribosylation factor 1, was significantly reduced after the intervention. ADP ribosylation factor 1 was recently linked to rapamycin (mTOR) complex 2 (mTORC2) [22], which has been shown to be involved in exercise-dependent regulation of muscle glucose uptake in mice [23]. SIN3 transcription regulator family member A, encoded by SIN3A, has been linked to glucose metabolism in murine beta cells [24]. It has further been shown that SIN3A is an insulin-sensitive forkhead box protein O1 (FOXO1) corepressor of glucokinase in murine liver [25]. SIN3A was also shown to negatively regulate insulin receptor (Insr/INSR) mRNA in mice and human muscle [26]. Finally, ST3GAL2, which encodes ST3 beta-galactoside alpha-2,3-sialyltransferase 2, was the only gene identified to positively predict exercise response at high expression levels. Mice lacking this protein have been shown to develop obesity and insulin resistance after 7–9 months of age [27]. In summary, three of the four predictive genes that we identified have already been linked to insulin resistance and diabetes but their predictive potential has not yet been explored.
In conclusion, we identified novel markers for predicting impaired insulin sensitivity in human muscle and found four markers that predict individual exercise intervention responses in participants with diabetes. These findings may help to classify and characterise individuals with obesity, impaired glucose tolerance or diabetes more precisely than using state-of-the-art variables such as GIR and FG alone. Additionally, we anticipate that these findings may also help to develop precise and individualised intervention strategies for patients at risk of obesity and type 2 diabetes.
## Supplementary information
ESM 1(PDF 2365 kb)
## Authors’ relationships and activities
SS is an employee of Cellarity and has stakeholder interests. The present work was carried out as an employee of the Helmholtz Zentrum München, HMGU.
## Contribution statement
DL, BCB and SMH conceived the research question and designed and planned the study. DL, SS, MW, AK, LP, DEK, ADW and MK contributed to the acquisition of data. DL analysed the data. DL, BCB and SMH interpreted the results and wrote the manuscript. All authors contributed to revision of the manuscript and approved the final manuscript prior to submission. DL is responsible for the integrity of the work as a whole.
## References
1. 1.Fryar CD, Carroll MD, Afful J (2020) Prevalence of overweight, obesity, and severe obesity among adults aged 20 and over: United States, 1960–1962 through 2017–2018. NCHS Health E-Stats. Available from https://www.cdc.gov/nchs/products/. Accessed Mar 2021
2. 2.IDFIDF diabetes atlas20199Brussels, BelgiumIDF. *IDF diabetes atlas* (2019.0)
3. Zhou Z, Macpherson J, Gray SR. **Are people with metabolically healthy obesity really healthy? A prospective cohort study of 381,363 UK Biobank participants**. *Diabetologia* (2021.0) **64** 1963-1970. DOI: 10.1007/s00125-021-05484-6
4. Goni L, Qi L, Cuervo M. **Effect of the interaction between diet composition and the PPM1K genetic variant on insulin resistance and beta cell function markers during weight loss: results from the Nutrient Gene Interactions in Human Obesity: implications for dietary guidelines (NUGENOB) randomized trial**. *Am J Clin Nutr* (2017.0) **106** 902-908. DOI: 10.3945/ajcn.117.156281
5. Huls A, Wright MN, Bogl LH. **Polygenic risk for obesity and its interaction with lifestyle and sociodemographic factors in European children and adolescents**. *Int J Obes (Lond)* (2021.0) **45** 1321-1330. DOI: 10.1038/s41366-021-00795-5
6. Kriska AM, Rockette-Wagner B, Edelstein SL. **The impact of physical activity on the prevention of type 2 diabetes: evidence and lessons learned from the Diabetes Prevention Program, a long-standing clinical trial incorporating subjective and objective activity measures**. *Diabetes Care* (2021.0) **44** 43-49. DOI: 10.2337/dc20-1129
7. Pearson ER. **Dorothy Hodgkin lecture 2021: drugs, genes and diabetes**. *Diabet Med* (2021.0) **38** e14726. DOI: 10.1111/dme.14726
8. DeFronzo RA, Ferrannini E, Sato Y, Felig P, Wahren J. **Synergistic interaction between exercise and insulin on peripheral glucose uptake**. *J Clin Investig* (1981.0) **68** 1468-1474. DOI: 10.1172/JCI110399
9. Garneau L, Aguer C. **Role of myokines in the development of skeletal muscle insulin resistance and related metabolic defects in type 2 diabetes**. *Diabetes Metab* (2019.0) **45** 505-516. DOI: 10.1016/j.diabet.2019.02.006
10. Correa-de-Araujo R, Addison O, Miljkovic I. **Myosteatosis in the context of skeletal muscle function deficit: an interdisciplinary workshop at the National Institute on Aging**. *Front Physiol* (2020.0) **11** 963. DOI: 10.3389/fphys.2020.00963
11. Boettcher M, Machann J, Stefan N. **Intermuscular adipose tissue (IMAT): association with other adipose tissue compartments and insulin sensitivity**. *J Magn Reson Imaging* (2009.0) **29** 1340-1345. DOI: 10.1002/jmri.21754
12. Sachs S, Zarini S, Kahn DE. **Intermuscular adipose tissue directly modulates skeletal muscle insulin sensitivity in humans**. *Am J Physiol Endocrinol Metab* (2019.0) **316** E866-E879. DOI: 10.1152/ajpendo.00243.2018
13. Majer M, Popov KM, Harris RA, Bogardus C, Prochazka M. **Insulin downregulates pyruvate dehydrogenase kinase (PDK) mRNA: potential mechanism contributing to increased lipid oxidation in insulin-resistant subjects**. *Mol Genet Metab* (1998.0) **65** 181-186. DOI: 10.1006/mgme.1998.2748
14. 14.T2DKP (2018) Type 2 Diabetes Knowledge Portal. Available from http://www.type2diabetesgenetics.org. Accessed Feb 2021
15. Covington JD, Galgani JE, Moro C. **Skeletal muscle perilipin 3 and coatomer proteins are increased following exercise and are associated with fat oxidation**. *PLoS One* (2014.0) **9** e91675. DOI: 10.1371/journal.pone.0091675
16. Zhao X, Han Q, Liu Y, Sun C, Gang X, Wang G. **The relationship between branched-chain amino acid related metabolomic signature and insulin resistance: a systematic review**. *J Diabetes Res* (2016.0) **2016** 2794591. DOI: 10.1155/2016/2794591
17. Abdelmoez AM, Sardon Puig L, Smith JAB. **Comparative profiling of skeletal muscle models reveals heterogeneity of transcriptome and metabolism**. *Am J Physiol Cell Physiol* (2020.0) **318** C615-C626. DOI: 10.1152/ajpcell.00540.2019
18. Jeon JH, Thoudam T, Choi EJ, Kim MJ, Harris RA, Lee IK. **Loss of metabolic flexibility as a result of overexpression of pyruvate dehydrogenase kinases in muscle, liver and the immune system: Therapeutic targets in metabolic diseases**. *J Diabetes Investig* (2021.0) **12** 21-31. DOI: 10.1111/jdi.13345
19. Goldstein BJ. **Insulin resistance as the core defect in type 2 diabetes mellitus**. *Am J Cardiol* (2002.0) **90** 3G-10G. DOI: 10.1016/s0002-9149(02)02553-5
20. Stratton IM, Adler AI, Neil HA. **Association of glycaemia with macrovascular and microvascular complications of type 2 diabetes (UKPDS 35): prospective observational study**. *BMJ* (2000.0) **321** 405-412. DOI: 10.1136/bmj.321.7258.405
21. Sacksteder KA, Biery BJ, Morrell JC. **Identification of the alpha-aminoadipic semialdehyde synthase gene, which is defective in familial hyperlysinemia**. *Am J Hum Genet* (2000.0) **66** 1736-1743. DOI: 10.1086/302919
22. Luciano AK, Korobkina ED, Lyons SP. **Proximity labeling of endogenous RICTOR identifies mTOR complex 2 regulation by ADP ribosylation factor ARF1**. *J Biol Chem* (2022.0) **298** 102379. DOI: 10.1016/j.jbc.2022.102379
23. Kleinert M, Parker BL, Fritzen AM. **Mammalian target of rapamycin complex 2 regulates muscle glucose uptake during exercise in mice**. *J Physiol* (2017.0) **595** 4845-4855. DOI: 10.1113/JP274203
24. Yang X, Graff SM, Heiser CN. **Coregulator Sin3a promotes postnatal murine beta-cell fitness by regulating genes in Ca(2+) homeostasis, cell survival, vesicle biosynthesis, glucose metabolism, and stress response**. *Diabetes* (2020.0) **69** 1219-1231. DOI: 10.2337/db19-0721
25. 25.Langlet F, Haeusler RA, Linden D et al (2017) Selective inhibition of FOXO1 activator/repressor balance modulates hepatic glucose handling. Cell 171(4):824–835 e818. 10.1016/j.cell.2017.09.045
26. Cen HH, Hussein B, Botezelli JD. **Human and mouse muscle transcriptomic analyses identify insulin receptor mRNA downregulation in hyperinsulinemia-associated insulin resistance**. *FASEB J* (2022.0) **36** e22088. DOI: 10.1096/fj.202100497RR
27. Lopez PH, Aja S, Aoki K. **Mice lacking sialyltransferase ST3Gal-II develop late-onset obesity and insulin resistance**. *Glycobiology* (2017.0) **27** 129-139. DOI: 10.1093/glycob/cww098
|
---
title: 'Islet Autoantibody Standardization Program: interlaboratory comparison of
insulin autoantibody assay performance in 2018 and 2020 workshops'
authors:
- Ilaria Marzinotto
- David L. Pittman
- Alistair J. K. Williams
- Anna E. Long
- Peter Achenbach
- Michael Schlosser
- Beena Akolkar
- William E. Winter
- Vito Lampasona
journal: Diabetologia
year: 2023
pmcid: PMC10036445
doi: 10.1007/s00125-023-05877-9
license: CC BY 4.0
---
# Islet Autoantibody Standardization Program: interlaboratory comparison of insulin autoantibody assay performance in 2018 and 2020 workshops
## Abstract
### Aims/hypothesis
The Islet Autoantibody Standardization Program (IASP) aims to improve the performance of immunoassays measuring autoantibodies in type 1 diabetes and the concordance of results across laboratories. IASP organises international workshops distributing anonymised serum samples to participating laboratories and centralises the collection and analysis of results. In this report, we describe the results of assays measuring IAA submitted to the IASP 2018 and 2020 workshops.
### Methods
The IASP distributed uniquely coded sera from individuals with new-onset type 1 diabetes, multiple islet autoantibody-positive individuals, and diabetes-free blood donors in both 2018 and 2020. Serial dilutions of the anti-insulin mouse monoclonal antibody HUI-018 were also included. Sensitivity, specificity, area under the receiver operating characteristic curve (ROC-AUC), partial ROC-AUC at $95\%$ specificity (pAUC95) and concordance of qualitative/quantitative results were compared across assays.
### Results
Results from 45 IAA assays of seven different formats and from 37 IAA assays of six different formats were submitted to the IASP in 2018 and 2020, respectively. The median ROC-AUC was 0.736 (IQR 0.617–0.803) and 0.790 (IQR 0.730–0.836), while the median pAUC95 was 0.016 (IQR 0.004–0.021) and 0.023 (IQR 0.014–0.026) in the 2018 and 2020 workshops, respectively. Assays largely differed in AUC (IASP 2018 range 0.232–0.874; IASP 2020 range 0.379–0.924) and pAUC95 (IASP 2018 and IASP 2020 range 0–0.032).
### Conclusions/interpretation
Assay formats submitted to this study showed heterogeneous performance. Despite the high variability across laboratories, the in-house radiobinding assay (RBA) remains the gold standard for IAA measurement. However, novel non-radioactive IAA immunoassays showed a good performance and, if further improved, might be considered valid alternatives to RBAs.
### Supplementary Information
The online version contains peer-reviewed but unedited supplementary material available at 10.1007/s00125-023-05877-9.
## Introduction
Antibody reactivity to insulin was first described in individuals undergoing exogenous insulin administration [1] but people with different autoimmune diseases can also produce IAA in the absence of prior treatment with the hormone. IAA were first described in the insulin autoimmune syndrome (IAS, Hirata’s disease) in 1970 [2] and then in type 1 diabetes in 1983 [3]. The development of insulin autoimmune hypoglycaemia has been linked to exposure to environmental triggers (e.g. drugs and food supplements) in individuals with IAS [4], while no environmental factors have been confirmed in type 1 diabetes.
IAA are among the first biomarkers to appear during type 1 diabetes natural history. In asymptomatic individuals at risk of diabetes, the appearance of IAA associates with a faster progression to overt type 1 diabetes and younger age at onset [5, 6]. Additionally, specific features of IAA, such as binding affinity, targeted epitopes and titre, are associated with a higher risk of rapid type 1 diabetes development [7–9]. For these reasons, the measurement of IAA has become a cornerstone of screening strategies for type 1 diabetes [10–13] and a particular focus has been given to the improvement of IAA assays.
The first attempts at the standardisation of IAA measurement highlighted a large variability of results across centres and assays [14–18]. More recent interlaboratory comparisons studies organised by the Diabetes Autoantibody Standardization Program (DASP) showed only partial improvements in assay performance and results concordance [19, 20].
The Islet Autoantibody Standardization Program (IASP) has superseded the DASP in promoting the continuous improvement of type 1 diabetes autoantibody assays and disseminating empirically tested best-practice protocols, state-of-the-art reagents and serum standards [21]. The IASP is a collaborative effort supported by the Immunology of Diabetes Society (IDS) and the US NIH, which is run by the University of Florida Pathology Laboratories, Endocrine Autoantibody Laboratory and coordinated by an IDS nominated committee. The IASP pursues its goals through the establishment of a periodic interlaboratory comparison of type 1 diabetes-associated autoantibody measurements, aimed at providing an unbiased assessment of assay performance and improving the concordance of results across laboratories around the world. In IASP workshops the participating laboratories test type 1 diabetes autoantibodies in anonymised type 1 diabetes patient, ‘at-risk’ person, and control serum samples. An unbiased comparison of assay performance is provided through the centralised collection and analysis of results by the IASP committee.
In this report, we present the results of the 2018 and 2020 IASP IAA assays interlaboratory comparison studies that were preliminarily presented at the IASP 2018 and IASP 2020 workshops, held at the 16th and 17th IDS Congress, respectively.
## Study design
The study was aimed at comparing assay performance across laboratories. Participating laboratories received the same sera in anonymised sets, each labelled with an aliquot-specific unique code. Sera were obtained from the following individuals: individuals with new-onset type 1 diabetes (contributed by several centres around the world), collected within 14 days of the first insulin treatment; multiple islet autoantibody-positive first-degree relatives (FDR) of individuals with type 1 diabetes (enrolled in the TrialNet Ancillary Study – Pathway to Prevention and showing a transiently altered GTT during screening); blood donors without diabetes, collected in the USA. Due to their heterogeneous origin and to the difficulty of procuring sufficiently large serum volume from young children, these samples were only partially representative of incident new-onset diabetes and age-matched control individuals.
All samples were collected upon written informed consent, with the approval of local ethics committees and according to the ethical principles for medical research involving human subjects of the Declaration of Helsinki [22].
In 2018, the set included samples from 43 individuals with new-onset diabetes, seven multiple islet autoantibody-positive FDRs, 90 blood donor control individuals, six serial dilutions (156, 20, 5, 1.2, 0.6 and 0 ng/ml) in normal human serum of HUI-018, an IgG1 anti-insulin mouse monoclonal antibody (mAb) targeting a conformational epitope spanning the insulin A and B chains [23], and four additional samples from non-diseased individuals with type 1 diabetes autoantibodies.
Individuals with type 1 diabetes had a median age of 14 years (range 8–47) and included 15 female individuals and 28 male individuals; 37 were White, two Black, two of mixed ancestry and two of undisclosed ancestry. The FDRs had a median age of 16 years (range 12–53) and included four female and three male individuals, all of White ancestry. Demographic data were available only for 88 non-diabetic blood donors, who had a median age of 20 years (range 18–30), and included 44 female and 44 male individuals, of whom 69 were White and 19 Black.
In 2020, the set included samples from 38 individuals with new-onset diabetes, 12 multiple islet autoantibody-positive FDRs, 90 control blood donors, HUI-018 mAb serial dilutions, and four standards from the National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK) consortium, corresponding to 235, 5.8, 2 and 0 DK units, made from a mixture of GADA- and IA-2A-positive sera and used as reference material to harmonise antibody assays [24].
Individuals with type 1 diabetes had a median age of 14 years (range 8–47) and included 15 female and 23 male individuals, of whom 31 were White, two Hispanic, three Black, one of mixed ancestry and one of undisclosed ancestry. FDRs had a median age of 18 years (range 10–53) and included seven female and five male individuals, of whom 11 were White and one was of mixed ancestry.
Only a minority of samples were present in both 2018 and 2020 sample sets (21 type 1 diabetes, six FDRs, one blood donor) (electronic supplementary material [ESM] Table 1).
## Data analysis
Laboratory personnel were asked to report details of their assay protocol, assay raw data and results using uniform Excel reporting sheets. All data analyses were performed in the R language and environment for statistical computing and graphics [25].
Assays sensitivity and specificity was calculated as the percentage of case (new-onset plus FDRs) sera reported as IAA positive and as the percentage of blood donor sera reported as IAA-negative, respectively. Adjusted sensitivity at $95\%$ (AS95), $99\%$ (AS99) and $100\%$ specificity (AS100) were calculated after placing the threshold for positivity at the 95th, 99th and 100th percentiles of values observed in the blood donor samples in each assay, respectively.
Concordance of laboratory-assigned positive or negative scores across assays was expressed as average pairwise per cent agreement (APPA) between assays (i.e. the average number of times each possible combination of two assays agreed on IAA-positive/negative scores divided by the number of samples scored). We tested the occurrence of agreement by pure chance by calculating Gwet’s coefficient of inter-rater agreement reliability (AC1) [26] and Fleiss’ coefficient of inter-rater agreement reliability (k) [27] using the irr R package [28].
Assay performance in discriminating case from control samples was analysed using the area under the receiver operating characteristic curve (ROC-AUC) and the partial ROC-AUC at $95\%$ specificity (pAUC95) [29] using the pROC R package [30].
Interassay antibody titre concordance was analysed by calculating Kendall’s ranking agreement coefficient (W) [31], after ranking of case and control samples according to autoantibody levels in each assay. The significance of differences in mean ranking of selected case samples between different assays was tested using the Mann–Whitney test.
In a subset of assays with good performance (pAUC95 ≥0.015) and an immune-complex capture system compatible with the measurement of an anti-mouse IgG mAb, local units were converted into common units (HUI-018 ng/ml equivalents) by applying a linear model to the local units attributed to the provided HUI-018 dilutions followed by rescaling of local quantitative results. The concordance of antibody units was then evaluated by calculating the overall concordance correlation coefficient (OCCC) according to Barnhart using the epiR R package [32].
For all statistical analyses, two-tailed p values <0.05 were considered as significant.
## Summary of submitted IAA assay formats in the IASP 2018 and IASP 2020 workshops
In the IASP 2018 workshop, 23 laboratories from 13 countries submitted results from 45 IAA assays. The breakdown of assays according to format was as follows: radiobinding assay (RBA) [33]; antibody-dependent agglutination PCR (ADAP) [34]; luciferase immunoprecipitation system (LIPS) [35]; electrochemiluminescence (ECL) [36]; chemiluminescence immunoassay (CLIA); ELISA assay (of which some measured antibodies to oxidised insulin [37]); and Luminex bead immunoassay (LBI) (Fig. 1). Fig. 1IASP 2018 assay formats: main characteristics and performance. Single assays and their main characteristics are reported with the corresponding specificity (orange bars indicate the % specificity, with a maximum of $100\%$), sensitivity, AS95, AS99, AS100 (green bars indicate the % sensitivity, with a maximum of $100\%$), ROC-AUC, pAUC95 (green bars indicate the maximum value of 1 and 0.05, respectively) and LCSP. Assays are grouped by format. The median and the IQR values of each variable are reported for each group. aLocal units were not reported by the participating laboratory (only positivity scores). bThe NanoLuc reporter is alternatively placed near the insulin B or A chain. cPositivity scores in pan-Ig ECL assays for laboratory 1306 were assigned based on the combination (and/or) of positivity scores in the corresponding multiplexed Ig class-specific ECL assays (thus, no units were reported for either assay). dMultiplexed Ig class-specific insulin or proinsulin assays. NR, not reported All RBAs used recombinant insulin radiolabelled with 125I. LIPS assays used proinsulin ($$n = 5$$) or insulin antigens ($$n = 4$$) with the NanoLuc luciferase reporter tagged either at the C-terminus of the B chain ([pro]insulin-B-NLuc) or the N-terminus of the A chain (insulin-A-NLuc). Most RBA and LIPS assays were competitive assays performed with or without the presence of untagged insulin competitor (5.6 × 10−6 mol/l and 3.6 × 10−7 mol/l, respectively).
Antigen–antibody binding occurred in liquid phase in 35 assays (ADAP, ECL, LIPS, RBA) and was followed by the capture of immune complexes via recovery of immunoglobulins (LIPS, RBA) or tagged antigen (ADAP, ECL). Antigen–antibody binding occurred in hybrid solid/liquid phase in two assays (CLIA) and in solid phase for eight (ELISA, LBI). Major characteristics and metrics of each individual assay are reported in ESM Table 2.
In the IASP 2020 workshop, 22 laboratories from 11 countries submitted results from 37 IAA assays (Fig. 2) based on the following formats: RBA; ADAP; LIPS; ECL; CLIA; and Flow cytometric microsphere-based immunoassay (FloCMIA) [38]. Fig. 2IASP 2020 assay formats: main characteristics and performance. Single assays and their main characteristics are reported with the corresponding specificity (orange bars indicate the % specificity, with a maximum of $100\%$), sensitivity, AS95, AS99, AS100 (green bars indicate the % sensitivity, with a maximum of $100\%$), ROC-AUC, pAUC95 (green bars indicate the maximum value of 1 and 0.05, respectively) and LCSP. Assays are grouped by format. The median and the IQR values of each variable are reported for each group. aDuplex LIPS assays multiplexing IAA and IA-2A testing using a dual luciferase system. bMultiplexed with the measurement of IA-2A, GADA, ZnT8A and autoantibodies to transglutaminase (TGA). cMultiplexed Ig class-specific insulin or proinsulin assays. NR, not reported All RBAs used recombinant insulin radiolabelled with 125I. LIPS assays used either proinsulin ($$n = 6$$) or insulin ($$n = 5$$) antigens tagged with a NanoLuc luciferase reporter at the C-terminus of the insulin B chain. One LIPS assay (Duplex LIPS) combined individual measurement of IA-2A and IAA using a dual luciferase system. One laboratory used two different concentrations of competitor in LIPS (3.6 × 10−7 mol/l and 1.1 × 10−9 mol/l).
Antigen–antibody binding occurred in liquid phase in 36 assays (ADAP, ECL, FloCMIA, LIPS, RBA) or in hybrid solid/liquid phase (CLIA) and was followed by the capture of immune complexes through the recovery of immunoglobulins (LIPS, RBA) or antigens with different tags (ADAP, ECL, FloCMIA, CLIA). Major characteristics and metrics of each individual assay are reported in ESM Table 3.
## Assay sensitivity and specificity based on laboratory-assigned scores in the IASP 2018 and IASP 2020 workshops
In the IAA assays submitted to the IASP 2018 workshop, laboratory-assigned scores showed a median assay sensitivity of $32.0\%$ (IQR 16.0–46.0) and a specificity of $96.7\%$ (IQR 89.7–97.8), with a wide range for both (sensitivity 66.0–$2.0\%$; specificity 100.0–$45.6\%$) (Fig. 1, ESM Table 2 and ESM Fig. 1).
In the assays submitted to the IASP 2020 workshop, the median assay sensitivity was $46.0\%$ (IQR 26.0–56.0) and the specificity $98.9\%$ (IQR 96.7–100.0), with a wide range of both (sensitivity 66.0–$2.0\%$; specificity 100.0–$80.0\%$) (Fig. 2, ESM Table 3 and ESM Fig. 2).
## ROC-AUC analysis of assay format performance in the IASP 2018 and IASP 2020 workshops
We evaluated the performance of the IAA assays using the full ROC-AUC and the partial ROC-AUC after imposing a specificity of ≥$95\%$, as a more relevant proxy of assay performance [29] (Fig. 3). Fig. 3ROC curve analysis of assays submitted to the IASP 2018 (a–g) and IASP 2020 (h–m) workshops. ROC curves are shown for RBA (a: $$n = 14$$ and h: $$n = 13$$), LIPS (b: $$n = 9$$ and i: $$n = 11$$), ECL (c: $$n = 8$$ and j: $$n = 8$$), ADAP (d: $$n = 1$$ and k: $$n = 1$$), ELISA (e: $$n = 6$$), LBI (f: $$n = 2$$), CLIA (g: $$n = 2$$ and m: $$n = 1$$) and FloCMIA (l: $$n = 1$$); indicated assay variants within each format are shown by curve colour. Black lines, median ROC curve; grey rectangles, area corresponding to a specificity ≥$95\%$, where the pAUC95 is calculated; dashed lines, identity line. h.c., high concentration of unlabelled insulin competitor (3.6 × 10−7 mol/l); l.c., low concentration of unlabelled insulin competitor (1.1 × 10−9 mol/l) In the IASP 2018 workshop, IAA assays showed a median ROC-AUC of 0.736 (IQR 0.617–0.803, range 0.232–0.874) and a median pAUC95 of 0.016 (IQR 0.004–0.021, range 0–0.032), against a theoretical pAUC95 maximum of 0.05 (Figs 1, 4 and ESM Fig. 3a). A wide heterogeneity of performance was present both within and across assay formats. Fig. 4Distribution of the pAUC95 of IAA assays submitted to the IASP 2018 (a) and IASP 2020 (b) workshops. Results are grouped by assay format. LIPS assays using two alternative amounts of unlabelled insulin are labelled as either high (h.c., 3.6 × 10−7 mol/l) or low concentration (l.c., 1.1 × 10−9 mol/l). The grey half violin plots show the overall probability density estimate. Circles correspond to the pAUC95 value of each single assay, with colours indicating different assay variants. The vertical dashed lines correspond to the median pAUC95 of all assays (black) and to the ROC identity line (red), respectively In the IASP 2020 assays, the median ROC-AUC of the IAA assays was 0.790 (IQR 0.730–0.836, range 0.378–0.924), while the median pAUC95 was 0.023 (IQR 0.014–0.026, range 0–0.032) (Figs 2, 4 and ESM Fig. 3b). Additionally, in the IASP 2020 workshop, assay performance varied widely both within and across formats.
## Qualitative concordance of laboratory-assigned positive/negative scores
In the assays submitted to the IASP 2018 workshop, the APPA of positive/negative scores assigned to type 1 diabetes cases across all assays of IAA was $67.1\%$, while the first-order AC1 of these scores was 0.415. The concordance analysis was greater in control samples (APPA $86.5\%$, AC1 0.863). When the analysis was limited to assays with a pAUC95 above the median (pAUC95 ≥0.013), these concordance variables improved in both case (APPA $73.4\%$; AC1 0.476) and control samples (APPA $96.1\%$; AC1 0.959).
In the assays submitted to the IASP 2020 workshop, in cases the APPA was $66.9\%$ and the AC1 0.360, while in controls the respective values were $93.7\%$ and 0.932. In assays with a pAUC95 above the median (pAUC95 ≥0.023), these values improved in both case (APPA $74.8\%$; AC1 0.497) and control samples (APPA $97.3\%$; AC1 0.972).
In both years, the observed low Fleiss’ k concordance coefficients were consistent, with most of the control samples scored IAA positive being sporadically so in only a small fraction of assays.
Overall, the concordance of IAA-positive/-negative scores was greater across assays using the same format (ESM Tables 4–7).
## Assay format-specific patterns of IAA recognition
In both IASP workshops, discrepancies in positive/negative scores showed format-specific patterns. In 2018, a subset of type 1 diabetes cases (IDS326, IDS322, IDS292, IDS298, IDS004, IDS309, IDS006, IDS290) was IAA positive predominantly in ADAP and LIPS and in few RBA and ECL assays. Conversely, a different subset of sera (IDS303, IDS301) was IAA positive in ADAP and most RBAs but only in a minority of LIPS and ECL assays. A subset of control sera (TS24176, N59807, N54153) was IAA positive mostly in LIPS and sporadically in other formats, while another control sample (C1401) was positive mostly in RBA and ELISA (Figs 5, 6 and ESM Figs 4, 5). Fig. 5Tilemap of IAA positivity scores in case sera submitted to the IASP 2018 workshop. Tilemap of IAA-positive (dark grey) or -negative (light grey) scores assigned by laboratories to each new-onset type 1 diabetes, multiple autoantibody-positive and HUI-018 standard samples. Samples are sorted on the x-axis according to the median rank calculated in each group. The shown sample labels identify sera with format-specific patterns of reactivity described in the text. Assays on the y-axis are grouped by format or format variant and then sorted according to their median pAUC95. Ab+, autoantibody-positive; HUI-018 STD, HUI-018 standard dilutionsFig. 6Tilemap of IAA positivity scores in control sera submitted to the IASP 2018 workshop. Tilemap of IAA-positive (dark grey) or -negative (light grey) scores assigned by laboratories to each control sample. Samples are sorted on the x-axis according to the median rank calculated in each group. The shown sample labels identify sera with format-specific patterns of reactivity described in the text. Assays on the y-axis are grouped by format or format variant and then sorted according to their median pAUC95. Only samples with a positive score in at least one assay are shown In the sera submitted to the IASP 2020 workshop, two case samples (IDS359, IDS326) were IAA positive predominantly in LIPS but only in a minority of RBA and ECL assays. Conversely, another subset of case samples (IDS324, IDS303 and IDS351) was positive in most RBA and ECL but only in a minority of LIPS assays. Among control sera, two (S6320, S8768) were positive mostly in LIPS but only sporadically in other formats, while two other serum samples (S6389, LQ22722) were positive exclusively in ECL. In LIPS assays using two different concentrations of unlabelled insulin competitor, IAA positivity was sometimes conditional upon the use of a high concentration of competitor (IDS326), suggesting the presence of low-affinity IAA. Always in LIPS, one serum sample (IDS351) resulted positive only when insulin antigen was used, suggesting that antigen recognition was conditional on removal of the C-peptide (Figs 7, 8 and ESM Figs 6, 7). Fig. 7Tilemap of IAA positivity scores in case sera submitted to the IASP 2020 workshop. Tilemap of IAA-positive (dark grey) or -negative (light grey) scores assigned by laboratories to each sample of new-onset type 1 diabetes, multiple autoantibody-positive, DK standard and HUI-018 standard sera. The shown sample labels identify sera with format-specific patterns of reactivity described in the text. Samples are sorted on the x-axis according to the median rank calculated in each group. Assays on the y-axis are grouped by format or format variant and then sorted according to their median pAUC95. LIPS assays using two alternative amounts of unlabelled insulin are labelled as either high (h.c., 3.6 × 10−7 mol/l) or low concentration (l.c., 1.1 × 10−9 mol/l). Ab+, autoantibody-positive. DK STD, test standards distributed by the NIDDK consortiumFig. 8Tilemap of IAA positivity scores in control sera submitted to the IASP 2020 workshop. Tilemap of IAA-positive (dark grey) or -negative (light grey) scores assigned by laboratories to each control sample. Samples are sorted on the x-axis according to the median rank calculated in each group. The shown sample labels identify sera with format-specific patterns of reactivity described in the text. Assays on the y-axis are grouped by format or format variant and then sorted according to their median pAUC95. LIPS assays using two alternative amounts of unlabelled insulin are labelled as either high (h.c., 3.6 × 10−7 mol/l) or low concentration (l.c., 1.1 × 10−9 mol/l) The results also highlighted the selective recognition of some case sera by assays in which antibody–antigen binding occurs in liquid phase (ADAP, ECL, LIPS, RBA) vs solid phase (CLIA, ELISA, FloCMIA, LBI) in both IASP 2018 (IDS312, IDS009, IDS268, IDS328, IDS334, IDS317) and IASP 2020 (IDS372, IDS345, IDS317, IDS312, IDS365, IDS310, IDS358, IDS367) workshops.
## Ranking of autoantibody levels
Quantitative interassay concordance of IAA levels was evaluated by ranking sera in each assay and then calculating the Kendall’s W ranks agreement coefficient.
In the IASP 2018 workshop, the W coefficient across all assays was 0.354 for case sera and 0.081 for control sera. In the IASP 2020 workshop, the W coefficient across all assays was 0.458 for case sera and 0.047 for control sera.
In both workshops, excluding assays with low performance (i.e. low pAUC95) from the analysis led to a modest increase of the agreement coefficient in both case and control samples (IASP 2018, $W = 0.431$ and 0.086, respectively; IASP 2020, $W = 0.587$ and 0.057, respectively). Limiting the analysis to assays with higher performance (i.e. greater than the median pAUC95) showed a further increase of W for case sera but only a marginal improvement for control sera (IASP 2018, $W = 0.656$ and 0.098, respectively; IASP 2020, $W = 0.617$ and 0.086, respectively).
Concordance of IAA level ranks increased among assays using the same format for both case and control sera (ESM Figs 8–15 and ESM Tables 4–7). The comparison of assay formats’ W coefficients showed that ranking agreement was higher in LIPS than in both local and commercial RBAs, CLIA and ECL assays.
IAA level ranking also confirmed the previously observed preferential recognition of some case samples by assays in which antibody–antigen binding occurs in liquid phase vs solid phase (Mann–Whitney test, all p≤0.001 in IASP 2018 workshop and all $p \leq 0.05$ in IASP 2020 workshop).
## Autoantibody levels in 2018 vs 2020
Using the 28 samples that were distributed in both the IASP 2018 workshop and the IASP 2020 workshop, we analysed the correlation of quantitative results in 24 assays (nine local RBAs, one commercial RBA, one ADAP, five LIPS, one CLIA and seven ECL). Correlation was highest in RBA, LIPS and ADAP (median R2=0.97 [IQR 0.94–0.98]) and lower in ECL and CLIA (median R2=0.25 [IQR 0.01–0.50]) (ESM Figs 16, 17).
## Conversion of local units into common HUI-018 units
Using the provided HUI-018 anti-insulin mAb serial dilutions, we converted local arbitrary units into common units (HUI-018 ng/ml) by applying a log–log linear regression model. In both the IASP 2018 and the IASP 2020 workshop, the correlation of local and common units was high (R2 range 0.85–1.00) but the slopes of the regression curves varied considerably across assays (slope range 0.406–1.000) (ESM Figs 18, 19). We then compared calculated common HUI-018 units in each assay with the true concentration of the HUI-018 serial dilutions (ESM Figs 20, 21). The observed high degree of variability of the assigned concentrations was suggestive of potentially large discrepancies in the linear range of the different assays.
## Autoantibody levels in common HUI-018 units
We converted local arbitrary units into common HUI-018 units for assays with good performance and a detection system compatible with the measurement of a mouse IgG mAb (Fig. 9). We then calculated the OCCC of HUI-018 units as another measure of interassay quantitative concordance. In both workshops, the OCCC was relatively low in both case (IASP 2018, 0.285; IASP 2020, 0.203) and control sera assays (IASP 2018, 0.013; IASP 2020, 0.012). After stratification of assays according to format and antigen, the quantitative concordance of common units improved only for some assays (ESM Figs 22, 23 and ESM Tables 4–7). Fig. 9Stripchart of common HUI-018 units in selected assays submitted to the IASP 2018 (a) and IASP 2020 (b) workshops. Samples from blood donors, individuals with new-onset type 1 diabetes, multiple autoantibody positivity, DK and HUI-018 standards are sorted on the x-axis according to the median rank calculated in each group. Circles show the HUI-018 units attributed to each sample; circle colour indicates assay format. Ab+, autoantibody-positive; AU, arbitrary units calculated as HUI-018 ng/ml equivalents. DK STD, test standards distributed by the NIDDK consortium
## Analytical sensitivity of IAA assays
As a proxy of the IAA assays’ analytical sensitivity, we determined the lowest concentration scored positive (LCSP) by each assay for HUI-018 mAb (Figs 1, 2, 5, 7). The median LCSP was 1.2 ng/ml in both workshops but exhibited extreme variability across assays (range 0.6–156). In addition, some assays recognised the diluent normal serum as weakly positive.
## Discussion
Early interlaboratory comparison studies demonstrated that the detection of disease-specific IAA was crucially dependent on assay format choice and led to the establishment of the liquid-phase immunoprecipitation RBA assay as the de facto gold standard for IAA measurement [14–20].
The more recent IASP 2018 and IASP 2020 interlaboratory IAA measurement comparison studies saw not only the continued implementation of RBAs by most laboratories but also an increasingly wider adoption of alternative non-radioactive formats. Therefore, in this paper we were able to perform a comprehensive comparison of RBAs and other assay formats in terms of diagnostic performance, sensitivity, specificity and concordance.
RBAs submitted to the IASP workshop in both 2018 and 2020 were predominantly in-house assays derived from the IAA micro-assay originally described in 1997 by Williams et al [33]. The performance of these micro-assays varied widely in both workshops but compared with more recent non-radioactive IAA immunoassays most RBAs showed similar or better sensitivity and specificity, as well as pAUC95 and AS95, two metrics aimed at excluding regions of low assay specificity and poor clinical relevance from the analysis. Of note, commercial RBAs, while showing good specificity, had a sensitivity less than half that of most in-house micro-assays, possibly because of a lower capacity to immuno-precipitate IAA immune complexes by the anti-human IgG polyclonal antibody used in commercial kits compared with the protein A/G coated Sepharose beads adopted by the micro-assay RBA.
Among non-radioactive IAA immunoassays, the only submitted ADAP assay, the most recently developed IAA assay format, showed the highest or second-highest pAUC95 and AS95.
In both workshops, among non-radioactive assay formats, the most widely adopted were LIPS and ECL. Submitted LIPS assays implemented a variety of alternative protocols and antigens and showed variable performance. Nevertheless, LIPS assays showed the overall second-highest median pAUC95 and AS95 after RBA.
ECL assays comprised two major variants, the first measuring IAA potentially of any Ig class (pan-Ig ECL) and the second multiplexing and discriminating Ig of different classes (IgM, IgA and IgG multiplexed-ECL). In both 2018 and 2020, pan-Ig ECL assays showed a relatively good performance, although lower than that of the best RBAs. Multiplexed IgG, IgM and IgA ECL assays instead were not only less sensitive, as might have been expected, but also less specific, as signalled by the scoring as IgM and/or IgA IAA positive of some of the included dilutions of HUI-018, a mouse anti-insulin mAb of the IgG1 class.
The remaining non-radioactive assays submitted to the workshops comprised a variety of formats such as CLIA, FloCMIA, LBI, ELISAs, and ELISAs using oxidised insulin as antigen. All of these assays showed a poor performance, with drastically reduced sensitivity and specificity compared with RBA and their ROC-AUCs demonstrated an inability to discriminate case sera from control sera.
In both workshops, we observed discrepancies of positive/negative scores and ranking of antibody levels across assays and formats, even when the analysis was limited to assays with overall good performance. The underlying reasons for the selective recognition of some case and control samples as IAA positive in certain assay formats but not others remain to be fully clarified. However, two main mechanisms can be put forward to explain these discrepancies: the first is simply linked to the difficulty of lower performance assays in identifying as positive sera presenting with low level IAA; and the second presumes a differential recognition of insulin epitopes by antibodies present only in a subset of sera. Multiple causes might underly this potential second mechanism, such as possible alteration of some insulin epitopes in certain assay formats caused by the addition of tags (e.g. biotin residues in ECL or luciferase enzyme in LIPS) or by potential alternative antigen post-translational modification(s) in the different expression systems used for their production. This potential format-associated selective recognition of some epitopes remains to be evaluated but might have an important impact on autoantibody-based screening strategies, which currently are still based on RBA.
In both the IASP 2018 and IASP 2020 workshops, the comparison of quantitative IAA results was complicated by the variety of local non-standardised arbitrary units and calculation algorithms into which results were expressed. In the absence of a WHO-recognised IAA standard serum, we explored the possibility of using an anti-insulin mouse monoclonal antibody as reference. For this reason, we distributed anonymised serial dilutions of the HUI-018 mAb. Most liquid-phase assays showed a clear ability to detect HUI-018 binding to insulin but the correct ranking of the mAb dilutions was challenging for assays with lower performance. While neither IASP study was designed to determine rigorously the analytical sensitivity of participating assays, using the differential recognition of the HUI-018 mAb dilution as a proxy, we could infer the presence of important differences in analytical sensitivity across IAA assays.
Moreover, the conversion of local units into common HUI-018 units further confirmed the lack of good quantitative concordance of IAA assays even when using the same format, a phenomenon that could not be conclusively clarified within the limits of the current study design.
Legislative and logistic pressure against the use of radioactive substances spur the development and validation of novel non-radioactive immunoassays. Furthermore, the expected future implementation of antibody-based population screening programmes for type 1 diabetes would benefit from the implementation of high-performance IAA assays dispensing with the need for radio-isotopic tracers. In this context, while in-house micro-assay RBAs still constituted the majority of the best-performing assays in 2020, some non-radioactive formats could indeed achieve both high sensitivity and specificity (e.g. ADAP, LIPS and ECL). However, none of the classical immunoassay formats widely adopted in routine clinical diagnostics (e.g. ELISA, CLIA and Luminex bead-based assays) nor assays aimed at measuring Ig class-specific autoantibody responses can be currently recommended for IAA measurement in light of their poor demonstrated performance.
In conclusion, our research supports the value of type 1 diabetes autoantibody assay evaluation programmes. These programmes not only help to assess the accuracy of diagnostic tests objectively but also provide academic research laboratories and companies with the chance to learn and improve their immunoassays.
## Supplementary information
ESM 1(PDF 9.18 kb)ESM 2(XLSX 28 kb)ESM 3(XLSX 24 kb)
## Authors’ relationships and activities
PA is a member of the Editorial Board of Diabetologia. All other authors declare that there are no relationships or activities that might bias, or be perceived to bias, their work.
## Contribution statement
AJW, PA, MS, BA, DLP, WEW and VL contributed to the conception and design of the study. DLP and WEW organised and supervised sample preparation and distribution and data collection from participating laboratories. IM, DLP and VL contributed to the collection, preliminary analysis and interpretation of data. IM and VL drafted the initial manuscript. AJW, AEL, PA, MS, BA, DLP and WEW critically revised the data analysis and the manuscript. All authors approve of the publication. VL is the guarantor of the paper and accepts full responsibility for the work and/or the conduct of the study, had access to the data, and controlled the decision to publish.
## Participating laboratories and contacts for the IASP 2018 and IASP 2020 IAA interlaboratory comparison studies
S. N. Valdez, Facultad de Farmacia y Bioquímica, Universidad de Buenos Aires, Buenos Aires, Argentina; K. Watson, Royal Melbourne Hospital, Melbourne, VIC, Australia; K. Verhaeghen, Uzbrussel Vub, Clinical Biology of Diabetes-Diabetes Research Center, Brussels, Belgium; V. Chen, Laboratory of Snibe, Shenzhen, China;
J. Zhang, Shenzhen YHLO Biotech Co., Ltd, Shenzhen, China; Z. Zhou, G. Huang, Diabetes Center, Central South University, Changsha, China; R. Uibo, K. Reimand, University of Tartu, Department of Immunology, Tartu, Estonia; M. Knip, T. Härkönen, Children’s Hospital, Scientific Laboratory, University of Helsinki, Helsinki, Finland;
R. Veijola, Department of Pediatrics, Diabetes Research Laboratory, University of Oulu, Oulu, Finland; L. Chatenoud, Laboratoire d’Immunologie Biologique-Hôpital Necker-Enfants Malades Paris, Paris, France; D. Mueller, Preclinical Approaches to Stem Cell Therapy/ Diabetes, Dresden, Germany; P. Achenbach, Institute of Diabetes Research, Helmholtz Zentrum München, Neuherberg, Germany;
M. Schlosser, University Medical Center Greifswald, Karlsburg, Germany; V. Lampasona, Diabetes Research Institute, IRCCS Istituto Scientifico San Raffaele, Milano, Italy B. Almås, The Hormone Laboratory, Haukeland University Hospital, Bergen, Norway; K. S. Opsion, Hormone Laboratory, Oslo University Hospital, Oslo, Norway;
A. Ramelius, Diabetes And Celiac Disease Research Unit, Lund University, Malmö, Sweden; I. Johansson, Clinical & Experimental Research, Division of Pediatrics, Linköping, Sweden; M. R. Batstra, T. Cieremans, Reinier De Graaf Groep, Department of Medical Immunology, Delft, the Netherlands; S. Kostense, World Without Disease, Janssen, Leiden, the Netherlands;
A. E. Long, Diabetes & Metabolism, Learning & Research, University of Bristol, Bristol, UK; A. Mathew, Meso Scale Diagnostics, LLC., Rockville, MD, USA; C. Hampe, University of Washington, Seattle, WA, USA; C. Lu, Meso Scale Discovery, Gaithersburg, MD, USA;
C. Mann, Quest Diagnostics Nichols Institute, San Juan Capistrano, CA, USA; L. Yu, Barbara Davis Center, Aurora, CO, USA; M. Mamula, L2 Diagnostics at Yale University, New Haven, CT, USA; P. Robinson, Enable Biosciences, INC, San Francisco, CA, USA;
W. A. Hagopian, Pacific Northwest Diabetes Research Institute, Seattle, WA, USA.
## References
1. Moinat P, Marston E. **A quantitative estimation of antibodies to exogenous insulin in diabetic subjects**. *Diabetes* (1958.0) **7** 462-467. DOI: 10.2337/diab.7.6.462
2. Cappellani D, Macchia E, Falorni A, Marchetti P. **Insulin autoimmune syndrome (Hirata disease): a comprehensive review fifty years after its first description**. *Diabetes Metab Syndr Obes Targets Ther* (2020.0) **13** 963-978. DOI: 10.2147/DMSO.S219438
3. Palmer JP, Asplin CM, Clemons P. **Insulin antibodies in insulin-dependent diabetics before insulin treatment**. *Science* (1983.0) **222** 1337-1339. DOI: 10.1126/science.6362005
4. Censi S, Mian C, Betterle C. **Insulin autoimmune syndrome: from diagnosis to clinical management**. *Ann Transl Med* (2018.0) **6** 335-335. DOI: 10.21037/atm.2018.07.32
5. Vehik K, Bonifacio E, Lernmark Å. **Hierarchical order of distinct autoantibody spreading and progression to type 1 diabetes in the TEDDY study**. *Diabetes Care* (2020.0) **43** 2066-2073. DOI: 10.2337/dc19-2547
6. Krischer JP, Liu X, Lernmark Å. **Characteristics of children diagnosed with type 1 diabetes before vs after 6 years of age in the TEDDY cohort study**. *Diabetologia* (2021.0) **64** 2247-2257. DOI: 10.1007/s00125-021-05514-3
7. Achenbach P, Koczwara K, Knopff A, Naserke H. **Mature high-affinity immune responses to (pro) insulin anticipate the autoimmune cascade that leads to type 1 diabetes**. *J Clin Invest* (2004.0) **114** 589-597. DOI: 10.1172/JCI200421307
8. Schlosser M, Koczwara K, Kenk H. **In insulin-autoantibody-positive children from the general population, antibody affinity identifies those at high and low risk**. *Diabetologia* (2005.0) **48** 1830-1832. DOI: 10.1007/s00125-005-1864-6
9. Achenbach P, Schlosser M, Williams AJK. **Combined testing of antibody titer and affinity improves insulin autoantibody measurement: Diabetes Antibody Standardization Program**. *Clin Immunol* (2007.0) **122** 85-90. DOI: 10.1016/j.clim.2006.09.004
10. Elding Larsson H, Vehik K, Gesualdo P. **Children followed in the TEDDY study are diagnosed with type 1 diabetes at an early stage of disease: Intense follow-up enables early detection of type 1 diabetes**. *Pediatr Diabetes* (2014.0) **15** 118-126. DOI: 10.1111/pedi.12066
11. Krischer JP, Liu X, Vehik K. **Predicting islet cell autoimmunity and type 1 diabetes: an 8-year TEDDY study progress report**. *Diabetes Care* (2019.0) **42** 1051-1060. DOI: 10.2337/dc18-2282
12. Ziegler A-G, Kick K, Bonifacio E. **Yield of a public health screening of children for islet autoantibodies in Bavaria, Germany**. *JAMA* (2020.0) **323** 339. DOI: 10.1001/jama.2019.21565
13. 13.Karpen SR, Dunne JL, Frohnert BI et al (2022) Consortium-based approach to receiving an EMA qualification opinion on the use of islet autoantibodies as enrichment biomarkers in type 1 diabetes clinical studies. Diabetologia. 10.1007/s00125-022-05751-0
14. Wilkin T, Palmer J, Bonifacio E. **First international workshop on the standardisation of insulin autoantibodies**. *Diabetologia* (1987.0) **30** 676-677. DOI: 10.1007/BF00277328
15. Wilkin T, Palmer J, Kurtz A, Bonifacio E, Diaz JL. **The second international workshop on the standardisation of insulin autoantibody (IAA) measurement. Held in New York, 27-30 October 1987**. *Diabetologia* (1988.0) **31** 449-450. DOI: 10.1007/BF00271590
16. Greenbaum CJ, Palmer JP, Kuglin B, Kolb H. **Insulin autoantibodies measured by radioimmunoassay methodology are more related to insulin-dependent diabetes mellitus than those measured by enzyme-linked immunosorbent assay: results of the Fourth International Workshop on the Standardization of Insulin**. *J Clin Endocrinol Metab* (1992.0) **74** 1040-1044. DOI: 10.1210/jcem.74.5.1569152
17. Greenbaum CJ, Wilkin TJ, Palmer JP. **Fifth international serum exchange workshop for insulin autoantibody (IAA) standardization. The immunology and diabetes workshops and participating laboratories**. *Diabetologia* (1992.0) **35** 798-800. DOI: 10.1007/BF00429105
18. Wilkin TJ, Schoenfeld SL, Diaz JL, Kruse V, Bonifacio E, Palmer JP. **Systematic variation and differences in insulin-autoantibody measurements**. *Diabetes* (1989.0) **38** 172-181. DOI: 10.2337/diabetes.38.2.172
19. Bingley PJ, Bonifacio E, Mueller PW. **Diabetes antibody standardization program: first assay proficiency evaluation**. *Diabetes* (2003.0) **52** 1128-1136. DOI: 10.2337/diabetes.52.5.1128
20. Schlosser M, Mueller PW, Törn C, Bonifacio E, Bingley PJ. **Diabetes antibody standardization program: evaluation of assays for insulin autoantibodies**. *Diabetologia* (2010.0) **53** 2611-2620. DOI: 10.1007/s00125-010-1915-5
21. Lampasona V, Pittman DL, Williams AJ. **Islet autoantibody standardization program 2018 workshop: interlaboratory comparison of glutamic acid decarboxylase autoantibody assay performance**. *Clin Chem* (2019.0) **65** 1141-1152. DOI: 10.1373/clinchem.2019.304196
22. 22.World Medical Association (2001) World Medical Association declaration of Helsinki. Ethical principles for medical research involving human subjects. Bull World Health Organ 79(4):373–374
23. Johansson E, Wu X, Yu B. **Insulin binding to the analytical antibody sandwich pair OXI-005 and HUI-018: Epitope mapping and binding properties**. *Protein Sci* (2021.0) **30** 485-496. DOI: 10.1002/pro.4009
24. Bonifacio E, Yu L, Williams AK. **Harmonization of glutamic acid decarboxylase and islet antigen-2 autoantibody assays for national institute of diabetes and digestive and kidney diseases consortia**. *J Clin Endocrinol Metab* (2010.0) **95** 3360-3367. DOI: 10.1210/jc.2010-0293
25. 25.R Core TeamR: A Language and Environment for Statistical Computing2020Vienna, AustriaR Foundation for Statistical Computing. *R: A Language and Environment for Statistical Computing* (2020.0)
26. Wongpakaran N, Wongpakaran T, Wedding D, Gwet KL. **A comparison of Cohen’s Kappa and Gwet’s AC1 when calculating inter-rater reliability coefficients: a study conducted with personality disorder samples**. *BMC Med Res Methodol* (2013.0) **13** 61. DOI: 10.1186/1471-2288-13-61
27. Cicchetti DV, Feinstein AR. **High agreement but low kappa: II. Resolving the paradoxes**. *J Clin Epidemiol* (1990.0) **43** 551-558. DOI: 10.1016/0895-4356(90)90159-M
28. 28.Gamer M, Lemon J, Fellows I, Singh P (2019) irr: Various Coefficients of Interrater Reliability and Agreement. R package version 0.84.1. Available from: https://CRAN.R-project.org/package=irr (accessed on 26 January 2019)
29. Ma H, Bandos AI, Rockette HE, Gur D. **On use of partial area under the ROC curve for evaluation of diagnostic performance**. *Stat Med* (2013.0) **32** 3449-3458. DOI: 10.1002/sim.5777
30. Robin X, Turck N, Hainard A. **pROC: an open-source package for R and S+ to analyze and compare ROC curves**. *BMC Bioinformatics* (2011.0) **12** 77. DOI: 10.1186/1471-2105-12-77
31. Kendall MG, Babington Smith B. **The problem of m rankings**. *Ann Math Stat* (1939.0) **10** 275-287. DOI: 10.1214/aoms/1177732186
32. 32.Stevenson M, Sergeant E, Nunes T et al (2021) epiR: tools for the analysis of epidemiological data. R package version 2.0.26. https://CRAN.R-project.org/package=epiR (accessed on 15 July 2021)
33. Williams AJ, Bingley PJ, Bonifacio E, Palmer JP, Gale EA. **A novel micro-assay for insulin autoantibodies**. *J Autoimmun* (1997.0) **10** 473-478. DOI: 10.1006/jaut.1997.0154
34. Tsai C, Robinson PV, Spencer CA, Bertozzi CR. **Ultrasensitive antibody detection by agglutination-PCR (ADAP)**. *ACS Cent Sci* (2016.0) **2** 139-147. DOI: 10.1021/acscentsci.5b00340
35. Liberati D, Wyatt RC, Brigatti C. **A novel LIPS assay for insulin autoantibodies**. *Acta Diabetol* (2018.0) **55** 263-270. DOI: 10.1007/s00592-017-1082-y
36. Yu L, Dong F, Miao D, Fouts AR, Wenzlau JM, Steck AK. **Proinsulin/insulin autoantibodies measured with electrochemiluminescent assay are the earliest indicator of prediabetic islet autoimmunity**. *Diabetes Care* (2013.0) **36** 2266-2270. DOI: 10.2337/dc12-2245
37. Strollo R, Vinci C, Napoli N, Pozzilli P, Ludvigsson J, Nissim A. **Antibodies to post-translationally modified insulin as a novel biomarker for prediction of type 1 diabetes in children**. *Diabetologia* (2017.0) **60** 1467-1474. DOI: 10.1007/s00125-017-4296-1
38. Sabljic AV, Bombicino SS, Marfía JI. **Novel flow cytometric immunoassay for detection of proinsulin autoantibodies in diabetes mellitus employing a recombinant autoantigen expressed in E. coli**. *Front Immunol* (2021.0) **12** 648021. DOI: 10.3389/fimmu.2021.648021
39. **The environmental determinants of diabetes in the young (TEDDY) study**. *Ann N Y Acad Sci* (2008.0) **1150** 1-13. DOI: 10.1196/annals.1447.062
40. Skyler JS, Greenbaum CJ, Lachin JM. **Type 1 Diabetes TrialNet--an international collaborative clinical trials network**. *Ann N Y Acad Sci* (2008.0) **1150** 14-24. DOI: 10.1196/annals.1447.054
|
---
title: 'Health behaviors, obesity, and marital status among cancer survivors: a MEPS
study'
authors:
- Lixin Song
- Ting Guan
- Peiran Guo
- Xianming Tan
- Ashley Leak Bryant
- William A. Wood
- Anthony D. Sung
- Erin Elizabeth Kent
- Thomas C. Keyserling
journal: Journal of Cancer Survivorship
year: 2022
pmcid: PMC10036458
doi: 10.1007/s11764-022-01269-x
license: CC BY 4.0
---
# Health behaviors, obesity, and marital status among cancer survivors: a MEPS study
## Abstract
### Purpose
Promoting positive health behaviors helps improve cancer survivors’ health outcomes during survivorship; however, little is known about whether health behaviors differ by marital status. The purpose is to examine whether health behaviors and obesity among cancer survivors vary by marital status and whether the type of cancer and sociodemographic factors influence the relationship.
### Methods
We examined smoking, physical activity, and body mass index (BMI) among 1880 individuals diagnosed with prostate, breast, or colon cancer who were identified from the 2011–2017 Medical Expenditure Panel Survey (MEPS). We used Rao-Scott design-adjusted chi-square tests and weighted multivariable logistic regressions to achieve the research aims.
### Results
Current smoking behavior and BMI were significantly related to marital status. Survivors who had never married were the most likely to be current smokers across all cancer types. Married survivors were the most likely to be overweight or obese, while widowed survivors were the most likely to have a normal weight. The relationship between BMI and marital status varied by cancer type. Widowed colon cancer survivors were least likely to be overweight or obese; divorced/separated colon cancer survivors were most likely to be obese or overweight. Health behavior disparities were found among cancer survivors of different age, sex, race, and levels of education and income.
### Conclusions
There were relationships between marital status, health behaviors, and obesity among cancer survivors.
### Implications for Cancer Survivors
Our results suggested that relationship status and sociodemographic factors need to be considered in tailoring interventions to promote health behaviors among cancer survivors.
### Supplementary information
The online version contains supplementary material available at 10.1007/s11764-022-01269-x.
## Introduction
Health-promoting behaviors benefit cardiovascular wellbeing [1], prevent and manage cancer- and treatment-related side effects and complications [2], and prevent recurring cancer [3]. The American Cancer Society (ACS) and the National Comprehensive Cancer Network (NCCN) recommend improving cancer survivors’ health outcomes by promoting healthy behaviors during survivorship [4, 5]. Specifically, cancer survivors should engage in regular physical activity, maintain a healthy weight, achieve a healthy dietary pattern, and not smoke [4]. Overall adherence to these recommendations has been disappointing among cancer survivors [6]: 9–$10\%$ of cancer survivors are current smokers, less than half meet the physical activity recommendation [7], and about $30\%$ are obese [8].
Social networks and social support are related to health behavior change [9], which can ultimately influence health outcomes [10]. Marriage can provide social support that influences an individual’s health behaviors and related outcomes. As Waite indicated, “Marriage provides individuals—especially men—with someone who monitors their health and health-related behaviors and who encourages self-regulation” [11]. Marriage may mobilize social and emotional support to encourage health-promoting behaviors [10] and can provide meaning, purpose, and obligation, which may inhibit risky behaviors and promote healthy behaviors [12]. Those who are married are more likely to be non-smokers [13], consume more vegetables [14], and participate in exercise more than singles [15]. However, husbands and wives (couples) are more likely to be obese than singles [16]. Methodologically, research has often grouped participants into married vs. not married or single; thus, researchers cannot differentiate the effects of divorced, widowed, separated, and never married on health behaviors and related outcomes. In the context of cancer, research has shown that cancer impacts the physical and psychological well-being of both the survivor and their family members, particularly their spouses [7]. However, how cancer survivors’ health behaviors are related to their marital status (i.e., married, divorced, widowed, separated, and never married) has rarely been examined.
Furthermore, survivors manage cancer and related issues in a context—e.g., cancer types [17] and sociodemographic factors such as sex [17], age [18], race [19], education [20], and geographic location [21]—all of which influence health behaviors. However, little is known about whether these factors have the same influence on the health behaviors of cancer survivors who have different marital statuses. To fill these gaps, this study aimed to examine whether cancer survivors’ health behaviors varied by marital status and whether the health behaviors-marital status relationship differed by the cancer type and sociodemographic factors. Different from the general population, cancer survivorship creates a series of unique teachable moments for cancer survivors and their families when they frequently interact with healthcare professionals for treatment decision-making, symptom management, surveillance for cancer recurrence and second cancers, and management of their comorbid conditions [4, 5]. *Knowledge* generated from this study is critical for developing and implementing effective, targeted intervention programs to promote health behaviors among cancer survivors and their family members.
## Study population
We analyzed the Medical Expenditure Panel Survey (MEPS) (https://www.meps.ahrq.gov/mepsweb/) data between 2011 and 2017. MEPS is a set of large-scale questionnaires administered to noninstitutionalized populations in the USA. Yabroff et al. have reported detailed information on MEPS methodology previously [22]. The overall response rates for MEPS 2011–2017 ranged from 44.2 to $56.3\%$ [23]. We identified individuals as adult cancer survivors if they were ≥ 18 years old and responded “yes” to the question, “Have you ever been told by a doctor or other health professional that you had cancer or a malignancy of any kind?” Respondents were included in the analysis if they were diagnosed with prostate, breast, or colon cancer—three of the most common types of solid tumors in the USA [24].
## Measurement
Health behaviors assessed included smoking and physical activity. We obtained information on these behaviors based on a “yes” response to the survey items “currently smoke” and “currently spend half hour or more in moderate to vigorous physical activity (e.g., brisk walking, dancing, running, and fast swimming) at least five times a week” [25]. Obesity status was assessed according to BMI cutpoints outlined by the Centers for Disease Control and Prevention [26]: underweight (BMI < 18.5), normal (BMI = 18.5–24.9), overweight (BMI = 25.0–29.9), and obese (BMI ≥ 30.0) (Note: cancer survivors who were underweight were $1.12\%$ of the respondents, and thus, excluded from the analysis due to a small sample size).
Marital status was categorized in the survey data as married, widowed, divorced/separated, and never married. We obtained survivors’ demographic characteristics from the MEPS household component, including age, sex, race/ethnicity, education, family income, and insurance coverage [27]. We categorized race (White vs non-White), education (< high school degree or General Educational Development (GED), high school or GED, and > high school or GED), and insurance status (private, public, uninsured). Based on the federal poverty guideline, we classified family income into five categories: poor (< $100\%$); near poor (100– < $125\%$); low (125– < $200\%$); middle (200– < $400\%$); and high (≥ $400\%$) income. Data about the time from cancer diagnosis, cancer status, and treatment status are not available in the survey thus, are not included in the analyses.
## Data analysis
Using the Rao-Scott design-adjusted chi-square test—a design-adjusted version of the Pearson chi-square test [28], we calculated weighted percentages (SAS Proc surveyfreq) of survivor health behaviors and compared health behaviors among cancer survivors with different marital status. To examine whether the relationship between health behaviors and marital status differed by cancer type, we conducted weighted multivariable logistic regressions (SAS Proc surveylogistic) and modeled cancer type, marital status, and the cancer type*marital status interaction on health behaviors. We used a weighted regression approach with personal weights to account for MEPS’ design complexity. To examine whether sociodemographic factors influenced the relationship between health behaviors and marital status, we added survivors’ characteristics into the models obtained in the previous steps. We performed stepwise variable selection to obtain parsimonious models.
## Results
We identified 1880 individuals who reported being diagnosed with breast ($50.8\%$), prostate ($35.0\%$), and colon ($14.2\%$) cancers. The mean age was 68.7 and a majority were female ($57.4\%$); were White ($84.4\%$); had a greater than high school degree or GED education ($57.7\%$); were middle or high income ($25.5\%$ and $44.8\%$, respectively); and had private health insurance ($64.0\%$). These cancer survivors reported their marital status as married ($59.3\%$), widowed ($19.3\%$), divorced/separated ($16.0\%$), and never married ($5.5\%$) (Table 1).Table 1Sample characteristicsCharacteristicsN%Sex Female$107557.38\%$ Male$80542.62\%$Race White$133684.38\%$ Non-Whitea$54215.62\%$Hispanic Yes$2446.07\%$Poverty statusb Poor$28311.47\%$ Near poor$1024.67\%$ Low$30113.49\%$ Middle$51225.55\%$ High$68244.82\%$*Marital status* Married$102259.28\%$ Widowed$39219.25\%$ Divorced/separated$33915.97\%$ Never married$1275.50\%$Education ≤ 12th grade$33211.50\%$ GED or high school degree$57430.79\%$ > high school degree$95757.71\%$Insurance coverage Private$108864.02\%$ Publicc$76134.71\%$ Uninsured$311.27\%$Census region Northeast$33818.33\%$ Midwest$39522.64\%$ South$72537.83\%$ West$42221.20\%$Cancer type Breast$93150.80\%$ Prostate$65434.96\%$ Colon$29514.24\%$MeanSDAge68.720.36aNon-White included Black, American Indian/Alaska Native, Asian, Native Hawaiian/Pacific Islander, and multiple racesbFamily income was classified into five poverty categories: poor (< $100\%$), near poor (100– < $125\%$), low (125– < $200\%$), middle (200– < $400\%$), and high income (≥ $400\%$)cPersons identified as covered by public insurance are those reporting coverage under TRICARE, Medicare, Medicaid or SCHIP, or other public hospital/physician programs
## Health behaviors and marital status
Approximately $18.1\%$ of survivors were current smokers, and $41.4\%$ reported currently engaging in half an hour or more of moderate to vigorous physical activity at least five times a week. Approximately $32.6\%$ and $37.0\%$ of the survivors were overweight or obese, respectively. Current non-smoking behavior was significantly related to marital status ($p \leq 0.0001$). Widowed survivors were the least likely to be a current smoker, while never married were the most likely to currently smoke. Obesity was significantly related to marital status ($p \leq 0.01$): married survivors were the most likely to be overweight or obese, whereas widowed survivors were the most likely to have a normal weight (Table 2).Table 2The association between health behaviors and marital statusHealth behaviorsTotalMarriedWidowedDivorced/separatedNever marriedp valuecN%N%N%N%N%Currently non-smoke (yes/no)$154081.91\%$$86093.42\%$$33493.52\%$$25385.94\%$$9378.36\%$ <.0001Physical activity (yes/no)$77841.38\%$$44946.65\%$$13639.74\%$$14544.97\%$$4840.27\%$0.188BMI Normal$52127.71\%$$28830.92\%$$12136.94\%$$7924.51\%$$3331.37\%$0.0033 Overweight (BMI = 25.0–29.9)$61332.61\%$$29927.35\%$$12329.47\%$$14240.20\%$$4928.17\%$ Obese (BMI ≥ 30.0)$69637.02\%$$41140.63\%$$13131.43\%$$11234.45\%$$4238.84\%$cThe p values are based on Rao-Scott design-adjusted chi-square test to examine the association between health behaviors and marital status
## Health behaviors, marital status, and cancer type
The association between BMI and marital status differed by cancer type. Widowed colon cancer survivors had the lowest odds ratio (0.24) ($95\%$ CI [0.11–0.53]) of being obese or overweight, and divorced/separated colon cancer survivors had the highest odds ratio (1.18) ($95\%$ CI [0.60–2.31]) of being obese or overweight (Table 3).Table 3Health behaviors and marital status in the context of different cancer typesHealth behaviorsCurrent non-smoker(Ref: current smoker)Adequate physical activity(Ref: no physical activity)Increased or high BMI(Ref: normal BMI)βSEOdds ratio$95\%$ CIβSEOdds ratio$95\%$ CIβSEOdds ratio$95\%$ CIMarital status (ref: married) Divorced/separated − 1.100.510.33*0.12–0.90 − 0.740.380.480.22–1.010.170.341.180.60–2.31 Widowed − 0.780.650.460.13–1.62 − 0.670.470.510.20–1.29 − 1.410.390.24***0.11–0.53 Never married − 0.620.850.540.10–2.88 − 0.580.60.560.17–1.81 − 0.390.510.680.25–1.84Cancer type (ref: colon) Breast0.490.451.640.68–3.96 − 0.250.260.780.47–1.28 − 0.540.230.58*0.37–0.90 Prostate0.280.451.330.54–3.24 − 0.110.240.890.56–1.42 − 0.350.220.700.46–1.08Marital status * cancer type (ref: married * colon) Widowed*breast cancer1.070.762.930.65–13.100.440.521.550.56–4.331.730.455.64***2.34–13.63 Widowed*prostate cancer1.040.872.840.52–15.570.570.591.770.56–5.601.140.483.14*1.22–8.04 Divorced/separated*breast cancer0.350.661.410.39–5.130.750.442.110.88–5.030.470.411.600.71–3.55 Divorced/separated*prostate cancer0.290.641.330.38–4.650.880.482.420.95–6.170.210.461.230.50–3.01 Never married*breast cancer − 1.130.970.320.05–2.150.370.721.450.35–6.000.350.61.430.44–4.61 Never married*prostate cancer − 0.451.000.640.09–4.570.440.731.560.38–6.450.690.612.000.60–6.66*$p \leq .05$; **$p \leq .01$; ***$p \leq .001$
## Health behaviors and marital status in the context of sociodemographic factors
The results of the full and parsimonious models are displayed in the Online Appendix Table. In the parsimonious model that excluded the nonsignificant factors (Fig. 1A), the never married survivors had the highest odds ratio of being a smoker among all people with different marital status ($p \leq 0.01$), indicating never married survivors were the most likely to smoke currently. Current non-smoking behavior also differed by age, sex, and education. Older survivors were less likely to smoke currently ($p \leq 0.01$). Compared to females, males with cancer were more likely to smoke ($p \leq 0.01$). Survivors with a high school/GED or lower educational status were more likely to be current smokers as compared to survivors with more advanced education (both ps < 0.01).Fig. 1Influencing factors of smoking and physical activity behaviors Physical activity among survivors (Fig. 1B) differed by sex, education, and poverty status. Compared to male survivors, women were less likely to be physically active ($p \leq 0.05$). Survivors with a high school degree or GED were less likely to be active physically than those with a more than high school/GED education ($p \leq 0.01$). Survivors with low income were less likely to be physically active compared to those with high income (both ps < 0.05).
The relationships between BMI and marital status continued to be significant after controlling for the effects of cancer type and sociodemographic factors (Fig. 2A). Among survivors with different types of cancer, widowed colon cancer survivors had the lowest odds ratio (0.32) ($95\%$ CI [0.15, 0.68]) of being obese or overweight, and never married prostate cancer survivors had the highest odds ratio (1.09) ($95\%$ CI [0.76–7.74]) of being obese or overweight. Figure 2B shows that BMI differed by age, race, and education. Older survivors were less likely to have a normal BMI ($p \leq 0.001$). Non-White survivors were more likely to be overweight and obese than their White counterparts ($p \leq 0.01$). Survivors with a high school/GED or lower educational status were more likely to be overweight or obese than those with a higher than high school/GED education (both ps < 0.05).Fig. 2Influencing factors of BMI
## Discussion
To our knowledge, this study using MEPS data is the first to examine how selected health promoting and adverse health behaviors are associated with marital status (i.e., married, widowed, divorced/separated, and never married) among survivors with different types of cancer in the context of their sociodemographic factors. Current smoking behavior and BMI (a proxy for obesity) were related to marital status. Among patients with prostate, breast, and colon cancer, those who were never married had higher rates of smoking. Divorced/separated survivors were the most likely to be overweight, married survivors were the most likely to be obese, and those who were widowed were the most likely to have normal weight. The relationship between obesity and marital status varied by cancer type. We also identified disparities in health behaviors among cancer survivors by age, sex, race, education, and income. The results from this study inform development and implementation of tailored interventions to enhance healthy behaviors among cancer survivors with varying sociodemographic backgrounds.
Although cancer diagnosis and treatment offer survivors and their families the opportunity (e.g., education, skills training) to create healthy behavior change and promote positive outcomes, our findings indicate that some cancer survivors continue to engage in unhealthy behaviors. Approximately $18\%$ of survivors of breast, colorectal, and prostate cancer in this MEPS study were current smokers as compared to $20.8\%$ in the US general population [29]. Findings from other research that included survivors with breast, cervical, colorectal, and prostate cancer reported higher rates than our study [21]. About $41\%$ of cancer survivors reported that they spend at least half an hour in moderate to vigorous physical activity more than five times a week, meeting the physical activity recommendation of ACS [4]. This finding is within the range reported by LeMasters et al. ( 30.3–$46.6\%$) [17] but lower than that from a state-specific samples of randomly dialed telephone survey (~ $78\%$) using the Behavioral Risk Factor Surveillance System (BRFSS) [21]. The prevalence of overweight ($32.61\%$) and obese ($37.02\%$) survivors is similar to the general American adult population (overweight and obese > $70\%$) [26] but higher than previously reported estimates using data from the BRFSS [21] and the National Health Interview Survey (NHIS) [8]. Our findings suggest that smoking cessation, physical activity engagement, and, most importantly, weight loss remain challenging in cancer survivors. Researchers and healthcare providers can take better advantage of the teachable moments of cancer diagnosis, treatment, and follow-up visits to clearly communicate with cancer survivors about health behaviors and engage them in effective programs to promote positive outcomes.
We found that the never married survivors, regardless of cancer types, were most likely to smoke. This finding is different from that of the general population as recently noted in a Morbidity and Mortality Weekly Report [29], which showed that the current smoking rate was the highest among adults who were divorced, separated, or widowed, followed by adults who were single, never married, or not living with a partner; and the lowest among those who were married or living with a partner. The high prevalence of smoking among never married cancer survivors may be related to a lack of influence, support, and social control over risky behaviors from spouses. Marriage can influence health behaviors directly through sanctioning or impeding and indirectly through internalizing norms about the behaviors [30]. Furthermore, compared to unpartnered people, adults who had a non-smoking partner or whose partners quit smoking were more likely to quit smoking [31]. These findings emphasize that never-married cancer survivors may need additional support to quit smoking.
We also found that current smoking status differed by survivors’ age, sex, and educational attainment. Specifically, those who were male, younger, and whose educational status was lower than high school were more likely to smoke. Findings about the relationships between smoking status, age, and gender have been mixed; however, the inverse association between smoking prevalence and educational attainment has been consistent. A review of studies of the general public reported that men used tobacco products at higher rates than women, and the significant gender differences in smoking are prevalent among younger adults but absent among older smokers [32]. In contrast, an earlier study of US cancer survivors using NHIS data found that smoking status differed significantly by age and age at diagnosis among men and women and that females had higher rates of being a current smoker than males, particularly among those 40 years of age or younger [8]. Social, economic, personal, and political influences all impact smoking prevalence and cessation [33]. For example, a smoker’s age is related to the stage of the smoking cessation process, and thus, smoking cessation programs might be improved by matching intervention strategies to a smoker’s age and their stage of readiness [34]. Our findings indicate that, although being diagnosed with cancer may motivate people to quit smoking, many factors can impede smoking cessation, including social and personal factors. Our results may contribute to development of tailored smoking cessation interventions based on cancer survivors’ marital status and their sociodemographic backgrounds. Cancer survivors who are male, younger, and have less education may benefit from smoking cessation strategies that meet their unique needs.
We also found that obesity was related to marital status, even after considering the effects of sociodemographic factors. Married survivors were the most likely to be obese, while widowed survivors were the most likely to have a normal weight. Among all cancer survivors with different marital status and types of cancer, widowed colon cancer survivors were least likely to be obese or overweight, and divorced/separated colon cancer survivors were most likely to be obese or overweight. This finding is consistent with research on the general population [35]. Marriage may provide role obligations for eating regular meals [36]. Conversely, a spouse’s death is a stressful life event that may cause loss of appetite and regular meals. A lack of support, influence, and social control over risky behaviors (e.g., overeating) from spouses may be related to overweight and obesity among divorced/separated colorectal survivors [30]. Our findings highlight the need for interventions that encourage both cancer survivors and their spouses to establish healthy eating patterns and reduce obesity and for tailored interventions to target survivors with different types of cancer and with different marital status, especially among colon cancer patients.
Although the association between physical activity and marital status was non-significant, our findings show that physical activity differed by survivors’ educational attainment and family income. Specifically, survivors with high school or lower education and those with poor and low family income were more likely to be physically inactive. Our finding is congruent with the results from an ACS’ study ($$n = 1160$$) of patients with breast, colorectal, and prostate cancer, which also found that physically inactive survivors were more likely to have lower education (≤ high school) and household income [37]. This disparity in physical activity by educational status has been well documented [38]. High levels of education provide individuals with increased knowledge of the benefits of physical activity, greater access to resources, and healthier influences from their social networks, which all facilitate physical activity [39]. Similarly, individuals who have higher incomes have more resources and locations to exercise, which facilitates physical activity [40]. Effective interventions are needed to increase physical activity for cancer survivors with low socioeconomic status.
The limitations of our study are as follows. First, we used the marital status reported at one time point, making it impossible to assess marital status as a time-varying variable. We, therefore, could not investigate whether changes in marital status affected health behaviors. In the large-scale MEPS surveys of family and individuals, there are valid discrepancies in the case of persons who are married but not living with their spouse, separated but cohabitating, or unmarried partners living together (MEPS considers them as separate family units) [41], all of which largely limited our ability to tease out the information about how living arrangements are related to health behaviors among cancer survivors with different marital status. Our study using MEPS also has low percentage of never married survivors ($5.5\%$), which may be caused by the fact that cancer survivors skew older, and thus, have low never married proportion. Next, as several studies have indicated [42, 43], publicly available MEPS data do not include the time of cancer diagnosis; therefore, we could not examine the association between time since diagnosis and health behaviors tested about a decade ago [17]. And also, we used respondents’ self-report data, which may reduce the precision of estimation of health behaviors. Finally, the generalizability of our findings may be reduced because most respondents had a greater than high school education, middle to high income, and private health insurance, which is higher compared to cancer survivors using the Behavioral Risk Factor Surveillance System dataset [44].
Nonetheless, this study using MEPS extends the current research on the sociodemographic findings related to health behaviors in the general population to provide insight into how health behaviors are related to marital status among cancer survivors. Our findings suggest that interventions need to be tailored based on survivors’ marital status and sociodemographic characteristics in order to effectively promote healthy behaviors and to improve survivorship outcomes. Future research must develop and evaluate feasible supportive care interventions, especially for never-married cancer survivors to quit smoking and couple-based interventions for married survivors to reduce their weight and obesity. In addition, given that marital status is associated with smoking and BMI for cancer survivors, future research needs to address mechanisms that account for these associations, e.g., using dyadic analysis to understand the interpersonal influences and outcomes regarding health behaviors. Additionally, while this study grouped survivors based on their answer to a simple question of marital status, both married and unmarried survivors could have good or poor social support, living arrangement (living together or separately), and different time since cancer diagnosis. Future research must collect more comprehensive data to verify our study findings as well as to examine each of the subgroups—i.e., married vs unmarried with good vs poor support, living arrangement, and time since diagnosis —to further understand how social support influences health behaviors of individuals who have different marital status, living circumstance during the continuum of cancer survivorship. Future research should also include survivors who are partnered (vs married) so that the findings are more generalizable. Research using marital status as a time-varying variable may also help extrapolate how behaviors change over time. Lastly, the majority of the sample were White ($84.4\%$), and future studies should focus on recruiting and retaining people of color to investigate how marital status impacts health behaviors. As social determinants play a critical role in shaping health behaviors, research is also needed to identify strategies to promote health behaviors among disadvantaged cancer survivors (i.e., rural communities, LGBTQIA).
## Conclusions
We used the MEPS data to examine relationships between health behaviors, sociodemographic factors, and marital status among cancer survivors with prostate, breast, and colon cancers. Our findings suggest that relationship status and sociodemographic factors need to be considered in tailoring interventions to improve health behaviors among cancer survivors.
## Supplementary information
Below is the link to the electronic supplementary material. Supplementary file1 (DOCX 28.6 KB)
## References
1. McCullough ML, Patel AV, Kushi LH. **Following cancer prevention guidelines reduces risk of cancer, cardiovascular disease, and all-cause mortality**. *Cancer Epidemiol Biomarkers Prev* (2011.0) **20** 1089-1097. DOI: 10.1158/1055-9965.EPI-10-1173
2. Phillips SM, Awick EA, Conroy DE, Pellegrini CA, Mailey EL, McAuley E. **Objectively measured physical activity and sedentary behavior and quality of life indicators in survivors of breast cancer**. *Cancer* (2015.0) **121** 4044-4052. DOI: 10.1002/cncr.29620
3. Davies NJ, Batehup L, Thomas R. **The role of diet and physical activity in breast, colorectal, and prostate cancer survivorship: a review of the literature**. *Br J Cancer* (2011.0) **105** S52-73. DOI: 10.1038/bjc.2011.423
4. Rock CL, Doyle C, Demark-Wahnefried W. **Nutrition and physical activity guidelines for cancer survivors**. *CA Cancer J Clin* (2012.0) **62** 243-274. DOI: 10.3322/caac.21142
5. 5.National Comprehensive Cancer Network. Smoking cessation, version 1.2016, NCCN clinical practice guidelines in oncology. J Natl Compr Canc Netw. 2016;14(11):1430–1462.
6. Tollosa DN, Tavener M, Hure A, James EL. **Adherence to multiple health behaviours in cancer survivors: a systematic review and meta-analysis**. *J Cancer Surviv* (2019.0) **13** 327-343. DOI: 10.1007/s11764-019-00754-0
7. Song L, Guan T, Guo P, Keyserling T, Van Houtven C, Tan X. **Prevalence of cardiovascular disease and risk factors, quality of life, and health behaviors of cancer survivors and their spouses: findings from MEPS**. *J Cancer Surviv* (2019.0) **13** 739-748. DOI: 10.1007/s11764-019-00792-8
8. Shoemaker ML, White MC, Hawkins NA, Hayes NS. **Prevalence of smoking and obesity among U.S cancer survivors: estimates from the national health interview survey, 2008–2012**. *Oncol. Nurs. Forum.* (2016.0) **43** 436-441. DOI: 10.1188/16.ONF.43-04AP
9. Harvey IS, Alexander K. **Perceived social support and preventive health behavioral outcomes among older women**. *J Cross Cult Gerontol* (2012.0) **27** 275-290. DOI: 10.1007/s10823-012-9172-3
10. Reblin M, Uchino BN. **Social and emotional support and its implication for health**. *Curr Opin Psychiatry* (2008.0) **21** 201-205. DOI: 10.1097/YCO.0b013e3282f3ad89
11. Waite LJ. **Does marriage matter?**. *Demography* (1995.0) **32** 483-507. DOI: 10.2307/2061670
12. Gove WR. **Sex, marital status, and mortality**. *Am J Sociol* (1973.0) **79** 45-67. DOI: 10.1086/225505
13. Schoeppe S, Vandelanotte C, Rebar AL, Hayman M, Duncan MJ, Alley SJ. **Do singles or couples live healthier lifestyles? Trends in Queensland between 2005–2014**. *PLoS One* (2018.0) **13** e0192584. DOI: 10.1371/journal.pone.0192584
14. Eng PM, Kawachi I, Fitzmaurice G, Rimm EB. **Effects of marital transitions on changes in dietary and other health behaviours in US male health professionals**. *J Epidemiol Community Health* (2005.0) **59** 56-62. DOI: 10.1136/jech.2004.020073
15. Pettee KK, Brach JS, Kriska AM. **Influence of marital status on physical activity levels among older adults**. *Med Sci Sports Exerc* (2006.0) **38** 541-546. DOI: 10.1249/01.mss.0000191346.95244.f7
16. Teachman J. **Body Weight, marital status, and changes in marital status**. *J Fam Issues* (2016.0) **37** 74-96. DOI: 10.1177/0192513X13508404
17. LeMasters TJ, Madhavan SS, Sambamoorthi U, Kurian S. **Health behaviors among breast, prostate, and colorectal cancer survivors: a US population-based case-control study, with comparisons by cancer type and gender**. *J Cancer Surviv* (2014.0) **8** 336-348. DOI: 10.1007/s11764-014-0347-5
18. Bellizzi KM, Rowland JH, Jeffery DD, McNeel T. **Health behaviors of cancer survivors: examining opportunities for cancer control intervention**. *J Clin Oncol* (2005.0) **23** 8884-8893. DOI: 10.1200/JCO.2005.02.2343
19. O'Neill SC, DeFrank JT, Vegella P. **Engaging in health behaviors to lower risk for breast cancer recurrence**. *PLoS ONE* (2013.0) **8** e53607. DOI: 10.1371/journal.pone.0053607
20. Bluethmann SM, Basen-Engquist K, Vernon SW. **Grasping the ‘teachable moment’: time since diagnosis, symptom burden and health behaviors in breast, colorectal and prostate cancer survivors**. *Psychooncology* (2015.0) **24** 1250-1257. DOI: 10.1002/pon.3857
21. **Surveillance of demographic characteristics and health behaviors among adult cancer survivors—Behavioral Risk Factor Surveillance System, United States, 2009**. *MMWR* (2012.0) **61** 1-23
22. Yabroff KR, Dowling E, Rodriguez J. **The Medical Expenditure Panel Survey (MEPS) experiences with cancer survivorship supplement**. *J Cancer Surviv* (2012.0) **6** 407-419. DOI: 10.1007/s11764-012-0221-2
23. 23.Agency for Healthcare Research and Quality. MEPS-HC response rates by panel. Retrieved from https://meps.ahrq.gov/survey_comp/hc_response_rate.jsp. Accessed 6 Jan 2019.
24. 24.National Cancer Institution. Common cancer types. Retrived from https://www.cancer.gov/types/common-cancers. 2020. Accessed 6 Jan 2019.
25. 25.World Health Organizaition. What is moderate-intensity and vigorous-intensity physical activity? Retreived from https://www.who.int/dietphysicalactivity/physical_activity_intensity/en/. n.d. Accessed 6 Jan 2019.
26. 26.Centers for Disease Control and Prevention. Obesity and overweight. Retrevied from https://www.cdc.gov/nchs/fastats/obesity-overweight.htm. 2017. Accessed 6 Jan 2019.
27. 27.Agency for Healthcare Research and Quality. MEPS HC-138: 2010 full year consolidated data file. Retreived from https://meps.ahrq.gov/data_stats/download_data/pufs/h138/h138doc.shtml#254Income. 2012. Accessed 6 Jan 2019.
28. Rao JN, Scott AJ. **The analysis of categorical data from complex sample surveys: chi-squared tests for goodness of fit and independence in two-way tables**. *J Am Stat Assoc* (1981.0) **76** 221-230. DOI: 10.1080/01621459.1981.10477633
29. Cornelius ME, Wang TW, Jamal A, Loretan CG, Neff LJ. **Tobacco product use among adults — United States, 2019**. *MMWR Morb Mortal Wkly Rep* (2020.0) **69** 1736-1742. DOI: 10.15585/mmwr.mm6946a4
30. 30.Umberson D. Family status and health behaviors: social control as a dimension of social integration. J Health Soc Behav. 1987:306–319.
31. Margolis R, Wright L. **Better off alone than with a smoker: the influence of partner’s smoking behavior in later life**. *J Gerontol B Psychol Sci Soc Sci* (2016.0) **71** 687-697. DOI: 10.1093/geronb/gbu220
32. Higgins ST, Kurti AN, Redner R. **A literature review on prevalence of gender differences and intersections with other vulnerabilities to tobacco use in the United States, 2004–2014**. *Prev Med* (2015.0) **80** 89-100. DOI: 10.1016/j.ypmed.2015.06.009
33. Jarvis MJ. **Why people smoke**. *BMJ* (2004.0) **328** 277-279. DOI: 10.1136/bmj.328.7434.277
34. Clark MA, Rakowski W, Kviz FJ, Hogan JW. **Age and stage of readiness for smoking cessation**. *J Gerontol B Psychol Sci Soc Sci* (1997.0) **52** S212-S221. DOI: 10.1093/geronb/52B.4.S212
35. The NS, Gordon-Larsen P. **Entry into romantic partnership is associated with obesity**. *Obesity* (2009.0) **17** 1441-1447. DOI: 10.1038/oby.2009.97
36. Sobal J, Rauschenbach B, Frongillo EA. **Marital status changes and body weight changes: a US longitudinal analysis**. *Soc Sci Med* (2003.0) **56** 1543-1555. DOI: 10.1016/S0277-9536(02)00155-7
37. 37.Troeschel AN, Leach CR, Shuval K, Stein KD, Patel AV. Physical activity in cancer survivors during “re-entry” following cancer treatment. Prev. Chronic Dis. 2018;15.
38. Harper S, Lynch J. **Trends in socioeconomic inequalities in adult health behaviors among US states 1990–2004**. *Public Health Rep* (2007.0) **122** 177-189. DOI: 10.1177/003335490712200207
39. Shaw BA, Spokane LS. **Examining the association between education level and physical activity changes during early old age**. *J Aging Health* (2008.0) **20** 767-787. DOI: 10.1177/0898264308321081
40. Shuval K, Li Q, Gabriel KP, Tchernis R. **Income, physical activity, sedentary behavior, and the ‘weekend warrior’ among US adults**. *Prev Med.* (2017.0) **103** 91-97. DOI: 10.1016/j.ypmed.2017.07.033
41. 41.Agency for Healthcare Research and Quality. MEPS HC-201: 2017 full year consolidated data file. Retrieved from https://meps.ahrq.gov/data_stats/download_data/pufs/h201/h201doc.shtml. Accessed 6 Jan 2019
42. Vyas A, Babcock Z, Kogut S. **Impact of depression treatment on health-related quality of life among adults with cancer and depression: a population-level analysis**. *J Cancer Surviv* (2017.0) **11** 624-633. DOI: 10.1007/s11764-017-0635-y
43. Rim SH, Yabroff KR, Dasari S, Han X, Litzelman K, Ekwueme DU. **Preventive care service use among cancer survivors with serious psychological distress: an analysis of the medical expenditure panel survey data**. *Prev Med* (2019.0) **123** 152-159. DOI: 10.1016/j.ypmed.2019.03.024
44. Han X, Jemal A, Zheng Z, Sauer AG, Fedewa S, Yabroff KR. **Changes in noninsurance and care unaffordability among cancer survivors following the Affordable Care Act**. *J Natl Cancer Inst* (2020.0) **112** 688-697. DOI: 10.1093/jnci/djz218
|
---
title: 'Risk of kidney disease following a pregnancy complicated by diabetes: a longitudinal,
population-based data-linkage study among Aboriginal women in the Northern Territory,
Australia'
authors:
- Matthew J. L. Hare
- Louise J. Maple-Brown
- Jonathan E. Shaw
- Jacqueline A. Boyle
- Paul D. Lawton
- Elizabeth L. M. Barr
- Steven Guthridge
- Vanya Webster
- Denella Hampton
- Gurmeet Singh
- Roland F. Dyck
- Federica Barzi
journal: Diabetologia
year: 2023
pmcid: PMC10036460
doi: 10.1007/s00125-023-05868-w
license: CC BY 4.0
---
# Risk of kidney disease following a pregnancy complicated by diabetes: a longitudinal, population-based data-linkage study among Aboriginal women in the Northern Territory, Australia
## Abstract
### Aims/hypothesis
The aim of this work was to investigate the risk of developing chronic kidney disease (CKD) or end-stage kidney disease (ESKD) following a pregnancy complicated by gestational diabetes mellitus (GDM) or pre-existing diabetes among Aboriginal women in the Northern Territory (NT), Australia.
### Methods
We undertook a longitudinal study of linked healthcare datasets. All Aboriginal women who gave birth between 2000 and 2016 were eligible for inclusion. Diabetes status in the index pregnancy was as recorded in the NT Perinatal Data Collection. Outcomes included any stage of CKD and ESKD as defined by ICD-10 coding in the NT Hospital Inpatient Activity dataset between 2000 and 2018. Risk was compared using Cox proportional hazards regression.
### Results
Among 10,508 Aboriginal women, the mean age was 23.1 (SD 6.1) years; 731 ($7.0\%$) had GDM and 239 ($2.3\%$) had pre-existing diabetes in pregnancy. Median follow-up was 12.1 years. Compared with women with no diabetes during pregnancy, women with GDM had increased risk of CKD ($9.2\%$ vs $2.2\%$, adjusted HR 5.2 [$95\%$ CI 3.9, 7.1]) and ESKD ($2.4\%$ vs $0.4\%$, adjusted HR 10.8 [$95\%$ CI 5.6, 20.8]). Among women with pre-existing diabetes in pregnancy, $29.1\%$ developed CKD (adjusted HR 10.9 [$95\%$ CI 7.7, 15.4]) and $9.9\%$ developed ESKD (adjusted HR 28.0 [$95\%$ CI 13.4, 58.6]).
### Conclusions/interpretation
Aboriginal women in the NT with GDM or pre-existing diabetes during pregnancy are at high risk of developing CKD and ESKD. Pregnancy presents an important opportunity to identify kidney disease risk. Strategies to prevent kidney disease and address the social determinants of health are needed.
### Supplementary Information
The online version contains peer-reviewed but unedited supplementary material available at 10.1007/s00125-023-05868-w.
## Introduction
Indigenous populations impacted by colonisation are disproportionately affected by diabetes and related cardiometabolic conditions [1]. The prevalence of type 2 diabetes in Aboriginal people in the Northern Territory (NT), Australia, is among the highest of any population globally [2]. Onset is occurring at increasingly young ages, especially in women [3]. Rates of both gestational diabetes mellitus (GDM) and pre-existing type 2 diabetes in pregnancy are growing rapidly [4]. Of concern, Aboriginal women in the NT have 12.8 years shorter life expectancy than non-Aboriginal women [5], with diabetes being the most commonly attributed cause of death [6].
The burden of chronic kidney disease (CKD) for Aboriginal people in the NT is among the highest reported [7]; one-third of adults display biochemical evidence of CKD [8]. Furthermore, the age-standardised prevalence of end-stage kidney disease (ESKD) is estimated to be $1.8\%$ for Aboriginal people compared with $0.1\%$ for non-Aboriginal people [9]. Diabetes is a key contributing factor but there are multiple determinants with differing impacts across the life course [7, 10]. There is evidence that intergenerational risk factors, such as maternal nutrition and smoking, affect the intrauterine environment, leading to a reduction in nephron number from birth and greater susceptibility to CKD [11].
The prevalence of CKD is highest in remote and socioeconomically disadvantaged areas, where provision of renal replacement therapy is immensely challenging [7]. Aboriginal women in the NT have higher prevalence and incidence of ESKD than men [9]. The determinants of this difference are not known. Women are also less likely to access renal replacement therapy [9], possibly relating to reluctance to relocate hundreds of kilometres away from home due to family and community responsibilities [12].
Research into sex-specific risk factors for diabetes and kidney disease is warranted. Hypertensive disorders of pregnancy are known to predict future CKD and ESKD [13]. However, few studies have investigated the risk of subsequent CKD following a pregnancy complicated by diabetes [13–17]. Given near universal healthcare engagement and routine diabetes screening, pregnancy and the postpartum period present valuable opportunities for targeted prevention strategies.
The NT spans 1.35 million km2 (more than twice the size of France) but has a population density of just 0.2 people/km2 [18]. Almost one-third of the population identify as Aboriginal people and most live in remote communities [18]. Aboriginal people have lived in this region for more than 60,000 years. Traditional diet and lifestyle of many people have changed but great strength of culture persists, with more than 100 languages and dialects spoken [19].
In this context, we investigated risk of developing kidney disease following a pregnancy complicated by GDM or pre-existing diabetes among Aboriginal women.
## Study design and population
A longitudinal, population-based study was undertaken using linked healthcare datasets. All women, between 1 July 2000 and 31 December 2016, who identified as an Aboriginal person, gave birth at ≥20 weeks gestation and usually resided in the NT were eligible for inclusion. Women who gave birth at the one private hospital in the region were excluded as data could not be linked. Figure 1 summarises the cohort formation. For women with more than one pregnancy during the study period, only the earliest (‘index’) pregnancy was included. Fig. 1Study population flow chart showing inclusion and exclusion criteria *Baseline data* were extracted from the NT Perinatal Data Collection (NT PDC), a population-based register of all births, including non-hospital births. Investigators had access to the full database population and applied the inclusion and exclusion criteria to form the study population. The NT PDC contains detailed information about demographics, maternal health, antenatal care and perinatal outcomes. Follow-up outcome data between 1 July 2000 and 30 June 2018 were from the NT Hospital Inpatient Activity database. This provided ICD-10 (http://apps.who.int/classifications/icd10/browse/2016/en), Australian Modification (ICD-10-AM) codes for principal and secondary diagnoses for all admissions at every public (government-funded) hospital in the NT. Universal free access to hospital care is available in Australia. Loss to follow-up could not be identified in the study data. Nevertheless, the out-migration rate for Aboriginal people in the NT is low, with only $6\%$ of the population moving away over a 5 year period [20].
Individual-level records were deterministically linked using a de-identified linkage key derived from each individual’s unique hospital reference number. These reference numbers have been used across all public health services in the NT since the early 1990s and have been reliably used in previous data-linkage studies [9]. In this study, there was $100\%$ concordance in the sex of participants between the datasets and $99.0\%$ of dates of birth matched to the day, indicating highly accurate linkage.
## Diabetes in pregnancy
Diabetes status during the index pregnancy was as recorded in the NT PDC, including GDM and pre-existing diabetes (type 1, type 2, or other diabetes). Data on the type of pre-existing diabetes was only available for 2014–2016. Women without recorded GDM or pre-existing diabetes were presumed to have no diabetes in the index pregnancy. This clinical information was entered by referring clinicians, usually midwives, immediately after a birth and thus represents the clinically known diabetes status during the pregnancy. To ensure capture of known diabetes status, the NT PDC has additionally been centrally cross-referenced against hospital files from the birth admission since 2008 and, from 2014, reporting of diabetes has been strengthened by data from the NT Diabetes in Pregnancy Clinical Register [21]. Some cases of diabetes will have been undiagnosed due to incomplete uptake of screening. Universal screening for GDM was recommended throughout the study period but screening approaches and diagnostic criteria did change (electronic supplementary material [ESM] Table 1). Glucose thresholds for the diagnosis of type 2 diabetes were unchanged over the study period, but HbA1c, at a threshold of ≥48 mmol/mol (≥$6.5\%$), was adopted as a diagnostic option from 2012 [22].
## Ethnicity
First Nations peoples in Australia include both Aboriginal and Torres Strait Islander peoples. In the NT, $96\%$ of the First Nations population identify as Aboriginal people [23]. To avoid misinterpretation regarding the generalisability of this study, the cohort was restricted to Aboriginal women specifically. Data from the NT PDC did not differentiate between Aboriginal and/or Torres Strait Islander identification. This more detailed information was available in the hospital data. Therefore, following data linkage, women identifying as Torres Strait Islander or both Aboriginal and Torres Strait Islander people were excluded.
## Other baseline variable definitions
Age was calculated at the day women gave birth. Usual location of residence at the time of the index pregnancy was recorded as free text in the NT PDC. The 438 different place names were manually designated into statistical areas of the Australian Statistical Geography Standard (ASGS), including Statistical Areas Level 1 (SA1s) and Indigenous Areas (IAREs) [24]. Using each individual’s SA1 code, remoteness was classified according to standard remoteness areas using the Accessibility/Remoteness Index of Australia (ARIA+), which is based on road distance from a location to the nearest urban centre [25]. Three out of five remoteness levels exist in the NT: outer regional; remote; and very remote. Area-level socioeconomic status was classified using the Indigenous Relative Socioeconomic Outcomes (IRSEO) index according to IAREs [26]. The IRSEO index is calculated using data from the 2016 Australian Census relating to the usual resident Aboriginal and/or Torres Strait Islander population of an area. It draws upon nine variables, covering employment, education, income and housing, to create a percentile score of relative socioeconomic outcomes. Scores range from 1 (most advantaged) to 100 (most disadvantaged). The region variable classified women as being from the ‘Top End’ or ‘Central Australia’. The Aboriginal population of the NT encompasses numerous distinct people groups. These regions are known to have marked differences in health outcomes [2]. Smoking status during the index pregnancy was missing for 1430 ($13.6\%$) women. Many of these women had a subsequent pregnancy within the study period. For each woman missing smoking status in the index pregnancy, smoking status was assumed to be the same as in her most recent pregnancy. Following this imputation, smoking data were missing for 434 ($4.1\%$) women. History of kidney disease pre-dating the index pregnancy was as recorded in the NT PDC or if a woman met the study’s outcome definitions in the linked hospital data prior to the index pregnancy. History of GDM in a previous pregnancy (previous GDM) was as recorded by clinicians in the NT PDC during the index pregnancy. Hypertensive disorders of pregnancy were as recorded in the NT PDC and included hypertension, pre-eclampsia and eclampsia arising in pregnancy or labour.
## Outcome definitions
Outcomes were captured from hospital admission coding. The two key outcomes were any CKD and ESKD. Any CKD incorporated ICD-10-AM codes for any stage of CKD or dialysis. ESKD included coding for stage 5 CKD or dialysis. Rates of type 2 diabetes among women with a history of GDM, who developed CKD or ESKD, were also explored. Analyses relating to incident type 2 diabetes were exploratory in nature only due to limited capture of events when relying on hospitalisation data. Most diagnoses of type 2 diabetes are made in the outpatient, primary care setting. For details of ICD-10-AM codes, see ESM Table 2. Date of death was obtained from hospital discharge data. Out-of-hospital deaths were not captured.
## Statistical analysis
Baseline descriptive data are presented as n (%), mean ± SD or median (IQR) as appropriate. Comparisons across categories of diabetes status during the index pregnancy were made using χ2 tests, one-way ANOVA and the non-parametric equality of medians test. The numbers of outcome events and crude rates (%) are presented by categories of diabetes status.
Kaplan–Meier estimates were used to calculate cumulative incidence of the pre-specified outcomes with $95\%$ CIs. The risk of developing each outcome was compared using Cox proportional hazards models. Diabetes status in the index pregnancy was included as a three-level categorical variable, with both exposure categories assessed in the same regression models and compared against no known diabetes. Time origin was from the baby’s date of birth. Women with pre-existing kidney disease ($$n = 243$$) were excluded. Subsequently, women missing data for smoking ($$n = 419$$) and/or socioeconomic status ($$n = 159$$) were also excluded. The date of the first event for each outcome was taken as the date of admission to hospital for the earliest admission in which a relevant ICD-10-AM code was recorded. Women were censored at the end of the follow-up period (30 June 2018) or on the date of death. The assumption of proportional hazards was verified by plotting the log of negative log of the survival function against the log of time and also by plotting the observed Kaplan–Meier survival curves against the Cox predicted curves. Crude and adjusted HRs were calculated. Variables included in the adjusted Cox models were chosen based on the plausibility of a clinically relevant relationship and assessment of associations with CKD in univariable or minimally adjusted models. The final adjusted model for the primary analysis included age, hypertensive disorder of pregnancy, smoking status, region and socioeconomic status (IRSEO score as a continuous variable). Available but excluded variables were parity, which had no association with CKD after accounting for maternal age, and multiple pregnancy (twin or triplet) and alcohol intake in pregnancy, neither of which had an association with CKD in univariable or multivariable models. Separate Cox models were used to examine associations between each secondary exposure and CKD and ESKD, with inclusion of potential confounders specific to each exposure [27].
Due to existing evidence showing that the relative risk of end-organ complications associated with having diabetes lessens with age [28, 29], it was pre-specified to look for an interaction between diabetes status and age using an interaction term in Cox models for the two key outcomes. Sensitivity analyses were undertaken to investigate any potential impact of excluding women with missing smoking data. Two opposite, extreme scenarios were applied to the missing data. First, it was assumed that women with diabetes smoked during pregnancy and women without diabetes did not smoke. Second, it was assumed that women with diabetes did not smoke during pregnancy and women without diabetes did smoke.
All analyses were conducted in Stata (V17.0; StataCorp, USA).
## Ethics and governance
The study was approved by the Human Research Ethics Committee of NT Health and Menzies School of Health Research, including the Aboriginal Ethics Sub-Committee (Ref: 2018-3069), with reciprocal approval from the Central Australian Human Research Ethics Committee (Ref: CA-19-3412). In addition, the Aboriginal and Torres Strait Islander Advisory Group of the Diabetes across the Lifecourse: Northern Australia Partnership supported the planned analyses and provided feedback on results. Authors VW and DH represent the Advisory Group in this publication. The Advisory Group will play an ongoing role in guiding knowledge translation.
## Results
In total, 10,508 Aboriginal women were included in the study (see Fig. 1). Mean age was 23.1 (SD 6.1) years, 731 ($7.0\%$) women had GDM, 239 ($2.3\%$) had pre-existing diabetes in pregnancy and 3451 ($32.8\%$) were from Central Australia (rather than the Top End). Women with GDM or pre-existing diabetes were older, less likely to smoke and more likely to have a hypertensive disorder of pregnancy (Table 1). Pre-eclampsia accounted for most of the hypertensive disorders ($$n = 750$$, $72.9\%$). Pre-existing kidney disease in the index pregnancy was more common among women with pre-existing diabetes (n/$$n = 36$$/239, $15.1\%$) compared with women with GDM (n/$$n = 13$$/731, $1.8\%$) or no known diabetes (n/$$n = 194$$/9538, $2.0\%$). After a median (IQR) follow-up time of 12.1 (7.2–15.8) years, 334 ($3.3\%$) women developed CKD, 70 ($0.7\%$) developed ESKD and 54 ($0.5\%$) died. Table 1Characteristics of the study cohort stratified by diabetes status during the index pregnancyCharacteristicNNo known diabetes ($$n = 9538$$)Gestational diabetes ($$n = 731$$)Pre-existing diabetes ($$n = 239$$)p valueBaseline characteristics Age, years10,50822.6 ± 5.826.6 ± 6.731.5 ± 6.9<0.001 Central Australia region10,5083101 (32.5)238 (32.6)112 (46.9)<0.001 Remoteness10,4450.380 Outer regional1384 (14.6)94 (13.0)27 (11.3) Remote2046 (21.6)163 (22.5)59 (24.7) Very remote6051 (63.8)468 (64.6)153 (64.0) Socioeconomic index (IRSEO)10,33890.1 (77.0–95.3)90.4 (77.0–95.3)91.9 (79.2–96.1)0.110 Smoking during pregnancy10,0744416 (48.2)320 (46.3)87 (40.1)0.043 Pre-existing kidney disease10,508194 (2.0)13 (1.8)36 (15.1)<0.001 Previous GDM10,50829 (0.3)165 (22.6)24 (10.0)<0.001 Hypertensive disorder of pregnancy10,508880 (9.2)101 (13.8)48 (20.1)<0.001 Nulliparity10,5086129 (64.3)402 (55.0)78 (32.6)<0.001Follow-up information Follow-up, years10,50812.3 (7.4–15.9)10.4 (5.0–14.4)11.2 (6.5–16.1)<0.001 No. of hospitalisations10,5084 (2–6)4 (2–8)8 (4–16)<0.001Data are shown as n (%), mean ± SD or median (IQR) Compared with women without diabetes during their index pregnancy, the incidence of any CKD and ESKD during follow-up was higher for women with GDM and highest for women with pre-existing diabetes in pregnancy (Fig. 2, Table 2). After adjustment for potential confounding factors, there was strong evidence for increased risk of any CKD among women with GDM (HR 5.2 [$95\%$ CI 3.9, 7.1]) and women with pre-existing diabetes in pregnancy (HR 10.9 [$95\%$ CI 7.7, 15.5]) compared with women without diabetes in pregnancy. An even greater elevation in risk was seen for ESKD with both GDM (HR 10.8 [$95\%$ CI 5.6, 20.8]) and pre-existing diabetes in pregnancy (HR 28.0 [$95\%$ CI 13.4, 58.6]). Among women with GDM, 62 ($93.9\%$) of those who developed any CKD and 16 ($94.1\%$) of those who developed ESKD were known to have progressed to type 2 diabetes prior to onset of kidney disease. In 2014–2016, when data on pre-existing diabetes type were available, all 36 women with pre-existing diabetes in pregnancy had type 2 diabetes. Fig. 2Cumulative incidence ($95\%$ CI) of any CKD (a) and ESKD (b) according to diabetes status during the index pregnancy of Aboriginal women in the NT, AustraliaTable 2Absolute and relative risk of future kidney disease according to diabetes status during index pregnancy among Aboriginal women in NT, AustraliaKidney diseaseEvent rates, n (%)Crude HRvs no diabetes during index pregnancy ($95\%$ CI)Adjusted HRvs no diabetes during index pregnancy ($95\%$ CI)aNo known diabetes ($$n = 9344$$)GDM($$n = 718$$)Pre-existing diabetes ($$n = 203$$)GDMPre-existing diabetesGDMPre-existing diabetesAny CKD209 (2.2)66 (9.2)59 (29.1)6.1 (4.5, 8.1)17.2 (12.6, 23.4)5.2 (3.9, 7.1)10.9 (7.7, 15.4)ESKD33 (0.4)17 (2.4)20 (9.9)11.2 (6.0, 20.8)35.7 (19.2, 66.7)10.8 (5.6, 20.8)28.0 (13.4, 58.6)Women missing data for smoking ($$n = 419$$) and socioeconomic status ($$n = 159$$) were excluded from both the crude and adjusted Cox models ($$n = 9687$$ included)aAdjusted for maternal age, hypertensive disorder of pregnancy, socioeconomic status, region and smoking status during pregnancy In separate Cox regression analyses designed to assess the effects of secondary exposure variables, hypertensive disorders of pregnancy, maternal age, socioeconomic disadvantage (IRSEO score), residing in a very remote location and being from the Central Australia region were associated with increased risk of CKD (ESM Table 3). Maternal age, socioeconomic disadvantage and being from Central Australia were also associated with ESKD (ESM Table 4).
Clinical and demographic characteristics of women excluded due to missing data vs those included in the Cox regression models are presented in ESM Table 5. Excluded women were older, more likely to have diabetes, and more likely to live in a very remote area. Nevertheless, sensitivity analyses suggested no impact of their exclusion on the associations of GDM and pre-existing diabetes with future CKD or ESKD (ESM Table 6).
The relative risk inferred by having a history of pre-existing diabetes in pregnancy lessened with age for both CKD ($$p \leq 0.004$$ for interaction) and ESKD ($p \leq 0.001$ for interaction). Event rates and HRs among women with pre-existing diabetes in pregnancy, stratified by age, are presented in ESM Table 7. There were no significant interactions between age and GDM status.
## Discussion
In this relatively young cohort of parous Aboriginal women with a long duration of follow-up, there was strong evidence that both GDM and pre-existing diabetes (predominantly type 2 diabetes) in pregnancy are associated with high risk of future CKD and ESKD. Other risk factors identified for future CKD included age, hypertensive disorders of pregnancy, living in remote areas, socioeconomic disadvantage and being from the Central Australia region. While the relative risk of kidney disease compared with women without diabetes during pregnancy was remarkably high, the absolute event rates are equally concerning, with almost one in ten women with GDM developing CKD and a similar proportion of women with pre-existing diabetes developing ESKD.
This is the first study to investigate long-term kidney disease outcomes after diabetes in pregnancy among Aboriginal women in Australia. The findings suggest potential modifiable risk factors for kidney disease that are specific to parous women. Compared with Aboriginal men, women have a higher prevalence of ESKD ($2.1\%$ in women vs $1.5\%$ in men) and an accelerated decline in eGFR over time has been observed [9, 10]. This disparity may in part relate to dysglycaemia, which can be detected relatively early in the life course, during or prior to pregnancy.
In this study, almost all of the women with GDM, who later developed kidney disease, progressed to having type 2 diabetes prior to kidney disease onset. Therefore, the risk associated with GDM was not shown to be independent of future type 2 diabetes. Nevertheless, GDM provides an early risk marker that could potentially enable targeted screening and prevention activities. A previous data-linkage study from Ontario demonstrated increased risk of ESKD among women with GDM who subsequently developed type 2 diabetes [16]. Similar to our study, the relative risk in that cohort, compared with women with no history of GDM and no subsequent type 2 diabetes, was very high (HR 7.5 [$95\%$ CI 5.2, 10.8]). Another study examined associations between GDM and future CKD and ESKD in Swedish national registry data [17]. The relative risks among women with GDM and subsequent type 2 diabetes, compared with women without GDM, were remarkably high for CKD (adjusted HR 21.7 [$95\%$ CI 17.2, 27.4]) and ESKD (adjusted HR 112.4 [$95\%$ CI 61.2, 206.4]). In keeping with our findings, women with GDM who did not progress to type 2 diabetes were not at increased risk of kidney disease [17].
These Canadian and Swedish studies were not included in a 2020 systematic review, which did not find any studies showing an association between GDM and future ESKD [13]. With regard to earlier stages of CKD, a prospective study from the USA found that GDM was predictive of future albuminuria over 21 years of follow-up among Black women but not White women [15]. Similarly, in a cross-sectional analysis from the Kidney Early Evaluation Program (KEEP) in the USA, self-reported history of GDM without subsequent type 2 diabetes was associated with risk of microalbuminuria but not later stages of CKD [14]. When stratified by ethnicity, evidence for this association was present among Black participants but not White participants. The existence of similarities between minority groups in the USA and our study population likely reflect shared risks relating to socioeconomic disadvantage. The stronger associations with kidney disease observed in our study are presumably driven by higher rates of progression to chronic impaired glucose regulation among Aboriginal women in the NT. A prospective cohort study in the NT showed that within just 2.5 years’ follow-up postpartum, $22\%$ of Aboriginal women with GDM had developed type 2 diabetes and an additional $11\%$ had intermediate hyperglycaemia (HbA1c ≥$6.0\%$, impaired glucose tolerance or impaired fasting glycaemia) [30].
The high incidence of CKD and ESKD in this relatively young cohort with pre-existing diabetes during pregnancy is consistent with studies of young-onset type 2 diabetes [31]. The TODAY (Treatment Options for Type 2 Diabetes in Adolescents and Youth) trial enrolled participants with type 2 diabetes aged 10–17 years [32]. Over 15 years, the cumulative incidence of moderate or severe albuminuria was $54.7\%$. A relatively high incidence of kidney disease in individuals with young-onset type 2 diabetes has been observed in other Indigenous populations, including Akimel O’odham (Pima) people from Arizona, USA and First Nations people from Saskatchewan and Manitoba, Canada [33–35].
The additional risk factors associated with CKD in our study are consistent with existing evidence but should be interpreted with caution given that diabetes status in pregnancy was the primary exposure of interest. Our findings emphasise the importance of considering the social determinants of health and equity of access to culturally appropriate health services.
Strengths of our study include its whole-of-population approach, large sample size and duration of follow-up in a high-priority, marginalised population group. The accurate data-linkage methodology facilitated valuable research in a context where traditional epidemiological studies are difficult to undertake for numerous reasons, including the vast distances between many small population centres. Additionally, we were able to look at robust clinical outcomes rather than just surrogate markers of risk. Importantly, the study is embedded in a larger programme of work with strong Aboriginal representation and governance processes as well as meaningful collaborative partnerships that help facilitate translation of research findings into policy and practice [36].
The study has several limitations, largely relating to the retrospective approach and use of existing datasets. The classification of diabetes status during pregnancy was limited to known diagnoses and assessed categorically without any glycaemic measures available. There were changes in the screening and diagnosis of GDM during the study period. However, the proportionality of the HRs over time suggests that the risk inferred by a diabetes diagnosis at different time points across the study period was consistent. While there is universal access to free healthcare in Australia, uptake of GDM screening will not have been complete. There will have been some women with undiagnosed diabetes in the reference group, contributing to likely underestimation of risks associated with both GDM and pre-existing diabetes in pregnancy. Recording of kidney disease that pre-dated the index pregnancy, especially early CKD, may also have been incomplete. Our findings do not show that having a pregnancy complicated by diabetes is causative of heightened CKD risk, but rather that diabetes detected either during or prior to pregnancy strongly predicts future development of CKD. The same outcomes could potentially be seen among nulliparous women with young-onset diabetes.
The use of hospital coding data for outcome capture is highly reliable for ESKD, as dialysis initiation services are provided by the public hospital system and hospitalisation rates are high in people with ESKD irrespective of dialysis use. However, detection of early stages of CKD and future type 2 diabetes will not have been complete. Early CKD and type 2 diabetes are diagnoses predominantly made in the primary care setting, so capture from hospital data is largely dependent on women being admitted for another reason. The diagnosis of CKD is more likely to have been documented for later stages of CKD due to the clinical relevance to an acute admission. There is also potential for detection bias with regard to CKD and type 2 diabetes because the risk of hospitalisation may vary between the diabetes in pregnancy exposure groups. However, the median number of hospitalisations during the study period was the same for women with GDM and women with no diabetes in pregnancy. The median number of hospitalisations was greater for women who had pre-existing diabetes during the index pregnancy, which could have increased detection of CKD diagnoses among this group. Nevertheless, the consistency of the findings relating to CKD and ESKD (ESKD being reliably detected in hospital data) suggest that the associations observed with CKD are true. Mortality data were also reliant on hospital records. It is possible that some women may have died without being hospitalised. Given the young age of the cohort, it is unlikely that capturing out-of-hospital deaths would alter the findings.
Another limitation is the potential for unmeasured confounding in our analyses, in particular from obesity but also other known risk factors such as recurrent infections and family history of kidney disease. Existing evidence suggests that cumulative exposure to GDM in more than one pregnancy is associated with greater risk of CKD [37]. Due to the design of our linked study database, we were unable to reliably investigate this hypothesis. Finally, our findings are only generalisable to the Aboriginal population of the NT but may be of relevance to the broader Aboriginal population of Australia and other Indigenous populations globally with high rates of diabetes and kidney disease.
In conclusion, GDM and pre-existing diabetes during pregnancy are strongly associated with increased risk of developing CKD and ESKD among Aboriginal women in the NT of Australia. Pregnancy presents an important opportunity to identify kidney disease risk. Our findings also highlight the importance of ongoing screening for type 2 diabetes after a pregnancy complicated by GDM. Culturally appropriate strategies to prevent kidney disease in this high-risk population should be investigated alongside public health strategies that address the social determinants of health and improve access to healthcare services.
## Supplementary information
ESM(PDF 705 kb)
## Authors’ relationships and activities
Outside the submitted work, MJLH has received honoraria for lectures and consultancies from AstraZeneca, Eli Lilly and Novo Nordisk and JES has received honoraria for lectures and consultancies from AstraZeneca, Eli Lilly, Mylan, Novo Nordisk, Pfizer, Sanofi, Merck Sharp and Dohme, Abbott and Boehringer Ingelheim. All other authors declare that there are no relationships or activities that might bias, or be perceived to bias, their work.
## Contribution statement
MJLH contributed to ethics applications and data acquisition, designed and undertook the analyses, and wrote the manuscript. FB led funding, ethics and data access applications. FB, LJMB, SG and RFD conceived the project. FB, LJMB, JES and JAB supervised the design and interpretation of the analyses. PDL, SG, ELMB and GS contributed to the study design and assisted in interpreting the data. VW and DH coordinated intellectual input from the Aboriginal and Torres Strait Islander Advisory Group. All authors critically revised the manuscript for important intellectual content and approved the final version. MJLH is the guarantor of this work and had full access to the data and takes responsibility for the integrity of the data and accuracy of analysis.
## References
1. Harris SB, Tompkins JW, TeHiwi B. **Call to action: a new path for improving diabetes care for Indigenous peoples, a global review**. *Diabetes Res Clin Pract* (2017.0) **123** 120-133. DOI: 10.1016/j.diabres.2016.11.022
2. Hare MJL, Zhao Y, Guthridge S. **The prevalence and incidence of diabetes among Aboriginal people in remote communities of the Northern Territory, Australia: a retrospective, longitudinal data-linkage study**. *BMJ Open* (2022.0) **12** e059716. DOI: 10.1136/bmjopen-2021-059716
3. Titmuss A, Davis EA, O'Donnell V. **Youth-onset type 2 diabetes among First Nations young people in northern Australia: a retrospective, cross-sectional study**. *Lancet Diabetes Endocrinol* (2022.0) **10** 11-13. DOI: 10.1016/S2213-8587(21)00286-2
4. Hare MJL, Barzi F, Boyle JA. **Diabetes during pregnancy and birthweight trends among Aboriginal and non-Aboriginal people in the Northern Territory of Australia over 30 years**. *Lancet Reg Health West Pac* (2020.0) **1** 100005. DOI: 10.1016/j.lanwpc.2020.100005
5. 5.Department of the Prime Minister and Cabinet (2020) Closing the Gap Report. Commonwealth of Australia, Canberra. Available from https://ctgreport.niaa.gov.au/. Accessed 20 May 2022
6. 6.Australian Bureau of Statistics (2021) 3303.0 Causes of Death, Australia, 2020. Commonwealth of Australia, Canberra
7. Hoy WE, Mott SA, Mc Donald SP. **An expanded nationwide view of chronic kidney disease in Aboriginal Australians**. *Nephrology* (2016.0) **21** 916-922. DOI: 10.1111/nep.12798
8. 8.Australian Institute of Health and WelfareProfiles of Aboriginal and Torres Strait Islander people with kidney disease2020CanberraAustralian Government. *Profiles of Aboriginal and Torres Strait Islander people with kidney disease* (2020.0)
9. Li L, Guthridge S, Li SQ, Zhao Y, Lawton P, Cass A. **Estimating the total prevalence and incidence of end-stage kidney disease among Aboriginal and non-Aboriginal populations in the Northern Territory of Australia, using multiple data sources**. *BMC Nephrol* (2018.0) **19** 15. DOI: 10.1186/s12882-017-0791-3
10. Maple-Brown LJ, Hughes JT, Ritte R. **Progression of kidney disease in indigenous Australians: the eGFR follow-up study**. *Clin J Am Soc Nephrol* (2016.0) **11** 993-1004. DOI: 10.2215/CJN.09770915
11. Hoy WE, Hughson MD, Singh GR, Douglas-Denton R, Bertram JF. **Reduced nephron number and glomerulomegaly in Australian Aborigines: a group at high risk for renal disease and hypertension**. *Kidney Int* (2006.0) **70** 104-110. DOI: 10.1038/sj.ki.5000397
12. 12.Gorham G (2003) Prevention and treatment options for renal disease in the Northern Territory (with particular reference to the Barkly region). Cooperative Research Centre for Aboriginal and Tropical Health, Darwin
13. Barrett PM, McCarthy FP, Kublickiene K. **Adverse pregnancy outcomes and long-term maternal kidney disease: a systematic review and meta-analysis**. *JAMA Netw Open* (2020.0) **3** e1920964. DOI: 10.1001/jamanetworkopen.2019.20964
14. Bomback AS, Rekhtman Y, Whaley-Connell AT. **Gestational diabetes mellitus alone in the absence of subsequent diabetes is associated with microalbuminuria: results from the Kidney Early Evaluation Program (KEEP)**. *Diabetes Care* (2010.0) **33** 2586-2591. DOI: 10.2337/dc10-1095
15. Dehmer EW, Phadnis MA, Gunderson EP. **Association between gestational diabetes and incident maternal CKD: the Coronary Artery Risk Development in Young Adults (CARDIA) study**. *Am J Kidney Dis* (2018.0) **71** 112-122. DOI: 10.1053/j.ajkd.2017.08.015
16. 16.Retnakaran R, Shah BR (2017) Role of type 2 diabetes in determining retinal, renal, and cardiovascular outcomes in women with previous gestational diabetes mellitus. Diabetes Care 40(1):101–108. 10.2337/dc16-1400
17. 17.Barrett PM, McCarthy FP, Evans M et al (2022) Does gestational diabetes increase the risk of maternal kidney disease? A Swedish national cohort study. PLoS ONE 17(3):e0264992. 10.1371/journal.pone.0264992
18. 18.Northern Territory Government Department of Treasury and Finance (2022) Northern Territory Economy. Northern Territory Government, Darwin. Available from https://nteconomy.nt.gov.au/population. Accessed 20 May 2022
19. 19.Northern Territory Government Aboriginal Interpreter Service (2022) Aboriginal languages in NT. Northern Territory Government, Darwin. Available from https://nt.gov.au/community/interpreting-and-translating-services/aboriginal-interpreter-service/aboriginal-languages-in-nt. Accessed 20 May 2022
20. 20.Taylor A, Dyrting S (2019) Northern Territory Contemporary Indigenous Migration Trends. Northern Institute, Darwin
21. Kirkham R, Whitbread C, Connors C. **Implementation of a diabetes in pregnancy clinical register in a complex setting: Findings from a process evaluation**. *PLoS One* (2017.0) **12** e0179487. DOI: 10.1371/journal.pone.0179487
22. D’Emden MC, Shaw JE, Colman PG. **The role of HbA1c in the diagnosis of diabetes mellitus in Australia**. *Med J Aus* (2012.0) **197** 220-221. DOI: 10.5694/mja12.10988
23. 23.Australian Bureau of Statistics (2017) 2075.0 Census of Population and Housing - Counts of Aboriginal and Torres Strait Islander Australians, 2016. Commonwealth of Australia, Canberra
24. 24.Australian Bureau of StatisticsAustralian Statistical Geography Standard (ASGS) Edition 32021CanberraCommonwealth of Australia. *Australian Statistical Geography Standard (ASGS) Edition 3* (2021.0)
25. 25.Australian Bureau of Statistics (2018) 1270.0.55.005 Australian Statistical Geography Standard (ASGS): Volume 5 - Remoteness Structure, July 2016. Commonwealth of Australia, Canberra
26. 26.Biddle N, Markham F (2017) Areal Level Socioeconomic Outcomes for Aboriginal and Torres Strait Islander Australians, 2016. Austaxpolicy: Tax and Transfer Policy Blog. Available from: https://www.austaxpolicy.com/area-level-socioecononomic-outcomes-aboriginal-torres-strait-islander-australians-2016. Accessed 4 Jan 2022
27. Westreich D, Greenland S. **The table 2 fallacy: presenting and interpreting confounder and modifier coefficients**. *Am J Epi* (2013.0) **177** 292-298. DOI: 10.1093/aje/kws412
28. Hoffmann F, Haastert B, Koch M, Giani G, Glaeske G, Icks A. **The effect of diabetes on incidence and mortality in end-stage renal disease in Germany**. *Nephrol Dial Transplant* (2011.0) **26** 1634-1640. DOI: 10.1093/ndt/gfq609
29. Singh GM, Danaei G, Farzadfar F. **The age-specific effects of metabolic risk factors on cardiovascular diseases and diabetes: a pooled analysis**. *PLoS ONE* (2013.0) **8** e65174. DOI: 10.1371/journal.pone.0065174
30. Wood AJ, Boyle JA, Barr ELM. **Type 2 diabetes after a pregnancy with gestational diabetes among first nations women in Australia: The PANDORA study**. *Diabetes Res Clin Pract* (2021.0) **181** 109092. DOI: 10.1016/j.diabres.2021.109092
31. Magliano DJ, Sacre JW, Harding JL, Gregg EW, Zimmet PZ, Shaw JE. **Young-onset type 2 diabetes mellitus - implications for morbidity and mortality**. *Nat Rev Endocrinol* (2020.0) **16** 321-331. DOI: 10.1038/s41574-020-0334-z
32. **Long-term complications in youth-onset type 2 diabetes**. *N Engl J Med* (2021.0) **385** 416-426. DOI: 10.1056/NEJMoa2100165
33. Pavkov ME, Bennett PH, Knowler WC, Krakoff J, Sievers ML, Nelson RG. **Effect of youth-onset type 2 diabetes mellitus on incidence of end-stage renal disease and mortality in young and middle-aged Pima Indians**. *JAMA* (2006.0) **296** 421-426. DOI: 10.1001/jama.296.4.421
34. Dyck RF, Jiang Y, Osgood ND. **The long-term risks of end stage renal disease and mortality among First Nations and non-First Nations people with youth-onset diabetes**. *Can J Diabetes* (2014.0) **38** 237-243. DOI: 10.1016/j.jcjd.2014.03.005
35. Dart AB, Sellers EA, Martens PJ, Rigatto C, Brownell MD, Dean HJ. **High burden of kidney disease in youth-onset type 2 diabetes**. *Diabetes Care* (2012.0) **35** 1265-1271. DOI: 10.2337/dc11-2312
36. Maple-Brown LJ, Graham S, McKee J, Wicklow B. **Walking the path together: incorporating Indigenous knowledge in diabetes research**. *Lancet Diabetes Endocrinol* (2020.0) **8** 559-560. DOI: 10.1016/S2213-8587(20)30188-1
37. Beharier O, Shoham-Vardi I, Pariente G. **Gestational diabetes mellitus is a significant risk factor for long-term maternal renal disesae**. *J Clin Endocrinol Metab* (2015.0) **100** 1412-1416. DOI: 10.1210/jc.2014-4474
|
---
title: Estradiol-mediated protection against high-fat diet induced anxiety and obesity
is associated with changes in the gut microbiota in female mice
authors:
- Kalpana D. Acharya
- Madeline Graham
- Harshini Raman
- Abigail E. R. Parakoyi
- Alexis Corcoran
- Merzu Belete
- Bharath Ramaswamy
- Shashikant Koul
- Ishneet Sachar
- Kevin Derendorf
- Jeremy B. Wilmer
- Srikanth Gottipati
- Marc J. Tetel
journal: Scientific Reports
year: 2023
pmcid: PMC10036463
doi: 10.1038/s41598-023-31783-6
license: CC BY 4.0
---
# Estradiol-mediated protection against high-fat diet induced anxiety and obesity is associated with changes in the gut microbiota in female mice
## Abstract
Decreased estrogens during menopause are associated with increased risk of anxiety, depression, type 2 diabetes and obesity. Similarly, depleting estrogens in rodents by ovariectomy, combined with a high-fat diet (HFD), increases anxiety and adiposity. How estrogens and diet interact to affect anxiety and metabolism is poorly understood. Mounting evidence indicates that gut microbiota influence anxiety and metabolism. Here, we investigated the effects of estradiol (E) and HFD on anxiety, metabolism, and their correlation with changes in gut microbiota in female mice. Adult C57BL/6J mice were ovariectomized, implanted with E or vehicle-containing capsules and fed a standard diet or HFD. Anxiety-like behavior was assessed and neuronal activation was measured by c-fos immunoreactivity throughout the brain using iDISCO. HFD increased anxiety-like behavior, while E reduced this HFD-dependent anxiogenic effect. Interestingly, E decreased neuronal activation in brain regions involved in anxiety and metabolism. E treatment also altered gut microbes, a subset of which were associated with anxiety-like behavior. These findings provide insight into gut microbiota-based therapies for anxiety and metabolic disorders associated with declining estrogens in menopausal women.
## Introduction
Estrogens have profound effects on energy homeostasis in humans and rodents by acting as an anorectic, preventing fat weight gain, and increasing physical activity1–5. Lower levels of circulating estrogens in postmenopausal women increase their risk for obesity, type 2 diabetes, cardiovascular disease, and stroke2,6–9. In rodents, ovariectomy decreases physical activity and increases food intake3,10, while 17β-estradiol (E) treatment in ovariectomized mice fed a HFD prevents weight gain11–14, suggesting that E protects against HFD-induced obesity.
In addition to their effects on energy homeostasis, estrogens also exert effects on mood and anxiety in women15,16. Postmenopausal women experience an increased rate of depressive and anxious mood17–20, particularly during the onset of menopause, that is ameliorated by hormone replacement therapy21–23. Similarly, ovariectomy in rodents increases depressive- and anxiety-like behavior24–27, while physiological doses of E decrease anxiety-like behavior in ovariectomized rodents, indicating that estrogens have anxiolytic effects in females28,29. Estrogen receptor (ER)-specific effects have been demonstrated using ER\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha$$\end{document}α and ER \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\beta$$\end{document}β knockout mice or receptor-specific agonists, which have shown ER\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\beta$$\end{document}β as a primary mediator of anxiety behavior in females30–32.
The lower gastrointestinal (GI) tract is inhabited by a collection of bacteria, viruses, archaea, protozoa, and fungi, known as the gut microbiota33. These microbes, along with their genomes, comprise the gut microbiome33. Mounting evidence indicates that the gut microbiome is integral in maintaining healthy physiology in humans and rodents34–36. Dysbiosis of these microbiota has been implicated in metabolic diseases, including obesity37 and type 2 diabetes38,39. Gut microbiota also primes the development of the innate immune system and host immune response to pathogenic bacteria, and in turn modulate the production of cytokines and lymphokines40. Gut immune alterations mediated by bacteria and their metabolites exert effects on the central nervous system via the gut-brain axis41,42. Dysbiosis of the gut microbiota is associated with anxiety43–45 and depression46–48 in laboratory animals and humans. In male rodents, differences in stress sensitivity across strains have been linked to distinct changes in gut microbiota-dependent stress-induced changes in lipid and energy metabolism49,50, although these effects have not been explored in females. However, in a different study, female mice challenged with a HFD and chronic mild stress showed less anxiety and a different gut microbial composition compared to males51. Collectively, these findings suggest that the gut microbiome profoundly impacts anxiety and metabolism.
Estrogens and diet are independently associated with changes in gut microbiota52–55. In rats and mice, ovariectomy shifts the relative abundances of the major phyla, increasing the ratio of Firmicutes to Bacteroidetes. A high relative abundance of Firmicutes relative to *Bacteroidetes is* associated with metabolic disorders56–58. HFD and Western diet (high in fat and sucrose) feeding also alter gut microbiota composition, including changes in relative abundances of Bacteroidetes and Firmicutes in both humans and rodents59–62. While HFD-induced alterations in the gut microbiota are associated with changes in anxiety-like and depression-like behavior in male mice38,63, it is unknown if E- and diet-induced changes in gut microbiota similarly alter these behaviors in adult female mice.
The aim of this study was to investigate the effects of E and HFD on anxiety-like behavior and energy metabolism and associations with changes in the gut microbiota in female mice. We tested the hypotheses that: [1] E treatment reduces HFD-induced anxiety-like behavior, [2] E alters activity in anxiety-relevant brain regions following stress in female mice on a HFD, and [3] E and HFD-mediated changes in anxiety are associated with alterations in gut microbiota composition.
## Estradiol prevents HFD-induced obesity and reduces food intake
Eight-week-old C57BL/6J female mice, fed a standard diet (SD) or HFD, were implanted with either 50 μg of 17β-estradiol (E) or oil (V) (Supplementary Fig. 1). Body weight and food intake were measured every three days. Both E and V mice were cohoused with either similar partners (E–E, E-treated mice cohoused with E-treated mice, or V–V) or with mice from a different treatment group (E–V, V–E). Because there were no effects of cohousing on anxiety-like behavior or body weight, cohoused groups were collapsed and effects of diet (SD vs. HFD) and hormone treatment (V vs. E) were analyzed. For the days prior to the diet switch, a one-way repeated measures (RM) ANOVA showed an effect of E (F1,62 = 4.88, $$p \leq 0.03$$) on body weight gain. Following the diet switch, using a two-way RM ANOVA, we found main effects of hormone treatment (F1,57 = 61, $p \leq 0.001$), diet (F1,57 = 89.6, $p \leq 0.001$) and their interaction (F1,57 = 24, $p \leq 0.001$) over time during HFD. During SD feeding, E mice weighed more, starting on D4 ($$p \leq 0.018$$) and continued to weigh more on D7 ($p \leq 0.001$) and D10 ($$p \leq 0.009$$), likely due to a faster recovery mediated by E following surgery. On D13, within each hormone condition, HFD groups differed from their SD counterparts. The HFD-V animals weighed more than all other groups starting from day 16 (Tukey’s HSD, $p \leq 0.001$) (Fig. 1A). The SD-V animals weighed more than the SD-E animals on days 22, 25, and 28 (Tukey’s HSD, $p \leq 0.05$). The HFD-E mice weighed more than the SD-E animals on days 22, 25, 28, and 36 (Tukey’s HSD, $p \leq 0.05$).Figure 1Estradiol prevents HFD-induced obesity and reduces HFD intake in female mice. ( A) E prevented average body weight gain in mice on a HFD ($$n = 16$$/group). ( B) E decreased caloric intake per cage in mice on a HFD ($$n = 4$$ cages/group). Larger circles show mean (± SEM) and smaller circles show individual data points. * Indicate days when HFD-V mice differ from all other groups. $P \leq 0.05$, Tukey’s HSD.
During the 10 days on SD, one-way RM ANOVA showed an effect of E (F1,14 = 5.34, $$p \leq 0.037$$) on food intake. In particular, E mice ate less than V mice on D10. After switching to HFD, a two-way RM ANOVA found main effects on food intake for diet (F1,12 = 83.9, $p \leq 0.001$), hormone treatment (F1,12 = 51.3, $p \leq 0.001$), and an interaction between the two (F1,12 = 24.2, $p \leq 0.001$). Following the switch to HFD on D10, HFD-V animals consumed the most calories (Tukey’s HSD, $p \leq 0.05$) except on D31, when food intake for HFD-V showed a strong trend towards an increase compared to HFD-E ($$p \leq 0.052$$) and D34 (Fig. 1B). HFD-E animals ate a similar amount to SD groups throughout the study, except on D13, when they ate more than the SD groups (Tukey’s HSD, $p \leq 0.05$). Overall, V mice on a HFD weighed more and consumed more calories than all other groups.
## Estradiol reduces anxiety-like behavior
In order to assess the effects of E treatment, diet, and their associations with gut microbiota on anxiety-like behavior, mice were subjected to Light–Dark (LD), Elevated Plus Maze (EPM) and Open Field (OF) testing starting on day 30. The OF test was also administered as a control to confirm that any effects observed in differences in anxiety-like behavior were independent of locomotor activity64,65.
## Light–dark test
The LD test was administered to examine anxiety-like behavior based on the anxiogenic effect of light on rodents, as previously described66,67. A two-way ANOVA showed main effects for hormone treatment (F1, 60 = 25.81, $p \leq 0.0001$) and diet (F1, 60 = 10.33, $$p \leq 0.0021$$) on time spent in the light compartment. E-treated mice spent more time in the light compartment than their vehicle-treated counterparts on SD (Tukey’s HSD, $$p \leq 0.0002$$) or HFD (Tukey’s HSD, $$p \leq 0.045$$), suggesting that E decreased anxiety-like behavior in both diet conditions (Fig. 2A). HFD-fed E-treated mice spent less time in the light compartment than SD-fed counterparts (Tukey’s HSD, $$p \leq 0.012$$), suggesting that HFD had an anxiogenic effect. Figure 2Estradiol reduces anxiety-like behavior in both SD and HFD-fed mice in Light–Dark test ($$n = 15$$–16/group). ( A) Percent (%) time spent in the light compartment. ( B) The number of light/dark transitions. Black lines show mean (middle line) and $83\%$ CI (whiskers). Green lines show median (middle line) and the 1st (lower line) and the 3rd quartile (upper line). Data points greater than 1.5 times the interquartile range are shown within open circles. Different letters denote differences across groups, $P \leq 0.05$, Tukey’s HSD.
There was a main effect of hormone treatment (F1,56 = 8.68, $$p \leq 0.0046$$), but not diet or their interaction, on the number of light–dark transitions. However, Tukey’s HSD post-hoc test found no significant differences between individual groups. These findings suggest the groups did not differ in terms of their tendency to explore the compartments, and thus differences in time spent in the light may be due to differences in aversion to the light (Fig. 2B).
The anxiogenic effects of light on rodents were measured using the LD test as previously described66,67. The LD box (40 × 40 × 35 cm) is divided into two compartments of equal size separated by a wall of black Perspex with a small opening connecting them (#63101, Stoelting, Wood Dale, IL). The light compartment is made of clear Perspex and was illuminated to 300 lx. The dark compartment is made of black Perspex and was illuminated to < 5 lx. Mice were tested during the dark phase, beginning 2 h after lights off. The 10-min test was initiated by placing a mouse in the center of the light compartment facing the dark compartment. Number of transitions between light–dark compartments, latency to enter the dark compartment, latency to reenter the light, distance traveled in the light compartment, total distance traveled, average speed, time spent in either compartment, and average freezing score were measured.
## Elevated plus maze
The EPM test was used to measure anxiety-like behavior based on rodents’ inherent aversion to open spaces and heights, as previously described68–70. While there was an effect of diet (Kruskal–Wallis, $$p \leq 0.022$$) on the number of open arm entries, suggesting that HFD exerted anxiogenic effects in female mice, individual groups did not differ from each other (Fig. 3A). Similarly, there was an effect of diet (Kruskal–Wallis, $$p \leq 0.027$$) on the percent time spent in the open arms, but the individual groups did not differ from each other (Fig. 3B). No effects of diet or hormone treatment were detected on distance traveled on the EPM, suggesting that there were no differences in locomotion that might confound the other measures (Fig. 3C). It should be noted that two animals fell off the maze and were excluded from analysis. Figure 3HFD increases anxiety-like behavior in the Elevated Plus Maze test. HFD decreased (A) the number of open arm entries, (B) time spent in the open arms ($$n = 15$$–16/group). For (A) and (B), a main effect of diet was detected (Kruskal–Wallis, $p \leq 0.05$), although individual groups did not differ from each other. ( C) Total distance traveled. Black lines show mean (middle line) and $83\%$ CI (whiskers). Green lines show median (middle line) and the 1st (lower line) and the 3rd quartile (upper line). Data points greater than 1.5 times the interquartile range are shown within open circles.
## Open field test
There were no effects of diet or hormone treatment on the number of entries to, or time in the center of the arena, suggesting differences in anxiety-like behavior were not detected by the OF test. Distance traveled was also measured in the OF test to ensure that anxiogenic stimuli were not affecting general locomotor activity as previously described64,65. While no main effects of hormone treatment were detected, there was a trend towards an effect of diet (F1,56 = 3.64, $$P \leq 0.061$$) on locomotor activity, as measured by distance traveled in the OF arena. The four treatment groups did not differ in locomotor activity (Supplementary Fig. 2), suggesting there were no differences in general activity across treatment groups. Furthermore, the distance travelled in OF was comparable to previous studies in mice, suggesting the present test parameters did not cause additional stress to the mice71.
In order to assess locomotor activity and anxiety-like behavior, mice were subjected to the OF test as previously described64,65. The OF apparatus is a beige ABS plastic enclosure divided into four 50 × 50 × 38 cm compartments (Wellesley College Machine Shop, Wellesley, MA). The center zone of the apparatus was defined as the 30 × 30 cm square area in the center of the apparatus. On day 33 post-OVX/implant, mice were tested two at a time in two adjacent compartments, beginning 2 h after lights off. The apparatus was illuminated at 120 lx. The 10 min test was initiated by placing a mouse in the center of the apparatus. The behavior was recorded and total distance traveled, entries to the center zone and time and distance travelled in the center zone were measured.
## Estradiol reduces neural activity in discrete brain regions in HFD-fed animals
In order to elucidate the effects of E on neural activity in mice fed a HFD, HFD-E and HFD-V mice were perfused immediately after EPM, the last of the three anxiety tests administered, and brains were immunolabeled for c-fos using iDISCO (Fig. 4). Labeled whole brains were imaged with light sheet microscopy and immunolabeled cells were measured within regions of interest (ROIs) or evenly spaced voxel (each 2 × 2 × 3 µm). E treatment decreased the number of c-fos immunoreactive cells in the paraventricular nucleus (PVH), particularly the medial dorsal parvicellular part (PVHmpd) (q < 0.0001), medial preoptic area (MPO) ($q = 0.03$), lateral amygdala nucleus (LA) ($q = 0.01$), and the subparafascicular area (SPA) ($q = 0.0002$), including the subparafasicular nucleus (SPF) ($q = 0.02$) of the thalamus. E also reduced the number of c-fos-immunoreactive cells in the magnocellular nucleus of the anterior bed nucleus of the stria terminalis (BSTmg) ($q = 0.045$).Figure 4iDISCO reveals that estradiol reduces c-fos immunoreactivity in brain regions involved in anxiety-like behavior and energy homeostasis in HFD-fed female mice. c-fos immunoreactivity (green) in the (A) paraventricular nucleus of the hypothalamus (PVH), (B) medial preoptic area (MPO), (C) subparafascicular nucleus (SPF) of the thalamus ($$n = 4$$/group). Some non-specific c-fos immunoreactivity (red) is also observed. The box limits are the first and third quartiles. The whiskers are at 1.5 times the interquartile range below the first quartile and above the third quartile. * Denotes q < 0.05 between the two groups.
## Gut microbiota associate with body weight gain
The 16S rRNA from fecal samples was analyzed using Quantitative Insights Into Microbial Ecology (QIIME2) to identify the organisms present in each sample. The taxonomy of each OTU was established by matching to the GreenGenes (v13_8, $97\%$ clustered OTUs) (https://greengenes.secondgenome.com/). Feature tables were used to calculate alpha diversity and a phylogenetic tree was constructed to measure beta diversity metrics.
Longitudinal composition of gut microbiota based on samples taken from days 16–28 correlated with body weight (Fig. 5). The relative abundances of Clostridiales (order) and its families Peptostreptococcaceae and Clostridiaceae, and Eubacterium were positively correlated with body weight gain, of which Peptostreptococcaceae was also positively associated with HFD feeding. Relative abundances of Ruminococcaceae, Anaerotruncus, and Coprococcus were negatively associated with weight gain, but positively associated with HFD feeding (Fig. 5C).Figure 5HFD and estradiol alter gut microbial composition. ( A) Time-longitudinal graph of gut microbiota taxa over time. ( B) Bray–Curtis dissimilarity show that gut microbiota from SD-fed mice (blue) cluster differently from HFD-fed mice (red) and (C) multiple bacterial taxa correlate with E treatment, diet, and body weight in female mice based on longitudinal samples through the full length of the study ($$n = 16$$/group). Positive correlation with SD, increased days, V treatment, increased body weight, or cohousing with V mice are shown as red in the heatmap.
## Estradiol treatment and HFD feeding alter gut microbiota
16 s rRNA sequence analysis revealed differences in fecal microbiota composition between SD and HFD-fed mice. Longitudinal analysis of gut microbiota revealed a noticeable change in taxa abundance over time, with the largest shift detected on the 5th day on HFD, (D16, Fig. 5A). In particular, there was a sharp decrease in the relative abundances of Muribaculaceae and Streptophyta, whereas Coprobacillus and Oscillospira increased in response to HFD (Fig. 5A,C). Diet-dependent effect on gut microbiota was additionally confirmed by cluster analysis, using Bray–Curtis distance measure (PERMANOVA, Fig. 5B). Gut microbiota community also clustered differently between E and V groups ($p \leq 0.02$, Supplementary Fig. 3).
Relative abundances of multiple taxa correlated with E treatment (Fig. 5C). Multivariable correlational analysis found that E treatment was positively associated with Muribaculaceae, Sutterella, Roseburia, B. ovatus, A. muciniphila, and R. gnavus. In contrast, Peptostreptococcaceae, Mogibacteriaceae, Coprococcus, and cc_115 were negatively correlated with E treatment, but positively correlated with HFD feeding. Additional taxa including Clostridiales, Lachnospiraceae, Anaeroplasma, and RF39 were also negatively correlated with E treatment.
HFD-feeding was associated with profound changes in gut microbiota. In addition to the taxa mentioned above that negatively associated with E but positively associated with HFD, Ruminococcaceae, Erysipelotrichaceae, Coprobacillus, Lactococcus, Anaerotruncus, A. muciniphila, Oscillospira, Adlercreutzia, Dorea, rc 4.4, and R. gnavus were also increased during HFD feeding. In contrast, Clostridiales, Streptophyta, the families Clostridiaceae, Christensenellaceae, Muribaculaceae, Lachnospiraceae, Turicibacter, Lactobacillus, Lachnospira, Ruminococcus, Anaeroplasma, Roseburia, RF39, and C. piliforme were decreased during HFD intake. Of the microbes that were increased by HFD feeding, Clostridiales and Peptostreptococcaceae were positively associated, while Ruminococcaceae, Coprococcus, and Anaerotruncus were negatively associated with weight gain.
Cohousing was also associated with alterations in gut microbiota, although, as expected, the effect was not as pronounced when compared with the effect of HFD or E. Sutterella was positively associated with cohousing with E mice, while Peptostreptococcaceae and Clostridiaceae were negatively associated (Fig. 5C).
## Gut microbiota associate with anxiety-like behavior
*Multiblock* generalized canonical correlations were computed between measures of the LD, OF, and EPM tests indicative of anxiety-like behavior, longitudinal microbiome features which include abundances on D28, change in abundances between D28 and D13, and change in abundances between D10 and D7, and the treatment binary variables E, diet and cohousing, using the Regularized and Sparse Generalized Canonical Correlation Analysis (RGCCA)72. The top 3 canonical components between the three blocks were explored for significant associations (Supplementary Fig. 4A). The first canonical component of the treatment block, which was strongly defined by HFD (adj $$p \leq 0.029$$, CI [0.32, 0.99], Supplementary Fig. 4B), was associated with anxiety (Fig. 6A and Supplementary Table 1) and explained $28\%$ and $12.3\%$ variance on anxiety-like behavior and gut microbiota, respectively (Supplementary Fig. 5A,B). Canonical components 2 and 3 were influenced by E treatment (adj $$p \leq 0.11$$, CI [− 0.22, 0.98]) and cohousing (adj $$p \leq 0.023$$, CI [− 0.99, − 0.01]) respectively (Supplementary Fig. 4C,D and Supplementary Fig. 5C) and each explained $9\%$ and $12\%$ of variance in behavior, respectively (Fig. 6B,C, and Supplementary Figs. 1A and 4A). The correlation of diet and microbiome with increased anxiety behavior was strongly captured by canonical component 1 (Fig. 6A), but to a lesser extent by canonical components 2 and 3 (Fig. 6B,C).Figure 6HFD most strongly contributes to anxiety-like behavior in female mice. Canonical loadings of behavioral tests Open Field, Light–Dark, and Elevated Plus Maze tests, were computed using the RGCCA method, of which (A) the 1st canonical component that optimized correlation between anxiety behavior and HFD component of the treatment block. ( B,C) The 2nd and 3rd canonical components were orthogonal canonical correlations of anxiety behavior and microbiome with E treatment and cohousing, respectively ($$n = 8$$/group). Color gradings depict the statistical significance levels of canonical loadings and whiskers show $95\%$ CI (mean + SEM) which measure the significance and stability of the block-weight vectors on 1000 bootstrap samples. The direction (+/−) of canonical loadings depict the direction (+/−) of variable (anxiety, microbiome, treatment) correlations with canonical variates (anxiety, microbiome, treatment).
To examine the association between changes in bacterial community over time and manifestation of anxiety, changes in microbial abundance during SD (D10 minus D7), and during HFD (D28 minus D13), were correlated with composite anxiety components. Relative abundance of microbial taxa on D28 only (just before anxiety-like behavior testing) were also examined. Increased relative abundances of Muribaculaceae, Turicibacter, Lachnospira, C. piliforme, RF39, Ruminococcus, Lactobacillus, and Anaeroplasma ($p \leq 0.0001$) and Roseburia ($p \leq 0.01$) on D28 correlated with decreased anxiety-like behavior, as evidenced by their inverse correlation with the canonical component 1 (Fig. 7A). All these microbes were also increased during SD, suggesting that a depletion of SD-associated gut microbial community could increase anxiety. Figure 7HFD, estradiol and cohouse-induced changes in gut microbiota correlate with anxiety behavior. Canonical loadings of bacterial OTUs of the top 3 canonical components that were associated with diet, estradiol treatment, and/or cohousing, respectively, and were correlated with components of anxiety behavior derived using individual anxiety measures from the Open Field, Light–Dark, and Elevated Plus Maze tests. ( A) Canonical loadings of microbial taxa that correlated with the 1st canonical component. ( B,C) Canonical loadings of microbes that were correlated with the 2nd and 3rd canonical components of anxiety ($$n = 8$$/group). Color gradings depict the statistical significance levels of canonical loadings and whiskers show $95\%$ CI (mean + SEM) which measure the significance and stability of the block-weight vectors on 1000 bootstrap samples. The direction (+/−) of canonical loadings depict the direction (+/−) of variable (anxiety, microbiome, treatment) correlations with canonical variates (anxiety, microbiome, treatment).
A net increase in Lachnospiraceae and Ruminococcus during HFD feeding, between D13–D28 ($p \leq 0.05$), was also associated with reduced anxiety-like behavior (Fig. 7A), suggesting that increases in these two taxa may attenuate the detrimental effect of HFD intake on anxiety response.
Multiple microbial taxa were identified to be associated with increased anxiety-like behavior. For example, Ruminococcaceae, Mogibacteraceae, Peptostreptococcaceae, Erysipelotrichaceae, Lactococcus, A. muciniphila, Coprococcus, rc4_4, Coprobacillus, and Adlercreutzia ($p \leq 0.001$), and Oscillospira, R. gnavus, Dorea, and Anaerotruncus ($p \leq 0.01$) on D28 were positively associated with the anxiety component that was most correlated with both diet and microbiome (Fig. 7A and Supplementary Table 2). These taxa were also associated with HFD, further suggesting that HFD-associated gut microbiota may contribute to anxiety. Similarly, an increase in abundances of Erysipelotrichaceae, Adlecreutzia ($p \leq 0.001$), Christensenellaceae ($p \leq 0.05$) on D28 compared to D13 was also associated with increased anxiety-like behavior (Fig. 7A and Supplementary Table 2).
The canonical components 2 and 3, which were mostly affected by E treatment and cohousing with E-mice, explained $9\%$ and $12\%$ of the variance in behavior, respectively (Supplementary Fig. 3C,D and Supplementary Fig. 5A). Most gut microbes associated with component 2 were the microbes affected by E, suggesting that microbial alteration resulting from E treatment may have less impact on anxiety (Fig. 7B). Only change in the abundance of Mogibacteriaceae during HFD (D28 minus D13) was notable for having a positive association with the anxiety component that was most correlated with E treatment ($p \leq 0.01$). However, the confidence interval was wide, suggesting a small sample size. Similarly, cohousing-induced alterations in gut microbiota may also have a limited effect on anxiety (Fig. 7C). None of the alterations were found to be significant based on $p \leq 0.01$ and had wide confidence intervals.
The first two canonical loadings of the microbiome (Supplementary Fig. 5B) showed a separation based on diet in the first component and E treatment in the second component. The second and third canonical loadings of the microbiome (Supplementary Fig. 5C) showed a separation based on E treatment and cohousing with E-treated mice, respectively. This was also evident in the canonical covariates of the microbiome block, where the first-three components showed a separation based on diet, E treatment and cohousing with E treated mice (Supplementary Fig. 5D–F).
The most significant canonical loadings of the behavior block (Fig. 6A–C) showed a higher representation by LD tests in the first component, mixed representation of LD, EPM and OF in the second component and higher representation of EPM in the third component. Only the first and the third components showed significant correlations with the behavior tests. The first component, that is most associated with the HFD, had a positive association with anxiety behaviors. The third component, negatively correlated with cohousing with E-treated mice, was negatively correlated with number of entries in the open arms zone in EPM. The first canonical covariate of the behavior blocks showed separation based on diet (Supplementary Fig. 5D) but not as wide as the first canonical covariate of the microbiome, possibly due to the difference in correlation between microbiome and behavior blocks with the diet (Supplementary Fig. 4A). Additionally, there was a different trend between the interaction of E treatment and cohousing in the HFD compared to SD in the first canonical covariate of the behavior block. The second and third canonical covariates of the behavior block showed a reverse trend when compared to the second and third canonical covariates of the microbiome block (Supplementary Fig. 5E,F), possibly due to the different direction of correlation between microbiome and behavior blocks with the treatment block (Supplementary Fig. 4A): the microbiome components had a positive correlation, and the behavior components had a negative correlation.
In addition to the central effects of estrogens on anxiety discussed above, there is evidence that estrogens and HFD impact the gut microbiota to alter communication between the gut and central nervous system via the gut-brain axis44,45. Thus, it is important to consider the possible effects of the changing gut microbiome on anxiety. In the present study, microbes increased by HFD, including Ruminococcaceae, Mogibacteraceae, Peptostreptococcaceae, Erysipelotrichaceae, Lactococcus, A. muciniphila, Coprococcus, Coprobacillus, Adlercreutzia, and Oscillospira, were associated with increased anxiety-like behavior. Of these, Mogibacteraceae, Peptostreptococcaceae, and Coprococcus were decreased in E mice, suggesting that HFD-induced proliferation of these microbial communities increases anxiety risk and is exacerbated by reduction in estrogens. In particular, Erysipelotrichaceae and its genus Coprobacillus, which were increased by HFD and E-deficiency have been shown to be associated with weight gain58,73,92, suggesting a mechanism by which menopause leads to weight gain and increased anxiety.
Akkermansia is a gut microbe that has been associated with metabolic health and as having a beneficial role in the stress response. Male mice exposed to chronic social defeat stress had a lower abundance of Akkermansia, suggesting that decreased levels of this microbe are associated with increased anxiety93,94. However, interestingly, Akkermansia was positively associated with increased anxiety in female mice in the current study. These differing results between previous studies and the present findings could be due to a sex difference95. Since *Akkermansia is* increased by HFD feeding, the detrimental effects of HFD feeding on anxiety could be augmented by Akkermansia and other microbial communities as a result of HFD intake. Interestingly, *Akkermansia is* increased in a rat model of Type 1 diabetes mellitus in females, providing further support for its association with metabolic and inflammatory perturbation in females96. In future studies it will be important to investigate sex differences in the function of Akkermansia and other microbes in behavior.
Gut microbes that were dominant in SD-fed mice, including Muribaculaceae, Turicibacter, Lachnospira, C. piliforme, Ruminococcus, Lactobacillus, Roseburia, and Anaeroplasma were negatively correlated with anxiety-like behavior. Turicibacter, which was associated with decreased anxiety, was associated with SD in the current study and a previous one73, suggesting it as a promising candidate in low-fat diet-dependent mediation of metabolic health and mood. Moreover, a Lactobacillus species, L. rhamnosus, decreased anxiety and increased sociability in female pups. The L. rhamnosus-dependent decrease in anxiety was accompanied by a decrease in Erysipelotrichaceae97. In the present study, Erysipelotrichaceae was increased due to HFD and E-deficiency, and positively associated with anxiety. Quorum sensing, or cross-talk among microbial species, may be the mechanism by which these bacterial population densities are regulated98.
## Discussion
In the present study, we investigated the impact of E and diet on body weight, gut microbiota, and anxiety-like behavior. E treatment reduced HFD intake and prevented HFD-induced obesity in female mice, consistent with our previous findings11,12,58,73 and those of others14,74,75. Moreover, the present findings reveal that E reduces anxiety-like behavior and neural activity in discrete brain regions associated with energy homeostasis and anxiety processing. Consistent with our previous studies and others51,58,76, HFD dramatically alters gut microbiota composition in female mice. E treatment associated with attenuation of some of these HFD-induced changes in gut microbiota. Moreover, HFD-induced changes in gut microbiota were associated with increased anxiety-like behavior. These findings suggest that E functions to maintain energy metabolism and modulate anxiety in female mice challenged with a HFD by attenuating the HFD-induced changes in gut microbiota.
## HFD and estradiol affect anxiety-like behavior
In the present study, E reduced anxiety-like behavior in mice on SD and HFD in the LD test, suggesting E has an anxiolytic effect in both diet conditions. This increase in anxiety-like behavior was not due to altered locomotor activity64,65, as locomotor activity was not different between hormone groups in the OF test. However, no differences in anxiety-like behavior were detected in mice on SD or HFD in the EPM or OF tests. In support of E reducing anxiety-like behavior as detected in the LD test, estrogens influence anxiety in humans and rodents by modulating the activity of the hypothalamic–pituitary–adrenal axis and serotonergic system28. Both anxiogenic and anxiolytic effects of E have been reported in female rodents, depending on timing, dose, and the nature of the behavioral testing and the anxiogenic stimuli used25,77,78. These opposing effects of E administration on anxiety-like behavior are likely due to the diverging action of different estrogen receptor (ER) subtypes, with ERβ and ERα activation having mostly anxiolytic and anxiogenic effects, respectively31,79. The present findings revealed that HFD had an anxiogenic effect in female mice, consistent with other studies in female and male mice80,81. Furthermore, we found that numerous microbial communities increased by HFD were associated with increased anxiety-like behavior, suggesting a mechanism by which HFD influences anxiety-like behavior. In support, changes in gut microbiota have been implicated as a mediating factor in diet-induced obesity (DIO)-related changes in anxiety-like behavior38.
## HFD and estradiol alter neuronal response in brain regions involved in anxiety
Using iDISCO, whole brains from animals on HFD were analyzed for c-fos expression, an indicator of neuronal response82, at the end of the experiment following the three anxiety-like behavior tests. As discussed below, E decreased c-fos expression in a number of brain regions involved in anxiety-like behavior and the stress response. While it is not known if this reduction in c-fos expression by E is affecting anxiety-like behavior, these findings do suggest that E altered neuronal responses in these brain regions involved in anxiety following the three behavior tests. E treatment decreased c-fos expression in the PVH, which functions in coordinating many autonomic processes including the stress response, energy homeostasis and reproduction83–85. In particular, compared to control mice, E-treated mice had a robust decrease in c-fos expression in the medial parvicellular part of the dorsal zone (PVHmpd), a brain region critical in driving the release of adrenocorticotropic hormone and activating the hypothalamic–pituitary–adrenal axis85. In rats, ERβ activation by an isoform-specific ligand decreases neural activity in the PVH in response to stress86,87, suggesting that ERβ activation decreased anxiety-like behavior and c-fos expression in the present study. E treatment also decreased c-fos expression in the subparafasicular nucleus (SPF). The SPF projects to areas involved in anxiety, including the amygdala, BST, hypothalamus and medial prefrontal cortex88. Further implicating the SPF in anxiety, SPF neurons produce tuberoinfundibular peptide of 39 residues, a peptide involved in anxiety-like behavior in rodents89,90. E treatment also decreased c-fos expression in the MPO of mice fed a HFD. ERα activation in the MPO has been implicated in anxiety-like behavior in female rodents, with ERα knockdown decreasing anxiety-like behavior in rats91. Taken together, these findings suggest that E-induced reduction of neural activity in the SPF and MPO decreases anxiety-like behavior in female mice. In future studies, it will be important to compare c-fos expression following anxiety-like behavior tests in animals fed standard diet vs. HFD.
## Specific gut microbial taxa associate with body weight gain
In the present study, a total of 33 bacterial taxa associated with weight gain, E treatment, cohousing, and diet. Among these are key taxa including Muribaculaceae, which negatively correlate with HFD feeding and overweight phenotypes58,73,99,100. Many of the constituent microbes of Muribaculaceae produce short-chain fatty acids (SCFA) that protect from inflammation and metabolic endotoxemia through bacterial fermentation of dietary fiber, suggesting these microbes promote a healthy metabolic milieu and function in mediating E-induced protection from diet-induced obesity101,102. In contrast, an increase in Peptostreptococcaceae in V mice during HFD intake and its association with weight gain is consistent with previous findings in female mice58,73, suggesting this microbe contributes the detrimental effects of high fat diet intake.
The relative abundance of A. muciniphila positively associated with E treatment and HFD feeding, suggesting it plays a role in mediating estrogenic protection from HFD-induced obesity. In support, A. muciniphila, the only species of the family Verrucomicrobiaceae cultured from intestinal contents, is associated with healthy body weight and energy metabolism in humans and rodents103–105. In the gut, A. muciniphila digests mucin and produces SCFAs, suggesting that it has an important function in maintaining gut epithelial integrity and function106. Furthermore, A. muciniphila abundance correlated with low adiposity, insulin sensitivity, and improved other markers of energy metabolism in male and ovariectomized female mice treated with E57. Administration of either live or heat-killed A. muciniphila protected from and rescued DIO and metabolic endotoxemia in male mice, indicating a causal role in preventing weight gain and inflammation107,108. Considering the correlation of Akkermansia with increased anxiety in the current study, and given that much of the previous work has been done in males, it will be important for future work to explore the effects of A. muciniphila administration on metabolic physiology and mood in female mice.
## Estradiol alters neural activity in brain regions involved in metabolism
Work from our lab and others has shown that E protects female mice from HFD-induced obesity11–14. Estrogens act centrally in the brain14,109,110 and peripherally13,75,111 to regulate energy homeostasis and protect against HFD-induced obesity. In the present study, E-mediated protection from HFD-induced obesity was in part due to its anorectic effects as E-treated mice consumed less HFD than vehicle mice. Although their food intake did not differ, the SD-V mice weighed more than the SD-E mice at the end of the study (e.g. days 22, 25, and 28), suggesting that E alters energy homeostasis on a standard diet. In support, ovariectomized mice gain more weight than intact female mice even when food intake is similar or less than intact animals112,113.
Estrogenic action in the hypothalamus plays an integral role in regulating many aspects of metabolism, including feeding3. In the present study, E-treated HFD-fed mice had a decreased number of c-fos immunoreactive cells in the PVH, and MPO compared to V control mice, suggesting E reduced neural activity in these brain regions. The PVH is critical in regulating many aspects of energy homeostasis, including food intake. Lesions to the PVH lead to hyperphagia and obesity in female rodents114,115, while E implants in this region reduce food intake in ovariectomized rodents116. Similarly, the MPO is important in energy homeostasis, with infusion of E into the MPO reducing food intake in ovariectomized rats117,118. We have previously shown that E reduces food intake and weight gain in ovariectomized mice fed a HFD and that E and HFD alter hypothalamic neurogenesis11,12. Taken together, the present findings suggest that E-mediated changes in activity in these key hypothalamic regions protect female rodents from HFD-induced obesity. The present study identified changes in gut microbiota and anxiety-like behavior in E-treated groups. In addition, E replacement persistently attenuated calorie intake during HFD feeding. Thus, the possibility that some of these changes in gut microbiota and anxiety are driven by excess calorie intake should be addressed in future studies using pair-feeding paradigms.
## Summary
The present findings provide evidence that E and gut microbiota mediate HFD-induced obesity and anxiety in female mice. The identification of microbial taxa that correlate with protection from diet-induced obesity and anxiety is critical in understanding the peripheral mechanisms by which E regulates mood and energy homeostasis. Given mounting evidence that psychiatric119 and metabolic39,120 disorders entail gut microbial dysbiosis, understanding how estrogenic modulation of mood and energy metabolism associates with gut microbial composition and function will enhance our understanding of host physiology. Taken together, these findings provide a basis for further exploration of these gut microbes through functional studies to elucidate their specific roles. Most importantly, these results allow identification of microbial targets for the comprehensive treatment of metabolic and mental health disorders, as well as other endocrinopathies that affect women’s health.
## Animals
Eight-week-old C57BL6 female mice (Jackson Laboratories, Bar Harbor, ME) were housed two per cage and maintained under a 12:12 h light/dark cycle (lights on 100 h to 1300 h) and fed a standard chow diet (SD) consisting of $13.5\%$ kcal from fat (catalog #5001; Purina, St. Louis, MO). Mice were anaesthetized with $2.5\%$ isoflurane, bilaterally ovariectomized (OVX) and implanted with a capsule made of Silastic tubing (Dow Corning, Midland, MI) capped with silicone sealant121 containing either 50 μg of 17β-estradiol (E) dissolved in 25 μl of $5\%$ ethanol/sesame oil (E mice, $$n = 32$$) or vehicle ($5\%$ ethanol/sesame oil) (V mice, $$n = 32$$)122,123. The capsules were implanted subcutaneously just below the left scapular blade. Nine days after surgery, mice were started on a high fat diet (HFD) consisting of $60\%$ kcal from fat in the form of lard and soybean oil ($26.2\%$ protein, $25.6\%$ carbohydrate, $34.9\%$ fat by weight) (catalog #D12492, Research Diets Inc., New Brunswick, NJ) or maintained on SD. In order to assess the effects of fecal microbiota exchange via coprophagy, mice were cohoused in three different configurations; E and E, E and V, and V and V.
Body weight and food intake were measured every three days at 900–1100 h. Fresh fecal samples were collected on the same days when mice were weighed and were immediately stored at − 80 °C until DNA extraction. On days 30–36, anxiety-like behavior of mice was assessed by: Light–Dark (LD), Open Field (OF), and Elevated Plus Maze (EPM) tests (Supplementary Fig. 1). All animal procedures were approved by the Institutional Animal Care and Use Committee of Wellesley College and were done in accordance with the NIH Animal Care and Use Guidelines.
## Anxiety-like behavior tests
In order to assess the effects of E, HFD and the resulting changes in gut microbiota on anxiety-like behavior, mice were subjected to a battery of anxiety tests over 7 days at the end of the study. All mice were habituated in the testing room for 2 h prior to testing. All tests were recorded and analyzed by ANY-maze software (version Stoelting, Wood Dale, IL). All testing apparatuses were cleaned with hypochlorous water and allowed to dry between trials to disinfect and reduce odors from the mice tested in the previous trials.
## Elevated plus maze test
Anxiety-like behavior was assessed using the EPM test based on rodents’ inherent aversion to open spaces and heights as previously described68–70. The EPM is a beige ABS plastic plus-shaped maze elevated 50 cm with two open arms and two closed arms (35 × 5 cm) with 15 cm walls that intersect at the open center (Wellesley College Machine Shop). Testing was conducted during the light phase, beginning 2 h after lights-on in normal room lighting (100 lx)70. The 5 min test was initiated by placing a mouse in the center of the maze facing an open arm. The behavior was recorded and open arm entries, distance traveled, and percent time spent in the open arms, average duration of and longest visit to the open arms, latency to first enter the closed arms, total distance travelled, and average freezing score were measured.
## c-fos Immunolabeling by iDISCO+
In order to observe the effects of E on neural responses to anxiety in mice on a HFD, brains of HFD-V and HFD-E ($$n = 4$$/group) mice were labelled for c-fos, a marker for neuronal response82 by modified immunolabeling-enabled three-dimensional imaging of solvent-cleared organs (iDISCO+, Certerra, Inc., Farmingdale, NY)124. On day 36, mice were deeply anesthetized with an intraperitoneal injection of Fatal-Plus (sodium pentobarbital; 390 mg/ml, 100 μl) and perfused transcardially with 0.01 M phosphate buffered saline (PBS) pH 7.2, for one minute followed by $4\%$ paraformaldehyde (w/v) for 8 min at a flow rate of 8 ml/min, 1.5–3 h after EPM testing. Brains were dissected out, post-fixed in $4\%$ paraformaldehyde overnight, transferred to 0.1 M glycine in phosphate buffer for 48 h, and stored in 0.01 M PBS at 4 °C until processing.
Brains were then dehydrated in a graded series of methanol/water solutions ($20\%$, $40\%$, $60\%$, $80\%$), and washed with $100\%$ methanol twice for 1 h each. Brains were bleached overnight in $5\%$ hydrogen peroxide (H2O2) in methanol (1 volume of $30\%$ H2O2 for 5 volumes of methanol, ice cold) at 4 °C. Brains were rehydrated with a graded series of methanol/water solutions ($80\%$, $60\%$, $40\%$, $20\%$), then washed with 0.01 M PBS with $0.2\%$ TritonX-100 twice for 1 h each at room temperature. Tissue was permeabilized with 0.1 M PBS with $0.2\%$ TritonX-100, $20\%$ dimethyl sulfoxide (DMSO), and 0.3 M glycine for 36 h at 37 °C. Brains were blocked in 0.1 M PBS and $0.2\%$ TritonX-100 with $10\%$ DMSO and $6\%$ donkey serum for 2 days at 37 °C. Brains were incubated with rabbit anti-c-fos (9F6) monoclonal antibody (#mAb2250, Cell Signaling Technology, Beverly, MA) at 1:400 in PBS-Tween $0.2\%$ with heparin 10 µg/ml with $5\%$ DMSO and $3\%$ donkey serum at 37 °C for 7 days. The brains were washed with the PBS-Tween $0.2\%$ with heparin 10 µg/ml for 24 h (5 changes) and incubated with donkey anti-rabbit-Alexa 560 at 1:450 in PBS-Tween $0.2\%$ with heparin 10 µg/ml and $3\%$ donkey serum at 37 °C for 7 days. Finally, brains were washed with PBS-Tween $0.2\%$ with heparin 10 µg/ml before being cleared.
Brains were once again dehydrated with a graded series of methanol/water solutions ($20\%$, $40\%$, $60\%$, $80\%$, $100\%$ twice) for 1 h each at room temperature. Brains were then incubated in $66\%$ dichloromethane/$33\%$ methanol at room temperature for 3 h and then incubated in $100\%$ dichloromethane to wash away the methanol with shaking for 15 min, twice. To finish clearing, brains were incubated with dibenzyl ether (DBE, #108014, Sigma) for 30 min, and then stored in DBE.
## Imaging and quantification of c-fos positive cells
Cleared and labeled whole brains were imaged in the sagittal orientation with a light sheet fluorescence microscope (LSFM) (Ultramicroscope II, LaVision Biotech, Bielefeld, Germany) with a sCMOS camera (Andor Neo) and a 4x/0.5 objective lens (MVPLAPO 4x) with a 6 mm working distance dipping cap. Brains were imaged in 3 µm optical sections at 2 × 2 × 3 µm voxel resolution. The brain images were aligned and c-fos-positive cells were detected and counted. iDISCO c-fos positive cell counts were statistically compared by negative binomial regression. Statistical comparisons between the two groups were run based on either regions of interest (ROIs) or evenly spaced voxels. Voxels are overlapping 3D spheres with 100 μm diameter each and spaced 20 μm apart from each other. The cell counts at a given location, Y, were assumed to follow a negative binomial distribution whose mean is linearly related to one or more experimental conditions, X: E[Y] = α + βX. For example, when testing an experimental group versus a control group, the X is a single column showing the categorical classification of mouse sample to group id, i.e. 0 for the control group and 1 for the experimental group125,126. The maximum likelihood coefficients α and β through iterative reweighted least squares were determined, obtaining estimates for sample standard deviations in the process, from which the significance of the β coefficient was obtained. A significant β means the group status is related to the cell count intensity at the specified location. Z-values correspond to this β coefficient normalized by its sample standard deviation, which under the null hypothesis of no group effect, has an asymptotic standard normal distribution. The p-values reveal the probability of obtaining a β coefficient as extreme as the one observed by chance assuming this null hypothesis is true. To account for multiple comparisons across all voxel/ROI locations, p-values were thresholded at 0.1 and false discovery rates with the Benjamini–Hochberg procedure were reported as adjusted q-values127.
## 16s rRNA sequencing and analysis
DNA was extracted from frozen fecal samples collected on days 1, 4, 7, 10, 13, 16, 22, 28, 31, 34, and 36 using the DNeasy PowerSoil Kit (Qiagen, Germantown, MD) with minor adjustments to the provided protocol. The lysis step was performed using a FastPrep-24™ 5G Instrument (MP Biomedicals, Santa Ana, CA). A 5-min incubation with 50 µl of elution buffer C6 before centrifugation was added to increase final concentration. Quality and quantity of DNA were measured using a Nanodrop spectrophotometer (Thermo Scientific, Waltham, MA). Extracts were stored at − 20 °C before amplification and sequencing.
The V3-V4 region of the 16S rRNA gene was amplified in the samples using the forward Nextera Meta_V4_515 (5′-TCGTCGGCAGCGTCAGATGTGTATAAGAGACAGCCTACGGGA GGCAGCAG-3′) and reverse Meta_V4_806 (5′-GTCTCGTGGGCTCGGAGATGTGTATAAGA GACAGGGACTACHVGGGTWTCTAAT-3′) primers with flow cell adapters on each. The indexed PCR products were quantified using PicoGreen (Invitrogen, Carlsbad, CA). Once quantified, the amplicons were normalized to 10 ng/µl, pooled, and purified using SPRI purification. The cleaned amplicons were then quantified by Qubit (Invitrogen, Carlsbad, CA) and underwent fragment analysis by Agilent TapeStation (Agilent, Santa Clara, CA). The pooled samples were sequenced with a paired-end Illumina MiSeq 600 cycle v3 kit (Illumina, San Diego, CA).
Forward and reverse reads were merged for each sample. The demultiplexed raw amplicon sequences were processed using an open-source software package of Quantitative Insights Into Microbial Ecology (QIIME2)128. Denoising and dereplication of paired-end sequences including chimera removal and trimming of reads based on positional quality scores were performed using the Divisive Amplicon Denoising Algorithm 2 (DADA2) built in QIIME2, an amplicon-specific error-correction method that models and corrects Illumina-sequenced amplicon errors129. Briefly, a feature table containing counts of each unique sequence variant in the samples was constructed using DADA2 as an operational taxonomic unit (OTU). An OTU is a cluster of sequences that differ by less than $3\%$ dissimilarity in our analysis.
In order to calculate alpha diversity metrics including observed feature counts (or observed OTUs) and Shannon and Simpson diversity indices, the feature table containing OTUs was rarefied. We further calculated microbial beta diversity (Bray–Curtis distance) using PERMANOVA. A summary of beta diversity relationships was visualized using principal coordinate analysis as PCoA plots.
Taxonomic composition analysis was performed to identify the organisms present in each sample. The taxonomy of each OTU was established by matching to the GreenGenes (v13_8, $97\%$ clustered OTUs), {https://greengenes.secondgenome.com/, #988} based on a naive Bayesian classifier with default parameters of QIIME2130–132.
## Body weight, food intake, and anxiety-like behavior data analyses
Body weight, food intake, behavior measures, and microbiota data were analyzed in Jamovi (v 1.8.4.0) or R (The R Foundation of Statistical Computing, v3.5.1) using the ‘lme’ (from the package ‘nlme’, v 3.1-152) and ‘anova (base R)’ functions with a repeated measures design in both one-way (by groups), two-way (diet and hormone treatment), or three-way (diet, hormone, and cohousing) analyses. For analyzing normally distributed single timepoint body weight, food intake, and behavior measures, a three-way (diet, treatment, and cohouse) or a two-way ANOVA (diet and treatment) were used (Jamovi). When there was a significant difference ($P \leq 0.05$), Tukey’s honestly significant difference (HSD) post-hoc test was used for comparisons. Some behavioral measures violated assumptions of normality and homogeneity of variance, thus were analyzed using non-parametric Kruskal–Wallis test followed by DCSF pairwise comparisons with family-wise corrections (Jamovi) and confidence intervals were generated using bootstrapping with 10,000 replicates, using ShowMyData.org (v.2.0)133.
## Longitudinal microbiota correlational analysis
The optimization problem is defined in Eq. [ 1] (Supplementary Fig. 6). Xi and ai denote the block data matrix and the weights, and the subscripts t, b and m denote treatment, behavior and microbiome respectively. Absolute value of covariance represents the centroid scheme. Each block is connected, i.e., the design matrix is an identity function (the coefficients of the covariances in the optimization are equal to 1). τi denotes the shrinkage that can be adjusted between maximum correlation (for τi = 0) and maximum covariance (for τi = 1).1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{array}{c}\underset{{a}_{t},{a}_{b},{a}_{m}}{\mathrm{max}}\left|{\text{cov}}\left({\mathbf{X}}_{t}{\mathbf{a}}_{t},{\mathbf{X}}_{b}{\mathbf{a}}_{b}\right)\right|+\left|{\text{cov}}\left({\mathbf{X}}_{b}{\mathbf{a}}_{b},{\mathbf{X}}_{m}{\mathbf{a}}_{m}\right)\right|+\left|{\text{cov}}\left({\mathbf{X}}_{t}{\mathbf{a}}_{t},{\mathbf{X}}_{m}{\mathbf{a}}_{m}\right)\right|\\ \text{s.t.}\left(1-{\tau }_{i}\right){\text{var}}\left({\mathbf{X}}_{i}{\mathbf{a}}_{i}\right)+{\tau }_{i}\parallel {\mathbf{a}}_{i}{\parallel }^{2}=1\text{ where }i\in \{t,b,m\}\end{array}.$$\end{document}maxat,ab,amcovXtat,Xbab+covXbab,Xmam+covXtat,Xmams.t.1-τivarXiai+τi‖ai‖2=1wherei∈{t,b,m}.
The treatment block, Xt, comprised of estradiol treatment (E), high-fat diet (HFD) and cohousing (C) binary vectors. Individual components for the LD, OF, and EPM behavioral tests indicative of anxiety-like behavior comprised the behavior block, Xb. Microbiome block, Xm, comprised of 3 derived features from observed longitudinal microbiome abundances: pre-HFD diet relative abundance as derived by change between day 10 and day 7 taxa abundances, post-HFD diet relative abundance as derived by change between day 28 and day 13 taxa abundances and lastly, relative abundance on day 28.
The centroid scheme was used as the scheme function to enable two components to be negatively correlated as well as to ensure fairness such that all blocks contribute equally to the solution in opposition to a model dominated by only a few blocks. The tuning parameter, τi, was selected using the permutation scheme as proposed in Ref.134. We have used the RGCCA R permute function to automatically select the hyper-parameters. Bootstrap confidence intervals and p-values were computed for evaluating the significance and stability of the block-weight vectors on 1000 bootstrap samples. The p-value was computed by assuming that the ratio of the blocks weight values to the bootstrap estimate of the standard deviation follows the standardized normal distribution. For a random selection of the variable within the block, the number of occurrences (0 or 1) follows a Bernoulli distribution with the parameter, p = proportion of selected variables in the block. This proportion was estimated by the average number of selected variables over all bootstraps divided by the total number of variables in each block (pi). On a larger number of bootstrap samples, the number of occurrences follows a binomial distribution B(n,p) with n = number of bootstraps. The test was based on the following null hypothesis: "the variable is randomly selected according to B(n,p)". This hypothesis was rejected when the number of occurrences is higher than the 1-(0.05/pi)th quantile.
## Limitations of the study
While this study provides important insights into the association of gut microbiota community with estradiol treatment, HFD and anxiety-like behavior, the functional implications of the associations of these microbes identified here remain to be investigated by manipulation studies. The present analyses do not discriminate between effects of HFD-feeding in vehicle-treated animals and the potential independent effects of excess caloric intake in this group. In addition, gut microbes, such as Akkermansia, had different associations with anxiety-like behavior in females in the current study compared to those previously reported in males, suggesting a sex difference in their functions and interactions with the host. In future studies, it will be critical to explore these potentially important sex differences in studies using male and female mice.
## Inclusion and diversity
One or more of the authors of this paper self-identifies as an underrepresented ethnic minority in science. One or more of the authors of this paper self-identifies as a member of the LGBTQ + community. One or more of the authors of this paper received support from a program designed to increase minority representation in science.
## Supplementary Information
Supplementary Information. The online version contains supplementary material available at 10.1038/s41598-023-31783-6.
## References
1. Asarian L, Geary N. **Cyclic estradiol treatment normalizes body weight and restores physiological patterns of spontaneous feeding and sexual receptivity in ovariectomized rats**. *Horm. Behav.* (2002.0) **42** 461-471. DOI: 10.1006/hbeh.2002.1835
2. Wade GN, Gray JM. **Gonadal effects on food intake and adiposity: A metabolic hypothesis**. *Physiol. Behav.* (1979.0) **22** 583-593. DOI: 10.1016/0031-9384(79)90028-3
3. Clegg DJ. **Minireview: The year in review of estrogen regulation of metabolism**. *Mol. Endocrinol. (Baltimore, Md)* (2012.0) **26** 1957-1960. DOI: 10.1210/me.2012-1284
4. Asarian L, Geary N. **Modulation of appetite by gonadal steroid hormones**. *Philos. Trans. R. Soc. Lond. B Biol. Sci.* (2006.0) **361** 1251-1263. DOI: 10.1098/rstb.2006.1860
5. Mamounis KJ, Hernandez MR, Margolies N, Yasrebi A, Roepke TA. **Interaction of 17beta-estradiol and dietary fatty acids on energy and glucose homeostasis in female mice**. *Nutr. Neurosci.* (2018.0) **21** 715-728. DOI: 10.1080/1028415X.2017.1347374
6. Guo SS, Zeller C, Chumlea WC, Siervogel RM. **Aging, body composition, and lifestyle: The Fels Longitudinal Study**. *Am. J. Clin. Nutr.* (1999.0) **70** 405-411. DOI: 10.1093/ajcn/70.3.405
7. Atsma F, Bartelink ML, Grobbee DE, van der Schouw YT. **Postmenopausal status and early menopause as independent risk factors for cardiovascular disease: A meta-analysis**. *Menopause* (2006.0) **13** 265-279. DOI: 10.1097/01.gme.0000218683.97338.ea
8. Brand JS. **Age at menopause, reproductive life span, and type 2 diabetes risk**. *Results EPIC-InterAct Study* (2013.0) **36** 1012-1019. DOI: 10.2337/dc12-1020
9. Hedlin H. **Development of a comprehensive health-risk prediction tool for postmenopausal women**. *Menopause* (2019.0) **26** 1385-1394. DOI: 10.1097/GME.0000000000001411
10. Wade GN. **Gonadal hormones and behavioral regulation of body weight**. *Physiol. Behav.* (1972.0) **8** 523-534. DOI: 10.1016/0031-9384(72)90340-X
11. Bless EP, Reddy T, Acharya KD, Beltz BS, Tetel MJ. **Oestradiol and diet modulate energy homeostasis and hypothalamic neurogenesis in the adult female mouse**. *J. Neuroendocrinol.* (2014.0) **26** 805-816. DOI: 10.1111/jne.12206
12. Bless EP. **Adult neurogenesis in the female mouse hypothalamus: Estradiol and high-fat diet alter the generation of newborn neurons expressing estrogen receptor alpha**. *eNeuro* (2016.0). DOI: 10.1523/ENEURO.0027-16.2016
13. Camporez JP. **Cellular mechanism by which estradiol protects female ovariectomized mice from high-fat diet-induced hepatic and muscle insulin resistance**. *Endocrinology* (2013.0) **154** 1021-1028. DOI: 10.1210/en.2012-1989
14. Xu Y. **Distinct hypothalamic neurons mediate estrogenic effects on energy homeostasis and reproduction**. *Cell Metab.* (2011.0) **14** 453-465. DOI: 10.1016/j.cmet.2011.08.009
15. Graham BM, Milad MR. **Blockade of estrogen by hormonal contraceptives impairs fear extinction in female rats and women**. *Biol. Psychiatry* (2013.0) **73** 371-378. DOI: 10.1016/j.biopsych.2012.09.018
16. Solomon MB, Herman JP. **Sex differences in psychopathology: Of gonads, adrenals and mental illness**. *Physiol. Behav.* (2009.0) **97** 250-258. DOI: 10.1016/j.physbeh.2009.02.033
17. Sagsoz N, Oguzturk O, Bayram M, Kamaci M. **Anxiety and depression before and after the menopause**. *Arch. Gynecol. Obstet.* (2001.0) **264** 199-202. DOI: 10.1007/s004040000108
18. Freeman EW. **Hormones and menopausal status as predictors of depression in womenin transition to menopause**. *JAMA Psychiat.* (2004.0) **61** 62-70. DOI: 10.1001/archpsyc.61.1.62
19. Wittchen HU, Hoyer J. **Generalized anxiety disorder: Nature and course**. *J. Clin. Psychiatry* (2001.0) **62** 15-19. PMID: 11414546
20. Sahingoz M, Uguz F, Gezginc K. **Prevalence and related factors of mood and anxiety disorders in a clinical sample of postmenopausal women**. *Perspect. Psychiatr. Care* (2011.0) **47** 213-219. DOI: 10.1111/j.1744-6163.2010.00296.x
21. Cagnacci A. **Depression and anxiety in climacteric women: Role of hormone replacement therapy**. *Menopause* (1997.0) **4** 206-211
22. Campbell S, Whitehead M. **Oestrogen therapy and the menopausal syndrome**. *Clin. Obstet. Gynaecol.* (1977.0) **4** 31-47. DOI: 10.1016/S0306-3356(21)00374-5
23. Demetrio FN. **Effect of estrogen replacement therapy on symptoms of depression and anxiety in non-depressive menopausal women**. *Arch. Women's Ment. Health* (2011.0) **14** 479-486. DOI: 10.1007/s00737-011-0241-3
24. Schoenrock SA. **Ovariectomy results in inbred strain-specific increases in anxiety-like behavior in mice**. *Physiol. Behav.* (2016.0) **167** 404-412. DOI: 10.1016/j.physbeh.2016.09.026
25. Raghavan NS. **Prepubertal ovariectomy exaggerates adult affective behaviors and alters the hippocampal transcriptome in a genetic rat model of depression**. *Front. Endocrinol. (Lausanne)* (2017.0) **8** 373. DOI: 10.3389/fendo.2017.00373
26. Bekku N, Yoshimura H, Araki H. **Factors producing a menopausal depressive-like state in mice following ovariectomy**. *Psychopharmacology* (2006.0) **187** 170-180. DOI: 10.1007/s00213-006-0395-2
27. Renczés E. **The role of estrogen in anxiety-like behavior and memory of middle-aged female rats**. *Front. Endocrinol.* (2020.0). DOI: 10.3389/fendo.2020.570560
28. Borrow AP, Handa RJ. **Estrogen receptors modulation of anxiety-like behavior**. *Vitam. Horm.* (2017.0) **103** 27-52. DOI: 10.1016/bs.vh.2016.08.004
29. Xu Y. **NLRP3 inflammasome activation mediates estrogen deficiency-induced depression- and anxiety-like behavior and hippocampal inflammation in mice**. *Brain Behav. Immun.* (2016.0) **56** 175-186. DOI: 10.1016/j.bbi.2016.02.022
30. Lund TD, Rovis T, Chung WC, Handa RJ. **Novel actions of estrogen receptor-beta on anxiety-related behaviors**. *Endocrinology* (2005.0) **146** 797-807. DOI: 10.1210/en.2004-1158
31. Tetel MJ, Pfaff DW. **Contributions of estrogen receptor-α and estrogen receptor-ß to the regulation of behavior**. *Biochem. Biophys. Acta.* (2010.0) **1800** 1084-1089. DOI: 10.1016/j.bbagen.2010.01.008
32. Imwalle DB, Gustafsson JA, Rissman EF. **Lack of functional estrogen receptor beta influences anxiety behavior and serotonin content in female mice**. *Physiol. Behav.* (2005.0) **84** 157-163. DOI: 10.1016/j.physbeh.2004.11.002
33. Marchesi JR, Ravel J. **The vocabulary of microbiome research: A proposal**. *Microbiome* (2015.0) **3** 31. DOI: 10.1186/s40168-015-0094-5
34. Desbonnet L, Clarke G, Shanahan F, Dinan TG, Cryan JF. **Microbiota is essential for social development in the mouse**. *Mol. Psychiatry* (2014.0) **19** 146-148. DOI: 10.1038/mp.2013.65
35. Thursby E, Juge N. **Introduction to the human gut microbiota**. *Biochem. J.* (2017.0) **474** 1823-1836. DOI: 10.1042/BCJ20160510
36. Cho I, Blaser MJ. **The human microbiome: At the interface of health and disease**. *Nat. Rev. Genet.* (2012.0) **13** 260-270. DOI: 10.1038/nrg3182
37. Ridaura VK. **Gut microbiota from twins discordant for obesity modulate metabolism in mice**. *Science* (2013.0) **341** 1241214. DOI: 10.1126/science.1241214
38. Soto M. **Gut microbiota modulate neurobehavior through changes in brain insulin sensitivity and metabolism**. *Mol. Psychiatry* (2018.0). DOI: 10.1038/s41380-018-0086-5
39. Hartstra AV, Bouter KE, Backhed F, Nieuwdorp M. **Insights into the role of the microbiome in obesity and type 2 diabetes**. *Diabetes Care* (2015.0) **38** 159-165. DOI: 10.2337/dc14-0769
40. Cryan JF, Dinan TG. **Mind-altering microorganisms: The impact of the gut microbiota on brain and behaviour**. *Nat. Rev. Neurosci.* (2012.0) **13** 701-712. DOI: 10.1038/nrn3346
41. De Palma G, Collins SM, Bercik P, Verdu EF. **The microbiota–gut–brain axis in gastrointestinal disorders: Stressed bugs, stressed brain or both?**. *J. Physiol.* (2014.0) **592** 2989-2997. DOI: 10.1113/jphysiol.2014.273995
42. Bravo JA. **Ingestion of Lactobacillus strain regulates emotional behavior and central GABA receptor expression in a mouse via the vagus nerve**. *Proc. Natl. Acad. Sci. U. S. A.* (2011.0) **108** 16050-16055. DOI: 10.1073/pnas.1102999108
43. Pirbaglou M. **Probiotic supplementation can positively affect anxiety and depressive symptoms: A systematic review of randomized controlled trials**. *Nutr. Res.* (2016.0) **36** 889-898. DOI: 10.1016/j.nutres.2016.06.009
44. Mayer EA, Knight R, Mazmanian SK, Cryan JF, Tillisch K. **Gut microbes and the brain: Paradigm shift in neuroscience**. *J. Neurosci.* (2014.0) **34** 15490-15496. DOI: 10.1523/JNEUROSCI.3299-14.2014
45. Tetel MJ, de Vries GJ, Melcangi RC, Panzica G, O'Mahony SM. **Steroids, stress and the gut microbiome-brain axis**. *J. Neuroendocrinol.* (2018.0). DOI: 10.1111/jne.12548
46. Kelly JR. **Transferring the blues: Depression-associated gut microbiota induces neurobehavioural changes in the rat**. *J. Psychiatr. Res.* (2016.0) **82** 109-118. DOI: 10.1016/j.jpsychires.2016.07.019
47. Foster JA, McVey Neufeld K-A. **Gut–brain axis: How the microbiome influences anxiety and depression**. *Trends Neurosci.* (2013.0) **36** 305-312. DOI: 10.1016/j.tins.2013.01.005
48. Strandwitz P. **GABA-modulating bacteria of the human gut microbiota**. *Nat. Microbiol.* (2019.0) **4** 396-403. DOI: 10.1038/s41564-018-0307-3
49. Bassett SA. **Metabolome and microbiome profiling of a stress-sensitive rat model of gut–brain axis dysfunction**. *Sci. Rep.* (2019.0) **9** 14026. DOI: 10.1038/s41598-019-50593-3
50. Chi Y. **PCBs-high-fat diet interactions as mediators of gut microbiota dysbiosis and abdominal fat accumulation in female mice**. *Environ. Pollut.* (2018.0) **239** 332-341. DOI: 10.1016/j.envpol.2018.04.001
51. Bridgewater LC. **Gender-based differences in host behavior and gut microbiota composition in response to high fat diet and stress in a mouse model**. *Sci. Rep.* (2017.0) **7** 10776. DOI: 10.1038/s41598-017-11069-4
52. Org E. **Sex differences and hormonal effects on gut microbiota composition in mice**. *Gut Microbes.* (2016.0). DOI: 10.1080/19490976.2016.1203502
53. Murphy EA, Velazquez KT, Herbert KM. **Influence of high-fat diet on gut microbiota: A driving force for chronic disease risk**. *Curr. Opin. Clin. Nutr. Metab. Care* (2015.0) **18** 515-520. DOI: 10.1097/MCO.0000000000000209
54. Shin J-H. **Serum level of sex steroid hormone is associated with diversity and profiles of human gut microbiome**. *Res. Microbiol.* (2019.0) **170** 192-201. DOI: 10.1016/j.resmic.2019.03.003
55. Graham ME. **Gut and vaginal microbiomes on steroids: Implications for women's health**. *Trends Endocrinol. Metab.* (2021.0) **32** 554-565. DOI: 10.1016/j.tem.2021.04.014
56. Moreno-Indias I. **Neonatal androgen exposure causes persistent gut microbiota dysbiosis related to metabolic disease in adult female rats**. *Endocrinology* (2016.0). DOI: 10.1210/en.2016-1317
57. Kaliannan K. **Estrogen-mediated gut microbiome alterations influence sexual dimorphism in metabolic syndrome in mice**. *Microbiome* (2018.0) **6** 205. DOI: 10.1186/s40168-018-0587-0
58. Acharya KD, Gao X, Bless EP, Chen J, Tetel MJ. **Estradiol and high fat diet associate with changes in gut microbiota in female ob/ob mice**. *Sci. Rep.* (2019.0) **9** 20192. DOI: 10.1038/s41598-019-56723-1
59. Hildebrandt MA. **High-fat diet determines the composition of the murine gut microbiome independently of obesity**. *Gastroenterology* (2009.0) **137** 1716-1724.e1242. DOI: 10.1053/j.gastro.2009.08.042
60. Turnbaugh PJ, Backhed F, Fulton L, Gordon JI. **Diet-induced obesity is linked to marked but reversible alterations in the mouse distal gut microbiome**. *Cell Host Microbe* (2008.0) **3** 213-223. DOI: 10.1016/j.chom.2008.02.015
61. Wu GD. **Linking long-term dietary patterns with gut microbial enterotypes**. *Science* (2011.0) **334** 105. DOI: 10.1126/science.1208344
62. Daniel H. **High-fat diet alters gut microbiota physiology in mice**. *ISME J.* (2014.0) **8** 295-308. DOI: 10.1038/ismej.2013.155
63. Hassan AM. **High-fat diet induces depression-like behaviour in mice associated with changes in microbiome, neuropeptide Y, and brain metabolome**. *Nutr. Neurosci.* (2018.0). DOI: 10.1080/1028415X.2018.1465713
64. Seibenhener ML, Wooten MC. **Use of the open field maze to measure locomotor and anxiety-like behavior in mice**. *J. Vis. Exp. JoVE* (2015.0). DOI: 10.3791/52434
65. Bale TL. **Mice deficient for corticotropin-releasing hormone receptor-2 display anxiety-like behaviour and are hypersensitive to stress**. *Nat. Genet.* (2000.0) **24** 410-414. DOI: 10.1038/74263
66. McEuen JG, Semsar KA, Lim MA, Bale TL. **Influence of sex and corticotropin-releasing factor pathways as determinants in serotonin sensitivity**. *Endocrinology* (2009.0) **150** 3709-3716. DOI: 10.1210/en.2008-1721
67. Bourin M, Hascoet M. **The mouse light/dark box test**. *Eur. J. Pharmacol.* (2003.0) **463** 55-65. DOI: 10.1016/S0014-2999(03)01274-3
68. Howerton AR, Roland AV, Bale TL. **Dorsal raphe neuroinflammation promotes dramatic behavioral stress dysregulation**. *J. Neurosci.* (2014.0) **34** 7113-7123. DOI: 10.1523/JNEUROSCI.0118-14.2014
69. Teegarden SL, Bale TL. **Decreases in dietary preference produce increased emotionality and risk for dietary relapse**. *Biol. Psychiatry* (2007.0) **61** 1021-1029. DOI: 10.1016/j.biopsych.2006.09.032
70. Komada M, Takao K, Miyakawa T. **Elevated plus maze for mice**. *JoVE* (2008.0). DOI: 10.3791/1088
71. Shoji H, Takao K, Hattori S, Miyakawa T. **Age-related changes in behavior in C57BL/6J mice from young adulthood to middle age**. *Mol. Brain* (2016.0) **9** 11. DOI: 10.1186/s13041-016-0191-9
72. Tenenhaus A, Tenenhaus M. **Regularized generalized canonical correlation analysis**. *Psychometrika* (2011.0) **76** 257. DOI: 10.1007/s11336-011-9206-8
73. Acharya KD. **Distinct changes in gut microbiota are associated with estradiol-mediated protection from diet-induced obesity in female mice**. *Metabolites* (2021.0) **11** 499. DOI: 10.3390/metabo11080499
74. Mamounis KJ, Hernandez MR, Margolies N, Yasrebi A, Roepke TA. **Interaction of 17beta-estradiol and dietary fatty acids on energy and glucose homeostasis in female mice**. *Nutr. Neurosci.* (2017.0). DOI: 10.1080/1028415x.2017.1347374
75. Bryzgalova G. **Mechanisms of antidiabetogenic and body weight-lowering effects of estrogen in high-fat diet-fed mice**. *Am. J. Physiol. Endocrinol. Metab.* (2008.0) **295** E904-E912. DOI: 10.1152/ajpendo.90248.2008
76. Choi S, Hwang Y-J, Shin M-J, Yi H. **Difference in the gut microbiome between ovariectomy-induced obesity and diet-induced obesity**. *J. Microbiol. Biotechnol.* (2017.0) **27** 2228-2236. DOI: 10.4014/jmb.1710.10001
77. Morgan MA, Pfaff DW. **Estrogen's effects on activity, anxiety, and fear in two mouse strains**. *Behav. Brain Res.* (2002.0) **132** 85-93. DOI: 10.1016/S0166-4328(01)00398-9
78. Kastenberger I, Lutsch C, Schwarzer C. **Activation of the G-protein-coupled receptor GPR30 induces anxiogenic effects in mice, similar to oestradiol**. *Psychopharmacology* (2012.0) **221** 527-535. DOI: 10.1007/s00213-011-2599-3
79. Boulware MJ. **Estradiol activates group I and II metabotropic glutamate receptor signaling, leading to opposing influences on cAMP response element binding protein**. *J. Neurosci.* (2005.0) **25** 5066-5078. DOI: 10.1523/JNEUROSCI.1427-05.2005
80. Krishna S. **Time-dependent behavioral, neurochemical, and metabolic dysregulation in female C57BL/6 mice caused by chronic high-fat diet intake**. *Physiol. Behav.* (2016.0) **157** 196-208. DOI: 10.1016/j.physbeh.2016.02.007
81. Kang SS. **Diet and exercise orthogonally alter the gut microbiome and reveal independent associations with anxiety and cognition**. *Mol. Neurodegener.* (2014.0) **9** 36. DOI: 10.1186/1750-1326-9-36
82. Morgan JI, Curran T. **Stimulus-transcription coupling in the nervous system—Involvement of the inducible proto-oncogenes fos and jun**. *Annu. Rev. Neurosci.* (1991.0) **14** 421-451. DOI: 10.1146/annurev.ne.14.030191.002225
83. Swanson LW, Sawchenko PE. **Paraventricular nucleus: a site for the integration of neuroendocrine and autonomic mechanisms**. *Neuroendocrinology* (1980.0) **31** 410-417. DOI: 10.1159/000123111
84. Williams G, Harrold JA, Cutler DJ. **The hypothalamus and the regulation of energy homeostasis: Lifting the lid on a black box**. *Proc. Nutr. Soc.* (2000.0) **59** 385-396. DOI: 10.1017/s0029665100000434
85. Antoni FA. **Hypothalamic control of adrenocorticotropin secretion: Advances since the discovery of 41-residue corticotropin-releasing factor**. *Endocr. Rev.* (1986.0) **7** 351-378. DOI: 10.1210/edrv-7-4-351
86. Lund TD, Hinds LR, Handa RJ. **The androgen 5alpha-dihydrotestosterone and its metabolite 5alpha-androstan-3beta, 17beta-diol inhibit the hypothalamo–pituitary–adrenal response to stress by acting through estrogen receptor beta-expressing neurons in the hypothalamus**. *J. Neurosci.* (2006.0) **26** 1448-1456. DOI: 10.1523/JNEUROSCI.3777-05.2006
87. 87.Weiser, M. J. & Handa, R. J. Estrogen impairs glucocorticoid dependent negative feedback on the hypothalamic-pituitary-adrenal axis via estrogen receptor alpha within the hypothalamus. Neuroscience (2009).
88. Calhoon GG, Tye KM. **Resolving the neural circuits of anxiety**. *Nat. Neurosci.* (2015.0) **18** 1394-1404. DOI: 10.1038/nn.4101
89. Wang J, Palkovits M, Usdin TB, Dobolyi A. **Forebrain projections of tuberoinfundibular peptide of 39 residues (TIP39)-containing subparafascicular neurons**. *Neuroscience* (2006.0) **138** 1245-1263. DOI: 10.1016/j.neuroscience.2005.12.022
90. Fegley DB. **Increased fear- and stress-related anxiety-like behavior in mice lacking tuberoinfundibular peptide of 39 residues**. *Genes Brain Behav.* (2008.0) **7** 933-942. DOI: 10.1111/j.1601-183X.2008.00432.x
91. Spiteri T, Ogawa S, Musatov S, Pfaff DW, Agmo A. **The role of the estrogen receptor alpha in the medial preoptic area in sexual incentive motivation, proceptivity and receptivity, anxiety, and wheel running in female rats**. *Behav. Brain Res.* (2012.0) **230** 11-20. DOI: 10.1016/j.bbr.2012.01.048
92. Kulecka M. **Prolonged transfer of feces from the lean mice modulates gut microbiota in obese mice**. *Nutr. Metab.* (2016.0) **13** 57. DOI: 10.1186/s12986-016-0116-8
93. Bharwani A. **Structural & functional consequences of chronic psychosocial stress on the microbiome & host**. *Psychoneuroendocrinology* (2016.0) **63** 217-227. DOI: 10.1016/j.psyneuen.2015.10.001
94. McGaughey KD. **Relative abundance of**. *Sci. Rep.* (2019.0) **9** 3281. DOI: 10.1038/s41598-019-40140-5
95. 95.Acharya, K. D., Parakoyi, A. E. R. & Tetel, M. J. In Endocrine Disruption and Human Health (ed. Darbre, P. D.) 355–376 (Academic Press, 2022).
96. Diviccaro S. **Gut microbiota composition is altered in a preclinical model of type 1 diabetes mellitus: Influence on gut steroids, permeability, and cognitive abilities**. *Neuropharmacology* (2023.0) **226** 109405. DOI: 10.1016/j.neuropharm.2022.109405
97. Leclercq S. **Low-dose penicillin in early life induces long-term changes in murine gut microbiota, brain cytokines and behavior**. *Nat. Commun.* (2017.0) **8** 15062. DOI: 10.1038/ncomms15062
98. Lebeer S. **Functional analysis of luxS in the probiotic strain**. *J. Bacteriol.* (2007.0) **189** 860-871. DOI: 10.1128/jb.01394-06
99. Evans CC. **Exercise prevents weight gain and alters the gut microbiota in a mouse model of high fat diet-induced obesity**. *PLoS ONE* (2014.0) **9** e92193-e92193. DOI: 10.1371/journal.pone.0092193
100. Ziętak M. **Altered microbiota contributes to reduced diet-induced obesity upon cold exposure**. *Cell Metab.* (2016.0) **23** 1216-1223. DOI: 10.1016/j.cmet.2016.05.001
101. Ormerod KL. **Genomic characterization of the uncultured Bacteroidales family S24–7 inhabiting the guts of homeothermic animals**. *Microbiome* (2016.0) **4** 36. DOI: 10.1186/s40168-016-0181-2
102. Chambers ES, Preston T, Frost G, Morrison DJ. **Role of gut microbiota-generated short-chain fatty acids in metabolic and cardiovascular health**. *Curr. Nutr. Rep.* (2018.0) **7** 198-206. DOI: 10.1007/s13668-018-0248-8
103. Yassour M. **Sub-clinical detection of gut microbial biomarkers of obesity and type 2 diabetes**. *Genome Med.* (2016.0) **8** 17. DOI: 10.1186/s13073-016-0271-6
104. Schneeberger M. **Akkermansia muciniphila inversely correlates with the onset of inflammation, altered adipose tissue metabolism and metabolic disorders during obesity in mice**. *Sci. Rep.* (2015.0) **5** 16643. DOI: 10.1038/srep16643
105. Clarke SF. **Targeting the microbiota to address diet-induced obesity: A time dependent challenge**. *PLoS ONE* (2013.0) **8** e65790. DOI: 10.1371/journal.pone.0065790
106. Derrien M, Vaughan EE, Plugge CM, de Vos WM. *Int. J. Syst. Evol. Microbiol.* (2004.0) **54** 1469-1476. DOI: 10.1099/ijs.0.02873-0
107. Plovier H. **A purified membrane protein from**. *Nat. Med.* (2017.0) **23** 107-113. DOI: 10.1038/nm.4236
108. Everard A. **Cross-talk between**. *Proc. Natl. Acad. Sci. U. S. A.* (2013.0) **110** 9066-9071. DOI: 10.1073/pnas.1219451110
109. Brown LM, Gent L, Davis K, Clegg DJ. **Metabolic impact of sex hormones on obesity**. *Brain Res.* (2010.0) **1350** 77-85. DOI: 10.1016/j.brainres.2010.04.056
110. Musatov S. **Silencing of estrogen receptor-alpha in the ventromedial nucleus of hypothalamus leads to metabolic syndrome**. *Proc. Natl. Acad. Sci. U.S.A.* (2007.0) **104** 2501-2506. DOI: 10.1073/pnas.0610787104
111. Ribas V. **Skeletal muscle action of estrogen receptor α is critical for the maintenance of mitochondrial function and metabolic homeostasis in females**. *Sci. Transl. Med.* (2016.0) **8** 334ra354. DOI: 10.1126/scitranslmed.aad3815
112. da Silva RP. **Leptin resistance is not the primary cause of weight gain associated with reduced sex hormone levels in female mice**. *Endocrinology* (2014.0) **155** 4226-4236. DOI: 10.1210/en.2014-1276
113. Rogers NH, Perfield JW, Strissel KJ, Obin MS, Greenberg AS. **Reduced energy expenditure and increased inflammation are early events in the development of ovariectomy-induced obesity**. *Endocrinology* (2009.0) **150** 2161-2168. DOI: 10.1210/en.2008-1405
114. Leibowitz SF, Hammer NJ, Chang K. **Hypothalamic paraventricular nucleus lesions produce overeating and obesity in the rat**. *Physiol. Behav.* (1981.0) **27** 1031-1040. DOI: 10.1016/0031-9384(81)90366-8
115. Butera PC, Beikirch RJ. **Central implants of diluted estradiol: independent effects on ingestive and reproductive behaviors of ovariectomized rats**. *Brain Res.* (1989.0) **491** 266-273. DOI: 10.1016/0006-8993(89)90062-0
116. Butera PC, Willard DM, Raymond SA. **Effects of PVN lesions on the responsiveness of female rats to estradiol**. *Brain Res.* (1992.0) **576** 304-310. DOI: 10.1016/0006-8993(92)90694-5
117. Dagnault A, Richard D. **Involvement of the medial preoptic area in the anorectic action of estrogens**. *Am. J. Physiol. Regul. Integr. Comp. Physiol.* (1997.0) **272** R311-R317. DOI: 10.1152/ajpregu.1997.272.1.R311
118. Santollo J, Torregrossa AM, Eckel LA. **Estradiol acts in the medial preoptic area, arcuate nucleus, and dorsal raphe nucleus to reduce food intake in ovariectomized rats**. *Horm. Behav.* (2011.0) **60** 86-93. DOI: 10.1016/j.yhbeh.2011.03.009
119. Dinan TG, Cryan JF. **Brain–gut–microbiota axis and mental health**. *Psychosom. Med.* (2017.0) **79** 920-926. DOI: 10.1097/PSY.0000000000000519
120. Tremaroli V, Bäckhed F. **Functional interactions between the gut microbiota and host metabolism**. *Nature* (2012.0) **489** 242-249. DOI: 10.1038/nature11552
121. Ingberg E, Theodorsson A, Theodorsson E, Strom JO. **Methods for long-term 17beta-estradiol administration to mice**. *Gen. Comp. Endocrinol.* (2012.0) **175** 188-193. DOI: 10.1016/j.ygcen.2011.11.014
122. Rissman EF, Heck AL, Leonard JE, Shupnik MA, Gustafsson JA. **Disruption of estrogen receptor beta gene impairs spatial learning in female mice**. *Proc. Natl. Acad. Sci. U. S. A.* (2002.0) **99** 3996-4001. DOI: 10.1073/pnas.012032699
123. Kudwa AE, Harada N, Honda SI, Rissman EF. **Regulation of progestin receptors in medial amygdala: Estradiol, phytoestrogens and sex**. *Physiol. Behav.* (2009.0) **97** 146-150. DOI: 10.1016/j.physbeh.2009.02.023
124. Renier N. **Mapping of brain activity by automated volume analysis of immediate early genes**. *Cell* (2016.0) **165** 1789-1802. DOI: 10.1016/j.cell.2016.05.007
125. 125.Venables, W. & Ripley, B. http://www.stats.ox.ac.uk/pub/MASS4 (Springer, 2002).
126. O'Hara RB, Kotze DJ. **Do not log-transform count data**. *Methods Ecol. Evol.* (2010.0) **1** 118-122. DOI: 10.1111/j.2041-210X.2010.00021.x
127. Benjamini Y, Hochberg Y. **Controlling the false discovery rate: A practical and powerful approach to multiple testing**. *J. R. Stat. Soc. Ser. B Methodol.* (1995.0) **57** 289-300
128. Bolyen E. **Reproducible, interactive, scalable and extensible microbiome data science using QIIME 2**. *Nat. Biotechnol.* (2019.0) **37** 852-857. DOI: 10.1038/s41587-019-0209-9
129. Callahan BJ. **DADA2: High-resolution sample inference from Illumina amplicon data**. *Nat. Methods* (2016.0) **13** 581-583. DOI: 10.1038/nmeth.3869
130. Yarza P. **Uniting the classification of cultured and uncultured bacteria and archaea using 16S rRNA gene sequences**. *Nat. Rev. Microbiol.* (2014.0) **12** 635-645. DOI: 10.1038/nrmicro3330
131. Wang J, Palkovits M, Usdin TB, Dobolyi A. **Afferent connections of the subparafascicular area in rat**. *Neuroscience* (2006.0) **138** 197-220. DOI: 10.1016/j.neuroscience.2005.11.010
132. Mallick H. **Multivariable association discovery in population-scale meta-omics studies**. *PLoS Comput. Biol.* (2021.0) **17** e1009442. DOI: 10.1371/journal.pcbi.1009442
133. 133.Wilmer, J. B. Data Visualization Web Apps (Version 2.0). [Web Apps]. ShowMyData https://showmydata.org (2021).
134. Witten DM, Tibshirani R, Hastie T. **A penalized matrix decomposition, with applications to sparse principal components and canonical correlation analysis**. *Biostatistics* (2009.0) **10** 515-534. DOI: 10.1093/biostatistics/kxp008
|
---
title: Satureja khuzistanica Jamzad essential oil and pure carvacrol attenuate TBI-induced
inflammation and apoptosis via NF-κB and caspase-3 regulation in the male rat brain
authors:
- Elham Abbasloo
- Sedigheh Amiresmaili
- Sara Shirazpour
- Mohammad Khaksari
- Firas Kobeissy
- Theresa Currier Thomas
journal: Scientific Reports
year: 2023
pmcid: PMC10036533
doi: 10.1038/s41598-023-31891-3
license: CC BY 4.0
---
# Satureja khuzistanica Jamzad essential oil and pure carvacrol attenuate TBI-induced inflammation and apoptosis via NF-κB and caspase-3 regulation in the male rat brain
## Abstract
Traumatic brain injury (TBI) causes progressive dysfunction that induces biochemical and metabolic changes that lead to cell death. Nevertheless, there is no definitive FDA-approved therapy for TBI treatment. Our previous immunohistochemical results indicated that the cost-effective natural Iranian medicine, *Satureja khuzistanica* Jamzad essential oil (SKEO), which consists of $94.16\%$ carvacrol (CAR), has beneficial effects such as reducing neuronal death and inflammatory markers, as well as activating astrocytes and improving neurological outcomes. However, the molecular mechanisms of these neuroprotective effects have not yet been elucidated. This study investigated the possible mechanisms involved in the anti-inflammatory and anti-apoptotic properties of SKEO and CAR after TBI induction. Eighty-four male Wistar rats were randomly divided into six groups: Sham, TBI, TBI + Vehicle, TBI + CAR (100 and 200 mg/kg), and TBI + SKEO (200 mg/kg) groups. After establishing the “Marmarou” weight drop model, diffuse TBI was induced in the rat brain. Thirty minutes after TBI induction, SKEO & CAR were intraperitoneally injected. One day after TBI, injured rats exhibited significant brain edema, neurobehavioral dysfunctions, and neuronal apoptosis. Western blot results revealed upregulation of the levels of cleaved caspase-3, NFκB p65, and Bax/Bcl-2 ratio, which was attenuated by CAR and SKEO (200 mg/kg). Furthermore, the ELISA results showed that CAR treatment markedly prevents the overproduction of the brain pro-inflammatory cytokines, including IL-1β, TNF-α, and IL-6. Moreover, the neuron-specific enolase (NSE) immunohistochemistry results revealed the protective effect of CAR and SKEO on post-TBI neuronal death. The current study revealed that the possible neuroprotective mechanisms of SKEO and CAR might be related to (at least in part) modulating NF-κB regulated inflammation and caspase-3 protein expression. It also suggested that CAR exerts more potent protective effects than SKEO against TBI. Nevertheless, the administration of SKEO and CAR may express a novel therapeutic approach to ameliorate TBI-related secondary phase neuropathological outcomes.
## Introduction
It is estimated that 69 million traumatic brain injuries (TBIs) occur each year globally which impose a significant therapeutic burden on human societies. Traffic accidents and TBI are the third most common causes of death worldwide, according to data provided by the World Health Organization1,2. The incidence of TBI is highest among high-income countries, while the burden is greatest in low-income and middle-income countries3. Therefore, cost-effective treatments for TBI are still important4.
It has been well established that TBI's highly complex pathophysiological process involves primary disruption of brain tissue due to direct mechanical trauma and secondary injury. While the primary phase occurs at the injury site within minutes to hours post-injury, the secondary injury occurs hours to days later. The secondary injury involves a series of neuropathological events, including blood–brain barrier (BBB) disruption, inflammation, excitotoxic damage, mitochondrial dysfunction, oxidative stress, lipid peroxidation, necrotic, and apoptotic cell death5. Among these events, persistent and excessive inflammation can worsen the neurological disruption during the secondary insult process by secretion of pro-inflammatory mediators, such as tumor necrosis factor-α (TNF-α), which plays an essential role in releasing interleukin-6 (IL-6) and interleukin-1β (IL-1β) by T-cells6. Increased IL-6 production causes neuronal impairment, BBB damage, and other acute neurological complications7. Many experimental data suggested that nuclear factor kappa B (NF-κB) activation enhances the transcription of pro-inflammatory cytokines8, and the cytokines are known to, in turn, activate NF-κB9. In this regard, the pro-apoptotic family members, such as B-cell lymphoma protein 2 (Bcl-2)-associated (Bax), triggering cytokine release, leading to caspase family activation10–12.
Our previous study findings showed that *Satureja khuzistanica* Jamzad essential oil (SKEO) with a high percentage of carvacrol (CAR) exerts a neuroprotective effect against the TBI pathophysiology in rats13. SKEO can reduce BBB permeability and edema, regulating astrocytes, neurons, and blood origin-infiltrated cells that produce cytokines, and decrease neuronal death followed by improved neurological function13. However, the protective mechanisms of SKEO against TBI complications are still unknown.
Satureja khuzistanica Jamzad (SKJ) (also known as Marzeh Khuzestan in Persian) is a member of the Satureja genus, which belongs to the Lamiaceae family and the Nepetoidae subfamily. It is found in the southwestern and southern sections of Iran. This plant is utilized as a dental anesthetic drop and mouth disinfectant in traditional medicine, as well as in the food and pharmaceutical industries14,15. This plant contains more than $4.5\%$ essential oil14, and CAR is the most abundant compound in SKEO (90.08–$94.16\%$) (see Table 1)16,17. Accumulating evidence has reported anti-nociceptive18, antioxidant19, anti-allergic, anti-apoptotic, neuroprotective, and anti-inflammatory effects of this plant essential oil and extract20–22.Table 1Composition of *Satureja khuzistanica* Jamzad essential oil. RtRIKICompoundArea%3.971846.9895847Ethyl 2-methylbutanoate0.075.405928.0449929α-Thujene0.375.613937.3933937α-Pinene0.266.694985.9775986Beta-pinene0.096.802990.8315990Myrcene0.77.3631013.2221013α-Phellanderene0.057.6191022.7041023α-Terpinene0.127.8791032.3331034P-Cymene2.297.9551035.1481035Limonene0.088.0451038.4811037β-Phellandrene0.068.7281063.7781063γ-Terpinene0.279.1791080.4811085Cis sabinene hydrate0.429.891106.6261106l-linalool0.6810.0911113.8641116Trans sabinene hydrate0.1212.2821192.7621193Endo-borneol0.1412.4241197.87512004-Terpineol0.5512.8781213.2511217α-Terpinol0.0812.9341248.6751244Iso thymol methyl ether0.1415.6921308.16911308Thyme camphor0.316.2531328.26941327Carvacrol91.3319.0981431.81131431Caryophyllene0.1821.251513.65261510Beta-Bisabolene0.5823.5121603.35711606Caryophyllene oxide0.17RI1; Retention indices determined relative to n-alkanes (C6–C24) on a DB-5GC column. RI: Retention indices; MS: mass spectra; Col: co-injection. Significant values are in bold.
Carvacrol (2-methyl-5-(1-methyl ethyl)-phenol) is found in the genus of plants belonging to the Lamiaceae family, such as Thym, Satureja, and Origanum, so oils derived from these plants can contain 85–$90\%$ carvacrol23,24. Because of its low molecular mass and lipophilic characteristics, this molecule can easily cross the BBB25. CAR possesses various biological and pharmacological properties in vitro and in vivo, including antioxidant, anticancer, antibacterial, antifungal, anti-inflammatory, and hepatoprotective effects26–28. Previous studies showed that CAR provided neuroprotection against ischemia reperfusion-induced cerebral injury29, spinal cord injury30, and neurodegenerative disease31. In 2012 Peter and colleagues declared that inhibition of transient receptor potential channel (TRPC1) by CAR enhanced neurological recovery after a TBI in mice32. However, the underlying mechanisms of these neuroprotective processes have yet to be fully clarified in TBI.
Based on the previously mentioned evidence about the protective activity of SKEO and CAR, the current study evaluates the potential neuroprotective mechanisms of CAR and SKEO post-TBI induction. In this regard, we investigated whether CAR and SKEO could regulate apoptosis-related proteins; caspase-3, Bax, and Bcl2, as well as the pro-inflammatory regulatory cytokine, NF-κB, to ameliorate the neurological deficits observed during the acute phase of TBI.
## Animals and experimental protocol
Male Wistar rats (200–250 g) were maintained in a temperature-controlled room (22–25 °C) with a regular 12-h light/dark cycle and free access to food and water. Rats were randomly divided into six groups, as follows: [1] Sham: these rats underwent all preliminary procedures for TBI except for the TBI induction (weight dropped), [2] TBI: these rats were exposed to the brain trauma and received no treatment, [3] TBI + Veh: these rats intraperitoneally injected with vehicle (tween 20, $1\%$ i.p)33,34, [4] TBI + CAR100: these rats received CAR (100 mg/kg, i.p), [5] TBI + CAR200: these rats received CAR (200 mg/kg, i.p), [6] TBI + SKEO200: these rats received SKEO (200 mg/kg, i.p). All treatments and vehicles were administered 30 min after TBI induction. To achieve $90\%$ power- to detect statistical significance at $95\%$ confidence interval, six rats were assigned to each group. The rats were sacrificed 24 h after the TBI35 (Table 1 and Supplemental Fig. 1).
A dose of 100 mg/kg of CAR was shown to be ineffective based on data obtained from the veterinary coma scale (VCS) and brain water content (BWC), but a dose of 200 mg/kg of CAR was associated with improved VCS and reduced BWC. As a result, the following experiments were carried out using a 200 mg/kg CAR and SKEO. In this regard, our previous dose–response analysis revealed that the SKEO (200 mg/kg) is the most beneficial dose13. Amanlou et al. found that intraperitoneal treatment of dosages greater than 200 mg/kg of SKJ was related to sleepiness and decreased physical activity19. In a pilot study, we observed that CAR (250 mg/kg) caused drowsiness (somnolent-like behavior) and staggering in rats by a similar mechanism to SKEO (250 mg/kg). As a result, we avoided utilizing higher doses due to the potential risk of interfering with neurological consciousness-dependent outcomes (e.g., VCS).
Furthermore, vehicle administration (Tween® 20, $1\%$) caused no significant difference between the TBI and TBI + Veh groups in the water content, VCS, and ELISA tests; hence, we only examined the indicated proteins expression and neural death in the TBI + Veh group. As a result, the number of utilized animals and associated research costs were reduced.
All experiments were performed in accordance with relevant guidelines and regulations and adhered to the ARRIVE guidelines (https://arriveguidelines.org/) which was used for reporting of animal experiments. The study was reviewed and approved by the local ethical committee of the Kerman University of Medical Sciences (Ethics code No. 1395.703).
## Chemicals and essential oil preparation
Tween® 20 and CAR were purchased from Merck Millipore (Darmstadt, Germany) and Sigma-Aldrich (Germany).
## Preparing of the essential oil
SKJ was obtained from a cultivated source (Khorraman Farm, Khorramabad, Iran) during the plant's flowering period, identified by the Department of Botany of the Research Institute of Forests and Rangelands (TARI; Tehran, Iran), and assigned a specimen voucher number (No. 58416) at the Herbarium of TARI. The obtained materials were crushed after drying in the air and then boiled in distilled water for 5 h using a Clevenger machine before being purified. The evaporated essence mixture and the water were separated based on the difference in mass between the water and the essence. A sodium sulfate solution was used to absorb the suspended water particles after the yellow oil (essence) was obtained. Carvacrol ($91.33\%$), p-Cymene ($2.29\%$), and l-Linalool ($0.68\%$) made up the majority of the essence, according to previous gas chromatography-mass spectroscopy (GC-Mass) analysis16,21 (Supplemental Fig. 2). All procedures of collecting, boiling, preparing of essential oil and determining its component were done according to the Khorraman Company standard guideline. The essential oil was diluted in $1\%$ Tween® 20 and administered 30 min after TBI induction, similar to previous studies13.
## Diffuse traumatic brain injury (TBI) induction
All rats were intubated before the TBI induction. Based on the Marmarou approach36–38, the adopted TBI method was a moderate diffuse brain injury. Briefly, all the animals were anesthetized with ketamine and xylazine in preparation for TBI. Then, a 300 g weight was dropped from a height of 2 m onto the anesthetized rat's head, while a metal disc (10 mm in diameter, 3 mm thick) was affixed to the animal’s skull. After trauma induction, all rats were immediately connected to the breathing pump (Animal Respiratory Compact, TSE systems, Germany). The intra-tracheal tube was removed once the spontaneous breathing had returned, and rats were housed in individual cages. The site of brain damage was clearly visible on H&E-stained slides (Supplemental Fig. 3).
## Brain water content (BWC) determination
Each rat BWC was measured 24 h after the TBI induction to quantify cerebral edema. As previously mentioned, anesthetized rats were sacrificed, and their brains were rapidly removed and subjected to wet-dry weight comparisons to measure water content percent (%). Briefly, the brain was weighed after placement in pre-weighed glass vials (wet weight). Then, the vials were placed in a 100 °C incubator (Memmert, Germany) for 24 h before being weighed again (dry weight). The percentage of water in each brain specimen was then calculated using the following formula39:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\text{Brain water content }}\left(\% \right) \, = \, \left[{\left({{\text{wet weight}} - {\text{dry weight}}} \right)/{\text{wet weight}}} \right] \, \times { 1}00$$\end{document}Brain water content%=wet weight-dry weight/wet weight×100
## Neurological outcomes assessment
The veterinary coma scale (VCS) (3–15), the sum of the respiratory response score (1–3), the motor response score (1–8), and the visual response score (1–4) were used to assess neurological outcomes. The definitions of the stated scoring systems are presented in Table 2. A higher score indicates better neurological outcomes. The VCS was measured before TBI induction and 4 and 24 h afterward40.Table 2Veterinary coma scale. Veterinary coma scale variableScoreMotor function Normal movement8 Mildly drowsy with spontaneous purposeful movements7 Lethargic, unable to stand, but maintains sternal recumbency6 Lethargic, withdraws to pinch, and lifts head with attention to visual stimuli; no sternal recumbency5 Withdraws or pedals to pinch4 Spontaneous pedaling3 Extensor posturing (spontaneous or to stimuli)2 Flaccid to stimuli1Eye function Open4 Open on stimulation3 Normal eyelid reflexes2 No eyelid response to stimuli1Respiration Normal3 Ataxic2 Apneic1
## Western blot analysis of apoptotic and inflammatory markers
The brain was removed from the skull and divided into two hemispheres. The right hemispheres of the brain were homogenized in 700 μl ice-cold RIPA lysis buffer (sigma; R0278), 1 mM protease inhibitors (sigma; P2714-IBTL), and 1 mM sodium orthovanadate using a tissue homogenizer (Hielscher UP200-Germany). The homogenate was centrifuged at 17,000×g for 15 min at 4 °C13,41,42. The supernatant was collected, and its protein concentrations were determined using the Bradford method (Bio-Rad Laboratories, München, Germany)41. On a $10\%$ SDS-PAGE gel, equal amounts of protein (40 g) were separated electrophoretically and transferred to nitrocellulose membranes (Hybond ECL, GE Healthcare Bio-Sciences Corp., NJ, USA). The membranes were blocked for 2 h with $5\%$ non-fat milk, then incubated overnight at 4 °C with the following antibodies; Bax (1:200, B-9: sc-7480), Bcl-2 (1:200, N-19: sc-492), NF-κB p65 (1:200, F-6: sc-8008), and caspase-3 (1:200, E-8: sc-7272). The membranes were incubated for 2 h at room temperature with the secondary antibodies (1:5000, mouse anti-rabbit IgG-HRP: sc-2357) and (1:5000, m- IgGκBP-HRP: sc-516102) followed by three times washing with tris-buffered saline with Tween® (TBST; 15 min each). All antibodies were diluted in a blocking buffer containing $5\%$ non-fat dried milk and TBST. Control of loading was performed using β-actin immunoblotting (antibody from Cell Signaling Technology Inc., Beverly, MA, USA; 1:200). Antibody-antigen complex was visualized via the Enhanced Chemiluminescence (ECL) method using the Gel Documentation System (BIO-RAD). The density of the bands was calculated using Lab Works software43,44.
## Enzyme-Linked immunosorbent assay (ELISA) of inflammatory cytokines
The tissue levels of TNF-α, IL-1β, and IL-6 were determined using commercially available ELISA kits (Eastibiopharm, USA). Briefly, the left cerebral hemispheres of rats were extracted and homogenized in phosphate buffer (pH 7.4, 0.1 M) using a homogenizer (Hielscher UP200-Germany) 24 h after TBI induction. The samples were then centrifuged at 3500 rpm for 15 min at 4 °C. The supernatant was collected and kept at − 80 °C until the analysis of TNF-α, IL-1β, and IL-6 levels as described by the kits manufacturer’s instructions45.
## Immunohistochemistry (IHC)
To determine neuronal activation, neuron-specific enolase (NSE) levels were assessed46 at 24 h after injury in paraffin-embedded sections (4 μm) using a Leica DM500-Germany microscope and ICC50 HD digital camera. Positive IHC-stained cells were counted in five different high-power fields (HPF; 40X). Neuron-specific enolase (NSE) IHC (NSE-IHC) protocols in this lab have been previously published13.
## Statistical analysis
To test the hypothesis that acute CAR treatment after diffuse TBI can reduce inflammatory and apoptotic properties at 24 h, associated with CARs ability to improve VCS scores similar to our previous publication testing SKEO13, statistical analyses were performed using SPSS, version 24.0 (SPSS, Inc., Chicago, IL, USA). Independent data are in structured data sets that are normally distributed (Shapiro–Wilk test) with similar variances (Brown-Forsythe test) meeting assumptions for the use of analysis of variance (ANOVA)47,48. One-way analysis of variance (ANOVA) was used to examine BWC and cytokines data, followed by a Tukey’s honest significance test (HSD) for post-hoc analysis. Repeated measure ANOVA was done on the logarithmic transformation of VCS scores to compare VCS scores between groups at different times by Bonferroni post-hoc test, similar to previous publications from our group49. Protein density and IHC data were compared between studied groups using one-way ANOVA followed by Tukey’s post-hoc test. Data are presented as the mean ± the standard error from the mean (SEM). Differences were considered statistically significant at the $P \leq 0.05$ level.
## SKEO and CAR treatments improved veterinary coma scale (VCS) scores after TBI
VCS changes in trial groups at different times after TBI are shown in Table 3. There were no significant differences in VCS between groups before TBI. A significant decrease was shown in the VCS scores in the TBI group at 4 h (9.33 ± 0.33) and 24 h (11.33 ± 0.33) post-TBI in comparison with the sham group (15 ± 0.0) ($P \leq 0.001$ and $P \leq 01$, respectively). The VCS scores in the SKEO (200 mg/kg) group at 4 h (12.00 ± 0.58) and 24 h (13.17 ± 0.31) post-TBI increased as compared to the Veh group ($P \leq 0.01$ and ns, respectively). Also, VCS scores in the CAR (200 mg/kg) group significantly increased at 4 h (12.33 ± 0.33) and 24 h (14.5 ± 0.22) post-TBI compared to the Veh group ($P \leq 0.05$ and $P \leq 0.05$, respectively). There was no significant difference in VCS between the TBI and TBI + Veh groups at any time after TBI.Table 3Veterinary coma scale (VCS) scores. Animal groupsVCS [-1]VCS [0]VCS [4]VCS [24]Sham15 ± 015 ± 015 ± 015 ± 0TBI15 ± 015 ± 09.33 ± 0.33###11.33 ± 0.33##TBI + Veh15 ± 015 ± 09.17 ± 0.31###11.00 ± 0.37##CAR (100 mg/kg)15 ± 015 ± 09.33 ± 0.17###11.67 ± 0.49#CAR (200 mg/kg)15 ± 015 ± 012.33 ± 0.33#*14.50 ± 0.22*SKEO (200 mg/kg)15 ± 015 ± 012.00 ± 0.58**13.17 ± 0.31#Data are expressed as mean ± SEM.TBI: traumatic brain injury; SKEO: Satureja Khuzistanica essential oil; CAR: Carvacrol.#$P \leq 0.05$, ##$P \leq 0.01$ & ###$P \leq 0.001$ represents significant difference with sham; *$P \leq 0.05$,*$P \leq 0.01$,*$P \leq 0.001$ represents significant difference with TBI + Veh.
## SKEO and CAR treatments reduced brain edema after TBI
Changes in BWC 24 h post-TBI have been shown in Fig. 1. The BWC increased from 70.83 ± 0.23 to 76.88 ± 0.58 and 76.02 ± 0.51 in the TBI and TBI + Veh groups compared to the sham group ($P \leq 0.001$). The BWC in the TBI + SKEO200 and TBI + CAR200 groups was significantly lower than in the TBI + Veh group ($P \leq 0.001$), whereas there was no significant difference between the TBI and TBI + Veh groups. The administration of CAR100 had no significant effects on BWC. The BWC was significantly decreased in the TBI + CAR200 group (69.55 ± 0.78) compared to the TBI + SKEO200 group (72.02 ± 0.59) ($P \leq 0.05$). Based on the dose–response investigation, we discovered that a 100 mg/kg CAR dosage was ineffective, but a 200 mg/kg dose significantly decreased BWC (Fig. 1) and improved VCS at 4 and 24 h after TBI (Table 3).Figure 1Effects of SKEO and CAR on the percent of brain water content 24 h after TBI induction in the different experimental groups ($$n = 6$$ rats per group). Data are expressed as mean ± SEM. ## $P \leq 0.01$ and ###$P \leq 0.001$ represents significant difference with sham; ***$P \leq 0.001$ represents significant difference with TBI + Veh; and $$P \leq 0.05$ represents significant difference with TBI + SKEO. TBI: Traumatic brain injury; SKEO: *Satureja khuzistanica* essential oil; CAR: Carvacrol.
## Effects of SKEO and CAR on the apoptotic-related proteins after TBI induction
The proenzyme (32 kDa) and the active fragment (17 kDa) of caspase-3 were evaluated using the western blot technique. The proactive form of caspase-3 was reduced in the brain 24 h after injury, suggesting its possible conversion to the active form. Cleaved Caspase-3 protein expression was significantly elevated in the vehicle-treated animals (TBI + Veh) (4.8 ± 0.25, $F = 35.42$) compared to the sham group ($P \leq 0.001$) and was significantly diminished after CAR200 (1.53 ± 0.3) ($P \leq 0.001$) and moderately suppressed by SKEO200 (3.08 ± 0.4) ($P \leq 0.01$) (Fig. 2).Figure 2Effects of SKEO and CAR on the cleaved caspase-3 protein expression were assessed by western blotting 24 h after TBI induction in the different experimental groups ($$n = 6$$ rats per group). Cropped representative bands from the western blot test from the same run are shown above. Data are expressed as mean ± SEM. ### $P \leq 0.001$ represents significant difference with sham; **$P \leq 0.01$ and ***$P \leq 0.001$ represent significant differences with TBI + Veh; $$$P \leq 0.01$ represents significant difference between TBI + CAR and TBI + SKEO. TBI: Traumatic brain injury; SKEO: *Satureja khuzistanica* essential oil; CAR: Carvacrol.
Expression of apoptotic-related proteins (Bax and Bcl-2) was used to investigate the protective effects of SKEO and CAR against TBI-induced neuronal death. The proapoptotic factor Bax (26 kDa) expression was low in the brains of sham rats, but the level of Bax expression in TBI + Veh was increased at 24 h post-TBI (3.11 ± 0.75, $F = 18.74$, $P \leq 0.001$) compared to sham group as shown in Fig. 3A. After SKEO and CAR treatment, Bax protein expression levels were significantly reduced (1.94 ± 0.13, $P \leq 0.01$) (1.44 ± 0.12, $P \leq 0.001$); respectively. Our findings indicated that SKEO and CAR administration was capable of modulating Bax expression following TBI induction, where CAR was more effective at lowering proapoptotic Bax protein compared to SKEO ($P \leq 0.05$) in this TBI model (Fig. 3A).Figure 3Effects of SKEO and CAR on the (A) Bax, (B) Bcl-2 protein levels, and (C) Bax:Bcl-2 ratio 24 h after TBI induction in the different experimental groups ($$n = 6$$ rats per group). Cropped representative bands from the western blot test from the same run are shown above. Data are expressed as mean ± SEM band density ratio for each group. β-actin was used as an internal control. ### $P \leq 0.001$ represents significant difference with sham; *$P \leq 0.05$, **$P \leq 0.01$, and ***$P \leq 0.001$ represent significant differences with TBI + Veh; $$P \leq 0.05$ represent significant differences between TBI + CAR and TBI + SKEO. TBI: Traumatic brain injury; SKEO: *Satureja khuzistanica* essential oil; CAR: Carvacrol.
Moreover, the Western blot analysis showed that he expression of Bcl-2 was dramatically decreased in the TBI + Veh group (0.32 ± 0.06, $F = 30.95$, $P \leq 0.001$) 24 h after injury compared to the sham group, but this effect was significantly restored in the TBI-induced rats treated with CAR (0.68 ± 0.07, $P \leq 0.05$) but not by SKEO (0.41 ± 0.06) (Fig. 3B).
Furthermore, the Bax:Bcl-2 ratio alterations were compared between the studied groups. Injured animals in the TBI + Veh group (9.99 ± 1.01) had a significantly higher Bax:Bcl-2 protein ratio than those in the sham group ($P \leq 0.001$). The elevated Bax:Bcl-2 ratio was significantly reduced in injured rats treated with 200 mg/kg SKEO (4.96 ± 0.67) and CAR (2.26 ± 0.48) compared to the TBI + Veh group ($P \leq 0.01$ and $P \leq 0.001$, respectively) (Fig. 3C). There was a significant difference between CAR and SKEO treatment groups ($P \leq 0.05$).
## SKEO and CAR reduced the inflammatory cytokines after TBI induction
The tissue levels of IL-1β, TNF-a, and IL-6 in rat brains were significantly increased 24 h after TBI induction compared to the sham group. No significant differences were observed between the TBI and TBI + Veh groups. Figure 4A indicates a significant decrease in IL-1β levels in the TBI + SKEO (862.33 ± 21.88) and TBI + CAR (715.66 ± 12.99) groups compared to the TBI + Veh (1186.6 ± 46.75) group ($P \leq 0.001$). There was also a significant difference between the TBI + SKEO and TBI + CAR groups ($P \leq 0.05$).Figure 4Effects of SKEO and CAR on the (A) IL-1β, (B) TNF-α, and (C) IL-6 levels 24 h after TBI induction in the different experimental groups ($$n = 6$$ rats per group). Data are expressed as mean ± SEM. ## $P \leq 0.01$ & ###$P \leq 0.001$ represent significant differences with sham; ***$P \leq 0.001$ represents significant difference with TBI + Veh; and $$P \leq 0.05$ represents significant difference between TBI + CAR and TBI + SKEO. TBI: Traumatic brain injury; SKEO: *Satureja khuzistanica* essential oil; CAR: Carvacrol.
TNF-α levels were also decreased significantly in the TBI + SKEO (79.61 ± 5.2) and TBI + CAR (62.08 ± 1.5) groups ($P \leq 0.001$) compared to the TBI + Veh group (118.89 ± 1.23) (Fig. 4B). A significant difference in TNF-α levels was found between the TBI + SKEO and TBI + CAR groups ($P \leq 0.05$).
Figure 4C also shows that SKEO (99.68 ± 0.42) and CAR (80.13 ± 3.83) treatment after TBI induction significantly decreased the level of IL-6 compared to the TBI + Veh group (132.25 ± 7.81) ($P \leq 0.001$). The results also showed that CAR significantly reduced IL-6 levels compared to the SKEO ($P \leq 0.05$).
Western blot assays were used to determine whether the neuroprotective effect of CAR and SKEO was due to the inhibition of NF-κB signaling. TBI substantially enhanced NF-κB p65 expression (4.18 ± 0.26) compared to the sham group ($P \leq 0.05$). In contrast, SKEO (2.72 ± 0.41) and CAR (1.53 ± 0.22) administration significantly reduced the postinjury NF-κBp65 protein expression in the brain when compared to the TBI + Veh group ($P \leq 0.01$ & $P \leq 0.001$, respectively) (Fig. 5).Figure 5Effects of SKEO and CAR on the inflammatory protein NF-κB p65 expression 24 h after TBI induction in the different experimental groups ($$n = 6$$ rats per group). Cropped representative bands from the western blot test from the same run are shown above. Data are expressed as mean ± SEM band density ratio for each group. β-actin was used as an internal control. ### $P \leq 0.001$ represents a significant difference with sham; **$P \leq 0.05$ and ***$P \leq 0.01$ represent significant differences with TBI + Veh. TBI: Traumatic brain injury; SKEO: *Satureja khuzistanica* essential oil; CAR: Carvacrol.
## SKEO and CAR inhibited neuron injury
We performed IHC staining for NSE to assess for neuronal loss/injury. NSE-IHC staining showed that the neuronal nucleus and the intact, brown-stained neural cytoplasm and membrane were present in neurons in the sham group (100 ± $1.55\%$). In contrast, in the TBI + Veh group, a few non-injured neurons were seen, accompanied by a large number of malformed or squeezed neurons in comparison with the sham group (28.42 ± $1.33\%$) ($P \leq 0.001$). However, treatment with SKEO and CAR, respectively, resulted in the survival of (110.71 ± $1.2\%$) and (175.02 ± $1.62\%$) of neurons, positively stained, in comparison with the TBI + Veh group ($P \leq 0.001$). The results also showed that CAR potently and significantly reduced neural damage compared to the SKEO ($P \leq 0.05$) (Fig. 6).Figure 6(A–D) IHC staining for NSE in the rat brain (×40). ( A) Sham; arrow indicates the normal neurons. ( B) 24 h after TBI; arrow indicates the degenerated neurons. ( C,D) TBI + SKEO200& TBI + CAR200; arrow indicates the viable neurons. The scale bar illustrates the means ± SEM percentages of neuron-specific enolase (NSE from different groups. ### $p \leq 0.001$ compared to the sham group, ***$p \leq 0.01$ compared to the TBI group, and $$P \leq 0.05$ represents significant difference between TBI + CAR and TBI + SKEO. TBI: Traumatic brain injury; SKEO: *Satureja khuzistanica* essential oil; CAR: Carvacrol.
## Discussion
In the present study, we established a successful TBI weight-drop model and confirmed it by evaluating the rate of brain edema, neurological scores, and histological findings (Figs. 1, 6, Table 3, and Supplemental Fig. 1). Furthermore, the current study findings revealed that SKEO and CAR could reduce edema and have beneficial effects on regulating apoptotic and inflammatory signaling pathways following TBI model induction. Administration of SKEO and CAR reduced pro-apoptotic proteins such as Bax and caspase-3 and improved anti-apoptotic proteins such as Bcl-2 (Figs. 2 and 3). Moreover, after brain injury induction, the inflammatory signaling molecules, such as the NF-κB, IL-1β, TNF-α, and IL-6, were significantly reduced following CAR and SKEO treatments (Figs. 4 and 5). It should be noted that in most of the studied variables, the effects of CAR were more potent and significant than the SKEO.
Our previous findings showed that SKEO (200 mg/kg), which contains more than $94\%$ CAR, inhibits astrocytes, neurons, and blood origin-infiltrated cells, all of which are cytokines-producing sources, lowering pro-inflammatory cytokines such as IL-1β, TNF-α, and IL-6, and resulting in considerably reduced brain injuries13.
The inflammatory response, which plays a crucial role in developing secondary damage such as cell death, is related to the generation of pro-inflammatory cytokines. Acute reduction in the inflammatory response has been implicated in reducing the morbidity and mortality associated with trauma50. We found significant neuronal cell death after TBI in this work, as demonstrated by alterations in VCS scores, elevated cleaved caspase-3 along with neuron damage (NSE-IHC) (Table 3, Figs. 2 and 6), consistent with earlier findings10,51,52, while SKEO and CAR therapy mitigated these changes.
The overall pathophysiology of TBI in humans and animals is associated with apoptosis or programmed cell death of neurons and glia53 and is regulated by several interconnected mechanisms, including extrinsic and intrinsic signaling pathways. In the weight drop TBI model, stimulation of the cell-extrinsic pathway triggers the binding of ligands, such as TNF-α and FasL, to their receptors on the cell surface, whereas higher permeabilization of Bax/Bak channels on the outer mitochondria membrane are the central events in the intrinsic pathway, which leads to the cytochrome c release, a protein that plays a crucial role in cell death and is typically inhibited by the Bcl-2 protein. Then, cytochrome c forms the apoptosome complex in the cytosol. These two mechanisms converge at the level of effector caspases, such as caspase-3 and caspase-7, resulting in the cleavage of cellular proteins and apoptosis51,54–56. It has also been demonstrated that decreasing caspase-3 activation and apoptotic cell death promotes functional recovery in animals following TBI57,58.
Anti-apoptotic genes like Bcl-259 and pro-apoptotic genes like Bax60 belong to the Bcl-2 multigene superfamily. Following localized and global ischemia and TBI, damaged neurons showed decreased Bcl-2 and elevated Bax immunoreactivity61–63. The ratio of anti-apoptotic and pro-apoptotic Bcl-2 superfamily proteins appears to be essential for cell survival following CNS damage and in other cells20,64–68. After TBI in rat brains, changes in the cellular Bax:Bcl-2 ratio may mediate cell death via modulating the activity of the cell death-inducing caspase family of proteases69,70.
In the present study, TBI induces an increase in the Bax:Bcl-2 ratio in the TBI + Veh group, whereas SKEO and CAR treatments cause a more substantial decline in the Bax:Bcl-2 ratio in rat brains (Fig. 3). As a result, the decrease in the Bax:Bcl-2 protein ratio in the brains of rats treated with SKEO and CAR may be critical for reducing cleaved caspase-3 synthesis and, consequently, neurological scores improvement. Consistent with these results, Peter et al.32 reported neurological function improvement after CAR administration in a mouse model of weight drop closed head injury. Another research study showed that administration of *Satureja khuzistanica* extract reduced motor deficits in diabetic rats, which has been associated with a reduction in cleaved caspase-3 protein and Bax:Bcl-2 ratio in rat dorsal root ganglion21.
There is growing evidence that pro-apoptotic family members promote cytokine release into the cytoplasm, which leads to caspase activation10–12. CAR-rich SKEO has been demonstrated to reduce anti-inflammatory and neuroprotective properties, but the full mechanistic range of CAR and SKEO remains unknown31. Therefore, a rat model was used to investigate whether SKEO and CAR may help prevent neuroinflammation-induced apoptosis following TBI. Findings of our previous study71 and current results showed that TBI induced inflammation in the rat brains by overexpressing IL-1β, TNF-α, and IL-6, while SKEO200 and CAR200 treatment alleviated these inflammatory markers (Fig. 4). According to Li et al. findings, CAR decreased the levels of TNF-α and IL-1β, as well as the production of inducible nitric oxide synthase (iNOS) and cyclooxygenase-2 (COX-2) in ischemic cortical tissues. They also found that CAR inhibited tissue invasion by cellular markers of inflammatory enzymes72.
*Microglia* generate members of the pro-inflammatory cytokine family, which raise BBB permeability73. Brain edema and neuronal degeneration were linked to increased pro-inflammatory cytokines74. The findings of the present study, in agreement with our previous study11, demonstrated an increase in cerebral edema accompanied by up-regulation levels of pro-inflammatory cytokines following TBI in rat brains. The ability of CAR to minimize edema has been shown in previous research, which supports CAR to diminish edema and inflammatory responses by suppressing TNF-α75. Zhong et al. demonstrated the preventive effects of CAR on an intracerebral hemorrhage by reducing cerebral edema33. In the current study, CAR200 dramatically reduced cerebral edema and levels of pro-inflammatory cytokines, IL-1β, TNF-α, and IL-6. In our previous study, the Evans blue dye extravasation was used to quantify BBB permeability and revealed that SKEO200 reduced the water accumulation in the brain parenchyma due to improved vascular consistency13. Accordingly, we predict that the reduction in cerebral edema in the SKEO-treated group in this study was probably through a similar vascular mechanism (Fig. 1). Besides this, our recent finding showed that CAR potentially protects BBB via zonula occludens-1(ZO-1)/occludin, tight junction proteins, in the rat76. Consistent with our results, Park and colleagues revealed the efficacy of CAR in reducing vascular permeability following spinal cord injury by raising the tight junction and adhesion proteins levels77. Increased BBB permeability allows blood cells such as neutrophils, lymphocytes, monocytes, and macrophages to enter the central nervous system78, followed by releasing mediators such as prostaglandins, free radicals, and inflammatory cytokines, which trigger the production of chemokines and adhesion molecules and activate immune cells and glia79. As a result, it appears that a reduction in vascular permeability is one of the reasons for the reduction of edema and inflammatory cytokines by CAR in the current study.
TNF-α not only promotes the expression of other interleukins, like IL-6, which cause nerve impairment, BBB damage, and acute neurological problems in injury sites79,80, but it can also trigger programmed cell death in neurons81,82. Cytokines also have a role in apoptosis and inflammation by degrading IkappaB kinase (IκB) and activating the NF-κB signaling pathway, which is critical in initiating apoptosis83. NF-κB is a proinflammatory regulatory cytokine that can enhance the effects of other inflammatory signals84. Due to binding to the IκB inhibitor, the NF-κB transcription factor complex is generally retained in the cytoplasm in its inactive form. Upon activation, IκB is ubiquitinated and degraded, allowing the NF-B complex to be phosphorylated and transported into the nucleus, where it can bind to promoter sites to regulate the transcription of target genes85. Nearly the whole arsenal of immunological defenders is triggered by NF-κB, including chemokines, cytokines, adhesion molecules, inflammatory mediators, and apoptosis inhibitors, giving NF-κB a crucial role in global immunity86 that is conserved across species87. NF-κB is also implicated in the downregulation of pro-survival brain-derived neurotrophic factor (BDNF) at 24 h following exposure to high extracellular glutamate levels88, a common occurrence after mechanical depolarization during TBI. In an experimental model of closed-head injury, repression of the NF-κB inhibitor system enhanced neuronal cell death, worsened neurological outcomes, and accelerated post-traumatic mortality rate. It has been observed that NF-κB activation continues for a long time in glial cells and neurons following TBI, and it might regulate pro-apoptotic factors such as Bax protein89. In this regard, Li et al. reported the inhibitory effect of CAR on the NF-κB p65 (active form) that was generated after ischemia–reperfusion in rat brains72. In the present study, SKEO and CAR treatments efficiently reduced NF-κBp65 protein expression, indicating that the plant-origin essential oil, SKJ, and commercial CAR may attenuate post-traumatic inflammatory responses by modifying NF-κB signaling (Fig. 5) by decreasing Bax protein and ultimately caspase-3 protein activation causing improved VCS. Because CAR is the suspected principal active component in SKEO, it is plausible to assume that CAR is responsible for SKEO protective effects on post-traumatic outcomes and that the reduction in NF-κBp65 protein in the TBI + SKEO group was related to its CAR content performance. While NF-κB signaling is predominantly associated with pathological processes, there is growing evidence that it can mediate neuroprotective pathways through the upregulation of hemoxygenase 1 (HO-1), early growth response protein 1 (Egr-1), and heat shock protein β-1 of 27 KB (Hsp27) after excitotoxicity and stroke88,90, which requires further consideration for the impact on NF-κB signaling associated with chronic inflammation after experimental TBI91–94.
In the present study, although SKEO contains a significant amount of carvacrol ($91.33\%$), it showed weaker effects than pure CAR. *In* general, essential oils have multifunctional activities due to the presence of various components. Numerous comparative studies have examined the effects of essential oils with the most constituent components. They indicated that the active substances are not necessarily more potent than the essential oils. For example, Magierowicza et al. investigated the biopesticide activity of *Satureja hortensis* essential oil containing mainly CAR ($73.24\%$) and purified CAR. They found that the essential oil was more effective than its active ingredients alone95. Another comparative study showed that CAR-rich SKEO ($87.16\%$) has the same antimicrobial properties as commercial CAR96.
Extensive research has shown that the activity of essential oils is influenced by various factors such as seasonal variation and harvesting month. These factors are so important that they can lead to changes in potency and poor performance of the essential oil or existing compounds. For example, it is said that the famous essential oil called Ocimum gratissimum, popularly known as basil essential oil, has variable activities on the mouse central nervous system depending on the harvest season and essence extraction97. Other environmental factors can also affect the activity of essential oils, such as day/night temperature, rainfall and intensity of sunlight98. Fortunately, many of the above factors are compensable or adjustable99. Therefore, the difference between the TBI + SKEO and TBI + CAR groups in our study may be due to the aforementioned interfering factors.
According to the results found in the present study, it can be concluded that SKEO and CAR can reduce acute edema and inhibit the production of pro-inflammatory cytokines regulator NF-κB and apoptotic-related proteins caspase 3, resulting in the preservation of neuronal death and acute neurological functions in this model of TBI. The long-term outcomes require further investigation to confirm if a single treatment of SKEO or CAR can prevent persisting TBI-induced symptoms.
## Supplementary Information
Supplementary Legends. Supplementary Figure 1.Supplementary Figure 2.Supplementary Figure 3.Supplementary Figure 4. The online version contains supplementary material available at 10.1038/s41598-023-31891-3.
## References
1. Hyder AA, Wunderlich CA, Puvanachandra P, Gururaj G, Kobusingye OC. **The impact of traumatic brain injuries: A global perspective**. *NeuroRehabilitation* (2007.0) **22** 341-353. DOI: 10.3233/NRE-2007-22502
2. 2.Global status report on road safety 2018: summary. Geneva: World Health Organization; (WHO/NMH/NVI/18.20). License: CC BY-NC-SA 3.0 IGO). (2018).
3. Dewan MC. **Estimating the global incidence of traumatic brain injury**. *J. Neurosurg.* (2018.0) **130** 1080-1097. DOI: 10.3171/2017.10.JNS17352
4. Williams J. **Cost-effectiveness analysis of tranexamic acid for the treatment of traumatic brain injury, based on the results of the CRASH-3 randomised trial: A decision modelling approach**. *BMJ Glob. Health* (2020.0) **5** e002716. DOI: 10.1136/bmjgh-2020-002716
5. 5.Thapa, K., Khan, H., Singh, T. G. & Kaur, A. Traumatic brain injury: Mechanistic insight on pathophysiology and potential therapeutic targets. J. Mol. Neurosci.71(9), 1725–1742 (2021).
6. Jang C-H, Choi J-H, Byun M-S, Jue D-M. **Chloroquine inhibits production of TNF-α, IL-1β and IL-6 from lipopolysaccharide-stimulated human monocytes/macrophages by different modes**. *Rheumatology* (2006.0) **45** 703-710. DOI: 10.1093/rheumatology/kei282
7. Lenzlinger PM, Morganti-Kossmann M-C, Laurer HL, McIntosh TK. **The duality of the inflammatory response to traumatic brain injury**. *Mol. Neurobiol.* (2001.0) **24** 169-181. DOI: 10.1385/MN:24:1-3:169
8. Chio C-C. **Exercise attenuates neurological deficits by stimulating a critical HSP70/NF-κB/IL-6/synapsin I axis in traumatic brain injury rats**. *J. Neuroinflamm.* (2017.0) **14** 1-18. DOI: 10.1186/s12974-017-0867-9
9. Neurath MF, Pettersson S, ZumBüschenfelde K-HM, Strober W. **Local administration of antisense phosphorothiate olignucleotides to the p65 subunit of NF-κB abrogates established experimental colitis in mice**. *Nat. Med.* (1996.0) **2** 998-1004. DOI: 10.1038/nm0996-998
10. Zhang MH. **Neuroprotective effects of dexmedetomidine on traumatic brain injury: Involvement of neuronal apoptosis and HSP70 expression**. *Mol. Med. Rep.* (2018.0) **17** 8079-8086. PMID: 29693126
11. Xue Z. **Calcium-sensing receptor antagonist NPS2390 attenuates neuronal apoptosis though intrinsic pathway following traumatic brain injury in rats**. *Biochem. Biophys. Res. Commun.* (2017.0) **486** 589-594. DOI: 10.1016/j.bbrc.2017.03.097
12. Zhao W-Y. **Establishment of an ideal time window model in hypothermic-targeted temperature management after traumatic brain injury in rats**. *Brain Res.* (2017.0) **1669** 141-149. DOI: 10.1016/j.brainres.2017.06.006
13. Abbasloo E. **The anti-inflammatory properties of**. *Sci. Rep.* (2016.0) **6** 31866. DOI: 10.1038/srep31866
14. Hadian J, Hossein Mirjalili M, Reza Kanani M, Salehnia A, Ganjipoor P. **Phytochemical and morphological characterization of**. *Chem. Biodivers.* (2011.0) **8** 902-915. DOI: 10.1002/cbdv.201000249
15. Abdollahi M. **Antioxidant, antidiabetic, antihyperlipidemic, reproduction stimulatory properties and safety of essential oil of**. *Med. Sci. Monit.* (2003.0) **9** 331-335
16. Khosravinia H. **Hypolipidemic effects of carvacrol in relation with sex hormones in broiler chicken**. *Biotechnol. Anim. Husb.* (2014.0) **30** 89-102. DOI: 10.2298/BAH1401089K
17. Khosravinia H. **Hypolipidemic effects of**. *Kafkas Univ. Vet. Fak. Derg.* (2015.0) **21** 203-209
18. 18.Saberi, A. et al.Satureja khuzestanica extract elicits antinociceptive activity in several model of pain in rats. (2013).
19. Amanlou M, Dadkhah F, Salehnia A, Farsam H, Dehpour AR. **An anti-inflammatory and anti-nociceptive effects of hydroalcoholic extract of**. *J. Pharm. Pharm. Sci.* (2005.0) **8** 102-106. PMID: 15946603
20. Al Seyedan A, Dezfoulian O, Alirezaei M. *Res. Pharm. Sci.* (2020.0) **15** 481. DOI: 10.4103/1735-5362.297851
21. Kaeidi A. *J. Nat. Med.* (2013.0) **67** 61-69. DOI: 10.1007/s11418-012-0646-y
22. Ghazanfari G. **Biochemical and histopathological evidences for beneficial effects of**. *Toxicol. Mech. Methods* (2006.0) **16** 365-372. DOI: 10.1080/15376520600620125
23. 23.Pirbalouti, A. G. & Moalem, E. Variation in antibacterial activity of different ecotypes of Satureja khuzestanica Jamzad, as an Iranian endemic plant. (2013).
24. Turgut K, Özyiğit Y, Tütüncü B, Sözmen EU. **Agronomic and chemical performance of selected**. *Turk. J. Agric. For.* (2017.0) **41** 272-277. DOI: 10.3906/tar-1612-75
25. Trabace L. **Estrous cycle affects the neurochemical and neurobehavioral profile of carvacrol-treated female rats**. *Toxicol. Appl. Pharmacol.* (2011.0) **255** 169-175. DOI: 10.1016/j.taap.2011.06.011
26. Melusova M, Slamenova D, Kozics K, Jantova S, Horvathova E. **Carvacrol and rosemary essential oil manifest cytotoxic, DNA-protective and pro-apoptotic effect having no effect on DNA repair**. *Neoplasma* (2014.0) **61** 690-699. DOI: 10.4149/neo_2014_084
27. 27.Samarghandian, S., Farkhondeh, T., Samini, F. & Borji, A. Protective effects of carvacrol against oxidative stress induced by chronic stress in rat’s brain, liver, and kidney. Biochem. Res. Int.2016, 1–7 (2016).
28. Suntres ZE, Coccimiglio J, Alipour M. **The bioactivity and toxicological actions of carvacrol**. *Crit. Rev. Food Sci. Nutr.* (2015.0) **55** 304-318. DOI: 10.1080/10408398.2011.653458
29. Hong DK. **Carvacrol attenuates hippocampal neuronal death after global cerebral ischemia via inhibition of transient receptor potential melastatin 7**. *Cells* (2018.0) **7** 231. DOI: 10.3390/cells7120231
30. Jiang ZS, Pu ZC, Hao ZH. **Carvacrol protects against spinal cord injury in rats via suppressing oxidative stress and the endothelial nitric oxide synthase pathway**. *Mol. Med. Rep.* (2015.0) **12** 5349-5354. DOI: 10.3892/mmr.2015.4045
31. 31.Zamanian, M. Y. et al. Carvacrol as a potential neuroprotective agent for neurological diseases: A systematic review article. CNS Neurol. Disord.-Drug Targets (Formerly Curr. Drug Targets-CNS Neurol. Disord.) (2021).
32. Peters M. **Carvacrol together with TRPC1 elimination improve functional recovery after traumatic brain injury in mice**. *J. Neurotrauma* (2012.0) **29** 2831-2834. DOI: 10.1089/neu.2012.2575
33. Zhong Z. **Carvacrol alleviates cerebral edema by modulating AQP4 expression after intracerebral hemorrhage in mice**. *Neurosci. Lett.* (2013.0) **555** 24-29. DOI: 10.1016/j.neulet.2013.09.023
34. Abbasloo E. **The anti-inflammatory properties of**. *Sci. Rep.* (2016.0) **6** 1-12. PMID: 28442746
35. Haidar MA. **Mitoquinone helps combat the neurological, cognitive, and molecular consequences of open head traumatic brain injury at chronic time point**. *Biomedicines* (2022.0) **10** 250. DOI: 10.3390/biomedicines10020250
36. O'Connor CA, Cernak I, Vink R. **Both estrogen and progesterone attenuate edema formation following diffuse traumatic brain injury in rats**. *Brain Res.* (2005.0) **1062** 171-174. DOI: 10.1016/j.brainres.2005.09.011
37. Keshavarzi Z, Amiresmaili S, Shahrokhi N, Bibak B, Shakeri F. **Neuroprotective effects of auraptene following traumatic brain injury in male rats: The role of oxidative stress**. *Brain Res. Bull.* (2021.0) **177** 203-209. DOI: 10.1016/j.brainresbull.2021.09.021
38. Farahani F. **Possible involvement of female sex steroid hormones in intracellular signal transduction mediated by cytokines following traumatic brain injury**. *Brain Res. Bull.* (2022.0) **178** 108-119. DOI: 10.1016/j.brainresbull.2021.11.013
39. Meymandi MS. **Effects of pregabalin on brain edema, neurologic and histologic outcomes in experimental traumatic brain injury**. *Brain Res. Bull.* (2018.0) **140** 169-175. DOI: 10.1016/j.brainresbull.2018.05.001
40. King DR, Cohn SM, Proctor KG. **Changes in intracranial pressure, coagulation, and neurologic outcome after resuscitation from experimental traumatic brain injury with hetastarch**. *Surgery* (2004.0) **136** 355-363. DOI: 10.1016/j.surg.2004.05.011
41. Abdi E. **Expression of IGF-1, IL-27 and IL-35 receptors in adjuvant induced rheumatoid arthritis model**. *Iran J. Immunol.* (2018.0) **15** 14-27. PMID: 29549229
42. Rostamzadeh F. **Heterodimerization of apelin and opioid receptors and cardiac inotropic and lusitropic effects of apelin in 2K1C hypertension: Role of pERK1/2 and PKC**. *Life Sci.* (2017.0) **191** 24-33. DOI: 10.1016/j.lfs.2017.09.044
43. Abbasloo E, Najafipour H, Esmaeili-Mahani S. **Induction of antinociceptive tolerance to the chronic intrathecal administration of apelin-13 in rat**. *Neuropeptides* (2016.0) **60** 7-12. DOI: 10.1016/j.npep.2016.08.001
44. Abbasloo E, Najafipour H, Vakili A. **Chronic treatment with apelin, losartan and their combination reduces myocardial infarct size and improves cardiac mechanical function**. *Clin. Exp. Pharmacol. Physiol.* (2020.0) **47** 393-402. DOI: 10.1111/1440-1681.13195
45. Jiao L. **Edaravone alleviates delayed neuronal death and long-dated cognitive dysfunction of hippocampus after transient focal ischemia in Wistar rat brains**. *Neuroscience* (2011.0) **182** 177-183. DOI: 10.1016/j.neuroscience.2011.01.017
46. Dash PK, Zhao J, Hergenroeder G, Moore AN. **Biomarkers for the diagnosis, prognosis, and evaluation of treatment efficacy for traumatic brain injury**. *Neurotherapeutics* (2010.0) **7** 100-114. DOI: 10.1016/j.nurt.2009.10.019
47. Song H. **The AMPK-SIRT1-FoxO1-NF-kappaB signaling pathway participates in hesperetin-mediated neuroprotective effects against traumatic brain injury via the NLRP3 inflammasome**. *Immunopharmacol. Immunotoxicol.* (2022.0) **44** 970-983. DOI: 10.1080/08923973.2022.2096464
48. 48.Pati, S. K. et al. In Advanced Machine Learning Approaches in Cancer Prognosis: Challenges and Applications (eds Nayak, J. et al.) 13–73 (Springer International Publishing, 2021).
49. Krishna G. **Traumatic brain injury-induced sex-dependent changes in late-onset sensory hypersensitivity and glutamate neurotransmission**. *Front. Neurol.* (2020.0) **11** 749. DOI: 10.3389/fneur.2020.00749
50. Wang D. **Dexmedetomidine attenuates traumatic brain injury: Action pathway and mechanisms**. *Neural Regen. Res.* (2018.0) **13** 819. DOI: 10.4103/1673-5374.232529
51. Li F, Wang X, Zhang Z, Zhang X, Gao P. **Dexmedetomidine attenuates neuroinflammatory–induced apoptosis after traumatic brain injury via Nrf2 signaling pathway**. *Ann. Clin. Transl. Neurol.* (2019.0) **6** 1825-1835. DOI: 10.1002/acn3.50878
52. Polcyn R. **Neuron specific enolase is a potential target for regulating neuronal cell survival and death: Implications in neurodegeneration and regeneration**. *Neuroimmunol. Neuroinflamm.* (2017.0) **4** 254. DOI: 10.20517/2347-8659.2017.59
53. Raghupathi R, Graham DI, McINTOSH TK. **Apoptosis after traumatic brain injury**. *J. Neurotrauma* (2000.0) **17** 927-938. DOI: 10.1089/neu.2000.17.927
54. Zhang X, Chen Y, Jenkins LW, Kochanek PM, Clark RS. **Bench-to-bedside review: Apoptosis/programmed cell death triggered by traumatic brain injury**. *Crit. Care* (2004.0) **9** 1-10. DOI: 10.1186/cc2950
55. Cernak I, Chapman SM, Hamlin GP, Vink R. **Temporal characterisation of pro-and anti-apoptotic mechanisms following diffuse traumatic brain injury in rats**. *J. Clin. Neurosci.* (2002.0) **9** 565-572. DOI: 10.1054/jocn.2002.1132
56. Xing P. **The protection effect and mechanism of hyperbaric oxygen therapy in rat brain with traumatic injury 1**. *Acta cirurgica brasileira* (2018.0) **33** 341-353. DOI: 10.1590/s0102-865020180040000006
57. Lu D. **Atorvastatin reduces neurological deficit and increases synaptogenesis, angiogenesis, and neuronal survival in rats subjected to traumatic brain injury**. *J. Neurotrauma* (2004.0) **21** 21-32. DOI: 10.1089/089771504772695913
58. Wu H. **Increase in phosphorylation of Akt and its downstream signaling targets and suppression of apoptosis by simvastatin after traumatic brain injury**. *J. Neurosurg.* (2008.0) **109** 691-698. DOI: 10.3171/JNS/2008/109/10/0691
59. Boise LH. **bcl-x, a bcl-2-related gene that functions as a dominant regulator of apoptotic cell death**. *Cell* (1993.0) **74** 597-608. DOI: 10.1016/0092-8674(93)90508-N
60. Oltval ZN, Milliman CL, Korsmeyer SJ. **Bcl-2 heterodimerizes in vivo with a conserved homolog, Bax, that accelerates programed cell death**. *Cell* (1993.0) **74** 609-619. DOI: 10.1016/0092-8674(93)90509-O
61. Gillardon F. **Altered expression of Bcl-2, Bcl-X, Bax, and c-Fos colocalizes with DNA fragmentation and ischemic cell damage following middle cerebral artery occlusion in rats**. *Mol. Brain Res.* (1996.0) **40** 254-260. DOI: 10.1016/0169-328X(96)00059-9
62. Krajewski S. **Upregulation of bax protein levels in neurons following cerebral ischemia**. *J. Neurosci.* (1995.0) **15** 6364-6376. DOI: 10.1523/JNEUROSCI.15-10-06364.1995
63. Zhang W, Hong J, Zheng W, Liu A, Yang Y. **High glucose exacerbates neuroinflammation and apoptosis at the intermediate stage after post-traumatic brain injury**. *Aging* (2021.0) **13** 16088. DOI: 10.18632/aging.203136
64. Korsmeyer SJ. **Regulators of cell death**. *Trends Genet.* (1995.0) **11** 101-105. DOI: 10.1016/S0168-9525(00)89010-1
65. Moorjani N. **Upregulation of Bcl-2 proteins during the transition to pressure overload-induced heart failure**. *Int. J. Cardiol.* (2007.0) **116** 27-33. DOI: 10.1016/j.ijcard.2006.04.037
66. Sakinah SS, Handayani ST, Hawariah LA. **Zerumbone induced apoptosis in liver cancer cells via modulation of Bax/Bcl-2 ratio**. *Cancer Cell Int.* (2007.0) **7** 1-11. PMID: 17331262
67. Cheng C-Y, Tang N-Y, Kao S-T, Hsieh C-L. **Ferulic acid administered at various time points protects against cerebral infarction by activating p38 MAPK/p90RSK/CREB/Bcl-2 anti-apoptotic signaling in the subacute phase of cerebral ischemia-reperfusion injury in rats**. *PLoS ONE* (2016.0) **11** e0155748. DOI: 10.1371/journal.pone.0155748
68. Chen W. **TRPM7 inhibitor carvacrol protects brain from neonatal hypoxic-ischemic injury**. *Mol. Brain* (2015.0) **8** 1-13. DOI: 10.1186/s13041-015-0102-5
69. Thornberry NA, Lazebnik Y. **Caspases: Enemies within**. *Science* (1998.0) **281** 1312-1316. DOI: 10.1126/science.281.5381.1312
70. Raghupathi R. **Temporal alterations in cellular Bax: Bcl-2 ratio following traumatic brain injury in the rat**. *J. Neurotrauma* (2003.0) **20** 421-435. DOI: 10.1089/089771503765355504
71. Khaksari M, Abbasloo E, Dehghan F, Soltani Z, Asadikaram G. **The brain cytokine levels are modulated by estrogen following traumatic brain injury: Which estrogen receptor serves as modulator?**. *Int. Immunopharmacol.* (2015.0) **28** 279-287. DOI: 10.1016/j.intimp.2015.05.046
72. Li Z, Hua C, Pan X, Fu X, Wu W. **Carvacrol exerts neuroprotective effects via suppression of the inflammatory response in middle cerebral artery occlusion rats**. *Inflammation* (2016.0) **39** 1566-1572. DOI: 10.1007/s10753-016-0392-5
73. Allan SM, Rothwell NJ. **Cytokines and acute neurodegeneration**. *Nat. Rev. Neurosci.* (2001.0) **2** 734-744. DOI: 10.1038/35094583
74. Ding K. **Melatonin reduced microglial activation and alleviated neuroinflammation induced neuron degeneration in experimental traumatic brain injury: Possible involvement of mTOR pathway**. *Neurochem. Int.* (2014.0) **76** 23-31. DOI: 10.1016/j.neuint.2014.06.015
75. Guimarães AG. **Carvacrol attenuates mechanical hypernociception and inflammatory response**. *Naunyn Schmiedebergs Arch. Pharmacol.* (2012.0) **385** 253-263. DOI: 10.1007/s00210-011-0715-x
76. Lee DK. **Characterization of apelin, the ligand for the APJ receptor**. *J. Neurochem.* (2000.0) **74** 34-41. DOI: 10.1046/j.1471-4159.2000.0740034.x
77. 77.Park, C. S., Lee, J. Y., Choi, H. Y., Ju, B. G. & Yune, T. Y. Suppression of TRPM7 by carvacrol protects against injured spinal cord by inhibiting blood-spinal cord barrier disruption. (2021).
78. Chodobski A, Zink BJ, Szmydynger-Chodobska J. **Blood–brain barrier pathophysiology in traumatic brain injury**. *Transl. Stroke Res.* (2011.0) **2** 492-516. DOI: 10.1007/s12975-011-0125-x
79. Lucas SM, Rothwell NJ, Gibson RM. **The role of inflammation in CNS injury and disease**. *Br. J. Pharmacol.* (2006.0) **147** S232-S240. DOI: 10.1038/sj.bjp.0706400
80. Habib P, Dreymueller D, Ludwig A, Beyer C, Dang J. **Sex steroid hormone-mediated functional regulation of microglia-like BV-2 cells during hypoxia**. *J. Steroid Biochem. Mol. Biol.* (2013.0) **138** 195-205. DOI: 10.1016/j.jsbmb.2013.06.003
81. Westmoreland SV, Kolson D, Gonzalez-scarano F. **Toxicity of TNFα and platelet activating factor for human NT2N neurons: A tissue culture model for human immunodeficiency virus dementia**. *J. Neurovirol.* (1996.0) **2** 118-126. DOI: 10.3109/13550289609146545
82. Shohami E, Gallily R, Mechoulam R, Bass R, Ben-Hur T. **Cytokine production in the brain following closed head injury: Dexanabinol (HU-211) is a novel TNF-α inhibitor and an effective neuroprotectant**. *J. Neuroimmunol.* (1997.0) **72** 169-177. DOI: 10.1016/S0165-5728(96)00181-6
83. Shojo H. **Genetic and histologic evidence implicates role of inflammation in traumatic brain injury-induced apoptosis in the rat cerebral cortex following moderate fluid percussion injury**. *Neuroscience* (2010.0) **171** 1273-1282. DOI: 10.1016/j.neuroscience.2010.10.018
84. Bortolotti P, Faure E, Kipnis E. **Inflammasomes in tissue damages and immune disorders after trauma**. *Front. Immunol.* (2018.0) **9** 1900. DOI: 10.3389/fimmu.2018.01900
85. Hoffmann A, Baltimore D. **Circuitry of nuclear factor κB signaling**. *Immunol. Rev.* (2006.0) **210** 171-186. DOI: 10.1111/j.0105-2896.2006.00375.x
86. Lenardo MJ, Fan C-M, Maniatis T, Baltimore D. **The involvement of NF-κB in β-interferon gene regulation reveals its role as widely inducible mediator of signal transduction**. *Cell* (1989.0) **57** 287-294. DOI: 10.1016/0092-8674(89)90966-5
87. Buhlman LM, Krishna G, Jones TB, Thomas TC. **Drosophila as a model to explore secondary injury cascades after traumatic brain injury**. *Biomed. Pharmacother.* (2021.0) **142** 112079. DOI: 10.1016/j.biopha.2021.112079
88. Caba E. **Excitotoxic stimulation activates distinct pathogenic and protective expression signatures in the hippocampus**. *J. Cell Mol. Med.* (2021.0) **25** 9011-9027. DOI: 10.1111/jcmm.16864
89. Sivandzade F, Prasad S, Bhalerao A, Cucullo L. **NRF2 and NF-қB interplay in cerebrovascular and neurodegenerative disorders: Molecular mechanisms and possible therapeutic approaches**. *Redox Biol.* (2019.0) **21** 101059. DOI: 10.1016/j.redox.2018.11.017
90. Rickhag M. **Comprehensive regional and temporal gene expression profiling of the rat brain during the first 24 h after experimental stroke identifies dynamic ischemia-induced gene expression patterns, and reveals a biphasic activation of genes in surviving tissue**. *J. Neurochem.* (2006.0) **96** 14-29. DOI: 10.1111/j.1471-4159.2005.03508.x
91. Sabetta Z, Krishna G, Curry T, Adelson PD, Thomas TC. **Aging with TBI vs. aging: 6-month temporal profiles for neuropathology and astrocyte activation converge in behaviorally relevant thalamocortical circuitry of male and female rats**. *bioRxiv* (2023.0). DOI: 10.1101/2023.02.06.527058
92. Thomas TC. **Does time heal all wounds? Experimental diffuse traumatic brain injury results in persisting histopathology in the thalamus**. *Behav. Brain Res.* (2018.0) **340** 137-146. DOI: 10.1016/j.bbr.2016.12.038
93. Bromberg CE. **Sex-dependent pathology in the HPA axis at a sub-acute period after experimental traumatic brain injury**. *Front. Neurol.* (2020.0) **11** 946. DOI: 10.3389/fneur.2020.00946
94. Beitchman JA. **Experimental traumatic brain injury induces chronic glutamatergic dysfunction in amygdala circuitry known to regulate anxiety-like behavior**. *Front. Neurosci.* (2019.0) **13** 1434. DOI: 10.3389/fnins.2019.01434
95. Magierowicz K, Górska-Drabik E, Sempruch C. **The insecticidal activity of**. *Pesticide Biochem. Physiol.* (2019.0) **153** 122-128. DOI: 10.1016/j.pestbp.2018.11.010
96. Mazarei Z, Rafati H. **Nanoemulsification of**. *LWT* (2019.0) **100** 328-334. DOI: 10.1016/j.lwt.2018.10.094
97. Freire CMM, Marques MOM, Costa M. **Effects of seasonal variation on the central nervous system activity of**. *J. Ethnopharmacol.* (2006.0) **105** 161-166. DOI: 10.1016/j.jep.2005.10.013
98. Soni U, Brar S, Gauttam VK. **Effect of seasonal variation on secondary metabolites of medicinal plants**. *Int. J. Pharm. Sci. Res.* (2015.0) **6** 3654-3662
99. Najafian S, Zahedifar M. **Antioxidant activity and essential oil composition of**. *J. Sci. Food Agric.* (2015.0) **95** 2404-2408. DOI: 10.1002/jsfa.6959
|
---
title: The inhibitive action of 2-mercaptobenzothiazole on the porosity of corrosion
film formed on aluminum and aluminum–titanium alloys in hydrochloric acid solution
authors:
- Abdel-Rahman El-Sayed
- Morad M. El-Hendawy
- Mohamed Sarwat El-Mahdy
- Fatma S. M. Hassan
- Adila E. Mohamed
journal: Scientific Reports
year: 2023
pmcid: PMC10036543
doi: 10.1038/s41598-023-31795-2
license: CC BY 4.0
---
# The inhibitive action of 2-mercaptobenzothiazole on the porosity of corrosion film formed on aluminum and aluminum–titanium alloys in hydrochloric acid solution
## Abstract
2-Mercaptobenzothiazole (2-MBT) in a solution of 0.5 M HCl is an effective corrosion inhibitor for aluminum and aluminum–titanium alloys. Tafel polarization and electrochemical impedance spectroscopy (EIS) were employed to assess this heterocyclic compound’s anticorrosive potential and complementary by scanning electron microscope (SEM) and calculating porosity percentage in the absence and presence of various inhibitor concentrations. Inhibition efficiency (IE%) was strongly related to concentration (10–6–10–3 M). Temperature’s effect on corrosion behavior was investigated. The data exhibited that the IE% decreases as the temperature increases. An increase in activation energy (Ea) with increasing the inhibitor concentration and the decrease in the IE% value of the mentioned compound with raising the temperature indicates that the inhibitor molecules are adsorbed physically on the surface. Thermodynamic activation parameters for Al and Al–Ti alloy dissolution in both 0.5 M HCl and the inhibited solution were calculated and discussed. According to Langmuir’s adsorption isotherm, the inhibitor molecules were adsorbed. The evaluated standard values of the enthalpy (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta {H}_{ads.}^{o})$$\end{document}ΔHads.o), entropy (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta {S}_{ads.}^{o})$$\end{document}ΔSads.o) and free energy changes (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta {G}_{ads.}^{o})$$\end{document}ΔGads.o) showed that \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta {H}_{ads.}^{o}$$\end{document}ΔHads.o and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta {G}_{ads.}^{o}$$\end{document}ΔGads.o are negative, while \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta {S}_{ads.}^{o}$$\end{document}ΔSads.o was positive. The formation of a protective layer adsorbed on the surfaces of the substrates was confirmed with the surface analysis (SEM). The porosity percentage is significantly reduced in the inhibitor presence and gradually decreased with increasing concentration. Furthermore, the density functional theory (DFT) and Monte Carlo (MC) simulations were employed to explain the variance in protecting the Al surface from corrosion. Interestingly, the theoretical findings align with their experimental counterparts. The planarity of 2-MBT and the presence of heteroatoms are the playmakers in the adsorption process.
## Introduction
For decades, lightweight alloys based on aluminum (Al) and titanium (Ti) have been debated and researched. The Al–Ti binary system’s resultant alloys with hard intermetallic phases typically show good microstructural and mechanical characteristics1. Al–Ti alloy is characterized by low densities, high specific strength, and reduced manufacturing cost. Aluminum alloys are frequently used in aircraft because of their high strength to weigh ratio2. Adding Ti to Al increases the material strength, but the presence of Ti in the Al–Ti alloy increases the corrosion rate compared with pure Al. Owing to the generation of intermetallic phase on the Al–Ti alloy surface, it is more susceptible to corrosive attacks1. Therefore, the presence of Al3Ti phases in the alloy can be catalyzed via chemical and electrochemical processes on the alloy surface. For this reason, working anodes and cathodes contribute to matrix dealloying3. However, the protective coating of Al2O3 that forms on the surface appears chemically unstable in both alkaline and acidic conditions. Subsequently, these environments are harmful to aluminum and its alloys4.
It is well known that Al and its alloys showed a high corrosion rate in acidic solutions. HCl is used to etch Al and its alloys chemically and electrochemically. Also, HCl solution is used to remove the oxides forming on the surfaces. Therefore, the corrosion rates of the investigated metal and its alloys should be retarded using organic inhibitors during the etching process. The corrosion processes were inhibited by the organic additives introduced to the acidic solution is due to the surface adsorption of the inhibitor molecules3,5. This adsorption occurs by physical adsorption through electrostatic action between the surface and organic molecules. However, chemisorption occurs by forming coordination bonds between the surface and inhibitor molecules. These studies aim to prevent the metal or alloy from a corrosive medium5. The first inhibiting mechanism in HCl solution is the organic compounds’ adsorption on the surface of metal or alloy. In order to explain the passive oxide layer breakdown that occurs when Cl− ions reach the film, many mechanisms have been mentioned6,7. One of such mechanism explains that Cl− ions may not enter the oxide layer; instead, they are chemisorbed on this layer5.
Generally, the selection of organic inhibitors containing sulfur, nitrogen and/or oxygen as polar groups, which are connected with double bonds in their compositions, have been recognized as good corrosion inhibitors for various alloys and metals in acidic solutions8–11. Some factors play an essential role in the adsorption process on the surface of metal or alloy, such as the nature and the type of charge on the surface, the type of the studied solution, and the chemical structure of the organic compound12.
Aluminum exposed to acidic or alkaline solutions should be treated with an efficient corrosion inhibitor. Corrosion inhibitor procedures are defined as either cathodic, anodic, or mixed inhibition based on their capacity to slow metal dissolution and reduction. Organic inhibitors generally have a dissimilar influence on both the cathodic and anodic processes13–15. The corrosion inhibition efficiency of imidazolium-based inhibitors in H2SO4 on mild steel was explored experimentally and theoretically16. The anticorrosive capabilities of benzimidazole and its derivatives are due to the π-electrons on the planar-fused moiety and the lone pair of electrons on the hetero-atoms. Using DFT and molecular dynamics simulations, Obot et al.17 investigated the adsorption mechanism of 2-mercaptobenzimidazole (2-MBI) as a corrosion inhibitor for Fe [110], Cu [111], and Al [111] surfaces. The data shows the higher anticorrosive performance of 2-MBI on steel corrosion compared to aluminum and copper.
Several investigations have already been conducted on the corrosion inhibition properties of 2-mercaptobenzothiazole (2-MBT) on AA6082 alloy, Cu, C-steel, and AA 2024-T3 alloy (Cu-rich intermetallic particles)6,18–20. The results of these studies show that 2-MBT can operate as a good corrosion inhibitor. However, due to the diversity of organic molecules, producing a highly efficient inhibitor faces significant obstacles. An extensive review of the relevant documented sources confirms that 2-mercaptobenzothiazole was not previously been used as a corrosion inhibitor for Al and Al–Ti alloy.
This work aims to introduce an electrochemical study of the new effect of (2-MBT) on corrosion inhibition of aluminum and aluminum–titanium alloys in HCl (0.5 M) solutions using Tafel polarization and electrochemical impedance spectroscopy (EIS) techniques at various concentrations and temperatures, as well as a scanning electron microscope (SEM) for surface characterization. On the other hand, the DFT and MC approaches were applied to relate the detected inhibition efficiency associated with the quantum chemical descriptors of the studied inhibitor and the adsorption parameters of the inhibitor/Al[111] complexes. Correspondingly, the porosity percentage on the surfaces of Al and Al–Ti alloy was evaluated using both SEM micrographs and Tafel polarization data in both the absence and presence of the tested inhibitor.
## Materials and solutions
Al and Ti were the starting components, both being $99.99\%$ pure. Al ingots were blasted to a temperature of more than 660 °C before various amounts of Ti (0, 1, 2, 5, and 8 wt%) were supplied. The molten solution was elevated to a pouring temperature (900–1400 °C) using a 200-kW medium frequency induction furnace (type ABB, Germany) with protective Ar gas after each Ti addition. Finally, each alloy melt was poured into a worm-shaped cast iron mold. The morphology and composition of the studied alloys have been assessed utilizing SEM and X-ray diffraction. The homogeneous composition of the solid solution phase was found to exist1.
Table 1 shows the compositions of Al and Al–Ti alloys. 2-Mercaptobenzothiazole (2-MBT) was obtained from Alfa Aesar (Fig. 1). It was employed at quantities ranging from 1 × 10−6 to 1 × 10−3 M without further purification. Diluting AR grade HCl with water yielded the corrosive solution (0.5 M HCl).Table 1Wt% of the prepared Al–Ti alloy. AlloyAlAT-1AT-2AT-3AT-4Wt%Al100Ti0Al98.7Ti1.3Al98.27Ti1.73Al94.7Ti5.3Al91.27Ti8.73Figure 1Structure of 2-mercaptobenzothiazole (2-MBT).
## Electrochemical measurements
Using emery paper grades 200–1000–2000–4000, the working electrodes (1 cm diameter) were polished to a mirror-like prior to each experiment. Before being placed in the polarization cell, the working electrodes were degreased in pure ethanol and acetone and washed in flowing bidistilled water. Electrochemical testing with a three-electrode setup was carried out using the VersaSTAT4 potentiostat. Platinum mesh, calomel electrodes, and Al–Ti alloys have been utilized as the counter, reference, and working electrodes, respectively. The working electrode’s diameter was one centimeter. Electrochemical tests began after the open circuit potential (OCP) had been stable for 30 min. Potentiodynamic polarization tests were performed using a scan rate of 1 mV/s and a voltage of ± 0.25 versus OCP. EIS was carried out using a 10 mV sinusoidal perturbation electric potential signal at 10 points per decade OCP increments at frequencies between 100 kHz and 10 mHz. For data fitting, ZsimpWin version 3.6 was utilized. All electrochemical experiments were performed between 25 and 55 ± 0.5 °C. A scanning electron microscope was used to examine the surface morphology of Al and Al–Ti alloys (FE-SEM, QUANTAFEG 250, Netherlands).
## Methods for assessing corrosion parameters
After 30 min of immersion, the investigated electrodes reached a steady-state of open-circuit corrosion potential (OCP) in the absence and presence of the investigated inhibitor. Tafel Polarization was used to calculate the corrosion current density (icorr) and (Ecorr) of the examined electrodes in the absence and presence of the inhibitor21.
The inhibition efficiency (IE%) is calculated from:1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$IE\text{\%}=\left[\left({i}_{\text{free}}-{i}_{\text{inh}}\right)/{i}_{\text{free}}\right] \times 100,$$\end{document}IE\%=ifree-iinh/ifree×100,where ifree and iinh denote to the corrosion current densities of uninhibited and inhibited solution, respectively.
The surface coverage degree is evaluated as follows:2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\theta = \left[\left({i}_{\text{free}}-{i}_{\text{inh}}\right)/{i}_{\text{free}}\right].$$\end{document}θ=ifree-iinh/ifree.
## Details of the computation
The isolated molecules were subjected to quantum chemical computations by using Gaussian 16 software’s implementation of the B3LYP-D$\frac{3}{6}$-311++ G(2d, 2p)/B3LYP-D$\frac{3}{6}$-311(d, p) modeling chemistry22. Utilizing the universal solvation model, full geometry optimization of the isolated molecule has been carried out in the water phase (SMD)23. Frequency calculations follow this stage to determine the characteristics of the stationary points. Several quantum chemical descriptors were generated in this situation to find a correlation with the experimental results. The highest occupied molecular orbital’s energy is one of their characteristics (EHOMO), the lowest unoccupied molecular orbital (ELUMO), energy gap (ΔE), dipole moment (μ), global electronegativity (χ), chemical hardness (η), and electrophilicity index (ω), the fractions of electrons transferred (ΔN), and the total energy change (ΔET). In the past, the literature has reported on the mathematical descriptors of these descriptors24.
The Adsorption Locator module, built-in Materials Studio 201725, used Metropolis Monte Carlo simulations26 to pinpoint the most stable arrangement of the adsorbates on the aluminum surface. The Al [111] surface is the most stable of the many aluminum surfaces27. Consequently, we decided to use it to model the adsorption behavior. ( 2.9 2.9 5.9 nm) is the size of the simulation box when periodic boundary conditions are used. The metallic surface comprises five layers to guarantee adequate depth, each having 121 aluminum atoms.
By adding 5 five molecules of HCl, and 139 molecules of H2O to one inhibitor molecule on the surface of Al, the medium of HCl (0.5 M) was simulated. The simulation was performed utilizing the COMPASS force field28, and the electrostatic and energetic Van der Waals components were calculated using the Ewald and atom-based summations, correspondingly.
## Electrochemical behavior of Al and Al–Ti alloys in the absence and presence of 2-MBT
The Ecorr values for Al and Al–Ti electrodes (Table 2) show that for greater titanium concentrations (alloys AT-1, AT-2, AT-3, and AT-4) compared to pure Al, Ecorr moves to a more positive direction. This shift is related to a rise in the rate of corrosion in the case of 0.5 M HCl29. On the other hand, Ecorr of Al and Al–Ti alloys in 0.5 M HCl solution comprising different concentrations of 2-MBT inhibitor at 25 °C exhibit different shifts. In the case of Al, AT-2 (except for 2-MBT shifts to a positive value in the presence of 0.001 mM), and AT-4, the change in the values of potential Ecorr to negative proves that the inhibitor is a mixed type, and mainly cathodic30; with 2-MBT concentrations, this tendency is attributed to variations in the potential of the hydrogen evolution process toward more negative potentials. This shift may indicate that active cathodic sites are more blocked than anodic ones31. It is also evident from AT-1 and AT-3 alloy that the addition of inhibitor shifted corrosion potential Ecorr in an anodic direction, indicating that the inhibitor’s adsorption was more successful at anodic than cathodic sites32. The observed variation in Ecorr values has been repeatedly reported33–35. Specifically, if the Ecorr displacement is greater than 85 mV, the inhibitor would be classified as a cathodic or anodic type inhibitor. On the other hand, if the displacement is lower than 85 mV, the inhibitor would be related as mixed type. In the current study, the maximum value of Ecorr displacement was detected to be much lower than 85 mV, suggesting that 2-MBT is associated to the mixed-type inhibitor. Table 2Corrosion parameters of pure Al, AT-1, AT-2, AT-3, and AT-4 alloys after 30 min of exposure to 0.5 M HCl solution comprising different concentrations of 2-MBT inhibitor at a temperature of 25 °C.Metal and alloysInhibitor conc. ( mM)icorr (µA cm−2) − Ecorr (mV) (SCE)ba (mV dec−1) − bc (mV dec−1)θIE%AlBlank223.00832.5919.40122.300.001125.91850.8121.90122.600.4443.540.01100.53851.5220.30122.100.5554.920.191.26845.1224.20122.100.5959.08161.80845.8726.20123.200.7272.29AT-1Blank400.82822.0425.40127.600.001237.268794.8931.5122.30.4140.800.01174.047803.12523.2124.30.5756.580.1129.238812.67926.3123.10.6867.76196.792781.90833.41240.7675.85AT-2Blank503.84822.1622.70120.100.001307.268799.9127.5126.30.3939.010.01225.214843.70222.4123.40.5555.300.1174.737842.80623.11220.6565.321121.64854.79630121.30.7675.86AT-3Blank1001.99808.6532.20133.100.001724.748788.18429.8122.80.2827.670.01612.642793.10528.5123.10.3938.860.1448.302769.45125.3121.90.5555.261366.465774.02526.6123.30.6363.43AT-4Blank896.17804.3632.5129.80.001452.896826.33924.71230.4949.460.01371.996825.09230.9121.10.5858.490.1294.623839.64424.6121.80.6767.121180.088840.92327123.40.8079.90 Anodic and cathodic potential against current density were used to calculate the corrosion parameters concerning the Tafel potential areas36,37. Figure 2a,b depicts the data obtained from the experimental curves of polarization in the presence of 2-MBT concentrations in the case of Al and AT-2 alloy. In this context, Table 2 shows the experimental results derived from the polarization curves with adding various concentrations of 2-MBT. The icorr for the tested metal and its alloys in the presence and absence of the inhibitor has been estimated by extrapolating the cathodic and anodic lines of the Tafel polarization curve to the Ecorr. The presence of 2-MBT causes a decrease in both the anodic and cathodic branches of the polarization curves. However, the data exhibited that the inhibition action in the cathodic process is higher than in the anodic one. The anodic and cathodic potentials against current density were employed to compute the corrosion parameters. Figure 2Tafel polarizations for (a) Al and (b) AT-2 alloy in 0.5 M HCl in the absence and presence of various concentrations of 2-MBT at 25 °C.
Because bc and ba remain nearly unchanged in Table 2 compared to uninhibited solutions, the inhibitory action of 2-MBT does not alter the hydrogen evolution mechanism38. On the other side, it has been discovered that the cathodic Tafel slopes (bc) are greater than the anodic Tafel slopes (ba). As a result, one may hypothesize that the overall corrosion kinetics of all studied electrodes were under cathodic control39. Additionally, the 2-MBT is a mixed-type inhibitor (that possesses the capacity to retard both cathodic and anodic reactions). In the meantime, it influences the reaction of hydrogen evolution and the dissolution of Al and Al–Ti electrodes. This was supported by the observed shift in the Ecorr value obtained in the solution, which was less than 85 mV after adding 2-MBT compared to the bare HCl solution39. The Al–Ti phase on the alloy surface may be responsible for this opposing impact observed at different inhibitor concentrations in contrast to the pure Al31.
With increasing the concentration of 2-MBT, the icorr lowers, and the inhibition efficiency (IE %) of the examined electrodes increases Fig. 3a,b. For the AT-3 ($5.3\%$ Ti) alloy, the IE % of the tested at all the analyzed inhibitor concentrations is smaller than the equivalent value for Al and the other alloys (AT-1, AT-2, and AT-4). This effect could be explained by inhibitor molecules adsorbing is less on the mentioned alloy surface, owing to blocking active sites40 (see Table 2). At higher Ti content ($5.3\%$) in AT-3, a relative drop in IE% value is found compared to pure Al and other alloys at various inhibitor concentrations. This is attributed to a reduction of active sites on the surface, which causes the inhibitor molecules to be less adsorbable. This is likely owing to the generation of the solid solution phase, which reduces the heterogeneity1,41.Figure 3Comparison between both (a) icorr and (b) IE% with titanium percent in the alloy in 0.5 M solutions of HCl containing different concentrations of 2-MBT at 25 °C.
However, the inhibitory efficiency values for several examined alloys (AT-1, AT-2, and AT-4) appear almost similar at high concentrations. This result could be attributed to the fact that the adsorbed inhibitor molecules mainly occupied the electrode surface of the alloys under investigation. This has led to the suppression of the vast majority of active sites. In the same context, increasing the inhibitor concentration consequently increases the coverage and adsorption amount of inhibitor molecules, resulting in a noticeable improvement in the inhibition efficiency. It is generally believed that corrosion inhibitors protect metal surfaces by adsorbing the inhibitor molecules onto the metal surfaces42,43.
The surface porosity percentage fraction was estimated by potentiodynamic polarization (Tafel Polarization) data. In this case, the porosity percentage (PR%) can be calculated using the following equation44,45:3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${R}_{P}= \frac{\beta }{{i}_{corr.}}\text{ and }\beta = \frac{{b}_{a}{b}_{c}}{2.303\left({b}_{a}+ {b}_{c}\right)},$$\end{document}RP=βicorr.andβ=babc2.303ba+bc,4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${P}_{R}= \frac{{R}_{P}^{o}}{{R}_{P}} \times 100\%,$$\end{document}PR=RPoRP×$100\%$,where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${P}_{R}$$\end{document}PR, corresponds to the total porosity, while \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${R}_{P}^{o}$$\end{document}RPo and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${R}_{P}$$\end{document}RP are polarization resistance of the uninhibited and inhibited substrates (Al and Al–Ti alloys), respectively.
Figure 4 exhibits the porosity% variation with an increasing 2-MBT concentration with Al and its investigated alloys at 25 °C. It has been shown that the porosity% drops dramatically with increasing 2-MBT concentration. Consequently, corrosion resistance is increased in the case of inhibited substrates compared to uninhibited ones. Consequently, the observed data from Tafel plot measurements support this pattern. Figure 4Porosity percentage as a function of 2-MBT concentrations for Al and Al–Ti alloys.
## Scanning electron microscopy (SEM) micrographs
As illustrated in Fig. 5a–f, SEM images of Al, AT-1, and AT-2 surfaces were taken to determine the level of corrosion damage done in the absence and presence of inhibitor (2-MBT) on the surfaces. As a result of the extensive electrochemical corrosion caused by the immersion in 0.5 M HCl, the surfaces of the specimens had a highly rough texture with numerous deep pits on the surface. These data can be explained in terms of the activity of the chloride (Cl−) ion. Creating a rough surface helps to promote pit development by eliminating surface layers, allowing Cl− to react directly with the metal, and consequently, the dissolving of the metal in the solution is speeded46. However, when exposed to the inhibitor-containing media, comparatively smooth metal surfaces with fewer pits and their depths are perceived. Therefore, examining the surface morphology indicates the formation of a protective inhibitor coating that serves as a barrier between the surfaces and the aggressive acidic environments. This adherent coating enables superior corrosion prevention of aluminum and its studied alloys, particularly when 2-MBT is present. Figure 5SEM micrographs of Al (a,b), alloy AT-1(c,d) and alloy AT-2 (e,f) after immersion for 10 h in 0.5 M HCl in the absence (a,c,e) and presence of 1 mM of 2-MBT (b,d,f) at 25 °C, respectively.
The porosity values obtained using the SEM micrograph are shown in Table 3. Using the ImageJ program (Java version) image processing, we assessed the uninhibited and inhibited surface porosity from SEM. It was found that the surface porosity calculated using inhibited SEM is lower than that calculated using uninhibited SEM. The decrease in porosity found in the presence of a 2-MBT inhibitor is consistent with the idea that fewer holes exist. The obtained values from Tafel polarization and SEM are nearly consistent. Table 3Porosity percentage from SEM micrographs for uninhibited and inhibited Al, AT-1 and AT-2 alloys in 0.5 M HCl without and with 1 mM of 2-MBT.SampleUninhibited porosity%Inhibited porosity%Al30.37 ± 1.2910.2 ± 1.58AT-135.76 ± 3.816.59 ± 1.82AT-237.86 ± 4.618.33 ± 1.55
## Effect of temperature
Potentiodynamic polarization measurements for Al and its investigated alloys in 0.5 M HCl solution without and with selected concentrations of 2-MBT were evaluated at a temperature range of 25 to 55 °C to provide detailed evidence about the type of adsorption of investigated inhibitor as well as its effectiveness. The corrosion parameters revealed that raising the solution temperature rises the icorr. These results demonstrate that the reaction of cathodic hydrogen evolution and the anodic dissolution of aluminum and its corresponding alloys are enhanced by increasing temperature47,48. Consequently, the increase in corrosion is pronounced with the rising temperature49. Conversely, the slopes of the bc and ba Tafel lines are nearly unaltered as the temperature rises. This means that the temperature activates the corrosion of the metal surface while the corrosion mechanism remains unchanged.
The research showed that when the temperature of Al and its investigated alloy rises, the inhibition efficiency falls. For example, Fig. 6a,b depicts the correlation between the efficiency of inhibition (IE%) and the 2-MBT concentration for Al and its alloys at various temperatures. This pattern could be explained by weakening the adsorption process at higher temperatures, implying that the inhibitor molecules are physically adsorbed. The adsorption of inhibitor on the electrode results in constructing a physical protective barrier, which decreases the metal reactivity in electrochemical reactions. Figure 6The relationship between the concentration of 2-MBT and the inhibition efficiency (IE %) for (a) Al and (b) AT-3 at various temperatures.
Figure 7a,b demonstrates Arrhenius graphs for Al and AT-3 alloy in 0.5 M HCl solution based on the presence and absence of the examined inhibitor. Arrhenius’s equation can be used to calculate activation energy21,50,51.5\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${i}_{\text{corr}.}=A\text{exp}\left(\frac{-{E}_{\text{a}}}{RT}\right),$$\end{document}icorr.=Aexp-EaRT,6\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\text{log}{i}_{corr}=\text{log}A- \frac{{E}_{a}}{2.303RT}.$$\end{document}logicorr=logA-Ea2.303RT.Figure 7Arrhenius plots for (a) Al and (b) AT-3 alloy after 30 min of exposure to 0.5 M HCl with different concentrations of 2-MBT at different temperatures.
Associating values of Ea in the absence and presence of 2-MBT can yield significant information regarding inhibitor adsorption.
Table 4 shows the apparent activation energy (Ea) for corrosion of Al and Al–Ti alloys calculated using plots of the slope of log icorr against 1/T. The Ea values rise when inhibitor concentration increases for the investigated electrodes in 0.5 M HCl and in the presence of 2-MBT inhibitor. These findings demonstrate that the studied electrodes’ Ea values in the HCl solution are higher in the presence of 2-MBT than they are in the uninhibited acid solution. As a result, the presence of an inhibitor raises the activation energy barrier of the tested electrodes’ corrosion, reducing the corrosion rate. Due to 2-MBT’s substantial physical adsorption, the greater activation energy (Ea) makes it harder to dissolve aluminum or its alloys in an acid solution52. An increase in the activation energy and temperature, accompanied by an additive concentration increase and an IE% decrease in the presence of an inhibitor, proposes that an inhibitor molecule ensures physisorption on the surface of Al and Al–Ti alloys. Table 4Activation thermodynamic parameters for Al and Al–Ti alloys in (0.5 M) HCl solution comprising different concentrations of 2-MBT inhibitor after 30 min of exposure. Metal and alloysConc. ( mM)Ea (kJmol-1)ΔSa (J mol−1 K−1)ΔHa (kJ mol−1)Ea − ΔHa = RT (kJ mol-1)AlBlank50.36 − 154.7147.760.00156.24 − 139.4153.652.60.0161.10 − 125.3358.502.60.160.19 − 129.6657.592.6164.09 − 119.1761.492.6AT-1Blank36.52 − 197.7433.270.00145.03 − 171.3442.432.60.0148.91 − 160.8346.312.60.152.43 − 150.9749.842.6158.26 − 134.4955.662.6AT-2Blank35.86 − 193.6133.920.00146.32 − 164.6743.722.60.0150.54 − 153.1447.952.60.154.50 − 142.0151.902.6158.77 − 130.8056.182.6AT-3Blank29.18 − 213.1926.580.00135.78 − 193.7633.192.60.0138.44 − 186.2235.842.60.144.27 − 169.3041.672.6147.80 − 159.1945.202.6AT-4Blank28.18 − 217.3625.590.00141.08 − 179.2938.482.60.0144.34 − 170.7141.742.60.147.42 − 162.1444.822.6159.63 − 124.9357.032.6 The temperature effect was also confirmed by utilizing the Eyring transition state equation to compute the changes in activation enthalpy ΔHa and entropy ΔSa.7\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${i}_{\text{corr}.}= \frac{RT}{Nh}\text{exp}\left(\frac{{\Delta S}_{a}}{R}\right)\text{exp}\left(\frac{-{\Delta H}_{a}}{RT}\right).$$\end{document}icorr.=RTNhexpΔSaRexp-ΔHaRT.
This equation can be expressed as:8\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\text{log}\left(\frac{{i}_{corr}}{T}\right)=\text{log}\left(\frac{R}{Nh}\right)+ \frac{{\Delta S}_{a}}{2.303R}- \frac{{\Delta H}_{a}}{2.303 RT},$$\end{document}logicorrT=logRNh+ΔSa2.303R-ΔHa2.303RT,where h the Planck constant, N the Avogadro number of the transition state plots of log (icorr/T) versus 1/T is given in Fig. 8. ΔHa and ΔSa were computed respectively from the slopes \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$- \frac{{\Delta H}_{a}}{2.303 R}$$\end{document}-ΔHa2.303R and intercepts \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\text{log}\left(\frac{R}{Nh}\right)+ \frac{{\Delta S}_{a}}{2.303R}$$\end{document}logRNh+ΔSa2.303R of the straight lines obtained. The attained values are recorded in Table 4.Figure 8Transition state plots for (a) pure Al and (b) AT-3 alloy in HCl (0.5 M) comprising different concentrations of 2-MBT at different temperatures after 30 min exposure.
Table 4 shows that Ea and ΔHa are both variables in the same manner. The listed results confirmed the well-known thermodynamic relationship between the two activation parameters; the following equation describes the unimolecular reactions:9\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta {H}_{a}= {E}_{a}-RT.$$\end{document}ΔHa=Ea-RT.
Ea > ΔHa by a value roughly equivalent to RT for 2-MBT. The activation energies (Ea) are positive, and the inhibited solutions have higher activation energies than the uninhibited ones, showing that a physisorption or mix process is taking place53. The endothermic nature of the aluminum and Al–Ti alloys dissolution process is reflected by the positive sign of change in activation enthalpy (ΔHa). With increasing 2-MBT concentrations, the change values in activation enthalpy are increased, indicating that dissolving aluminum and its studied alloys becomes more difficult and slow54. The decrease in disorder from the reactant to the activated complex is indicated by the negative sign of ΔSa55. In 2-MBT, the activation entropy (ΔSa) change increases with increasing concentration, implying that disordering increases as it progresses from reactants to activated complex56.
The heat of adsorption (Qads) was calculated using the kinetic thermodynamic model to understand better the adsorption mechanism57.10\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\text{log}\left(\frac{\theta }{1-\theta }\right)=\text{log}A+\text{log}C-\left(\frac{{Q}_{ads.}}{2.303 RT}\right),$$\end{document}logθ1-θ=logA+logC-Qads.2.303RT,where A is a constant, C is the inhibitor concentration, θ is occupied, and (1 − θ) is the vacant site not occupied by the inhibitor. The relationship between log(θ/1 − θ) and 1/T is shown in Fig. 9 of the electrodes under investigation in the presence of 1 mM 2-MBT inhibitor. Figure 9Plot of log (θ/1 − θ) against 1000/T for aluminum and Al–Ti alloys in 0.5 M HCl containing 1 mM of 2-MBT inhibitor.
Negative Qads values correspond to physisorption properties, with inhibition efficiency decreasing as temperature rises, and positive Qads values indicate enhanced inhibition efficiency as temperature rises. In the acid with 2-MBT, the calculated values of Qads (Table 5) for aluminum and its alloys were all negative, which agrees with the postulated inhibitory physisorption properties58.Table 5The heat of adsorption (Qads.) for Al and Al–Ti alloys in the presence of 1 mM of MBT inhibitor. Metal and alloysQads. ( kJ mol−1)Al − 21.60AT-1 − 37.14AT-2 − 36.41AT-3 − 39.79AT-4 − 51.69
## The corrosion process adsorption isotherm and the parameters of thermodynamics
Adsorption isotherms play a significant impact in understanding how organo-electrochemical systems function. For various concentrations of the organic inhibitor, the degree of surface coverage (θ) can be calculated using potentiodynamic polarization measurements59. The type of adsorption isotherm influences the surface coverage, adsorption equilibrium constant, and interaction between the organic molecule and the electrode surface. The nature of the inhibitor on the corroding surface has been determined based on its adsorption qualities during the corrosion inhibition of metals and alloys. The solvent (H2O) molecules are potentially adsorbed at the metal–solution contact. As a result, it is possible to conceptualize the adsorption of inhibitor molecules from aqueous solutions as a quasi-substitution process involving water molecules at the electrode surface [H2O(ads.)] and organic compounds in the aqueous phase [2-MBT(sol.)]31,60.11\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${2{\text{-}}MBT}_{(sol.)}+ {nH}_{2}{O}_{(ads.)} \leftrightharpoons {2{\text{-}}MBT}_{(ads.)}+ {nH}_{2}{O}_{(sol.)},$$\end{document}2-MBT(sol.)+nH2O(ads.)⇋2-MBT(ads.)+nH2O(sol.),where n is the number of water molecules replaced by one 2-MBT inhibitor molecule. The adsorption isotherm can provide fundamental information about the inhibitor–electrode surface interaction. To determine the isotherm of adsorption, a linear connection correlation between the surface coverage degree (θ) produced by Tafel polarization (θ = IE%/100) and inhibitor concentration (Cinh.) is computed. Langmuir, Freundlich, Temkin, El-Awady, and Frumkin are among the isotherms that have been managed to fit the values. Langmuir isotherm provided the best fit. This isotherm implies that every possible adsorption sites are equal and that binding of particle takes place regardless of whether the surrounding sites are occupied61.
Accordingly, θ is associated to Cinh using the relation:12\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\frac{{C}_{inh}}{\theta }= {C}_{inh}+ \frac{1}{{K}_{ads}},$$\end{document}Cinhθ=Cinh+1Kads,where *Kads is* the inhibitor adsorption equilibrium constant and *Cinh is* the inhibitor concentration. The inhibitor molecule adsorption on the surface of the electrode followed the Langmuir adsorption model, as indicated by straight lines in Cinh/θ versus Cinh graphs (Fig. 10). The fitted curves had regression coefficients close to unity (Table 6), demonstrating the adsorption of 2-MBT molecules on Al and its alloys follows the Langmuir adsorption approach. The tested trend of the inhibitor depends on the molecule’s adsorption on the surface of the electrode62. The straight lines intercept with the Cinh/θ axis63 were used to calculate Kads values, which were then associated with the standard free energy of adsorption (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\Delta G}_{ads}^{0}$$\end{document}ΔGads0) using the following equation64:13\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${K}_{ads}= \frac{1}{55.5}\text{exp}\left(-\frac{{\Delta G}_{ads}^{0}}{RT}\right).$$\end{document}Kads=155.5exp-ΔGads0RT.Figure 10Fitting of Langmuir adsorption model (Cinh/θ versus Cinh) based on the data attained from measurements of Tafel polarization for (a) Al and (b) AT-3 alloy in 0.5 M HCl solution comprising various concentrations of 2-MBT at different temperatures. Table 6Thermodynamic parameters of the inhibitor adsorption on Al and Al–Ti alloys in (0.5 M) HCl solution at 25 °C.Metal and alloysRegression coefficient, R2Kads, M−1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\Delta G}_{ads.}^{0}$$\end{document}ΔGads.0(kJ mol−1)\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\Delta H}_{ads.}^{0}$$\end{document}ΔHads.0(kJ mol−1)\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\Delta S}_{ads.}^{0}$$\end{document}ΔSads.0(J mol-−1 K−1)Al0.9993484,745.76 − 38.06 − 16.4572.52AT-10.99984141,843.97 − 39.34 − 15.8878.72AT-20.99969110,987.79 − 38.73 − 37.284.85AT-30.9997885,106.38 − 38.08 − 35.199.7AT-40.99954104,821.80 − 38.59 − 8.44101.99 In mol/L, the molar concentration of water in the solution was represented by 55.5 in the equation above38. The adsorption equilibrium constant (Kads; Table 6) exhibited relatively high values, indicating that this inhibitor has a large adsorption capacity on the electrode surface and, thus, a higher inhibition efficiency65,66.
As a general rule, the values of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\Delta G}_{ads.}^{0}$$\end{document}ΔGads.0 up to approximately − 20 kJ mol−1 are following the physisorption. In comparison, values around − 40 kJ mol−1 or higher are consistent with chemisorption, characterized by the electrons transfer from organic molecules to the metal surface, forming a coordinate bond67,68. The values of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\Delta G}_{ads.}^{0}$$\end{document}ΔGads.0 determined in this investigation for the examined 2-MBT inhibitor on pure aluminum and aluminum-titanium alloys in HCl (0.5 M) solution range of − 38.06 to − 39.34 kJ mol−1 (Table 6). The latter findings imply that the studied inhibitor adsorbs on the surface of the electrodes via both physical and chemical adsorption processes. Physical adsorption occurs due to electrostatic attraction among the inhibiting species’ dipoles or ions and the electrically charged surface of electrodes. Values of − 40 kJ mol−1 or more negative are consistent with charge sharing or transfer from the organic molecules to the metal surface, causing the formation of a coordinate type of bond (chemisorption)69. The adsorption tendency of 2-MBT on the metal surface is indicated by the negative sign of the standard free energy of adsorption, indicating that the inhibitor adsorption on the metal occurs spontaneously70.
According to thermodynamics, enthalpy \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\Delta H}_{ads.}^{0}$$\end{document}ΔHads.0 and entropy \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\Delta S}_{ads.}^{0}$$\end{document}ΔSads.0 of the adsorption process are related to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\Delta G}_{ads.}^{0}$$\end{document}ΔGads.0 by the following equations71,72:14\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\Delta G}_{ads.}^{0}= {\Delta H}_{ads.}^{0}-T{\Delta S}_{ads.}^{0},$$\end{document}ΔGads.0=ΔHads.0-TΔSads.0,15\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\text{ln}{K}_{ads.}= -\frac{{\Delta H}_{ads.}^{0}}{RT}+ \frac{{\Delta S}_{ads.}^{0}}{R}-\text{ln}\left(55.5\right). $$\end{document}lnKads.=-ΔHads.0RT+ΔSads.0R-ln55.5.
Figure 11 shows the plot of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\Delta G}_{ads.}^{0}$$\end{document}ΔGads.0 against T which provides a straight line with an intercept of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\Delta H}_{ads.}^{0}$$\end{document}ΔHads.0 and a slope of − \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\Delta S}_{ads.}^{0}$$\end{document}ΔSads.0. The values of entropy and heat of adsorption estimated in Table 6 are shown. The value of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta {H}_{ads}^{o}$$\end{document}ΔHadso can provide critical information regarding an inhibitor’s adsorption process. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\Delta H}_{ads.}^{0}$$\end{document}ΔHads.0 ˂ 40 kJ mol−1 indicates physisorption in an exothermic adsorption process, whereas \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\Delta H}_{ads.}^{0}$$\end{document}ΔHads.0 values approaching 100 kJ mol−1 indicate chemical adsorption71. Al and its alloys in the solution of 0.5 M HCl containing varying concentrations of the 2-MBT inhibitor were found to have an estimated \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\Delta H}_{ads.}^{0}$$\end{document}ΔHads.0 values of − 8.44 to − 37.28 kJ mol−1. In accordance with the temperature-dependent variation in inhibitory efficiency, the adsorption process was exothermic associated with physisorption mechanism (Fig. 6). A negative \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\Delta H}_{ads.}^{0}$$\end{document}ΔHads.0 values also indicate exothermic adsorption of inhibitor molecules73.Figure 11Variation of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\Delta G}_{ads.}^{0}$$\end{document}ΔGads.0 against T of pure Al and AT-3 alloy in 0.5 M HCl solution comprising 2-MBT inhibitor.
For Al and its studied alloys, the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\Delta S}_{ads.}^{0}$$\end{document}ΔSads.0 values were found to be + 4.85 to + 101.99 J mol−1 K−1. An increase in solvent energy and the water desorption entropy could explain the positive \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\Delta S}_{ads.}^{0}$$\end{document}ΔSads.0 value71. It can also show how more water molecules can be desorbed off the metal surface by one inhibitor molecule, resulting in a rise in disorders74.
## Electrochemical impedance measurements (EIS)
The EIS examination in 0.5 M HCl without and with varied amounts of inhibitor was carried out to obtain information about the surface passive films on Al and its alloys (immersion time is 30 min). The Nyquist plots show a capacitive loop at high frequency (HF) and an inductive loop at low frequency (LF). Similar graphs for the corrosion of aluminum and its alloys in acidic conditions have been reported by other studies75,76. In Al and Al–Ti alloys, the charge transfer resistance (Rct) of the oxide layer may be a result of the relaxation of the H+ ion during the HF capacitive loop and the corrosive ions adsorption (mostly anions), such as the chloride ion onto or into the oxide film during the LF capacitive loop77–79. At low frequencies, the dissolution of Al or the re-oxidation of the oxide layer on the surface can also cause an inductive loop80,81. Surface area modulation or salt film property modification, such as density, ionic conductivity, or thickness, can be attributed to inductive activity82. With increasing inhibitor concentrations, the both size of HF and LF loops grew noticeably (Fig. 12). The phase angles changed to higher values as the absolute impedance’s magnitude is raised. This might be a result of the surface of the alloy producing a layer83.Figure 12Nyquist plots for (a) Al, (b) AT-2 alloy and (c) AT-3 alloy in 0.5 M HCl comprising different concentrations of 2-MBT inhibitor at 25 °C.
In order to evaluate, the EIS data had to be fitted utilizing an equivalent electric circuit (EEC). As illustrated in Fig. 13, the most appropriate EEC was used to simulate all plots. The EEC was composed of five components: Rs represents the resistance of solution, Rct represents the resistance of the charge-transfer, CPE represents the element of constant phase corresponding to the capacitance of double-layer (Q), L represents an inductive element, and RL represents the related resistance. Because the obtained plots had depressed semicircles, CPE was used instead of actual capacitance. CPE is a term that refers to a collection of features connected with both the surface and the electroactive components. It is frequency-independent. The CPE is crucial due to the distribution of relaxation periods caused by inhomogeneities such as surface roughness/porosity, adsorption, and diffusion78,84,85. Moreover, the CPE contains an exponent, “n”, which is utilized to investigate variations in the metal/solution interface. The frequency dispersion produced by an arbitrarily distributed current on the electrode surface is responsible for the near-unity values of n85,86, demonstrating the predominant capacitive behavior87, as in the present study. High Rct values are related to a slower corroding process88; as a result, corrosion slows down even more with higher concentrations of the investigated inhibitor aluminum and its alloys. The inhibitory efficiency values estimated using EIS correspond well with those calculated using polarization curves. Figure 13Equivalent electric circuit for quantitative estimation of EIS spectra.
The following equation was used to get the inhibitory efficiency (IE%)89:16\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$IE\%= \frac{{R}_{ct}-{R}_{ct}^{^\circ } }{{R}_{ct}} \times 100.$$\end{document}IE%=Rct-Rct∘Rct×100.
The values of the charge transfer resistance in the solutions with and without inhibitor are \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${R}_{ct}$$\end{document}Rct and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${R}_{ct}^{^\circ }$$\end{document}Rct∘, respectively. Table 7 shows that as the inhibitor concentration increases, the charge transfer resistance (Rct.) is gradually increased and consequently the inhibitory power increases. A sluggish corroding mechanism is coupled with a high charge transfer resistance90. At high frequencies, there are noticeable large capacitive curves, then inductive curves at lower frequencies. The capacitive curve diameters are greater in the inhibitor solution than in the blank solution. Because the inhibitors produce an increase in the impedance, this shows that the addition of 2-MBT to the solution increases the impedance of the inhibited substrate. Li et al.86 proposed a similar analogy earlier. The capacitive curves are frequently linked to the corrosion process’s charge transfer. At low frequencies, the inductive curves are believed to be created by the process of relaxation, which occurs when species such as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${H}_{ads.}^{+}$$\end{document}Hads.+ or inhibitor species that were adsorbed on the surface of the electrode86. As a result, the Nyquist plots’ inductive curve (Fig. 12) may be highly correlated with the presence of a passive film on Al and its alloys91. Various inhibitor concentrations result in bigger inductive curves than their absence, indicating a substantial involvement in the adsorption of inhibitor species onto the investigated electrodes. Table 7Electrochemical parameters and inhibition efficiencies derived from EIS for Al, AT-1, AT-2, AT-3, and AT-4 alloys in 0.5 M HCl solution comprising various concentrations of 2-MBT at 25 °C.Metal and alloysConc. of inhibitor (mmol/L)R1 (Ω cm2)CPE (µF cm-2)nR2 (Ω cm2)L (H cm2)R3 (Ω cm2)ƟIE%AlBlank5.1259.490.925695.4173.4414.150.0013.1749.190.936133.70460.90317.900.2928.640.014.9482.240.907159.70465.20538.600.4040.260.14.2635.500.9433211.22521.20310.200.5554.8313.6679.230.8897238.0077.8948.380.6059.91AT-1Blank4.09107.500.895780.6947.8943.920.0014.0396.740.9134120.90389.70216.000.3333.260.014.0942.840.9518143.80490.40328.000.4443.890.14.2673.670.9246180.30261.0078.910.5555.2513.9249.950.9505218.70846.10191.100.6363.10AT-2Blank3.5969.530.968954.8838.0911.180.0014.3891.330.923377.1387.2337.290.2928.850.014.1249.920.940893.91137.9044.280.4241.560.15.4054.690.9294122.00413.10505.200.5555.0216.1172.220.9113158.90176.6079.240.6565.46AT-3Blank4.53235.100.889526.2110.573.280.0014.08542.100.880240.51121.20100.500.3535.300.013.06467.200.78553.5834.3710.320.5151.080.14.11326.500.845261.3371.5864.040.5757.2614.1886.310.951172.05126.0077.870.6463.62AT-4Blank4.94322.700.862439.8445.6613.390.0014.3084.450.921765.7687.3437.880.3939.420.015.37170.000.887377.0295.8150.200.4848.270.15.38179.500.889882.19130.0063.900.5251.5315.4765.740.9729102.4099.2552.410.6161.09
## Mechanism of inhibition of 2-MBT on Al and Al–Ti alloys
Adsorption phenomena are often impacted by the kind, the metal’s surface charge, and the organic inhibitor structure. The surface charge of the metal is caused by the electrical field that forms at the contact after immersion in the electrolyte. The location of the open circuit potential relative to the respective zero charge potential determines the surface charge of metals (PZC). The zero-charge potential is measured against a reference electrode when the metal has no charge. The ionic double layer at the electrode is nonexistent at this voltage. The electrodes can adsorb dissolved compounds in the electrolyte at zero charge potential. There is no net charge on the electrodes at PZC92.
The PZC of Al in 0.5 M HCl solution was determined after 30 min of exposure and was found to be − 0.4 V (vs. SHE). The values of ϕ potential for Al and its alloys were calculated according to the following equation41:17\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\varnothing ={E}_{corr}- {E}_{PZC},$$\end{document}∅=Ecorr-EPZC,where ϕ is Antropov’s ‘rational’ corrosion potential93. Hence, the values of the ϕ potential of Al, AT-1, AT-2, AT-3, and AT-4 are − 0.191, − 0.181, − 0.181, − 0.168, and − 0.163 V (vs. SHE), respectively. This indicates that the surface of Al and Al–Ti alloys is negatively charged at Ecorr. 2-MBT investigated in this study exists as protonated through nitrogen atoms (–C=N–) in HCl. The protonated inhibitor molecules could be adsorbed on investigated electrodes via electrostatic attraction, which forms between the negatively charged surfaces and protonated organic cations (Fig. 14). Adsorption involves the displacement of water molecules from the aluminum surface as well as the sharing of electrons between the hetero-atoms and metal. Furthermore, the inhibitor molecules can be adsorbed on the aluminum surface via donor–acceptor interactions between π–electrons of aromatic rings and unoccupied p-orbitals of surface aluminum atoms94.Figure 14Adsorption of 2-MBT on Al and Al–Ti alloys surfaces.
## DFT study
Quantum chemical calculations have been primarily employed to predict the anticorrosive properties of the isolated molecule toward the metallic surface. In this respect, we calculated several global reactivity descriptors for the studied inhibitor in the aqueous phase, as illustrated in Table 8 and Fig. 15. The 2-MBT geometry was optimized at B3LYP-D$\frac{3}{6}$-311(d,p) model chemistry. This planarity of 2-MBT facilitates the horizontal loading of inhibitor molecules on the metallic surface. Table 8The B3LYP/ 6–311 + + G(2d,2p) chemical descriptors of the studied 2-MBT in the aqueous phase. MoleculeEHOMO, eVELUMO, eV∆E, eVη, eVDM, Dχ, eVω, eVΔN, eΔET, eV2-MBT − 6.43 − 1.365.072.541.443.902.990.54 − 0.63Figure 15The estimated FMOs and MEP of the investigated inhibitor using B3LYP/6-311++ G(2d,2p) in the aqueous media.
Figure 15 illustrates frontier molecular orbitals’ (FMOs) electron density distribution and the molecular electrostatic potential. The findings showed that HOMO and LUMO orbitals are delocalized over the whole molecular skeleton of the investigated compound. Since the HOMO of all the studied 2-MBT inhibitor is a molecular orbital of the π-type, parallel adsorption on a metal surface is highly expected in addition to the planar geometry.
The 3D charge distribution of the molecule is shown by molecular electrostatic potential (MEP) map, Fig. 15. This map helps visualize the variably charged regions within the molecule to predict electrophilic and nucleophilic attacks on the molecule. In the MEP plot, the blue color represents the maximum positive region subject to nucleophile attack. In contrast, the red represents the negative region subject to electrophilic attack. The imine nitrogen atom of 2-MBT is observed to carry the largest electron density. Therefore, this atom will be anticipated to engage in the adsorption process on Al surface busily.
According to the frontier molecular orbital theory, the occupied orbitals of one molecule interact with the unoccupied orbitals of other species, causing attraction. In this context, the molecule with higher EHOMO and lower ELUMO are favorable electron-donating and electron-accepting abilities, respectively. While the ELUMO values correlate well with the experimental results, the EHOMO does not follow this trend. Moreover, ΔE is a critical parameter associated with chemical reactivity. For example, an inhibitor with a lower ΔE has a higher chemical reactivity increasing the number of collisions with the surface and thus increasing the chance of forming stable interactions.
The electronegativity (χ) of the studied molecule matches the experimental results, reflecting the discrepancy in their ability to attract the bond electrons formed with the surface. Furthermore, chemical hardness95 is defined as the chemical species’ resistance toward electron cloud polarization. Generally, the inhibitor molecule with a low chemical hardness value possesses high corrosion inhibition performance96. 2-MBT exhibits a high value of electrophilicity (Table 8), confirming its high capacity to accept electrons. The dipole moment (DM) is another parameter of the electronic distribution in a molecule and measures the polarity of a polar covalent bond. According to Khalil97, lower DM values will favor the inhibitor’s accumulation in the surface layer and, therefore, higher inhibition efficiency. Accordingly, the expected inhibition efficiency agrees with the experimental findings. The negative sign of ΔET indicates that the back donation’s charge transfer to the molecule is energetically favorable98.
## Monte Carlo simulations
To minimize the contact area between the corrosion-causing materials (such as H2O, acidic or alkaline media) and the metal surface, molecules must parallel their structures to the metal surface as nearly as possible. This process is known as adsorption. We, therefore, sought to differentiate between the examined inhibitor’s adsorption capacity to the surface of Al[111] by doing Metropolis Monte Carlo simulations and contrasting them with pertinent empirical findings. The total energy profile for Inhibitor/Al[111] system during the simulated adsorption process by the MC approach is provided in Fig. S1. Figure 16 illustrates the best adsorption mode of the studied 2-MBT on the aluminum surface. No unusual adsorption mode on the surface was observed, where all molecules were loaded on the surface horizontally. The formation of the horizontal orientation can be ascribed to the relatively equal distribution of populations of HOMO and LUMO on the whole molecules. This result emphasizes the high adsorption strength of the studied inhibitor. Figure 16The side and top view of the most stable configurations for adsorption of 2-MBT on the Al[111] surface in an acidic medium. The below figure shows the distance data between the active atom in the inhibitor molecule and the metal surface atoms.
The adsorption descriptors of the investigated molecule are shown in Table 9, including total energy, adsorption energy, stiff adsorption energy, and deformation energy. In our earlier work, we have described the definitions of these parameters99. The most important energy parameter for adsorption is called adsorption energy, which is calculated as the total rigid adsorption energy before and after an adsorbate’s surface relaxation. According to Table 9, the negative value of the adsorption energies shows that the inhibitor spontaneously adheres to the surface of Al[111].Table 9Adsorption descriptors for Al[111]/2-MBT/(5 HCl/139 H2O) systems. All values are in kcal/mol. Inh. Total energyAdsorption energyRigid adsorption energyDeformation energydEad/dNiInhHClH2O2-MBT − 630 − 746 − 80155 − 88 − 7 − 14 By taking the surface energy of *Al is* zero, the differential adsorption energy (dEad/dNi) is defined as the energy required or liberated to remove a component of the adsorbate, i.e., desorption energy. The adsorption process is strongly preferred according to the inhibitor’s adsorption energy of − 746 kcal/mol and its desorption energy of − 88 kcal/mol. Despite corrosive substances like H2O and HCl in the media, the inhibitor preferentially adsorbs on the Al[111] surface with little to no competition because it adsorbs with much less energy.
Back to Fig. 16, the distance data between the active atom in the inhibitor molecule and the metal surface atom are depicted to judge their adsorption mode. It is shown that all distances between the inhibitor atoms and the surface ones exceed the sum of covalent radii, where they are higher than 3 Å. This supports the experimental finding that the inhibitor prefers the physical adsorption on the surface, as indicated by the activation energy and adsorption-free energy change values. It is also observed that the atoms of the imine group form relatively short physical bonds with the surface due to the high negative charge on the nitrogen atom, as indicated by the DFT results.
## Conclusion
The main conclusions deduced from this research are summarized as follows:*In a* 0.5 M solution of HCl, the corrosion of aluminum and aluminum-titanium alloys is inhibited by 2-mercaptobenzothiazole. With an increase in inhibitor concentration, the inhibitor’s inhibition effectiveness rises. At lower temperatures, 2-MBT inhibition efficiency values are higher (25 °C). However, the IE% values of the examined inhibitor’s Al and Al–Ti alloys at higher temperatures decrease. The correlation between inhibitor concentration and the reported rise in activation energy and the decrease in the inhibitor’s inhibitory efficiency with rising temperature indicate its physical adsorption on the electrode surface. Activation Thermodynamic parameters and heat of adsorption for Al and Al–Ti alloys in (0.5 M) HCl solution comprising different concentrations of 2-MBT inhibitor \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${E}_{a}$$\end{document}Ea, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\Delta H}_{a}$$\end{document}ΔHa, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\Delta S}_{a}$$\end{document}ΔSa and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${Q}_{ads}$$\end{document}Qads are evaluated and interpreted. The data obtained from polarization curves fit well with the Langmuir adsorption isotherm. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\Delta H}_{ads}^{o}$$\end{document}ΔHadso, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\Delta S}_{ads}^{o}$$\end{document}ΔSadso and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\Delta G}_{ads}^{o}$$\end{document}ΔGadso are evaluated and interpreted. The calculated values of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\Delta H}_{ads}^{o}$$\end{document}ΔHadso and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\Delta G}_{ads}^{o}$$\end{document}ΔGadso are negative, while those for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\Delta S}_{ads}^{o}$$\end{document}ΔSadso are positive. In an acid chloride solution, 2-MBT functions as an inhibitor of mixed types of the examined aluminum and its alloys. SEM and porosity percentage showed the adsorption of 2-MBT on the investigated electrodes surface and the pits on the electrode surfaces are decreased. The DFT findings indicates the horizontal loading of 2-MBT on the aluminum surface. The MC simulation confirms that 2-MBT prefers to protect the Al-surface through physical adsorption.
## Supplementary Information
Supplementary Figures. The online version contains supplementary material available at 10.1038/s41598-023-31795-2.
## References
1. El-Sayed A-R, Mohamed AE, Hassan FSM, El-Mahdy MS. **Influence of titanium additions to aluminum on the microhardness value and electrochemical behavior of synthesized aluminum–titanium alloy in solutions of HCl and H3PO4**. *J. Mater. Eng. Perform.* (2022.0). DOI: 10.1007/s11665-022-07248-8
2. Eswara Prasad N, Gokhale AA, Wanhill RJH. *Aluminum–Lithium Alloys* (2014.0) 503-535
3. Bhaskara S, Fakrudeen SP, Desalegn T, Murthy HC, Bheemaraju V. **Evaluation of corrosion inhibition efficiency of aluminum alloy 2024 by diaminostilbene and Azobenzene Schiff bases in 1 M hydrochloric acid**. *Int. J. Corros.* (2021.0) **2021** 1-20. DOI: 10.1155/2021/5869915
4. 4.Vargel, C. Corrosion of Aluminium. (Elsevier Ltd, ed., Oxford, 2004). ISBN 0080-33395-4.
5. Fouda AS, Mohamed FS, El-Sherbeni MW. **Corrosion inhibition of aluminum–silicon alloy in hydrochloric acid solutions using carbamidic thioanhydride derivatives**. *J. Bio- and Tribo-Corros.* (2016.0) **2** 11. DOI: 10.1007/s40735-016-0039-y
6. Xhanari K, Finšgar M. **The corrosion inhibition of AA6082 aluminium alloy by certain azoles in chloride solution: Electrochemistry and surface analysis**. *Coatings* (2019.0) **9** 380. DOI: 10.3390/coatings9060380
7. Xhanari K, Finšgar M. **Electrochemical analysis of AA6082 aluminium alloy in chloride media**. *Int. J. Electrochem. Sci.* (2017.0) **12** 5845-5853. DOI: 10.20964/2017.07.71
8. Sherif EM, Park S-M. **Effects of 2-amino-5-ethylthio-1, 3, 4-thiadiazole on copper corrosion as a corrosion inhibitor in aerated acidic pickling solutions**. *Electrochim. Acta* (2006.0) **51** 6556-6562. DOI: 10.1016/j.electacta.2006.04.047
9. Sherif E-SM. **Corrosion and corrosion inhibition of aluminum in Arabian Gulf seawater and sodium chloride solutions by 3-amino-5-mercapto-1, 2, 4-triazole**. *Int. J. Electrochem. Sci.* (2011.0) **6** 1479-1492
10. Sherif E-SM. **Electrochemical and gravimetric study on the corrosion and corrosion inhibition of pure copper in sodium chloride solutions by two azole derivatives**. *Int. J. Electrochem. Sci.* (2012.0) **7** 1482-1495
11. Sherif EM, Park S-M. **2-Amino-5-ethyl-1, 3, 4-thiadiazole as a corrosion inhibitor for copper in 3.0% NaCl solutions**. *Corros. Sci.* (2006.0) **48** 4065-4079. DOI: 10.1016/j.corsci.2006.03.011
12. Sherif E-SM. **Effects of 2-amino-5-(ethylthio)-1, 3, 4-thiadiazole on copper corrosion as a corrosion inhibitor in 3% NaCl solutions**. *Appl. Surf. Sci.* (2006.0) **252** 8615-8623. DOI: 10.1016/j.apsusc.2005.11.082
13. Sasikumar Y. **Experimental, quantum chemical and Monte Carlo simulation studies on the corrosion inhibition of some alkyl imidazolium ionic liquids containing tetrafluoroborate anion on mild steel in acidic medium**. *J. Mol. Liq.* (2015.0) **211** 105-118. DOI: 10.1016/j.molliq.2015.06.052
14. Wang H-L, Fan H-B, Zheng J-S. **Corrosion inhibition of mild steel in hydrochloric acid solution by a mercapto-triazole compound**. *Mater. Chem. Phys.* (2003.0) **77** 655-661. DOI: 10.1016/S0254-0584(02)00123-2
15. Amin MA, Abd Ei-Rehim SS, El-Sherbini EEF, Hazzazi OA, Abbas MN. **Polyacrylic acid as a corrosion inhibitor for aluminium in weakly alkaline solutions. Part I: Weight loss, polarization, impedance EFM and EDX studies**. *Corros. Sci.* (2009.0) **51** 658-667. DOI: 10.1016/j.corsci.2008.12.008
16. Zhou Y. **Corrosion control of mild steel in 0.1 MH2 SO4 solution by benzimidazole and its derivatives: An experimental and theoretical study**. *RSC Adv.* (2017.0) **7** 23961-23969. DOI: 10.1039/C7RA02192E
17. Obot IB, Gasem ZM, Umoren SA. **Understanding the mechanism of 2-mercaptobenzimidazole adsorption on Fe (110), Cu (111) and Al (111) surfaces: DFT and molecular dynamics simulations approaches**. *Int. J. Electrochem. Sci.* (2014.0) **9** 2367-2378
18. Wu X, Wiame F, Maurice V, Marcus P. **Adsorption and thermal stability of 2-mercaptobenzothiazole corrosion inhibitor on metallic and pre-oxidized Cu (1 1 1) model surfaces**. *Appl. Surf. Sci.* (2020.0) **508** 145132. DOI: 10.1016/j.apsusc.2019.145132
19. Jafari H, Akbarzade K, Danaee I. **Corrosion inhibition of carbon steel immersed in a 1M HCl solution using benzothiazole derivatives**. *Arab. J. Chem.* (2019.0) **12** 1387-1394. DOI: 10.1016/j.arabjc.2014.11.018
20. Balaskas AC, Curioni M, Thompson GE. **Corrosion protection mechanism of 2-mercaptobenzothiazole and its potential synergistic effect with cerium ions for treatment of AA 2024–T3**. *J. Electroanal. Chem.* (2020.0) **863** 114081. DOI: 10.1016/j.jelechem.2020.114081
21. El-Sayed A-R, Mohran HS, Abd El-Lateef HM. **The inhibition effect of 2, 4, 6-tris (2-pyridyl)-1, 3, 5-triazine on corrosion of tin, indium and tin–indium alloys in hydrochloric acid solution**. *Corros. Sci.* (2010.0) **52** 1976-1984. DOI: 10.1016/j.corsci.2010.02.029
22. 22.Frisch, M. J. et al. Gaussian 16 Rev. B. 01 (Wallingford, CT, 2016). Google Scholar There is no Corresponding Record for This Reference (2016).
23. Marenich AV, Cramer CJ, Truhlar DG. **Universal solvation model based on solute electron density and on a continuum model of the solvent defined by the bulk dielectric constant and atomic surface tensions**. *J. Phys. Chem. B* (2009.0) **113** 6378-6396. DOI: 10.1021/jp810292n
24. Eddy NO, Ita BI. **QSAR, DFT and quantum chemical studies on the inhibition potentials of some carbozones for the corrosion of mild steel in HCl**. *J. Mol. Model.* (2011.0) **17** 359-376. DOI: 10.1007/s00894-010-0731-7
25. 25.Dassault SystèmesDassault Systèmes BIOVIA, Materials Studio, 17.1.0.482017Dassault Systèmes. *Dassault Systèmes BIOVIA, Materials Studio, 17.1.0.48* (2017.0)
26. Metropolis N, Rosenbluth AW, Rosenbluth MN, Teller AH, Teller E. **Equation of state calculations by fast computing machines**. *J. Chem. Phys.* (1953.0) **21** 1087-1092. DOI: 10.1063/1.1699114
27. Yamashita J, Nunomura N. **First-principles study of chlorine adsorption on clean Al (111)**. *Mater. Trans.* (2017.0) **58** 1356-1363. DOI: 10.2320/matertrans.M2017101
28. Sun H, Ren P, Fried JR. **The COMPASS force field: Parameterization and validation for phosphazenes**. *Comput. Theor. Polym. Sci.* (1998.0) **8** 229-246. DOI: 10.1016/S1089-3156(98)00042-7
29. El-Sayed A-R, Shaker AM, Abd El-Lateef HM. **Corrosion inhibition of tin, indium and tin–indium alloys by adenine or adenosine in hydrochloric acid solution**. *Corros. Sci.* (2010.0) **52** 72-81. DOI: 10.1016/j.corsci.2009.08.047
30. Hameed RSA. **Electrochemical techniques for evaluation of expired megavit drugs as corrosion inhibitor for steel in hydrochloric acid**. *Int. J. Electrochem. Sci.* (2021.0) **16** 210446. DOI: 10.20964/2021.04.15
31. El-Lateef HMA, El-Sayed A-R, Mohran HS, Shilkamy HAS. **Corrosion inhibition and adsorption behavior of phytic acid on Pb and Pb–In alloy surfaces in acidic chloride solution**. *Int. J. Ind. Chem.* (2019.0) **10** 31-47. DOI: 10.1007/s40090-019-0169-4
32. Khadraoui A. **Adsorption and inhibitive properties of**. *Int. J. Electrochem. Sci.* (2014.0) **9** 3334-3348
33. Raviprabha K, Bhat RS. **5-(3-Pryridyl)-4H-1, 2, 4-triazole-3-thiol as potential corrosion inhibitor for AA6061 aluminium alloy in 0.1 M hydrochloric acid solution**. *Surf. Eng. Appl. Electrochem.* (2019.0) **55** 723-733. DOI: 10.3103/S1068375519060103
34. Iroha NB, Nnanna LA. **Leucas martinicensis as an inhibitor of carbon steel corrosion in acidic medium**. *Int. J. Res.* (2020.0) **7** 19-26
35. Iroha N, Maduelosi N. **Pipeline steel protection in oil well acidizing fluids using expired pharmaceutical agent**. *Chem. Int.* (2020.0) **6** 267-276
36. Tremont R, De Jesus-Cardona H, Garcia-Orozco J, Castro RJ, Cabrera CR. **3-Mercaptopropyltrimethoxysilane as a Cu corrosion inhibitor in KCl solution**. *J. Appl. Electrochem.* (2000.0) **30** 737-743. DOI: 10.1023/A:1004072117517
37. Schultze JW, Wippermann K. **Inhibition of electrode processes on copper by AHT in acid solutions**. *Electrochim. Acta* (1987.0) **32** 823-831. DOI: 10.1016/0013-4686(87)85115-0
38. Abd El-Lateef HM, Abbasov VM, Aliyeva LI, Qasimov EE, Ismayilov IT. **Inhibition of carbon steel corrosion in CO2-saturated brine using some newly surfactants based on palm oil: Experimental and theoretical investigations**. *Mater. Chem. Phys.* (2013.0) **142** 502-512. DOI: 10.1016/j.matchemphys.2013.07.044
39. Abdel Aal MS, Abdel Wahab AA, Saied AE. **A study of the inhibiting action of benzene thiols and related compounds on the corrosion of zinc in acidic media**. *Corrosion* (1981.0) **37** 557-563. DOI: 10.5006/1.3577538
40. El-Sayed A-R, Mohran HS, Shilkamy HAS. **Role of indium alloying with lead as a means to reduce the passivation phenomena in lead/acid batteries**. *Int. J. Electrochem.* (2014.0) **2014** 1-16. DOI: 10.1155/2014/932654
41. Abd El-Lateef HM, El-Sayed A-R, Mohran HS, Shilkamy HAS. **Corrosion inhibition and adsorption behavior of phytic acid on Pb and Pb–In alloy surfaces in acidic chloride solution**. *Int. J. Ind. Chem.* (2019.0) **10** 31-47. DOI: 10.1007/s40090-019-0169-4
42. Zor S. **Sulfathiazole as potential corrosion inhibitor for copper in 0.1 M NaCl**. *Prot. Met. Phys. Chem. Surf.* (2014.0) **50** 530-537. DOI: 10.1134/S2070205114040200
43. Yan Y, Li W, Cai L, Hou B. **Electrochemical and quantum chemical study of purines as corrosion inhibitors for mild steel in 1 M HCl solution**. *Electrochim. Acta* (2008.0) **53** 5953-5960. DOI: 10.1016/j.electacta.2008.03.065
44. El-Sayed AE-R, Shilkamy HAE-S, Elrouby M. **Tracing the influence of small additions of antimony to zinc on the hydrogen evolution and anodic dissolution processes of zinc as anodes for alkaline batteries application**. *Int. J. Hydrogen Energy* (2021.0) **46** 31239-31252. DOI: 10.1016/j.ijhydene.2021.07.014
45. Yu Y. **Al2O3 coatings on zinc for anti-corrosion in alkaline solution by electrospinning**. *Coatings* (2019.0) **9** 692. DOI: 10.3390/coatings9110692
46. El Ashry ESH, El Nemr A, Esawy SA, Ragab S. **Corrosion inhibitors-Part II: Quantum chemical studies on the corrosion inhibitions of steel in acidic medium by some triazole, oxadiazole and thiadiazole derivatives**. *Electrochim. Acta* (2006.0) **51** 3957-3968
47. Aouine Y. **Temperature and time investigations on the adsorption behavior of isoindoline, tetrazole and isoindoline-tetrazole on corrosion of mild steel in acidic medium**. *Int. J. Electrochem. Sci.* (2012.0) **7** 5400-5419
48. Amin MA, Ibrahim MM. **Corrosion and corrosion control of mild steel in concentrated H2SO4 solutions by a newly synthesized glycine derivative**. *Corros. Sci.* (2011.0) **53** 873-885. DOI: 10.1016/j.corsci.2010.10.022
49. Amin MA, Abd El-Rehim SS, El-Sherbini EEF, Bayoumi RS. **The inhibition of low carbon steel corrosion in hydrochloric acid solutions by succinic acid: Part I. Weight loss, polarization, EIS, PZC, EDX and SEM studies**. *Electrochim. Acta* (2007.0) **52** 3588-3600. DOI: 10.1016/j.electacta.2006.10.019
50. Mobin M, Aslam R, Aslam J. **Non toxic biodegradable cationic gemini surfactants as novel corrosion inhibitor for mild steel in hydrochloric acid medium and synergistic effect of sodium salicylate: Experimental and theoretical approach**. *Mater. Chem. Phys.* (2017.0) **191** 151-167. DOI: 10.1016/j.matchemphys.2017.01.037
51. Lgaz H, Salghi R, Bhat KS, Chaouiki A, Jodeh S. **Correlated experimental and theoretical study on inhibition behavior of novel quinoline derivatives for the corrosion of mild steel in hydrochloric acid solution**. *J. Mol. Liq.* (2017.0) **244** 154-168. DOI: 10.1016/j.molliq.2017.08.121
52. Kiani MA, Mousavi MF, Ghasemi S, Shamsipur M, Kazemi SH. **Inhibitory effect of some amino acids on corrosion of Pb–Ca–Sn alloy in sulfuric acid solution**. *Corros. Sci.* (2008.0) **50** 1035-1045. DOI: 10.1016/j.corsci.2007.11.031
53. Umoren SA, Obot IB, Akpabio LE, Etuk SE. **Adsorption and corrosive inhibitive properties of**. *Pigm. Resin Technol.* (2008.0) **37** 98. DOI: 10.1108/03699420810860455
54. Lebrini M, Lagrenée M, Vezin H, Traisnel M, Bentiss F. **Experimental and theoretical study for corrosion inhibition of mild steel in normal hydrochloric acid solution by some new macrocyclic polyether compounds**. *Corros. Sci.* (2007.0) **49** 2254-2269. DOI: 10.1016/j.corsci.2006.10.029
55. Krishnegowda PM, Venkatesha VT, Krishnegowda PKM, Shivayogiraju SB. **Acalypha torta leaf extract as green corrosion inhibitor for mild steel in hydrochloric acid solution**. *Ind. Eng. Chem. Res.* (2013.0) **52** 722-728. DOI: 10.1021/ie3018862
56. Beda RHB, Niamien PM, Bile EBA, Trokourey A. **Inhibition of aluminium corrosion in 10 M HCl by caffeine: Experimental and DFT studies**. *Adv. Chem.* (2017.0) **2017** 6975248. DOI: 10.1155/2017/6975248
57. Ejikeme PM, Umana SG, Alinnor IJ, Onukwuli OD, Menkiti MC. **Corrosion inhibition and adsorption characteristics of**. *Am. J. Mater. Sci.* (2014.0) **4** 194-201
58. Bouyanzer A, Hammouti B, Majidi L. **Pennyroyal oil from**. *Mater. Lett.* (2006.0) **60** 2840-2843. DOI: 10.1016/j.matlet.2006.01.103
59. Babić-Samardžija K, Khaled KF, Hackerman N. **Investigation of the inhibiting action of O-, S-and N-dithiocarbamato (1, 4, 8, 11-tetraazacyclotetradecane) cobalt (III) complexes on the corrosion of iron in HClO4 acid**. *Appl. Surf. Sci.* (2005.0) **240** 327-340. DOI: 10.1016/j.apsusc.2004.07.015
60. Şahin M, Bilgic S, Yılmaz H. **The inhibition effects of some cyclic nitrogen compounds on the corrosion of the steel in NaCl mediums**. *Appl. Surf. Sci.* (2002.0) **195** 1-7. DOI: 10.1016/S0169-4332(01)00783-8
61. Abd El-Lateef HM, Adam MSS, Khalaf MM. **Synthesis of polar unique 3d metal-imine complexes of salicylidene anthranilate sodium salt. Homogeneous catalytic and corrosion inhibition performance**. *J. Taiwan Inst. Chem. Eng.* (2018.0) **88** 286-304. DOI: 10.1016/j.jtice.2018.04.024
62. Obot IB, Obi-Egbedi NO. **Ipomoea involcrata as an ecofriendly inhibitor for aluminium in alkaline medium**. *Port. Electrochim. Acta* (2009.0) **27** 517-524. DOI: 10.4152/pea.200904517
63. Tang F, Wang X, Xu X, Li L. **Phytic acid doped nanoparticles for green anticorrosion coatings**. *Colloids Surf. A* (2010.0) **369** 101-105. DOI: 10.1016/j.colsurfa.2010.08.013
64. Adam MSS, Abd El-Lateef HM, Soliman KA. **Anionic oxide-vanadium Schiff base amino acid complexes as potent inhibitors and as effective catalysts for sulfides oxidation: Experimental studies complemented with quantum chemical calculations**. *J. Mol. Liq.* (2018.0) **250** 307-322. DOI: 10.1016/j.molliq.2017.12.013
65. Refaey SAM, Taha F, Abd El-Malak AM. **Inhibition of stainless steel pitting corrosion in acidic medium by 2-mercaptobenzoxazole**. *Appl. Surf. Sci.* (2004.0) **236** 175-185. DOI: 10.1016/j.apsusc.2004.04.016
66. Abd El-Lateef HM, Tantawy AH. **Synthesis and evaluation of novel series of Schiff base cationic surfactants as corrosion inhibitors for carbon steel in acidic/chloride media: Experimental and theoretical investigations**. *RSC Adv.* (2016.0) **6** 8681-8700. DOI: 10.1039/C5RA21626E
67. Abd El-Lateef HM. **Experimental and computational investigation on the corrosion inhibition characteristics of mild steel by some novel synthesized imines in hydrochloric acid solutions**. *Corros. Sci.* (2015.0) **92** 104-117. DOI: 10.1016/j.corsci.2014.11.040
68. Oguzie EE. **Corrosion inhibition of aluminium in acidic and alkaline media by**. *Corros. Sci.* (2007.0) **49** 1527-1539. DOI: 10.1016/j.corsci.2006.08.009
69. Hoseizadeh AR, Danaee I, Maddahy MH. **Thermodynamic and adsorption behaviour of vitamin B1 as a corrosion inhibitor for AISI 4130 steel alloy in HCl solution**. *Z. Phys. Chem.* (2013.0) **227** 403-418. DOI: 10.1524/zpch.2013.0276
70. Roberge PR. *Corrosion Engineering* (2008.0) 49-50
71. Solomon MM, Gerengi H, Umoren SA. **Carboxymethyl cellulose/silver nanoparticles composite: Synthesis, characterization and application as a benign corrosion inhibitor for St37 steel in 15% H2SO4 medium**. *ACS Appl. Mater. Interfaces* (2017.0) **9** 6376-6389. DOI: 10.1021/acsami.6b14153
72. Zadeh ARH, Danaee I, Maddahy MH. **Thermodynamic and adsorption behaviour of medicinal nitramine as a corrosion inhibitor for AISI steel alloy in HCl solution**. *J. Mater. Sci. Technol.* (2013.0) **29** 884-892. DOI: 10.1016/j.jmst.2013.06.006
73. Noor EA, Al-Moubaraki AH. **Thermodynamic study of metal corrosion and inhibitor adsorption processes in mild steel/1-methyl-4 [4′(-X)-styryl pyridinium iodides/hydrochloric acid systems**. *Mater. Chem. Phys.* (2008.0) **110** 145-154. DOI: 10.1016/j.matchemphys.2008.01.028
74. Solmaz R, Kardaş G, Çulha M, Yazıcı B, Erbil M. **Investigation of adsorption and inhibitive effect of 2-mercaptothiazoline on corrosion of mild steel in hydrochloric acid media**. *Electrochim. Acta* (2008.0) **53** 5941-5952. DOI: 10.1016/j.electacta.2008.03.055
75. Pinto GM, Nayak J, Shetty AN. **Corrosion inhibition of 6061 Al–15 vol. pct. SiC (p) composite and its base alloy in a mixture of sulphuric acid and hydrochloric acid by 4-(N, N-dimethyl amino) benzaldehyde thiosemicarbazone**. *Mater. Chem. Phys.* (2011.0) **125** 628-640. DOI: 10.1016/j.matchemphys.2010.10.006
76. Yurt A, Ulutas S, Dal H. **Electrochemical and theoretical investigation on the corrosion of aluminium in acidic solution containing some Schiff bases**. *Appl. Surf. Sci.* (2006.0) **253** 919-925. DOI: 10.1016/j.apsusc.2006.01.026
77. Bessone J, Mayer C, Jüttner K, Lorenz WJ. **AC-impedance measurements on aluminium barrier type oxide films**. *Electrochim. Acta* (1983.0) **28** 171-175. DOI: 10.1016/0013-4686(83)85105-6
78. Brett CMA. **On the electrochemical behaviour of aluminium in acidic chloride solution**. *Corros. Sci.* (1992.0) **33** 203-210. DOI: 10.1016/0010-938X(92)90145-S
79. Metikoš-Huković M, Babić R, Grubač Z. **Corrosion protection of aluminium in acidic chloride solutions with nontoxic inhibitors**. *J. Appl. Electrochem.* (1998.0) **28** 433-439. DOI: 10.1023/A:1003200808093
80. Abd El Rehim SS, Hassan HH, Amin MA. **Corrosion inhibition of aluminum by 1, 1 (lauryl amido) propyl ammonium chloride in HCl solution**. *Mater. Chem. Phys.* (2001.0) **70** 64-72. DOI: 10.1016/S0254-0584(00)00468-5
81. Noor EA. **Evaluation of inhibitive action of some quaternary N-heterocyclic compounds on the corrosion of Al–Cu alloy in hydrochloric acid**. *Mater. Chem. Phys.* (2009.0) **114** 533-541. DOI: 10.1016/j.matchemphys.2008.09.065
82. Bessone JB, Salinas DR, Mayer CE, Ebert M, Lorenz WJ. **An EIS study of aluminium barrier-type oxide films formed in different media**. *Electrochim. Acta* (1992.0) **37** 2283-2290. DOI: 10.1016/0013-4686(92)85124-4
83. Li X, Deng S, Fu H. **Inhibition by tetradecylpyridinium bromide of the corrosion of aluminium in hydrochloric acid solution**. *Corros. Sci.* (2011.0) **53** 1529-1536. DOI: 10.1016/j.corsci.2011.01.032
84. Navarro-Flores E, Chong Z, Omanovic S. **Characterization of Ni, NiMo, NiW and NiFe electroactive coatings as electrocatalysts for hydrogen evolution in an acidic medium**. *J. Mol. Catal. A Chem.* (2005.0) **226** 179-197. DOI: 10.1016/j.molcata.2004.10.029
85. Yang D, Zhang M, Zheng J, Castaneda H. **Corrosion inhibition of mild steel by an imidazolium ionic liquid compound: The effect of pH and surface pre-corrosion**. *RSC Adv.* (2015.0) **5** 95160-95170. DOI: 10.1039/C5RA14556B
86. Li X, Deng S, Li N, Xie X. **Inhibition effect of bamboo leaves extract on cold rolled steel in Cl3CCOOH solution**. *J. Market. Res.* (2017.0) **6** 158-170
87. Halambek J, Berković K. **Inhibitive action of**. *Int. J. Electrochem. Sci.* (2012.0) **7** 8356-8368
88. Ahamad I, Prasad R, Quraishi MA. **Adsorption and inhibitive properties of some new Mannich bases of Isatin derivatives on corrosion of mild steel in acidic media**. *Corros. Sci.* (2010.0) **52** 1472-1481. DOI: 10.1016/j.corsci.2010.01.015
89. Abd El-Lateef HM, Abu-Dief AM, Abdel-Rahman LH, Sañudo EC, Aliaga-Alcalde N. **Electrochemical and theoretical quantum approaches on the inhibition of C1018 carbon steel corrosion in acidic medium containing chloride using some newly synthesized phenolic Schiff bases compounds**. *J. Electroanal. Chem.* (2015.0) **743** 120-133. DOI: 10.1016/j.jelechem.2015.02.023
90. Khaled KF. **The inhibition of benzimidazole derivatives on corrosion of iron in 1 M HCl solutions**. *Electrochim. Acta* (2003.0) **48** 2493-2503. DOI: 10.1016/S0013-4686(03)00291-3
91. Hachelef H, Benmoussat A, Khelifa A, Athmani D, Bouchareb D. **Study of corrosion inhibition by electrochemical impedance spectroscopy method of 5083 aluminum alloy in 1 M HCl solution containing propolis extract**. *J. Mater. Environ. Sci.* (2016.0) **7** 1751-1758
92. Danaee I, RameshKumar S, RashvandAvei M, Vijayan M. **Electrochemical and quantum chemical studies on corrosion inhibition performance of 2, 2'-(2-hydroxyethylimino) bis [N-(alphaalpha-dimethylphenethyl)-N-methylacetamide] on mild steel corrosion in 1M HCl solution**. *Mater. Res.* (2020.0) **23** 1. DOI: 10.1590/1980-5373-mr-2018-0610
93. Popova A, Christov M, Raicheva S, Sokolova E. **Adsorption and inhibitive properties of benzimidazole derivatives in acid mild steel corrosion**. *Corros. Sci.* (2004.0) **46** 1333-1350. DOI: 10.1016/j.corsci.2003.09.025
94. Fouda AS, Shalabi K, Mohamed NH. **Corrosion inhibition of aluminum in hydrochloric acid solutions using some chalcone derivatives**. *Int. J. Innov. Res. Sci. Eng. Technol. J.* (2014.0) **3** 9861-9875
95. Pearson RG. **The principle of maximum hardness**. *Acc. Chem. Res.* (1993.0) **26** 250-255. DOI: 10.1021/ar00029a004
96. Wazzan NA, Obot IB, Kaya S. **Theoretical modeling and molecular level insights into the corrosion inhibition activity of 2-amino-1, 3, 4-thiadiazole and its 5-alkyl derivatives**. *J. Mol. Liq.* (2016.0) **221** 579-602. DOI: 10.1016/j.molliq.2016.06.011
97. Khalil N. **Quantum chemical approach of corrosion inhibition**. *Electrochim. Acta* (2003.0) **48** 2635-2640. DOI: 10.1016/S0013-4686(03)00307-4
98. Niamien PM. **Copper corrosion inhibition in 1 M HNO3 by two benzimidazole derivatives**. *ISRN Mater. Sci.* (2012.0) **623754** 1-15. DOI: 10.5402/2012/623754
99. Ibrahim MM. **Efficacious alkaline copper corrosion inhibition by a mixed ligand copper (II) complex of 2, 2′-bipyridine and glycine: Electrochemical and theoretical studies**. *ChemElectroChem* (2021.0) **8** 2052-2064. DOI: 10.1002/celc.202100218
|
---
title: Genome mining unveils a class of ribosomal peptides with two amino termini
authors:
- Hengqian Ren
- Shravan R. Dommaraju
- Chunshuai Huang
- Haiyang Cui
- Yuwei Pan
- Marko Nesic
- Lingyang Zhu
- David Sarlah
- Douglas A. Mitchell
- Huimin Zhao
journal: Nature Communications
year: 2023
pmcid: PMC10036551
doi: 10.1038/s41467-023-37287-1
license: CC BY 4.0
---
# Genome mining unveils a class of ribosomal peptides with two amino termini
## Abstract
The era of inexpensive genome sequencing and improved bioinformatics tools has reenergized the study of natural products, including the ribosomally synthesized and post-translationally modified peptides (RiPPs). In recent years, RiPP discovery has challenged preconceptions about the scope of post-translational modification chemistry, but genome mining of new RiPP classes remains an unsolved challenge. Here, we report a RiPP class defined by an unusual (S)-N2,N2-dimethyl-1,2-propanediamine (Dmp)-modified C-terminus, which we term the daptides. Nearly 500 daptide biosynthetic gene clusters (BGCs) were identified by analyzing the RiPP Recognition Element (RRE), a common substrate-binding domain found in half of prokaryotic RiPP classes. A representative daptide BGC from Microbacterium paraoxydans DSM 15019 was selected for experimental characterization. Derived from a C-terminal threonine residue, the class-defining *Dmp is* installed over three steps by an oxidative decarboxylase, aminotransferase, and methyltransferase. Daptides uniquely harbor two positively charged termini, and thus we suspect this modification could aid in membrane targeting, as corroborated by hemolysis assays. Our studies further show that the oxidative decarboxylation step requires a functionally unannotated accessory protein. Fused to the C-terminus of the accessory protein is an RRE domain, which delivers the unmodified substrate peptide to the oxidative decarboxylase. This discovery of a class-defining post-translational modification in RiPPs may serve as a prototype for unveiling additional RiPP classes through genome mining.
RiPP discovery has expanded the scope of post-translational modification chemistry, but genome mining of RiPP classes remains an unsolved challenge. Here, the authors employed bioinformatics and synthetic biology approaches to discover and characterize an unknown class of RiPPs, defined by an unusual amino-modified C-terminus.
## Introduction
Ribosomally synthesized and post-translationally modified peptides (RiPPs) constitute a major family of natural products present in all domains of life1,2. RiPP biosynthesis starts with the expression of a ribosomal precursor peptide, usually consisting of an N-terminal leader region and a C-terminal core region (Fig. 1). The leader region recruits the biosynthetic enzymes that perform post-translational modifications (PTMs) on the core region. Ultimately, the leader region is released from the modified core, often by a peptidase, to yield the final RiPP product. Due to the high variability in the sequences of precursor peptides and the growing list of chemistries performed by the modification enzymes, RiPPs exhibit vast structural diversity and hence a broad range of biological functions. Unlike non-ribosomal peptides, which have large genomic footprints to encode the required megasynthetases, RiPPs are genomically compact, and their ribosome-dependent biosynthetic logic endows RiPPs with attractive engineering potential2–5. In addition, the frequently observed promiscuous activity of biosynthetic enzymes also brings particular interest to the characterization of biosynthetic gene clusters (BGCs) that produce RiPPs with unknown PTMs6,7.Fig. 1RRE-based discovery of a RiPP class.a General RRE-dependent RiPP biosynthetic pathway. b Stylized depiction for use of RRE-Finder to discover RiPP BGCs independent of known RiPP chemistry. c Daptide precursor peptide sequence logo ($$n = 184$$). d Frequency of common pHMM hits in daptide BGCs. Common best hit pHMMs were counted and grouped as follows: Aminotransferase – PF00202, TIGR00508, TIGR00707; RRE-peptidase – PF02163, Actino_DapP_RRE, Bacill_DapP_RRE; DUF-RRE – Actino_DapB_RRE, Bacill_DapB_RRE; Methyltransferase – PF13649, PF08241, PF13849, PF08242; Fe-dependent alcohol dehydrogenase (ADH) – PF00465, TIGR03405, PF13685; Short chain oxidoreductase – PF13561, TIGR01830, TIGR02638; YcaO – PF02624, TIGR01575; Azoline dehydrogenase – TIGR03605; Lanthipeptide dehydratase – PF00069, TIGR03897; Luciferase-like reductase – TIGR03564, PF00296; Acyltransferase – PF00583, TIGR01575. e Targeted BGC from Microbacterium paraoxydans DSM 15019 identified from the genome mining analysis of RREs.
Despite the success of genome mining, its direct application in RiPP discovery remains challenging, principally owing to the lack of a ubiquitous feature across RiPP biosynthetic pathways and low accuracy in precursor peptide prediction8. Among the 48 classes of RiPPs reported to date, nearly all of the founding members of each class were identified by bioassay-guided screening, and their ribosomal origin was only later identified2. While bioassay-guided isolation is a powerful discovery strategy, known shortcomings include high rediscovery rates, inability to identify compounds that differ in activity from those examined by the bioassay, and challenges in specifically targeting natural products from cryptic BGCs9,10. Therefore, significant effort has been spent to develop bioinformatics tools for RiPP genome mining, using both class-dependent and class-independent strategies8.
Broadly speaking, class-dependent RiPP genome mining relies on sequence homology to well-characterized biosynthetic enzymes, affording BGC predictions with high fidelity but potentially reduced novelty of chemical structures. New RiPP classes, such as the pearlins11,12, ranthipeptides13, spliceotides14, ryptides15, rotapeptides16, and tryglysins17,18 were discovered using this strategy, but the scope of these genome mining efforts was limited to lanthipeptide dehydratases and radical S-adenosylmethionine (SAM)-dependent enzymes that were already well established in RiPP biosynthesis. In addition, key RiPP biosynthetic enzymes, such as cytochrome P450 enzymes for cittilins19, atropitides20, and biarylitides21, or radical SAM enzymes for additional classes, are widely distributed outside of RiPP BGCs; therefore, the presence of genes encoding such enzymes cannot be used as a reliable RiPP biomarker. In contrast, class-independent genome mining tools have also been developed in recent years using features common across natural product classes, but these have yet to deliver new enzyme families to RiPP biosynthesis10.
One unexploited approach to identify first-in-class RiPPs is to leverage the RiPP precursor recognition element (RRE), a domain responsible for recruiting the precursor peptide to RiPP biosynthetic enzymes22–24. As the most common domain involved in RiPP biosynthesis, the RRE has been found in 19 of the 41 known prokaryotic RiPP classes, and its frequency suggests it could be used as a class-independent handle for RiPP genome mining. The prevalence of the RRE is obscured by its small size and high sequence variability, but the recent report of RRE-Finder, a bioinformatics tool that rapidly and accurately detects RRE domains, offers the potential to discover first-in-class RiPP BGCs by identifying high-confidence RRE domains from public genome databases23 (Fig. 1).
In this work, we employ RRE-Finder23 and RODEO25 in a class-independent manner to uncover a RiPP class. The identified BGCs are unlike any other reported RiPP and ubiquitously encode a domain of unknown function (DUF)-RRE fusion protein, oxidative decarboxylase, aminotransferase, and methyltransferase. Direct cloning and heterologous expression of a representative BGC from Microbacterium paraoxydans DSM 15019 yields peptides containing a native N-terminus and an (S)-N2,N2-dimethyl-1,2-propanediamine (Dmp)-modified C-terminus. These peptides are further identified as synergistic hemolysins, with their bioactivity putatively conferred by helical structure and net positive charge on both termini. Study of the biosynthetic pathway indicates that *Dmp is* derived from the invariant C-terminal Thr of the precursor peptide through successive oxidative decarboxylation, transamination, and dimethylation. Subsequent characterization shows that the oxidative decarboxylase is leader peptide-dependent and tolerates variability in the core region. Overall, the discovery of daptides [(S)-N2,N2-dimethyl-1,2-propanediamine-containing peptides] showcases class-independent genome mining for RiPP biosynthetic enzyme families and provides a logical route for nature to produce ribosomal peptides with two amino termini.
## RRE domains guide RiPP class discovery
RiPP genome mining has primarily focused on known RiPP-modifying enzymes for new BGCs2. While this approach continues to yield new biosynthetic pathways, the discovery potential is restricted to known enzyme families. To uncover RiPP classes independent of an established chemistry, we devised an orthogonal approach using the RRE domain22. Initial analysis of all identified RRE domains faced two main challenges: (i) the large number of RRE-containing proteins, and (ii) the lack of sequence similarity between disparate RRE-containing proteins. These features of RRE analysis made traditional genome mining approaches computationally difficult and not generalizable across all RRE-containing BGCs. To circumvent these challenges, a workflow was devised to leverage the co-occurrence of RRE domains with nearby encoded open-reading frames (ORFs) using the genome mining tool RODEO (Supplementary Fig. 1, Supplementary Note)25. All RRE-containing proteins predicted in the RRE-Finder exploratory mode dataset were analyzed by RODEO for local hypothetical short ORFs and gene co-occurrence23. RRE-containing proteins were then subjected to all-by-all BLASTP and sorted by sequence similarity into large RRE families. Each RRE family was recursively analyzed for highly similar co-occurring ORFs at a series of bitscore similarity thresholds. Using this approach, RRE families with highly similar co-occurring ORFs were generated without requiring any visualization of phylogenetic data or a uniform similarity threshold (Supplementary Fig. 1).
Output RRE families were analyzed, and one set of BGCs was selected for further characterization based on the uniqueness of the co-occurring genes. The selected RRE domains were N-terminally fused to an intramembrane site-2 protease. A sequence logo of the co-occurring ORFs was generated, showing a conserved N-terminal motif (ELExMEAP), a stretch of residues rich in branched-chain aliphatic amino acids, and an invariant C-terminal Thr (Fig. 1, Supplementary Fig. 2). This approach identified 184 short ORFs encoded near 80 RRE-peptidase fusion proteins (average of 2.3 ORFs/BGC). The encoding of multiple similar short ORFs within each BGC increased confidence in their prediction as precursor peptides, and the position of the N-terminal motif within the putative leader region suggested it may function as the recognition sequence for the associated biosynthetic enzymes. *Initial* gene co-occurrence analysis identified multiple highly co-occurring genes encoding multi-component ABC transporters, a pyridoxal phosphate (PLP)-dependent aminotransferase, an NAD(P)-dependent alcohol dehydrogenase, and a SAM-dependent methyltransferase (Fig. 1, Supplementary Table 1). Each BGC also contained an additional gene, which lacks sequence similarity to any known protein. However, structural similarity analysis using HHpred26 predicted a C-terminal RRE domain, suggesting its inclusion in the potential biosynthetic pathway (Supplementary Fig. 3).
We next sought to expand the putative class by gathering all predicted BGCs from the NCBI non-redundant database. To gather a comprehensive set of putative modifying enzymes, Position Specific Iterative-Basic Local Alignment Search Tool (PSI-BLAST) methods were applied to each predicted biosynthetic protein, and the resulting BGCs were compiled. The originally identified BGCs of this class were restricted to Actinomycetota, but PSI-BLAST searches using the DUF-RRE fusion protein as a query returned a small number of similar BGCs from the phylum Bacillota. A profile Hidden Markov Model (pHMM) was constructed from the 184 initially predicted precursor peptides from Actinomycetota and used to identify a BGC from *Bacillus cereus* strain B4082 (Supplementary Fig. 4, Supplementary Data 1). PSI-BLAST searches were conducted using the B. cereus DUF-RRE fusion protein as a new query to expand the number of Bacillota BGCs. A phylogenetic comparison of the BGCs revealed both similarities and differences. One difference pertains to the number of precursor peptides encoded per BGC, with some Bacillota BGCs encoding >10 precursor peptides (Supplementary Fig. 5). Actinomycetota BGCs also encode a predicted Fe-dependent alcohol dehydrogenase (Pfam27 identifier PF00465), while Bacillota BGCs encode a different class of oxidoreductase (PF13561) and a shorter protease domain (Supplementary Fig. 6). Despite similar RiPP contexts, biosynthetic proteins from Actinomycetota were typically as similar to characterized outgroup proteins as they were to their Bacillota counterparts (Supplementary Fig. 7, phyloXML data is provided as Supplementary Data 2). In total, 1441 putative precursor peptides were identified from 483 BGCs (Supplementary Data 1, Actinomycetota – 791 precursor peptides in 302 BGCs; Bacillota – 647 precursor peptides in 180 BGCs; Pseudomonadota – 3 precursor peptides in 1 BGC). *In* general, there is little conservation of gene order or precursor peptide number within these BGCs. Several potential secondary modifying genes were identified, such as YcaO proteins, LanKC/LanJC pairs, glycosyltransferases, and nucleotidyltransferases (Fig. 1, Supplementary Data 1, Supplementary Table 1). This class of BGCs combines a minimal set of primary modifying enzymes, as well as secondary (ancillary) tailoring enzymes from other classes, suggesting the final biosynthetic transformations may work in concert in other RiPP pathways (Supplementary Fig. 4). After filtering the dataset for available strains and for BGCs encoding only primary modifying enzymes, we chose the representative from Microbacterium paraoxydans DSM 15019 for experimental characterization (Fig. 1).
## Discovery of peptides with Dmp-modified C-termini
To examine the predicted BGC, we directly cloned the representative BGC from Microbacterium paraoxydans DSM 15019 (mpa) using the recently developed CAPTURE (Cas12a-assisted precise targeted cloning using in vivo Cre-lox recombination) method28. *Besides* genes in the predicted BGC, flanking genes were also included in the cloning region to ensure successful expression (Supplementary Fig. 8). The cloned mpa and the pBE45 empty vector were then individually conjugated29 into Streptomyces lividans TK24 and *Streptomyces albus* J1074. After cultivation for 5 d, colonies were picked, extracted with methanol, and analyzed by matrix-assisted laser desorption/ionization-time-of-flight mass spectrometry (MALDI-TOF-MS) (Fig. 2, Supplementary Fig. 9). A series of ions only observed in strains transformed with mpa corresponded to a loss of 17 Da from the three predicted core regions (MpaA1-A3), with proteolysis C-terminal to the conserved Pro (Fig. 1). This consistency suggested that the three products shared the same post-translational modification(s). Larger scale cultures in liquid media were then prepared to afford the material required for structural characterization. After growth for 5 d at 30 °C, bacteria were collected by centrifugation, metabolites extracted with MeOH, and purified by solid-phase extraction and HPLC (Supplementary Fig. 10).Fig. 2Heterologous expression and product characterization of mpa.a MALDI-TOF mass spectra of the methanol extracts of S. albus J1074 containing mpa (1, m/z 2363; 2, m/z 2232; 3, m/z 2339) and the pBE45 empty vector. Sodium and potassium ions are denoted as ‘+Na’ and ‘+K’. b Summary of the HR-MS/MS results for compound 1-3. c, key HMBC and COSY correlations for the Dmp moiety in 1. NMR spectra of 1 and assigned signals of Dmp are shown in Supplementary Fig. 14 and Supplementary Table 2, respectively.
The HPLC-purified products were analyzed by high-resolution mass spectrometry and tandem mass spectrometry (HR-MS/MS). Analysis of the collision-induced dissociation spectra indicated that the 17 Da mass loss of modified MpaA1-A3 was confined to the C-terminal Thr (Fig. 2, Supplementary Fig. 11-13). Considering the predicted functions of genes encoded by the mpa BGC, we proposed that Thr underwent conversion to Dmp. To evaluate this hypothesis, a sufficient amount of 1 was accumulated for structure determination, due to its higher post-purification yield than 2 and 3. The peptide was only sparingly soluble in commonly used NMR solvents (H2O, methanol, acetone, acetonitrile, chloroform, DMSO, and pyridine) but was highly soluble in hexafluoroisopropanol (HFIP) which is often used to dissolve aggregated proteins and peptides30. NMR analyses (1H, 13C, 1H-1H COSY, 1H-13C HSQC and 1H-13C HMBC) were then conducted in HFIP-d2 and a substructure corresponding to Dmp was deduced. The HMBC correlations from the N-methyl (H3-4 and H3-5) to C-2 showed the attachment of a dimethylamino group to C-2. Other data, including the HMBC correlations from the H3-3 to C-1 and C-2, the HMBC correlation from H2-1 to C-2, and the COSY data, revealed the C-1 to C-3 Dmp substructure. The amide bond between Dmp and Ala-20 is suggested by the HMBC correlation from H2-1 in Dmp, as well as H1-1’ and H3-2’ in Ala-20, to the carbonyl carbon (δC = 175.8) (Fig. 2, Supplementary Fig. 14, Supplementary Table 2, Supplementary Table 3).
Extensive signal overlap was observed in the NMR spectra of 1, so orthogonal methods were employed to confirm the structure of 1. Accordingly, 1 was hydrolyzed and derivatized by Marfey’s reagent31 to determine the identity of Dmp and the remaining residues present. While the amino acid standards were obtained commercially, authentic samples of (R)-Dmp and (S)-Dmp were chemically synthesized (Supplementary Fig. 15). Using an LC-MS, all assignable amino acids (Ser, Asp, Phe, Gln, Val, Ile, Leu, Ala, and Tyr) were determined to be in the proteinogenic (S)-configuration (Supplementary Fig. 16). UHPLC afforded sufficient separation of the derivatized (R)-Dmp and (S)-Dmp standards, with subsequent co-elution with hydrolyzed 1 confirming the presence of (S)-Dmp (Supplementary Fig. 17).
To further validate Dmp as the primary and class-defining PTM, a second BGC identified from *Streptomyces capuensis* NRRL B-3501 (sca) was selected for experimental characterization. In addition to the primary modifying enzymes, the sca BGC also encodes a LanKC/LanJC pair that was predicted to convert L-Ser to D-*Ala via* a dehydroalanine intermediate32–34 (Supplementary Fig. 18). The sca BGC underwent direct cloning by the CAPTURE method28, heterologous expression in S. albus J1074, and MALDI-TOF-MS analysis (Supplementary Fig. 19). The most prominent ion observed from the sca-transformed strain was fragmented using MALDI LIFT-TOF/TOF MS and confirmed to derive from ScaA2. The corresponding fragmentation pattern confirmed the C-terminus had been converted to Dmp and the five L-Ser residues had been converted to Ala (Supplementary Fig. 19). Marfey’s assay performed on a mixture of the two major products further showed the peptides contained D-Ala residues and (S)-Dmp (Supplementary Fig. 20). With multiple characterized BGCs producing Dmp-modified peptides, we anticipate that the Dmp modification will be conserved across the RiPP class, hereafter termed daptides.
## Examination of daptide bioactivity
As 1-3 contain regions of hydrophobic residues and showed limited solubility in most solvents, we theorized that 1-3 might form hydrophobic α-helical secondary structures. Secondary structure prediction of MpaA1-A3 and circular dichroism spectra of 1-3 support adoption of an α-helical conformation (Supplementary Fig. 21). Melittin, the major pain-producing component of honeybee venom, disrupts membranes via its hydrophobic α-helical structure and a group of positively charged residues near the amidated C-terminus (Supplementary Fig. 22), giving it additional antimicrobial and hemolytic activity. Given the positive charge at the (former) C-terminus of 1-3, we hypothesized that daptides may imitate this strategy to interact with membranes35. Daptides 1-3 were first analyzed for antimicrobial activity by standard agar diffusion assay, but no significant growth inhibition was observed for any of the tested Bacillota, Actinomycetota, Pseudomonadota, and Ascomycota strains (Supplementary Table 4). To further evaluate their potential membrane interactions, 1-3 were incubated with bovine erythrocytes and exhibited hemolytic activity (Supplementary Fig. 23). Because daptide BGCs mostly encoded multiple precursors (Supplementary Fig. 5), combinations of 1-3 were also examined for hemolytic activity. Some combinations (i.e., 1 + 3, 2 + 3, and 1 + 2 + 3) exhibited higher hemolytic activity than the individual peptides alone, underscoring a collaborative bioactivity (Supplementary Fig. 23).
## Biosynthesis of Dmp via a three-step pathway
To determine the minimal set of genes responsible for daptide 1-3 biosynthesis, a series of gene omissions in the mpa BGC (Supplementary Fig. 24) were constructed by the DNA assembler method36 and conjugated into S. albus J1074 for expression. The colony extracts were analyzed by MALDI-TOF-MS, which indicated that mpaABCDMP were sufficient for synthesizing the fully modified peptides, albeit with compromised signal intensity compared to the mpa BGC, suggesting other flanking genes may enhance productivity. No diagnostic intermediates were observed when conserved biosynthetic genes were omitted (Supplementary Fig. 24). Therefore, we elected to express the mpaABCDM pathway in E. coli. Genes were codon-optimized, commercially synthesized, and refactored using a previous method (Supplementary Table 5, Supplementary Table 6)37. The precursor peptides (MpaA1-A3) were individually N-terminal His-tagged to facilitate purification by immobilized metal affinity chromatography (IMAC). The purified products were then further desalted using C18 ZipTip and analyzed by MALDI-TOF-MS. During optimization of the expression conditions, we observed that the signals of fully modified products were significantly improved when using M9 medium with extended cultivation and co-expression of GroES/EL chaperones (Supplementary Fig. 25-28)38. MpaA1 routinely delivered the most intense MS peaks; therefore, the following experiments were carried out using mpaA1 exclusively (hereafter, mpaA).
A biosynthetic scheme was next proposed, employing each critical gene in a distinct step in the conversion of a C-terminal Thr to Dmp (Fig. 3). The unmodified peptide 4, containing an intact C-terminal Thr would undergo oxidative decarboxylation to ketone 5, followed by transamination to yield primary amine 6, and finally dimethylation to afford the final tertiary amine-containing Dmp 7. One advantage of a refactoring strategy is the opportunity to evaluate biosynthetic intermediates by rapidly generating gene omission constructs. Thus, we evaluated the roles of mpaBCDM in the proposed biosynthetic route with a panel of gene omissions. Expression of mpaABCD (i.e., omission of mpaM, predicted involvement in the last step) resulted in the accumulation of a product 45 Da lighter than starting peptide 4. This result was consistent with primary amine 6 and concurred with the expected role of MpaM in dimethylation. Expression of mpaABCM or mpaABC resulted in the accumulation of a product 46 Da lighter than starting peptide 4, consistent with ketone 5 and a transaminase activity for MpaD. Given the mass similarity to amine 6, and to further support the existence of ketone 5 as a biosynthetic intermediate, we employed the use of aminooxy probe 8, which would yield an oxime after reaction with aldehyde and ketone functional groups39. The MS data revealed that 5 was converted to oxime 9, whereas 4 was unreactive under identical conditions (Fig. 3). Expression of mpaABDM, mpaAB, and mpaAC constructs yielded only unmodified starting peptide 4, which taken together implicated MpaB (DUF-RRE fusion) and MpaC (annotated alcohol dehydrogenase) in the first reaction, oxidative decarboxylation of the C-terminal Thr. Fig. 3Characterization of the biosynthetic pathway.a Proposed biosynthetic route for the Dmp modified C-terminus. b MALDI-TOF mass spectra of IMAC-purified MpaA1 that were co-expressed with the indicated Mpa biosynthetic proteins in E. coli (4, m/z 5872; 5, m/z 5826; 6, m/z 5827; 7, m/z 5855). c Enlarged MALDI-TOF mass spectra of mpaABCD, mpaABCM, and mpaABC co-expressions in panel b. d Carbonyl reactivity using aminooxy probe 8. e MALDI-TOF MS results for the unmodified peptide 4 and ketone intermediate 5 (oxime 9, m/z 5931).
## Oxidative Decarboxylation of the C-terminal Thr
*The* gene omission studies showed that ketone 5 can only form in the presence of MpaB and MpaC. Although the RRE domain fused to the peptidase was used to identify the daptide BGCs, MpaB also contains a predicted RRE domain (Supplementary Fig. 3). Therefore, we hypothesized that the MpaB RRE domain might bind the MpaA1-A3 leader region and deliver the C-terminal Thr to MpaC for modification. To evaluate this interaction, Ala substitutions were introduced into the conserved motif of the MpaA1 (i.e., MpaA) leader region, and the resulting products after co-expression with MpaB and MpaC were analyzed by MALDI-TOF-MS (Fig. 4, Supplementary Fig. 29). The D[-6]A variant was processed at a level equal to wild-type, while E[-8]A, P[-5]A, E[-3]A, and P[-1]A yielded a reduced level of product formation. Two variants, L[-7]A and M[-4]A, were devoid of detectable oxidative decarboxylation. Removal of the C-terminal RRE domain from MpaB (i.e., expression of MpaB1-266), also was incapable of generating ketone 5, further validating a specific interaction between the leader region of MpaA and the RRE domain. Fig. 4Characterization of ketone intermediate formation by MpaB and MpaC.a Proposed interaction model for MpaA, MpaB, and MpaC. The RRE domain (dark grey) of MpaB interacts with the conserved leader region of MpaA. The C-terminus (depicted as “-OH”) of MpaA is directed into the active site of MpaC for modification. b Summary of MpaA variant processing upon E. coli co-expression with MpaB/C. c *Mass spectra* of MpaA after co-expression with MpaC, and MpaB or MpaB1-266. The RRE domain of MpaB is dark grey. d The structure of MpaA1-MpaB-MpaC predicted using AlphaFold-Multimer40. MpaA1, MpaB, and MpaC are shown as cartoons in green, orange, and blue, respectively. The positioning of cofactor NADP was imported using a homologous alcohol dehydrogenase (PDB ID: 6C76) as a template. e, View of the leader peptide-RRE domain binding interface40. Select residues of MpaA and MpaB are in green and orange, respectively. f View of the C-terminal Thr24 of MpaA in the substrate-binding pocket of MpaC.
To assess the substrate tolerance of MpaC, the C-terminal Thr of MpaA was replaced with Ser, Cys, Asp, Asn, Val, Ala, and Lys. No mass change was observed for any variant, indicating a stringent selectivity for Thr (Fig. 4, Supplementary Fig. 30). Another MpaA variant, T24insA, appended Ala to the C-terminus and was not a substrate for MpaC (Supplementary Fig. 31). However, variants that increased or decreased the length of the core region (i.e., A20dup and A20del) were readily tolerated along with all other tested substitutions (Supplementary Fig. 32). Thus, while MpaC is highly chemoselective for a C-terminal Thr, the exact positioning of the C-terminal Thr and the identity of the intervening residues are not critical. A high tolerance to core peptide variation was anticipated from the bioinformatics analysis, as high variation of the core peptide sequences was observed even for those within the same BGC, while the C-terminal residue remained a Thr across the daptide class (Fig. 1).
To shed more light on the formation of ketone 5, AlphaFold-Multimer, a recently reported structure prediction tool for protein complexes, was used to calculate the protein-protein interactions between MpaA, MpaB, and MpaC40. MpaC is predicted to use NAD(P) as a cofactor, so a structural alignment was used to model NADP from a homologous alcohol dehydrogenase, given the current inability of AlphaFold to predict ligands. The complex with NADP was then relaxed to yield a final predicted structure (Fig. 4, Supplementary Fig. 33). These results aligned well with our hypothesis that MpaB recognizes the leader region of MpaA and recruits MpaC to form ketone 5. Two leader region residues of MpaA, Leu[-7] and Met[-4], putatively interact with MpaB in adjacent hydrophobic pockets, supporting the experimental findings that ketone 5 formation was abolished with the L[-7]A and M[-4]A variants (Fig. 4). Structural and electrostatic potential analysis of the predicted MpaB-MpaC interface shows a high degree of charge and shape complementarity (Supplementary Fig. 34). Finally, the Alphafold prediction directs Thr24 toward the active site NADP at an appropriate distance for catalysis.
## Discussion
With recent advances in DNA sequencing and bioinformatics, natural products with diverse structures and bioactivities have been continuously uncovered41. RiPPs constitute one of the major classes of natural products, and the number of characterized compounds has grown rapidly in the last decade. However, unveiling new PTMs remains a major challenge in the genome mining of RiPPs, as class-independent genome mining methods have only recently received broad attention from the field. With RRE domains being present in $50\%$ of prokaryotic RiPP classes, we saw an opportunity to exploit the RRE as a RiPP biomarker. In this work, RRE domains identified by RRE-Finder were binned into families, guided by co-occurrence with nearby short ORFs. Employing this method led to the discovery of the daptide class from a group of RRE-containing peptidases. Enabled by synthetic biology methods for direct cloning and refactoring, pathway characterization revealed the biosynthetic route to Dmp. Given the generality of the methods used in this study, the success of the daptide discovery could be replicated for yet-undiscovered RiPPs.
To assess the frequency of Dmp-modified C-termini in known compounds, we performed a substructure search using Dmp with several adjoining peptide bonds on SciFinder (https://scifinder-n.cas.org/). This search identified hominicin, a peptidic natural product produced by *Staphylococcus hominis* MBBL 2-942. Besides a Dmp-modified C-terminus, hominicin also contains several dehydrobutyrines and a dimethylated N-terminus. Unfortunately, the genome of S. hominis MBBL 2-9 is not available, and no sequenced S. hominis strains contain a daptide-producing BGC. However, the predicted precursor peptide of hominicin is virtually identical to a precursor peptide identified from *Staphylococcus pseudintermedius* strain B32 (Supplementary Fig. 35). The S. pseudintermedius BGC contains DapBCDM homologs (Supplementary Table 7), a second methyltransferase, and a lanthipeptide dehydratase, likely accounting for the additional methylations and dehydrobutyrines observed in hominicin. Retrospectively, these similarities identify hominicin as a daptide. Hominicin is growth suppressive towards Staphylococcus aureus, an activity which was not observed for daptides 1-3.
In comparison to the hominicin example, genome mining was critical for the discovery of daptides 1-3 and offers leads for the discovery of further compounds. In our bioinformatics analysis, we found major subgroups of daptide BGCs that encode YcaOs, thiopeptide-like pyridine synthases, radical SAM enzymes, lanthipeptide dehydratases, luciferase-like monooxygenases, glycosyltransferases, and/or nucleotidyltransferases. These secondary PTMs of daptides are expected to confer structural and functional diversity. The bioinformatics survey of the daptide class also demonstrated different paradigms between Actinomycetota and Bacillota BGCs. Functionally equivalent daptide biosynthetic enzymes from different phyla are not phylogenetically similar, despite all surveyed daptide BGCs containing an NAD(P)-dependent oxidative decarboxylase, aminotransferase, and an RRE-containing DUF (now known to be critical for oxidative decarboxylation). The evolutionary distance between the daptide BGCs suggests that the pathways emerged individually in each phylum, and this may further divide the ecological role of daptides by phylum in addition to secondary modifications.
Peptidic natural products with amino-modified C-termini are rare, and the corresponding biosynthetic machinery remains elusive. Recently, the non-ribosomal aquimarins, which feature an amino-modified C-terminus, were isolated from sponge-derived bacterium Aquimarina sp. Aq13543. The amino-modified C-terminus was proposed to be generated by thioester reductase-catalyzed release of an aldehyde and subsequent transamination by an aminotransferase. In contrast, our work unveils a strategy for the amino-modification of ribosomal peptide C-termini. MpaB and MpaC collaboratively catalyze the oxidative decarboxylation of a C-terminal Thr into a ketone with concomitant loss of CO2. PLP-dependent transamination and SAM-dependent N,N-dimethylation by MpaD and MpaM, respectively, yield the terminal (S)-N2,N2-dimethyl-1,2-propanediamine. The leader peptide is then removed by an RRE-dependent intramembrane peptidase (MpaP), yielding the final peptide that has converted a negatively charged C-terminal carboxylate into a positively charged tertiary amine.
Within RiPP biosynthesis, there are many other examples of C-terminal modifications. Of particular relevance are the oxidative decarboxylation reactions observed in the biosynthesis of mycofactocin (radical SAM), bottromycin (cytochrome P450), micrococcin/thiocillin (Bacillota DapC homolog), and Avi(Me)Cys formation (flavin-dependent decarboxylase)2. C-terminal modifications extend to peptide hormones (e.g., oxytocin and vasopressin) and antimicrobial peptides, which display decreased function without C-terminal amidation44,45. Additionally, GTPases employ a C-terminal modification strategy, including esterification of the C-terminus, to enhance insertion into cell membranes46. Nature has evolved numerous ways to neutralize local charge at the C-terminus through PTMs, but daptide biosynthesis goes one step further: generating a positive charge at the terminus. The biophysical implications of peptides with two positively charged amino-termini are not yet understood, but the combined hydrophobicity, helicity, and charge modification of the terminus suggest daptides may share a common membrane-targeting mode of action. Further studies of the substrate scope for Dmp formation will determine whether this charge swapping strategy will be useful for engineering peptides that associate with or penetrate membranes.
## Initial Analysis by RODEO
All RiPP Recognition Element (RRE)-containing Uniprot identifiers (IDs) that were previously gathered using the exploratory mode of RRE-Finder23 were converted to GenBank coding sequence (CDS) accession IDs using the Uniprot Retrieve/ID mapping tool. To identify potential precursor peptides in intergenic open reading frames (ORFs), all GenBank IDs were subjected to the genome mining tool Rapid ORF Description and Evaluation Online (RODEO)25. As part of standard RODEO analysis, ‘.gbk’ files for each provided accession ID are downloaded from the National Center for Biotechnology Information (NCBI). From these ‘.gbk’ files, nucleotide sequences were subjected to CDS prediction using Prodigal-shorter to identify ORFs. Prodigal-shorter is an adaptation of Prodigal-short used in the RiPPER workflow47 that allows for CDSs to be predicted as short as five amino acids in length, based on recent reports of RiPP precursor peptides of that length21. CDS prediction of the RODEO selected region was performed in Prodigal’s meta mode for records with fewer than 100,000 nucleotides. For larger records, a CDS training step was performed on the whole record before CDS prediction on the RODEO selected region. Intergenic ORFs were dereplicated by highest Prodigal score. These Prodigal scoring results were output to a separate ‘.csv’ file, and a FASTA file was generated containing all genes and all possible predicted ORFs that were smaller than 120 amino acids and within 8 genes of an RRE domain.
## All-by-all BLAST analysis
The list of GenBank CDS IDs was mapped to protein sequences in FASTA format using epost and efetch commands from the NCBI Entrez E-utilities. The FASTA file was converted to a blastdb using DIAMOND makedb48. An all-by-all BLAST search was performed using DIAMOND blastp on very-sensitive mode, a faster version of NCBI’s Basic Local Alignment Search Tool (BLAST). The list of all ORFs was checked against the query list of RRE-containing proteins to remove any RRE domains from precursor peptide analysis. Each ORF was also compared to a set of pHMMs adapted from RRE-Finder’s precision mode, and all hits were removed. These ORFs were then converted to a blastdb, and an all-by-all BLAST was performed using DIAMOND.
## Prediction of paired RRE families and precursor families
All-by-all BLAST results for the set of all RRE domains were parsed and entered into the precursor prediction algorithm (Fig. S1, Note S1). All RRE pairwise BLAST results with a bitscore <50 were removed. ORF all-by-all BLAST results were parsed, and pairwise BLAST results with bitscore <10 were removed. All self-hits were also removed.
All RRE domains were sorted into families using the presence or absence of pairwise BLAST results above the bitscore threshold (starting with 50). Families were sorted by descending order of length. Any RRE families with fewer than 30 members were omitted for simplified analysis of only families with high membership in the dataset. To maximize potential leads, cases where proteins in the RRE family were highly divergent were re-entered into the algorithm at a higher bitscore, increasing the threshold bitscore by 10 with each iteration. RRE divergence was assessed by “enzyme connectivity”, which is the ratio of observed pairwise hits in an RRE family above the threshold bitscore to the maximum possible number of pairwise hits for the RRE family. For RRE families with enzyme connectivity ratios >0.2 and sufficient size, the size of the largest precursor family was calculated.
The list of ORF pairwise BLAST results was checked for results where both ORFs are encoded near RREs from the same family. Precursor families were generated from each list of related ORFs encoded near members of the same RRE family as above. The size of the precursor family was then compared to the size of the RRE family. When the size of the largest precursor family was within an order of magnitude of the size of the RRE family, the RRE family was output from the algorithm for manual evaluation. For other cases, the RRE family was re-entered into the algorithm recursively, with the bitscore threshold increased by 10. This recursive process resulted in the fractionation of RRE families, with each resultant family re-entered into the algorithm at the more stringent similarity threshold. Output paired families of RRE domains and precursor peptides were assessed based on enzyme connectivity, the relative size of precursor families, enzyme co-occurrence, and precursor peptide sequences for novelty and prioritization.
## Generation of pHMMs for daptide precursor peptides and RRE domains
Precursor peptides output from the analysis algorithm were manually analyzed within the daptide family. Within the precursor family, 184 ORFs displayed Thr as the C-terminal residue. The 184 ORFs were aligned by Multiple Alignment using Fast Fourier Transform (MAFFT)49 under the G-INS-I method. This alignment was used as input for HMMER3 hmmbuild50 to create the daptide precursor peptide pHMM, which was subsequently uploaded to Skylign51 for visualization.
Representative BGCs from Actinomycetota and Bacillota were selected. The last 100 amino acids (encompassing the entire predicted RRE domain) from the DapB proteins in these BGCs were copied into separate FASTA files by phylum. These files were then aligned using MAFFT under the G-INS-I method. These alignments were used as input for HMMER3 hmmbuild to create the Actino_DapB_RRE and Bacill_DapB_RRE pHMMs. The same approach was applied for DapP proteins from the same representative BGCs, using each protein’s first 100 amino acids. Alignment with MAFFT and pHMM generation with hmmbuild generated the Actino_DapP_RRE and Bacill_DapP_RRE pHMMs.
## Expansion of daptide BGC family
MpaB was submitted to PSI-BLAST analysis to identify additional BGCs within the daptide class. The first iteration was restricted to 100 protein hits, with subsequent iterations allowing up to 1000 protein results, using an e-value cutoff of 10. After retrieving 1000 sequences, the accession IDs were submitted to RODEO for analysis and compilation.
The daptide precursor peptide pHMM was uploaded to the hmmsearch webtool hosted at EMBL-EBI52. The search was performed against UniProtKB, and the output results were analyzed for taxonomy. Individual IDs were converted to GenBank IDs using UniProt Retrieve/ID Mapping and analyzed by RODEO to confirm membership in the daptide class using local gene conservation. After identifying BGCs from Bacillota, an example DapB protein from *Bacillus cereus* (Genbank ID: KZD28690.1 [https://www.ncbi.nlm.nih.gov/protein/1017038289]) was selected for PSI-BLAST analysis using the method described for MpaB. These BGCs were analyzed and compiled with the results from the MpaB PSI-BLAST analysis.
Representative aminotransferase proteins in previously identified daptide BGCs were subjected to BLAST-P analysis on the NCBI’s BLAST website. The top 2000 results were gathered for each aminotransferase, and duplicate results were removed. All resulting proteins were analyzed by RODEO. Putative BGCs were checked using local gene co-occurrence. BGCs containing a RODEO annotation for a DapB protein and either an iron-containing alcohol dehydrogenase (PF00465) or an enoyl-(acyl carrier protein) reductase (PF13561) were included as daptide BGCs, with all others removed. After using Prodigal dereplication strategies as described above, daptide precursor peptides were manually identified from ORF output to confirm membership in the daptide class.
## Generation of aminotransferase phylogenetic tree
Outgroups for the phylogenetic tree were selected by analyzing the SwissProt database for class III aminotransferases. All daptide aminotransferases and three selected outgroups were aligned by MAFFT using the L-INS-I option. FastTree 2.1.10 was used to generate a maximum likelihood tree with the JTT + CAT model. The tree was visualized using the Interactive Tree of Life (iTOL) website (https://itol.embl.de/).
## Direct cloning of mpa and sca BGCs
Direct cloning of mpa and sca BGCs was achieved by the CAPTURE method28. Both M. paraoxydans DSM 15019 and S. capuensis NRRL B-3501 were recovered on ISP2 agar medium (malt extract 10 g/L, yeast extract 4 g/L, glucose 4 g/L, agar 20 g/L, pH 7.2–7.4) at 37 °C until colony appears (about 3 days). A single colony was inoculated into 5 mL ISP2 liquid medium as seed culture and grown at 37 °C 250 rpm until saturation (3 d). 1 mL of the seed culture was then transferred into 50 mL fresh ISP2 liquid medium and cultivated for 18-20 h. Cells were then harvested by centrifugation at 3000 × g for 15 min and resuspended in 12 mL of cell resuspension buffer (50 mM Tris-HCl pH 8.0, 25 mM EDTA). Cell lysis enzymes (lysozyme 30 mg, mutanolysin 300 U), and 0.6 mg RNase A) were added and the sample was incubated at 37 °C for 18 h. Afterwards, 6 mg proteinase K was added and the sample was incubated at 37 °C for another hour. Following the incubation, 1.2 mL of $10\%$ sodium dodecyl sulphate (SDS) was added and the sample was incubated at 50 °C for 2 h. Genomic DNA was recovered by phenol-chloroform extraction. The cell lysate was gently mixed with 15 mL phenol-chloroform-isoamyl alcohol (25:24:1, v/v, pH 8.0) until the aqueous phase became completely white. The sample was then centrifuged at 22,000 × g for 45 min at room temperature, and the aqueous phase was mixed with 10 mL chloroform. Following the same mixing procedure as described above, the sample was centrifuged at 22,000 × g for 10 min and the aqueous phase was aliquoted into 1.7 mL centrifuge tubes. DNA was recovered using isopropanol and sodium acetate precipitation, washed 3 times with $70\%$ ethanol (v/v), and rehydrated in 10 mM Tris-HCl pH 8.0.
Guide RNA was prepared through in vitro transcription. Briefly, 5 µM forward and reverse template oligonucleotides were mixed in NEBuffer 3.1. The mixture was incubated at 98 °C for 5 min, followed by slowly reducing the temperature with the rate of 0.1 °C/s until 10 °C was reached. A 2 µL aliquot of the annealed oligonucleotides was used as template for in vitro RNA transcription by HiScribe T7 quick high yield RNA Synthesis kit (New England Biolabs, MA). The transcribed RNA was purified using RNA clean and concentrator kit (Zymo Research, CA). Sequences of the DNA templates for the in vitro gRNA transcription were shown in Supplementary Table 1 and synthesized by Integrated DNA Technologies (Coralville, IA).
Genomic DNA was digested by FnCas12a in a 300 µL reaction containing 15 µg purified genomic DNA, 2.1 µg of each guide RNA, 60 pmol FnCas12a, and 30 µL of 10× NEBuffer 3.1. The reaction was performed at 37 °C for 2 h followed by 65 °C for 30 min. RNase A was then added to a final concentration of 0.1 mg/mL and the sample was incubated at 37 °C for another 30 min. Following RNase treatment, 3 µL of 20 mg/mL proteinase K solution was added and the sample was incubated at 50 °C for 30 min. The reaction was then transferred into 5PRIME PLG light phase-lock gel tubes (Quantbio, MA) and mixed with 300 µL of phenol-chloroform-isoamyl alcohol (25:24:1, v/v, pH 8.0). After proteins precipitated in the aqueous phase, the sample was centrifuged at 20,000 × g for 30 s. The aqueous phase was transferred into a new phase-lock tube and the extraction step was repeated. Ethanol and sodium acetate precipitation was then used to recover the DNA in the aqueous phase. The DNA pellets were washed 2 times with $70\%$ ethanol (v/v) and rehydrated in 15 µL of 10 mM Tris-HCl pH 8.0.
The digested genomic DNA was assembled with DNA receivers by the T4 DNA polymerase exo + fill-in method. DNA receivers were PCR amplified using primers listed in Supplementary Table 1. A 15 µL assembly reaction consists of 3–3.75 µg of digested genomic DNA, 15 ng of DNA receiver amplified from plasmid pBE44, 35 ng of DNA receiver amplified from plasmid pBE45, 1.5 µL of NEBuffer 2.1, and 0.75 U of T4 DNA polymerase. Before adding T4 DNA polymerase, the sample was first incubated at 65 °C for 10 min followed by 25 °C hold without any mixing. T4 DNA polymerase was then added and gently mixed using wide-bore pipette tips. The reaction was performed at 25 °C for 1 h, 75 °C for 20 min, and 50 °C for 30 min. Next, 1 µL of 1 mM NAD+, 0.4 µL of 10 mM dNTPs, 1 µL (3 U) of T4 DNA polymerase, and 1 µL of E. coli DNA ligase were added to the reaction, which was incubated at 37 °C for 1 h, 75 °C for 20 min, and stored at 10 °C until transformation.
The assembled DNA fragments were then circularized by Cre-lox in vivo recombination. E. coli NEB10β cells containing the pBE14 helper plasmid were grown overnight at 30 °C in modified SOB medium (20 g/L Bacto tryptone, 5 g/L Bacto yeast extract, 10 mM NaCl, 2.5 mM KCl) supplemented with 8 µg/mL tetracycline hydrochloride. A 100 µL aliquot of the overnight culture was used to inoculate 10 mL modified SOB medium supplemented with 8 µg/mL tetracycline hydrochloride and cultured at 30 °C. After 2 h growth (OD600 of ~0.2), 100 µL of 1 M L-arabinose was added to the culture and cells continued to grow at 30 °C until the OD600 reached 0.45-0.55 (~1.5 h after induction). Cells were then harvested by centrifugation at 3220 × g for 7 min at room temperature and washed using 1 mL of $10\%$ (v/v) glycerol for three times. Cells were then resuspended in a final volume of 70 µL $10\%$ (v/v) glycerol and gently mixed with 2.5 µL assembly reaction using wide-bore pipette tips. Electroporation was performed in 1 mm cuvettes using Gene Pulser XCell Electroporation system (Bio-Rad, CA) set at 1250 V, 100 Ω, 25 µF. Afterwards, cells were resuspended by 1 mL LB medium supplemented with 5 mM MgCl2 and recovered at 37 °C for 75 min with shaking at 250 rpm. All the cells were plated on LB agar plates containing 50 µg/mL apramycin with blue/white screening and incubated at 37 °C until colonies appeared. White colonies were picked and grown overnight in 5 mL LB medium supplemented with 50 µg/mL apramycin at 37 °C. The plasmid DNA was purified from the cultures using Qiaprep Spin Miniprep Kit (Qiagen, Germany), digested by appropriate restriction enzymes, and analyzed by agarose gel electrophoresis. Correct plasmids were transformed into WM6026 (supplemented with 2,6-diaminopimelic acid at the final concentration 40 µg/mL for growth) and conjugated into S. lividans TK24 and S. albus J1074 for expression53.
## Product analysis from colony extracts
The exconjugants of S. lividans TK24 and S. albus J1074 containing mpa, sca, or the empty pBE45 vector were picked and restreaked onto fresh MS (mannitol 20 g/L, soybean flour 20 g/L, agar 20 g/L) and ISP4 (soluble starch 10 g/L, K2HPO4 1 g/L, MgSO4·7H2O 1 g/L, NaCl 1 g/L, (NH4)2SO4 2 g/L, CaCO3 2 g/L, FeSO4 1 mg/L, MnCl2 1 mg/L, ZnSO4 1 mg/L, agar 20 g/L) agar medium supplied with apramycin at a final concentration of 50 μg/mL and incubated under 30 °C for 5 d. A portion of cell mass (pinhead-sized) was picked from the plate, placed in 20 μL methanol and incubated at room temperature for 1 h. Methanol extract was then mixed equally with 1 μL of 15 mg/mL $70\%$ aq. MeCN solution of α-cyano-4-hydroxycinnamic acid (CHCA) with $0.1\%$ trifluoroacetic acid (TFA) (v/v) on a ground steel MALDI target, and the droplet was dried under ambient conditions. Samples were analyzed using a Bruker UltrafleXtreme MALDI-TOF MS using manufacturer methods for reflector positive mode. The MALDI LIFT-TOF/TOF mass spectra were acquired in the positive ion mode. Metastable fragmentation was induced by a nitrogen laser (337 nm) without the further use of collision gas. Precursor ions were accelerated to 8 kV and selected in a timed ion gate. In the LIFT-cell the fragments were further accelerated to 19 kV. The reflector potential was 29 kV.
## Heterologous expression and product isolation
Freshly obtained exconjugants of S. albus J1074 containing mpa or sca were individually restreaked onto MS plates with apramycin and incubated under 30 °C for 5 d. Colonies were verified for producing the daptides by MALDI-TOF MS by the method described above. The colonies were then scratched individually from the plate by sterile cotton swabs, spread on fresh MS medium with apramycin, and allowed to grow for another 5 d. The spores obtained from a single 150 × 20 mm plate were then transferred into 1 mL sterile water. An 100 μL aliquot of the spore solution was used to inoculate 50 mL bottromycin production medium (BPM; 10 g/L glucose, 15 g/L soluble starch, 5 g/L yeast extract, 10 g/L soy flour, 5 g/L NaCl, 3 g/L CaCO3) supplied with 50 μg/mL apramycin in a 250 mL flask, which was shaken at 250 rpm for 4 d at 30 °C. Each 50 mL seed culture was then transferred into 500 mL fresh BPM supplied with 50 μg/mL apramycin in a 2 L flask and shaken at 250 rpm for another 5 d at 30 °C.
The cells were then harvested by centrifugation and mixed with methanol ($\frac{1}{10}$ volume of the liquid culture) to homogeneity with continuous stirring under room temperature for at least 1 h. Cell debris was then removed by centrifugation, and the methanol extract was mixed with an equal volume of water and loaded onto an Agilent Bond Elut C18 Solid Phase Extraction (SPE) column (bed mass, 10 g; volume, 60 mL; particle size 120 μm), which was pre-equilibrated by 50 mL $5\%$ B (solvent $A = 0.1$% TFA in water; solvent $B = 0.1$% TFA in acetonitrile). The compounds were then eluted using a step gradient with increasing percentage of solvent B in 150 mL volumes: $5\%$, $20\%$, $30\%$, $40\%$, $50\%$, $60\%$, $70\%$, $80\%$, and $100\%$ B. The eluted compounds were monitored by MALDI-TOF MS. Fractions containing the compounds ($50\%$, $60\%$, and $70\%$ B) were lyophilized to dryness and powders were redissolved into methanol. Semi-reparative HPLC purification was performed using a 1290 Infinity II Preparative LC System equipped with a Phenomenex Luna C5 column (5 μm, 100 Å, 250 × 10 mm) equilibrated in $5\%$ B. Compounds were eluted by an increase to $100\%$ B over 20 min with a flow rate of 3 mL/min. Under these conditions, daptides 1, 2, and 3 were eluted at 16.7, 15.5, and 14.4 min, respectively. All fractions were analyzed by MALDI-TOF MS, lyophilized to dryness, and stored at -80 °C until further use. Typical yields were 0.2 – 0.3 mg per liter of BPM.
## Antimicrobial activity assay
Daptides were dissolved in DMSO to achieve a concentration of 10 μM. Agar plates were prepared by combining 20 mL of melted solid medium (cooled to 42 °C for 5 min) with 200 μL of stationary phase overnight cell culture. The seeded agar was poured into a sterile 100-mm round dish (VWR) and allowed to solidify at 25 °C for 10 min. Daptides were directly spotted on the solidified agar. Plates were incubated at various temperatures shown in Supplementary Table 4 for 16 h, and the antimicrobial activity was determined by the presence or absence of zones of growth inhibition.
## Hemolytic Assay
Fresh defibrinated whole bovine blood was obtained from Hemostat Laboratories. Whole blood was washed three times in PBS and diluted to a final concentration of 1:25 vol/vol in PBS. Prepared whole blood was then split into 50 μL aliquots in individual 1.7 mL Eppendorf Tubes. Next, stock solutions of concentrations at 1 mM and 200 μM were prepared for 1, 2, and 3 in DMSO, and 1 + 2, 2 + 3, 1 + 3 and 1 + 2 + 3 stock solutions at 1 mM and 200 μM were then prepared by mixing equal volumes of the single daptide stock solutions accordingly. Aliquots of 2.5 μL each stock solution were mixed with the blood, yielding final concentrations at 50 μM and 10 μM. An equal volume of DMSO and Triton X-100 were used as negative and positive controls. The mixtures were then incubated in Eppendorf Thermoxixer C with a heated lid for 18 h at 37 °C. After incubation, the samples were processed by centrifugation at 500 × g for 10 min, and the supernatants were measured for hemoglobin absorbance at 410 nm on a NanoDrop Spectrophotometer. Each measurement was performed in biological triplicate, and the percentage of hemolysis was calculated using Equation 1.1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\%{{{{{\rm{Hemolysis}}}}}}=\frac{{{{{{\rm{Abs}}}}}}\,{{{{{\rm{of}}}}}}\; {{{{{\rm{test}}}}}}\; {{{{{\rm{sample}}}}}}-{{{{{\rm{Abs}}}}}}\; {{{{{\rm{of}}}}}}\; {{{{{\rm{DMSO}}}}}} }{{{{{{\rm{Abs}}}}}}\; {{{{{\rm{of}}}}}}\; {{{{{\rm{Triton}}}}}}\; {{{{{\rm{X-}}}}}}100-{{{{{\rm{Abs}}}}}}\; {{{{{\rm{of}}}}}}\; {{{{{\rm{DMSO}}}}}}}.$$\end{document}%Hemolysis=Absoftestsample−AbsofDMSOAbsofTritonX-100−AbsofDMSO.
## Construction and characterization of the minimal mpa BGC in S. albus J1074
The complete or partial mpa operons were PCR amplified from the directly cloned mpa BGC as multiple DNA fragments. The primers were designed to generate an 80 bp overlapped region between adjacent DNA fragments (Supplementary Table 5). The E. coli-Streptomyces shuttle vector pSET152 was digested by XbaI and EcoRI. The S. cerevisiae helper fragment was amplified from the plasmid pRS416. All DNA fragments were gel-purified from $0.7\%$ agarose. For each construct, 200 ng vector was mixed with other DNA fragments at a 1:1 molar ratio, concentrated by DNA Clean & Concentrator-5 (Zymo Research, CA) and eluted into 6 μL of water. The DNA mixture was then assembled by DNA assembler36. Single colonies of S. cerevisiae YSG50 were inoculated in 2 mL YPAD medium (yeast extract 10 g/L, peptone 20 g/L, glucose 20 g/L, adenine sulfate 40 mg/L) and grown overnight at 30 °C and 250 rpm. A 0.5 mL aliquot of the overnight culture was transferred into 50 mL fresh YPAD medium and shaken at 250 rpm and 30 °C for 4–5 h until OD 600 reached 0.8–1.0. Cells were harvested by centrifugation at 3000 × g for 5 min at 4 °C. The cell pellet was washed by 50 ml ice-cold water, followed by another wash by cold 1 M sorbitol, and resuspended in 250 μL cold 1 M sorbitol. An aliquot of 50 μl of yeast cells was mixed with 4 μL DNA mixture and electroporated in a 0.2 cm cuvette at 1.5 kV. Afterwards, the cells were resuspended by 1 mL room temperature YPAD medium and recovered at 30 °C with shaking at 250 rpm for 1 h. Cells were then harvested by centrifugation, washed by 1 mL of 1 M sorbitol at room temperature for two times, spread on Sc-Ura plates, and incubated at 30 °C for 2–4 days until colonies appeared. Single colonies were grown in Sc-*Ura medium* for 1 d and plasmid DNA was purified using Zymoprep II Yeast plasmid Miniprep kit (Zymo Research, CA). Plasmids were then transformed into E. coli NEB10β, purified by Qiaprep Spin Miniprep Kit (Qiagen, Germany), and analyzed by restriction digestion. Correct plasmids were conjugated into S. albus J1074, and colony extracts were analyzed by MALDI-TOF MS as described above.
## Refactoring and gene omission of the mpa pathway
The mpaA1-3, mpaB, mpaC, mpaD, and mpaM genes were codon-optimized and synthesized and then subcloned onto helper plasmids for a plug-and-play refactoring strategy (Supplementary Table 5, Supplementary Table 6)37. Point mutations to mpaA1 were introduced by overlap extension PCR. The mpaA1 and its mutants obtained by PCR were then cloned into T7 His Helper-1 for constructing plasmids with other genes. The refactored pathway was built by the Golden Gate assembly37. The 4 bp adapters affected by gene omission were changed by PCR with primers that anneal to the T7 promoter and terminator region. The resulting PCR fragment was used as the insert in the Golden Gate assembly.
## Expression of the refactored mpa pathways and product purification
Unless otherwise specified, the following conditions were used for the co-expression of MpaA1 and variants thereof with other biosynthetic genes in E. coli. The plasmids containing refactored mpa pathways were co-transformed with the chaperone plasmid (pGro7) into E. coli BL21(DE3). Cells were grown on Luria–Bertani (LB) agar plates containing 50 μg/mL kanamycin and 20 μg/mL chloramphenicol at 37 °C overnight. Single colonies were picked to inoculate 5 mL LB supplied with the same amount of antibiotics and grown at 37 °C for another 16 h. This culture was used to inoculate 500 mL M9 medium supplied with 50 μg/mL kanamycin, 20 μg/mL chloramphenicol, and 0.5 mg/mL arabinose. The M9 culture was grown at 37 °C to an optical density of 600 nm (OD600) of 0.6. Isopropyl β-d-1-thiogalactopyranoside (IPTG) was then added to a final concentration of 0.5 mM. The cell culture was then cooled to 18 °C and grown for an additional 3 d.
Cells were then harvested by centrifugation and resuspended in lysis buffer (6 M guanidine hydrochloride, 20 mM NaH2PO4, 500 mM NaCl, 0.5 mM imidazole, pH 7.5) to a final volume of 20 mL and lysed by sonication. The cell lysate was then clarified by centrifugation at 23,700 × g for 30 min at 4 °C, and the supernatant was passed through a syringe filter (0.45 μm). The clarified lysates were loaded onto a 5 mL NiNTA HisTrap column (GE Healthcare). The column was then washed with 25 mL of wash buffer (4 M guanidine hydrochloride, 20 mM NaH2PO4, 500 mM NaCl, 30 mM imidazole, pH 7.5) and eluted with 15 mL of elution buffer (4 M guanidine hydrochloride, 20 mM Tris, 100 mM NaCl, 1 M imidazole, pH 7.5). The eluent was desalted by ZipTip or Bond Elut C18 SPE column and analyzed by MALDI-TOF MS.
## Derivatization of ketone intermediate
A refactored plasmid containing genes mpaA1, mpaB, and mpaC was heterologously expressed as described above. After desalting by SPE column, 1 mL (sample in aq. $70\%$ acetonitrile/$0.1\%$ TFA) was added each to two scintillation vials (reaction vs. control). The pH of the eluent was adjusted to pH 4 using 0.1 M NaOH and checked by pH paper. To one vial, O-benzylhydroxylamine was added to 10 mM and scintillation vials were left overnight (~16 h) at room temperature to ensure maximal oxime formation. Reaction products were mixed 1:1 with 50 mg/mL Super-DHB (Sigma-Aldrich) in aq. $60\%$ acetonitrile/$0.1\%$ formic acid and dried under ambient conditions on a polished steel MALDI target. Samples were analyzed using a Bruker UltrafleXtreme MALDI-TOF MS using manufacturer’s methods for reflector positive mode.
## AlphaFold-Multimer Structural Prediction of MpaA1-MpaB-MpaC Complex
AlphaFold-Multimer was used to predict the MpaB-MpaC-MpaA1 interactions, with each of the five trained model parameters40. The MSA generation, AlphaFold-Multimer predictions, and structure relaxation with Amber were run using the code of ColabFold, a publicly available Jupyter notebook54, on a Google Colab GPU cluster. The input included the query sequences of MpaB, MpaC, and MpaA1, with an MSA from MMseqs2 (UniRef+Environmental) not using any templates. The Pair mode “unpaired+paired” and the number of recycles, “6,” were selected in the advanced settings section. The structure of MpaC with ligand NADP was constructed using a homologous alcohol dehydrogenase (PDB ID: 6C76) as a template followed by energy minimization using software YASARA Structure version 17.8.1955–57. The protein-protein interaction was visualized and analyzed using PyMOL version 2.5.458 and Chimera version 1.1659.
## Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.
## Supplementary information
Supplementary Information Peer Review File Description of Additional Supplementary Files Supplementary Data 1 Supplementary Data 2 Reporting Summary The online version contains supplementary material available at 10.1038/s41467-023-37287-1.
## Source data
Source Data
## Peer review information
Nature Communications thanks Francisco (Paco) Barona-Gómez, Jan Claesen and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Peer reviewer reports are available.
## References
1. Arnison PG. **Ribosomally synthesized and post-translationally modified peptide natural products: overview and recommendations for a universal nomenclature**. *Nat. Prod. Rep.* (2013.0) **30** 108-160. DOI: 10.1039/C2NP20085F
2. Montalban-Lopez M. **New developments in RiPP discovery, enzymology and engineering**. *Nat. Prod. Rep.* (2021.0) **38** 130-239. DOI: 10.1039/D0NP00027B
3. King AM. **Selection for constrained peptides that bind to a single target protein**. *Nat. Commun.* (2021.0) **12** 1-12. DOI: 10.1038/s41467-021-26350-4
4. Yang X. **A lanthipeptide library used to identify a protein-protein interaction inhibitor**. *Nat. Chem. Biol.* (2018.0) **14** 375-380. DOI: 10.1038/s41589-018-0008-5
5. Schmitt S. **Analysis of modular bioengineered antimicrobial lanthipeptides at nanoliter scale**. *Nat. Chem. Biol.* (2019.0) **15** 437-443. DOI: 10.1038/s41589-019-0250-5
6. Goto Y, Suga H. **Engineering of RiPP pathways for the production of artificial peptides bearing various non-proteinogenic structures**. *Curr. Opin. Chem. Biol.* (2018.0) **46** 82-90. DOI: 10.1016/j.cbpa.2018.06.014
7. Wu CY, van der Donk WA. **Engineering of new-to-nature ribosomally synthesized and post-translationally modified peptide natural products**. *Curr. Opin. Biotech.* (2021.0) **69** 221-231. DOI: 10.1016/j.copbio.2020.12.022
8. Russell AH, Truman AW. **Genome mining strategies for ribosomally synthesised and post-translationally modified peptides**. *Comput Struct. Biotec* (2020.0) **18** 1838-1851. DOI: 10.1016/j.csbj.2020.06.032
9. Rutledge PJ, Challis GL. **Discovery of microbial natural products by activation of silent biosynthetic gene clusters**. *Nat. Rev. Microbiol* (2015.0) **13** 509-523. DOI: 10.1038/nrmicro3496
10. 10.Hemmerling, F. & Piel, J. Strategies to access biosynthetic novelty in bacterial genomes for drug discovery. Nat Rev Drug Discov. 1–20 (2022).
11. Ting CP. **Use of a scaffold peptide in the biosynthesis of amino acid-derived natural products**. *Science* (2019.0) **365** 280-284. DOI: 10.1126/science.aau6232
12. Daniels PN. **A biosynthetic pathway to aromatic amines that uses glycyl-tRNA as nitrogen donor**. *Nat. Chem.* (2022.0) **14** 71-77. DOI: 10.1038/s41557-021-00802-2
13. Hudson GA. **Bioinformatic mapping of radical S-adenosylmethionine-dependent ribosomally synthesized and post-translationally modified peptides identifies new C alpha, C beta, and C gamma-linked thioether-containing peptides**. *J. Am. Chem. Soc.* (2019.0) **141** 8228-8238. DOI: 10.1021/jacs.9b01519
14. Morinaka BI. **Natural noncanonical protein splicing yields products with diverse beta-amino acid residues**. *Science* (2018.0) **359** 779-782. DOI: 10.1126/science.aao0157
15. Caruso A, Martinie RJ, Bushin LB, Seyedsayamdost MR. **Macrocyclization via an arginine-tyrosine crosslink broadens the reaction scope of radical S-adenosylmethionine enzymes**. *J. Am. Chem. Soc.* (2019.0) **141** 16610-16614. DOI: 10.1021/jacs.9b09210
16. Clark KA, Bushin LB, Seyedsayamdost MR. **Aliphatic ether bond formation expands the scope of radical SAM enzymes in natural product biosynthesis**. *J. Am. Chem. Soc.* (2019.0) **141** 10610-10615. DOI: 10.1021/jacs.9b05151
17. Bushin LB, Clark KA, Pelczer I, Seyedsayamdost MR. **Charting an unexplored Streptococcal biosynthetic landscape reveals a unique peptide cyclization motif**. *J. Am. Chem. Soc.* (2018.0) **140** 17674-17684. DOI: 10.1021/jacs.8b10266
18. Rued BE. **Quorum sensing in Streptococcus mutans regulates production of tryglysin, a novel RaS-RiPP antimicrobial compound**. *mBio* (2021.0) **12** e02688. PMID: 33727351
19. Hug JJ. **Biosynthesis of cittilins, unusual ribosomally synthesized and post-translationally modified peptides from**. *Acs Chem. Biol.* (2020.0) **15** 2221-2231. DOI: 10.1021/acschembio.0c00430
20. Reisberg SH. **Total synthesis reveals atypical atropisomerism in a small-molecule natural product, tryptorubin A**. *Science* (2020.0) **367** 458-463. DOI: 10.1126/science.aay9981
21. Hug JJ. **Genome-guided discovery of the first Myxobacterial biarylitide myxarylin reveals distinct C-N biaryl crosslinking in RiPP biosynthesis**. *Molecules* (2021.0) **26** 7483. DOI: 10.3390/molecules26247483
22. Burkhart BJ, Hudson GA, Dunbar KL, Mitchell DA. **A prevalent peptide-binding domain guides ribosomal natural product biosynthesis**. *Nat. Chem. Biol.* (2015.0) **11** 564-570. DOI: 10.1038/nchembio.1856
23. Kloosterman AM, Shelton KE, van Wezel GP, Medema MH, Mitchell DA. **RRE-Finder: a genome-mining tool for class-independent RiPP discovery**. *Msystems* (2020.0) **5** e00267. DOI: 10.1128/mSystems.00267-20
24. 24.Shelton, K.E. & Mitchell, D.A. in Methods in Enzymology (Academic Press, 2022).
25. Tietz JI. **A new genome-mining tool redefines the lasso peptide biosynthetic landscape**. *Nat. Chem. Biol.* (2017.0) **13** 470-478. DOI: 10.1038/nchembio.2319
26. Zimmermann L. **A completely reimplemented MPI bioinformatics toolkit with a new HHpred server at its core**. *J. Mol. Biol.* (2018.0) **430** 2237-2243. DOI: 10.1016/j.jmb.2017.12.007
27. Mistry J. **Pfam: the protein families database in 2021**. *Nucleic Acids Res* (2021.0) **49** D412-D419. DOI: 10.1093/nar/gkaa913
28. Enghiad B. **Cas12a-assisted precise targeted cloning using in vivo Cre-lox recombination**. *Nat. Commun.* (2021.0) **12** 1-11. DOI: 10.1038/s41467-021-21275-4
29. Tong Y. **CRISPR-Cas9, CRISPRi and CRISPR-BEST-mediated genetic manipulation in streptomycetes**. *Nat. Protoc.* (2020.0) **15** 2470-2502. DOI: 10.1038/s41596-020-0339-z
30. Anguiano M, Nowak RJ, Lansbury PT. **Protofibrillar islet amyloid polypeptide permeabilizes synthetic vesicles by a pore-like mechanism that may be relevant to type II diabetes**. *Biochem.-Us* (2002.0) **41** 11338-11343. DOI: 10.1021/bi020314u
31. Bhushan R, Bruckner H. **Use of Marfey’s reagent and analogs for chiral amino acid analysis: assessment and applications to natural products and biological systems**. *J. Chromatogr. B* (2011.0) **879** 3148-3161. DOI: 10.1016/j.jchromb.2011.05.058
32. Kloosterman AM. **Expansion of RiPP biosynthetic space through integration of pan-genomics and machine learning uncovers a novel class of lanthipeptides**. *PLoS Biol.* (2020.0) **18** e3001026. DOI: 10.1371/journal.pbio.3001026
33. Walker MC. **Precursor peptide-targeted mining of more than one hundred thousand genomes expands the lanthipeptide natural product family**. *BMC Genomics* (2020.0) **21** 387. DOI: 10.1186/s12864-020-06785-7
34. Xu M. **Functional Genome Mining Reveals a Class V Lanthipeptide Containing a d-Amino Acid Introduced by an F(420) H(2) -Dependent Reductase**. *Angew. Chem. Int Ed. Engl.* (2020.0) **59** 18029-18035. DOI: 10.1002/anie.202008035
35. Huan Y, Kong Q, Mou H, Yi H. **Antimicrobial peptides: classification, design, application and research progress in multiple fields**. *Front Microbiol* (2020.0) **11** 582779. DOI: 10.3389/fmicb.2020.582779
36. Shao ZY, Zhao H, Zhao HM. **DNA assembler, an in vivo genetic method for rapid construction of biochemical pathways**. *Nucleic Acids Res* (2009.0) **37** e16. DOI: 10.1093/nar/gkn991
37. Ren HQ, Hu PF, Zhao HM. **A plug-and-play pathway refactoring workflow for natural product research in Escherichia coli and Saccharomyces cerevisiae**. *Biotechnol. Bioeng.* (2017.0) **114** 1847-1854. DOI: 10.1002/bit.26309
38. Nishihara K, Kanemori M, Kitagawa M, Yanagi H, Yura T. **Chaperone coexpression plasmids: differential and synergistic roles of DnaK-DnaJ-GrpE and GroEL-GroES in assisting folding of an allergen of Japanese cedar pollen, Cryj2 in Escherichia coli**. *Appl Environ. Micro.* (1998.0) **64** 1694-1699. DOI: 10.1128/AEM.64.5.1694-1699.1998
39. Maxson T. **Targeting reactive carbonyls for identifying natural products and their biosynthetic origins**. *J. Am. Chem. Soc.* (2016.0) **138** 15157-15166. DOI: 10.1021/jacs.6b06848
40. 40.Evans, R. et al. Protein complex prediction with AlphaFold-Multimer. BioRxiv (2021).
41. Miethke M. **Towards the sustainable discovery and development of new antibiotics**. *Nat. Rev. Chem.* (2021.0) **5** 726-749. DOI: 10.1038/s41570-021-00313-1
42. Kim PI. **Characterization and structure identification of an antimicrobial peptide, hominicin, produced by Staphylococcus hominis MBBL 2-9**. *Biochem Bioph Res Co.* (2010.0) **399** 133-138. DOI: 10.1016/j.bbrc.2010.07.024
43. 43.Dieterich, C.L. et al. Aquimarins, peptide antibiotics with amino-modified C-Termini from a sponge-derived Aquimarina sp. bacterium. Angew Chem Int Ed Engl, e202115802 (2021).
44. Prigge ST, Mains RE, Eipper BA, Amzel LM. **New insights into copper monooxygenases and peptide amidation: structure, mechanism and function**. *Cell Mol. Life Sci.* (2000.0) **57** 1236-1259. DOI: 10.1007/PL00000763
45. Strandberg E. **Influence of C-terminal amidation on the antimicrobial and hemolytic activities of cationic alpha-helical peptides**. *Pure Appl Chem.* (2007.0) **79** 717-728. DOI: 10.1351/pac200779040717
46. Ahearn IM, Haigis K, Bar-Sagi D, Philips MR. **Regulating the regulator: post-translational modification of RAS**. *Nat. Rev. Mol. Cell Bio* (2012.0) **13** 39-51. DOI: 10.1038/nrm3255
47. Santos-Aberturas J. **Uncovering the unexplored diversity of thioamidated ribosomal peptides in Actinobacteria using the RiPPER genome mining tool**. *Nucleic Acids Res* (2019.0) **47** 4624-4637. DOI: 10.1093/nar/gkz192
48. Buchfink B, Xie C, Huson DH. **Fast and sensitive protein alignment using DIAMOND**. *Nat. Methods* (2015.0) **12** 59-60. DOI: 10.1038/nmeth.3176
49. Katoh K, Misawa K, Kuma K, Miyata T. **MAFFT: a novel method for rapid multiple sequence alignment based on fast Fourier transform**. *Nucleic Acids Res* (2002.0) **30** 3059-3066. DOI: 10.1093/nar/gkf436
50. Eddy SR. **Accelerated profile HMM searches**. *Plos Comput Biol.* (2011.0) **7** e1002195. DOI: 10.1371/journal.pcbi.1002195
51. Wheeler TJ, Clements J, Finn RD. **Skylign: a tool for creating informative, interactive logos representing sequence alignments and profile hidden Markov models**. *Bmc Bioinforma.* (2014.0) **15** 1-9. DOI: 10.1186/1471-2105-15-7
52. Potter SC. **HMMER web server: 2018 update**. *Nucleic Acids Res* (2018.0) **46** W200-W204. DOI: 10.1093/nar/gky448
53. 53.Kieser, T., Bibb, M.J., Buttner, M.J., Chater, K.F. & Hopwood, D.A. Practical streptomyces genetics (John Innes Foundation, 2000).
54. 54.Mirdita, M. et al. ColabFold-making protein folding accessible to all. Research Square (2021).
55. Van Durme J. **A graphical interface for the FoldX forcefield**. *Bioinformatics* (2011.0) **27** 1711-1712. DOI: 10.1093/bioinformatics/btr254
56. Schymkowitz J. **The FoldX web server: an online force field**. *Nucleic Acids Res* (2005.0) **33** W382-W388. DOI: 10.1093/nar/gki387
57. Krieger E. **Improving physical realism, stereochemistry, and side-chain accuracy in homology modeling: four approaches that performed well in CASP8**. *Proteins* (2009.0) **77** 114-122. DOI: 10.1002/prot.22570
58. 58.Pymol: the PyMOL molecular graphics system, version 1.2r3pre, Schrödinger, LLC.
59. Pettersen EF. **UCSF Chimera-a visualization system for exploratory research and analysis**. *J. Comput Chem.* (2004.0) **25** 1605-1612. DOI: 10.1002/jcc.20084
|
---
title: Paternal methotrexate exposure affects sperm small RNA content and causes craniofacial
defects in the offspring
authors:
- Nagif Alata Jimenez
- Mauricio Castellano
- Emilio M. Santillan
- Konstantinos Boulias
- Agustín Boan
- Luisa F. Arias Padilla
- Juan I. Fernandino
- Eric L. Greer
- Juan P. Tosar
- Luisa Cochella
- Pablo H. Strobl-Mazzulla
journal: Nature Communications
year: 2023
pmcid: PMC10036556
doi: 10.1038/s41467-023-37427-7
license: CC BY 4.0
---
# Paternal methotrexate exposure affects sperm small RNA content and causes craniofacial defects in the offspring
## Abstract
Folate is an essential vitamin for vertebrate embryo development. Methotrexate (MTX) is a folate antagonist that is widely prescribed for autoimmune diseases, blood and solid organ malignancies, and dermatologic diseases. Although it is highly contraindicated for pregnant women, because it is associated with an increased risk of multiple birth defects, the effect of paternal MTX exposure on their offspring has been largely unexplored. Here, we found MTX treatment of adult medaka male fish (Oryzias latipes) causes cranial cartilage defects in their offspring. Small non-coding RNA (sncRNAs) sequencing in the sperm of MTX treated males identify differential expression of a subset of tRNAs, with higher abundance for specific 5′ tRNA halves. Sperm RNA methylation analysis on MTX treated males shows that m5C is the most abundant and differential modification found in RNAs ranging in size from 50 to 90 nucleotides, predominantly tRNAs, and that it correlates with greater testicular Dnmt2 methyltransferase expression. Injection of sperm small RNA fractions from MTX-treated males into normal fertilized eggs generated cranial cartilage defects in the offspring. Overall, our data suggest that paternal MTX exposure alters sperm sncRNAs expression and modifications that may contribute to developmental defects in their offspring.
Anti-folate drugs, such as methotrexate, have been largely prohibited for pregnant women because of the teratogenic effect on their descendant. Here, the authors report a intergenerational mechanism by why paternal methotrexate exposure causes craniofacial defects on their offspring.
## Introduction
Folate is a water-soluble vitamin obtained from the diet that is essential for vertebrates. It is incorporated as an essential cofactor for the synthesis of nucleotides and the generation of S-adenosylmethionine (SAM) which serves as a universal donor of methyl groups for DNA, RNA and proteins implicated in gene regulation during early development1–4. Maternal folate deficiency leads to severe neural tube defects and craniofacial anomalies of descendants5–7. Importantly, the prevalence of these defects is highly reduced by folic acid supplementation prior and during pregnancy8,9. Despite global efforts to supplement the maternal diets with folate, there is still a worldwide prevalence of these congenital defects10–12. Methotrexate (MTX) is a recognized teratogenic folic acid antagonist that has been linked to an elevated incidence of congenital anomalies in children born from exposed women. Intrauterine MTX exposure has been linked to craniofacial and limb defects, as well as developmental delays13,14. In addition to oral clefts, folic acid antagonists may raise the risk of cardiovascular, neural tube, and urinary tract abnormalities15. As a result, current recommendations urge that mothers stop using MTX at least three months before conception16. Prior research has also identified a variety of issues concerning MTX use and a probable genotoxic effect on sperm, which might result in chronic disease or congenital anomalies17. However, medical care recommendations for males taking MTX while trying to conceive are less clear.
For decades, the sperm genome has been considered transcriptionally quiescent and solely contributing to the restoration of the ploidy of the zygote. However, more recently, a set of functional RNAs have been characterized in mature spermatozoa that are delivered to the oocyte upon fertilization, contributing to early embryo development and thus, influencing the phenotypic outcome of the offspring18–24. Intriguingly, paternal folate concentrations can affect the sperm epigenome25,26. Whereas the direct impact of these changes is expected to be minimal given the protamine exchange and resetting of DNA methylation during spermatogenesis and early development27–29, we wondered whether paternal folate levels may also affect the RNA composition of mature sperm.
Small non-coding RNAs (sncRNAs) are a particularly attractive potential carrier of non-genetic information in the spermatozoa. In particular, tRNA-derived small RNAs (tsRNAs) and microRNAs (miRNAs) are the most abundant in mature spermatozoa30,31; and have been identified as molecular carriers of paternal experiences, including high fat diet22,24,32, low protein diet33, stress21,34, and odoriferous sensitivity to chemicals23. Small RNA biogenesis, stability and functionality are highly dependent on their post-transcriptional modification status, primarily methylation35–37. Furthermore, transmission of paternally acquired metabolic disorders is dependent on the presence of post-transcriptional modifications in sperm sncRNAs19,24.
Here, we explored the intriguing possibility that paternal folate deficiency impacts the offspring’s development, and that it may do so through changes in sncRNA abundance and methylation levels. We injected medaka male fish with the folate inhibitor methotrexate (MTX) and characterized their offspring’s developmental defects. Next, we analyzed and compared the abundance and modifications of sncRNAs present in the sperm of MTX-treated males to test the idea that they work as mediators of congenital developmental defects.
## Paternal folate deficiency induced cranial cartilage malformations in their offspring
To investigate the impact of paternal folate deficiency on the development of their progeny, we administered medaka male fish with an intraperitoneal injection of methotrexate (MTX), a well-known folate inhibitor38–40, at 10 mg of MTX per Kg of body weight (10MTX) and 50 mg/Kg MTX (50MTX)(Fig. 1a). After 7 days, we fertilized wild type oocytes with sperm extracted from treated and untreated males. None of our treatments had a significant impact on sperm fecundity, hatching time, or overall embryo hatching (Fig. S1).Fig. 1Paternal MTX injection affects offspring’s cranial cartilages lengths.a scheme of experimental design. b–j Violin plots represent the measurement of different cranial cartilages lengths, angles and areas on control (Ctrl), 10MTX and 50MTX. Statistical analyses were performed using ANOVA one-way followed by multiple comparison Tukey’s test. Numbers of analyzed embryos: Control ($$n = 14$$), 10MTX ($$n = 14$$), 50MTX ($$n = 26$$). * $$P \leq 0.0164$$, ***$$P \leq 0.0008$$, ****$$P \leq 0.0001.$$
Several studies have shown that folate is an important vitamin for neural and neural crest development in several vertebrate species including humans5,41,42. Moreover, maternal folate deficiency during pregnancy leads to abnormal development of neural crest derivatives such as cranial cartilages43–46. Taking this into account, we first evaluated the effect of paternal folate deficiency on the development of the offspring’s cranial cartilages by performing alcian blue staining at 3 days post hatching-stage (3dph). We measured the length of three dorsal cartilages (anterior orbital, epiphyseal bar and posterior orbital), four ventral cartilages (Meckel, ceratohyal, basibranchial and palatoquadrate), and the Meckel’s area and ceratohyal angle (Fig. 1b–j). From the dorsal cartilages, we found a significant reduction in the length of the anterior orbital (also known as taenia marginalis anterior) in the 50MTX group (115.02 µm ± 9.03, one-way ANOVA followed by multiple comparison Tukey’s test $$p \leq 0.0164$$) when compared to the 10MTX (130.02 µm ± 10.22) and control (125.98 µm ± 15.77). On the ventral side, the basibranchial and Meckel’s cartilages were not affected. However, the ceratohyal was reduced to almost half the length, at both 10MTX (192.99 µm ± 7.55, $p \leq 0.0001$) and 50MTX (185.42 µm ± 8.71, $p \leq 0.0001$) compared with control (363.64 µm ± 18.96). Interestingly, when we looked at the morphology of those cranial cartilages, we found that two of them, the anterior orbital and basihyal, presented an abnormal shape (Fig. 2). In particular, the anterior orbital has an abnormal serpentine shape, compared with the normal curved shape (Fig. 2a, b). This phenotype was significantly prevalent ($$p \leq 0.0059$$) at the 50MTX group (Fig. 2c). However, one of the most drastically affected cartilages was the basihyal, whose phenotypes presented a curved trowel shape (mild) or hook shape (strong) (Fig. 2d). Quantitation of those phenotypes evidences a significant increase in the severity of them at both 10MTX ($$p \leq 0.0329$$) and 50MTX ($$p \leq 0.0006$$) compared with Control group (Fig. 2e). Overall, these findings support the notion that paternal MTX exposure affects the development of the offspring’s cranial cartilage, indicating that sperm may convey some information involved in the observed phenotypic inheritance. Fig. 2Paternal MTX injection produced offspring’s cranial cartilages malformations.a, b Dorsal view of normal and affected anterior orbital (ao) showing a meandering shape mostly observed on the offspring from MTX treated males. c Quantification of the percentage of embryos presenting affected or normal anterior orbital cartilages. Numbers in the graph represent the analyzed embryos. d Lateral view of larvae presenting normal (trowel shape), mild (bended shape), and strong (hook shape) deformities of the basihyal cartilage. e Quantification of the percentage of embryos presenting normal (non-affected), mild, or strong basihyal cartilage abnormalities observed on the offspring from MTX-treated males. Numbers in the graph represent the analyzed embryos. Statistical analyses were performed using a contingency table followed by two-sided Fisher’s exact test and each treatment were compared with the control independently. * $$P \leq 0.0329$$, **$$P \leq 0.0059$$, ***$$P \leq 0.0006.$$
## SncRNAs abundance is altered in the sperm of MTX-treated males
Epigenetic information, including histone modifications and DNA methylation, particularly from the paternal lineage, is largely reprogramed during germline and early embryo development. However, increasing evidence indicates that sncRNAs are a carrier of epigenetic information across generations and may act as mediators of paternally inherited traits18–23,47. To assess if paternal folate deficiency affects the small RNA content, we sequenced size selected (~18–30 nt long) RNAs from sperm of 10MTX and control males. Based on the analysis of three biological replicated for each group, we were able to identify different populations of sncRNAs including: ribosomal RNAs (rRNAs), small nuclear RNAs (snRNAs), small nucleolar RNAs (snoRNAs), micro RNAs (miRNAs), transfer RNA fragments (tRNAs), and other miscellaneous RNAs (miscRNA) (Fig. 3a). The comparative analysis showed that rRNAs and tRNA fragments were the most abundant on both control and 10MTX treatment. Some of the most abundant rRNAs mapped to 28S, 18S, and 5.8S rRNA. Reads mapping 5S rRNA and both 16S and 12S mitochondrial rRNA (mtrRNA) were also detected, albeit in a much lower proportion (Fig. 3b). Despite being the most abundant sncRNAs, we found no statistically significant difference in either cytoplasmic or mitochondrial rRNA expression between control and MTX-treated fish (Fig. 3c, d). On the other side, some of the most abundant tRNA fragments, aspartic acid (having the anticodon AspGUC), glutamic acid (GluCUC and GluUUC), lysine (LysCUU) and glycine (GlyGCC) (Fig. 3b), became further significantly enriched upon MTX treatment (Fig. 3c, d). Together, these results demonstrate that paternal MTX exposure affects the relative abundance of specific sncRNAs in the sperm, with tRNA fragments being the most affected population. Fig. 3Paternal MTX injection alters sperm sncRNAs.a Histogram representing the comparison from sperm sncRNA on control (Ctrl1-3) and 10MTX (MTX-1-3) treated males. b Histogram displaying biotypes of tRNAs and rRNAs from sperm of MTX treated males. See also Supplementary Data 1 for a and b. c Volcano plot of depicting the fold changes in sperm sncRNAs identified as being differentially expressed within control versus MTX-treated males. Analysis of differential expression was performed by using DESeq2 which use the two-tailed Wald test. d MA plot displaying normalized counts (base mean) for different sncRNAs. Dotted lines depict thresholds values for significantly up and down-regulated (±≥1 log2 fold change and -log10 P value ≥ 1.3). See also Supplementary Data 2 for c and d.
## 5′ halves of particular tRNAs are preferentially affected by methotrexate treatment
tRNAs can be cleaved into 5′ and 3′ halves, known as tsRNAs, in response to stress or other external factors19,24,35,48. Of particular interest in the sperm RNA content is the large abundance of those tsRNA fragments and their potential regulatory roles in early embryo development19,20,24,30,31. Interestingly, we observed that 5′ tsRNA fragments, product of the cleavage of three of the most abundant tRNAs (5′ tsRNA of AspGUC, LysCUU, and GluCUC), were significantly increased in MTX sperm, without a concomitant increase in their respective 3′ tsRNA fragments (Fig. 4a, b, e). Quantification of the proportion of 5′ halves relative to their corresponding 3′ halves showed a significant increase in the percentage of 5’ halves for tRNA AspGUC and GlyGCC in the MTX-treated samples (Fig. 4c, e). It is interesting to note that for some tRNAs (i.e., tRNA GluCUC, GluUUC, and GlyUCC) we mostly retrieve reads for their 5′ halves, but their corresponding 3′ halves are almost undetected for both control and MTX. On the other hand, the 5′ halves of many (most) other tRNAs did not show differences compared with their 3′ halves (tRNA ProUGG, ArgUCU), or a major proportion of their 3′ halves (tRNA SerGCU) (Fig. S2). These results suggest changes in processing or stability of specific tRNA fragments as a consequence of the MTX treatment. Fig. 4Paternal MTX injection induced the expression and cleavage of specific tRNAs.a Volcano plot of depicting the fold changes in sperm 5′ and 3′ tRNA halves as being differentially expressed within control versus MTX treated males. Analysis of differential expression was performed by using DESeq2 which use the two-tailed Wald test. b MA plot displaying normalized counts (base mean) for different 5′ and 3′tRNA halves. Dotted lines depict thresholds values for significantly up and down-regulated (±≥1 log2 fold change and -log10Pvalue ≥ 1.3). See also Supplementary Data 3 for a and b. c Histogram displaying percentage of 5′ halves relative to their corresponding 3′ halves from different tRNAs affected by MTX treatment. Asterisk indicated significant differences analyzed by multiple unpaired t-student’ test followed by a correction for multiple comparison (Holm-Sidak method, with alpha = 0.05)). d Histogram showing the length variation of mapped tRNA reads on control and MTX-treated males. Dotted lines represent the median length. Histogram showing the read coverage (e) and size (f) distribution for the most abundant and having a significant increase in the 5’tsRNA (tRNA-GluCUC, tRNA-AspGUC, and tRNA-GlyGCC) between control and MTX. Data on c, d and f represent three biologically independent replicates ($$n = 3$$) composed by a pool of 9 males’ sperm. Values are means ± SD. See also Supplementary Data 4 for c–f.
Production of 5′ tsRNAs fragments in the 15–22 nt range occurs in multiple tissues and cell lines49, whereas longer 5′ tsRNAs (31–40 nt long) are preferentially generated in response to different stresses50,51. We thus compared the length distribution of tRNAs-derived fragments in both conditions and observed a shift towards longer fragments in 10MTX relative to control samples (median length of 24 nt for control and 29 nt for MTX) (Fig. 4d). A similar situation is found for the tRNAs GluCUC, AspGUC and LysCUU where their 5’tsRNAs are significantly increased and their coverage lengths are greater in MTX-treated samples than in controls (Fig. 4f). There is a chance that bias size selection occurred when separating the small RNAs from the gel, resulting in these differences. However, we found that the read coverage and size distribution for the most abundant rRNA-derived fragments are slightly larger after MTX treatment (Fig. S3a), but this is not as dramatic as observed for tsRNAs (median length of 22 nt to 24 nt for rRNAs, vs. 24 nt to 29 nt for tsRNAs; Fig. S3b). These findings suggest that paternal MTX exposure alters the abundance and cleavage site of specific 5′tsRNAs in the sperm.
## m5C modifications are increased by methotrexate
Post-transcriptional modifications of tRNAs, including methylation, are important for their specific cleavage, stability, and functionality, as well as for the transmission of paternal experiences to the offspring35–37,52. We thus evaluated the methylation status of two populations of RNAs isolated from polyacrylamide: a 20–50 nt population (mostly enriched for miRNAs and tsRNAs), and a 50–90 nt population (mostly enriched for mature tRNAs). Within the 20–50 nt RNA population we did not observe significant differences in the abundance of any of the analyzed methylation events between MTX and control groups (Fig. 5a). Conversely, within the 50–90 nt population, we found that MTX treatment led to a significant increase in the relative abundance of several modifications (Fig. 5b). From the two most abundant modifications analyzed (m1A and m5C) only m5C was significantly increased ($$p \leq 0.0155$$) in MTX treated samples. From the other less abundant modifications only m2G, m7G and m2′2G presented a significant increase in MTX samples (p value = 0.0027; 0.0063; and 0.0090, respectively). Interestingly, the most abundant modification observed to be differentially detected in MTX samples has been described to be located at the 3′ ends of tRNAs53,54 (Fig. 5c).Fig. 5Paternal MTX injection alters smallRNA modifications in sperm tRNA fraction and the testicular expression of RNA-methyltransferases. Histogram comparing sperm RNA methylations on control and MTX analyzed by UHPLC-MS-MS in 50–90 nt (a) and 20–50 nt (b) fractions. Data represent two biologically independent replicates ($$n = 2$$) composed by RNA isolated from 9 males’ sperm. Statistical analysis was performed by two-tailed unpaired t-student’ test. Values are mean ± S.D. (c) Schematic representation of modified nucleotides in the tRNA at secondary and tertiary structure. d RT-qPCR for methyltransferases of m1A (TRMT6) and m5C (DNMT2 and NSUM2) on testis from control and MTX treated males, gene expression was normalized using Rpl7 and Ef1 as housekeeping genes. Statistical analysis was performed by using the two-tailed unpaired t-student’ test. Each dot represents a biologically independent sample composed by RNA isolated from individual male testis. Valures are means ± SEM.
Given the observed differences for certain RNA modifications, we examined the expression of specific RNA methyltransferases on the testis of control and 10MTX treated males (Fig. 5d). In agreement with our results, there was no significant change in the expression of Trmt6 which catalyzes m1A methylation. Conversely, the expression of the enzymes that catalyzed m5C was only significantly upregulated for Dnmt2 ($$p \leq 0.01$$), but not for Nsun2 ($$p \leq 0.46$$). These results suggest that an increase of RNA methyltransferase expression leads to changes in the methylation status of sperm tRNAs upon MTX treatment.
## Zygotic RNA injection derived from MTX-treated sperm partially recapitulates the craniofacial phenotype
To uncover the potential role of altered small RNAs on the sperm as the causes of offspring craniofacial defects, we isolated both 20–50nt and 50–90nt RNA fractions from MTX and control (DMSO) treated males and injected them independently or combined (20–90 nt) into fertilized wild-type eggs. Initially, we did not observe differences in the percentage of hatching and embryo survival between zygotic RNA injections derived from both control and MTX-treated sperm (Fig. S4).
We have focused our analysis on the ceratohyal and basihyal cartilages because they were most affected in our initial experiments. Importantly, we found that injection of 20–50 nt, 50–90 nt or the combination of both (20–90 nt) at the two doses of MTX-treated sperm showed a significant reduction ($p \leq 0.0001$) on the ceratohyal lengths compared to control (Fig. 6a). On the other side, when we analyze the basihyal phenotypes we were unable to evidence any hook shape malformation, but instead we evidenced embryos having curved trowel shape (affected) with the tip of the cartilage upward and downwards (Fig. 6b). Importantly, injection of 20–50 nt fraction from both 10MTX and 50MTX, but not the 50–90 nt, significantly increase the number of larvae having affected basyhial shape (Fig. 6b). Similarly, injection of 20–90 nt RNA fractions have a similar effect than the 20–50 nt. All these together suggest that RNAs from exposed males have the ability to alter the development of specific cranial cartilages on the offspring. Fig. 6Zygotic RNA injection derived from MTX-treated sperm partially recapitulates the craniofacial phenotype.a Violin plots represent the measurement of ceratohyal lengths on wild-type fertilized eggs injected with sperm-RNA fractions (20–50nt, 50–90nt, or both together) obtained from control, 10MTX and 50MTX treated males. Statistical analyses were performed using ANOVA one-way followed by multiple comparison Tukey’s test. b Lateral view of larvae presenting non-affected (trowel shape) or affected (bended upward or downward) basihyal cartilage shape. Quantification of the percentage of embryos presenting affected or non-affected basihyal cartilage abnormalities from zygotes injected with sperm-RNA fractions (20–50nt, 50–90nt, or both together) obtained from control, 10MTX and 50MTX treated males. Numbers in the graph represent the analyzed embryos. Statistical analyses were performed using a contingency table followed by two-sided Fisher’s exact test. c Proposed model summarizing the results.
## Discussion
MTX binds to and inhibits dihydrofolate reductase activity, preventing folic acid from performing its biological tasks. For more than 30 years, this drug has been used to treat immunological illnesses (including rheumatoid arthritis), blood and solid organ cancers, dermatologic diseases, and for pregnancy termination55,56. Despite the drug’s contraindication for pregnant women due to the risk of miscarriage and birth abnormalities, the paternal influence of MTX on their offspring was largely unknown. In addition to this, the vast majority of studies in fish models such as medaka and zebrafish has been performed during embryological stages57–60, while few have evaluated the effect on adults61 and the consequences on their offspring.
In agreement with other studies in mice62,63, we found that paternal MTX treatments had no effect on the fertility or survival of their progeny during the early embryonic stages. Lifetime exposure to folic acid-deficient diets, on the other hand, lead to lower sperm counts, negative consequences in progeny, and epigenetic changes62,63. However, this may be due to folate deficiency during embryonic and post-embryonic development, which could compromise early germ cell formation and adult spermatogenesis. Moreover, major epigenetic reprogramming occurs at these periods, and multiple imprinting areas may be altered as a result of the prolonged folate shortage.
For many years, paternal contribution to offspring’s health was thought to be restricted to the haploid genome of spermatozoa, whereas mother health and nutrition were linked to offspring’s wellness. However, multiple recent studies have revealed that spermatozoa carry a variety of RNAs18,64–66 capable to transmit paternal experiences19,22–24,32. In this regard, our work illustrated the critical significance of MTX therapy and its impact on sperm small non-coding RNA content as a possible mechanism underlying the observed craniofacial abnormalities or possibly other unexplored effects of this treatment. We discovered that tsRNAs and miRNAs are the most common small non-coding RNA in medaka sperm, which is consistent with past findings in mammals19,30,31,66,67. Furthermore, we revealed that tsRNAs halves changed significantly owing MTX treatment, which is in agreement with previous studies showing that tsRNAs are a dynamic population that responds to a variety of environmental stressors19,24,68. Particularly, we observed a higher abundance of certain 5’tsRNAs, where 5’tsRNA-AspGUC was the most abundant. This result is in concordance with several studies where external factors also modulated the abundance of 5’tsRNA-AspGUC20,24,31, thus highlighting the idea that certain tRNAs may be preferentially cleaved and their 5′ halves have a longer half-life compared to their respective 3’ halves.
tsRNAs can be generated through multistep cleavages, through the formation of various intermediates. Moreover, there is growing evidence that regulatory factors, such as RNA modifications and specific RNases, have a role in their specific cleavage and stability66. Interestingly, we found that 5’tsRNAs from AspGUC and GlyGCC are consistently longer (~35 nt) on the sperm of MTX-treated males. This is in agreement with the discovery that small 15–22 nucleotide long fragments are normally formed in multiple tissues and cell lines49, whereas longer 31–40 nucleotide tRNA halves are preferentially cleaved in response to different stresses50,51.
The tsRNA functions are very speculative, but have been associated to translation, ribosome biogenesis, retrotransposition, cell-cell communication, and epigenetic inheritance, as well as how tsRNA dysregulation are related to a variety of human disorders (summarized in recent reviews69,70). Importantly, both tsRNAs and their precursor tRNAs are heavily modified, which contributes to multiple aspects of their function, biogenesis, stability, amino acid charging, and translational accuracy71,72. Our initial hypothesis was that MTX treatment may reduce the tRNA methylation thus inducing their cleavage. This is based on previous studies where the addition of m5C, which is controlled by DNMT2 and NSUN2, increase tRNA stability in flies and mice, but its deletion makes them more likely to be cleaved into tsRNAs under stress conditions19,36,37. However, and contrarily to our predictions, we observed that tRNAs-enriched samples (~50–70 nt) derived from MTX-treated sperm showed significantly greater levels of methylation in m5C, m2G, m7G and m2’2G. Increased levels of m5C and m2G have been observed in the 30–40 nucleotide fraction of sperm RNAs (predominantly tsRNAs) in mice fed with high-fat diets compared with those from males fed with normal diets24. However, it is important to mention that in our MTX-treatment we fail to observe differences in the methylation levels from the tsRNAs/miRNAs fraction (~20–50 nt). We speculate that because the most abundant tRNA modifications found in our study (5mC) are stated to be positioned at the 3’ end of tRNAs (positions 38C, 48C, 49C, 50C)73, then the cleaved 3’tsRNAs halves, which accumulate the bulk of these methylations, may be preferentially degraded.
The high levels of 5mC in our 50–90nt fraction from MTX-treated males correlated with the higher expression of Dnmt2 (also known as Trdmt1), but not Nsun2. Dnmt2 is structurally close to other DNA methyltransferases but rather methylates only one tRNA, specifically at the cytosine 38 in the anticodon loop of aspartic acid (tRNA-Asp)74. The role of Dnmt2 in paternal non-genetic inheritance have been demonstrated in mice19,64, and recently associated with the Intergenerational effect of immune priming in insects, suggesting an evolutionary conservation of its functionality75. Interestingly, here we have found that 5’tsRNA-AspGUC was the second most abundant tRNA in MTX-treated males and presenting a significant increase respect to their 3’tsRNA-AspGUC half. In contradiction to our finding, Schaefe et al. 37. demonstrated that m5C modification mediated by DNMT2 improves tRNA stability, where tRNA-*Asp is* protected from angiogenin cleavage during the heat shock response in Drosophila. In mammals, it is well known that angiogenin activity, RNase that cleaves tRNAs, is also inhibited by the presence of 5mC19,36. However, it is important to mention that since the endonuclease targeting the anti-codon loop of Drosophila tRNAs has not been identified yet, the authors analyzed the cleavage of tRNA-Asp induced by the addition of recombinant human angiogenin into Drosophila S2 cells37, which may not reflect the truly physiological condition. Moreover, it has been shown that the presence of angiogenin is not mandatory for the generation of tsRNAs, and other RNases (Dicer, RNase T2, L) can also cleave tRNAs69,76,77. In that sense, fish does not have angiogenin, but instead orthologues genes with the capacity to cleave tRNAs have been found78–80 suggesting that the generation of tRNA fragments is an evolutive response against environmental stressors. In addition, it is important to mention that the activity78, structure79–81, and targeted dinucleotides for cleavage are different in between fish and mammals RNases81. These facts suggest that the overall generation of tRNA fragments is an ancient response where RNases have maintained their main role and have evolved as the organisms did it. On the other hand, the presence of 5mC, and/or other modifications, might affect their activity in a different way as it was speculated by Barraud and Tisné82. These authors stated that tRNA modifications are critical features of the cellular stress responses, and described the existence of a streaky crosstalk among them regulating the tRNAs stability82. As a result, modifications may act as a “barcode” to regulate the specific tRNA cleavage and stability resulting in the accumulation of specific tsRNAs in the sperm, which could affect the phenotype of their offspring. Finally, we showed that the zygotic injection of small RNAs isolated from MTX-treated sperm can partially reproduce the basihyal malformation and ceratohyal´s length reduction, which were also the most affected cartilages observed in the offspring of MTX-treated males. It is important to mention that the ceratohyal and basihyal cartilages form the ventral region of the hyoid arch required for the stabilization of the jaw83 which has evolved in mammals as a structure required for milk suckling84. Facial characteristics and growth deficiencies have been extensively linked to both folate deficiency and fetal alcohol syndrome. This phenotypic linkage is, in part, because chronic alcohol abuse affects the folate levels by reducing their initial hydrolysis and subsequent uptake into the cells85. Furthermore, fetal alcohol syndrome has also been connected with a weak sucking ability and other feeding difficulties in humans86, which may presume a hyoid malformation.
In summary, our data suggests that paternal MTX-exposure influenced sperm tRNA methylation, as a result of alterations in the expression of certain RNA methyltransferases. These epitranscriptomic changes may cause the selective tRNA cleavage and the maintenance of certain 5’ tRNA halves. These changes in the sperm RNA content and modifications might affect transcriptional cascades in the fertilized oocyte, with possible implications in cranial cartilage formation (see hypothetical model in Fig. 6c). The understanding of how tRNA modifications and their derived fragments impact on the transcriptional cascades occurring during early embryo will provide valuable insights into several diseases and it is expected that this will be a main focus of research in this field in the near future.
## Medaka Husbandry
All experiments were performed with medaka fish (Oryzias latipes) (strain hi-medaka, ID:MT835) supplied by the National BioResource Project (NBRP) Medaka (http://www.shingen.nig.ac.jp/medaka/). Fish were maintained and fed following standard protocols for medaka87. Fish were handled on the Care and Management of Laboratory Animals (http://www.ufaw.org.uk) and internal approved regulations (SICUAE-University of San Martín $\frac{33}{2022}$). Adult fishes were divided and acclimatized in 4L fish tank during 3 weeks under a constant photoperiod (14L:10D) and controlled temperature (26 ± 0.5 °C), prior to experimental procedures.
## Experimental design
Adults medaka fish were divided into 3 groups composed by 3 independent replicates having 9 males and 2 females per tank (all of them having a body weight of ~200 mg). After the acclimatizing period, each male was intraperitoneal injected with control solution (PBS/$1\%$DMSO), 10 mg of MTX per kg of body weight (10 mg/kg MTX) or 50 mg/kg MTX (A6770-SigmaAldrich, diluted in PBS/$1\%$DMSO). Briefly, males were anesthetized with $1\%$ benzocaine solution (Parafarm), gently dried with a paper towel, and placed in a dampened sponge ventral side up, with their anal fin and cloaca exposed. Immediately, using a 10ul syringe (Hamilton), fish were injected using a binocular stereoscope (Nikon SMZ745) and then returned to their tanks for 7 days until sperm collection for in vitro fertilization and small RNA extraction.
## In vitro fertilization
Sperm collection was carried by anesthetizing the fish and placed in a dampened sponge ventral side up following published protocols for medaka87. A micro-forceps was used to gently strip the fish and the released semen was collected by using a micropipette (~0.5 µl/fish) and pooled for the posterior in vitro fertilization and small RNA extraction. For the in vitro fertilization, 0.2 µl from obtained sperm were used to fertilize a pool of 24–28 eggs collected from mature untreated females. Fertilized eggs were immediately transferred and incubated in 60 mm petri dishes with embryo media (17 mM NaCl, 0.4 mM KCl, 0.27 mM CaCl2.2H2O, and 0.66 mM MgSO4; pH:7) until 3 days’ post hatching (dph). Incubation was monitored and the percentage of fertilization and survival until hatching was evaluated.
## Alcian blue staining
Cartilages from embryos were analyzed at 3 dph by using alcian blue staining. Larvae were fixed in $4\%$ paraformaldehyde overnight at 4 °C and washed three times with PBSw (PBS-$0.1\%$ tween20). After that, embryos were incubated in a bleaching solution (0.5X SSC, $5\%$ formamide, $10\%$ hydrogen peroxide) and exposed to light during 2 h. Larvae were washed several times with PBSw and immediately incubated in alcian Blue solution ($0.1\%$p/v alcian blue, $0.37\%$v/v HCl, $70\%$v/v EtOH) for 1 h on a nutator. Then, larvae were washed five times with $01\%$v/v HCl-$70\%$v/v EtOH for 30 min on a nutator; the last wash was left overnight at room temperature. Next, larvae were washed six times with $50\%$v/v glycerol-$0.5\%$v/v KOH for 30 min on a nutator and the last wash was left overnight. Finally, larvae were washed four times with the same solution and left in $90\%$ glycerol-$10\%$ ETOH for imaging processing and phenotype analysis. Larvae were photographed at ventral, dorsal and lateral view by using a trinocular stereoscope (SteREO Discovery v20. Zeiss) and analyzed using the ImageJ software88.
## Small RNA extraction and library preparation
Small RNAs were isolated from sperm following manufacturer’s instructions (illustra RNAspin Mini RNA isolation kit-GE Healthcare). The 3’ adapters (see Supplementary Table 2 for full list of adapter oligos utilized for library preparation) were ligated using SRBC barcode adapters for each sample, additionally 18-mer and 30-mer markers were ligated and used as control for the ligation process and markers for the product correct size. The 3’-ligated small RNAs were size selected using $15\%$ denaturing urea polyacrylamide gels at a constant power of 40–50 W for ~30 min and stained by using SYBR Gold $0.05\%$V/V in TBE 0.5X and the 3′ ligated RNAs ranging from 18 −30 mer were cut out. RNAs were purified using Zymo PAGE elution kit (ZRTM small RNA PAGE recovery kit) according to manufacturer’s instructions, the elute 3′-ligated small RNAs were elute in 5’ linker mix containing 5′ adaptor. The 3′-ligated RNAs + 5′ adapter were denaturated for 5 min at 70 °C, cooled on ice immediately, ligated with T4 RNA ligase (NEB) and incubated at 16 degrees overnight. Ligated small RNAs were purified by using MBS beads, briefly: MBS buffer, MBS bead slurry (beads + buffer), mixed by vortexing, added isopropanol and incubated at room temperature. Beads were separated on magnet and the supernatant was removed, after several washes with ethanol the RNA was eluted with ultrapure water and transferred into PCR strip. For reverse transcription, small-RNAseq RT primer to each sample were used and a negative control without reverse transcriptase was included, Superscript II reverse transcriptase was used to obtain the synthesis of the first strand. To amplify cDNA libraries, KAPA HiFi Real Time Library Amplification Kit (Roche) was used; PCR were performed using TruSeq Universal Adapter primer (Solexa_PCR_fwd) and TruSeq Index reverse primers (Solexa_IDX_rev), this latter includes barcodes assigned to each different sample. Briefly: master mix was added and TruSeq Index reverse primer were added to PCR strips containing cDNAs; then KAPA HiFi HS RM and Truseq Universal Adapter primer were added to the mix. The cycling program was: Denaturation at 98 °C for 45 s; 20 cycles of 98 °C for 15 s, 65 °C for 30 s, 72 °C for 30 s, 72 °C for 10 s; and a final extension at 72 °C for 1 min. The amplified cDNA was purified by using $3\%$ Low-Range Ultra Agarose gel (Bio-Rad) according to the manufacturer’s instructions at constant 80–100 V using GeneRuler 50 bp DNA Ladder (ThermoFischer Scientific) as molecular marker. Gel was visualized on a long wave UV transilluminator and DNA band between 150–200 bp were excised using a clean scalpel blade and put into a clean 15 ml Falcon tube; the DNA was purified using the Zymoclean Gel DNA recovery kit (Zymo Research) according to manufacturer’s instructions.
## Bioinformatics analysis
Adapters from reads were removed using CUTADAPT, the output were reads ≥15 bp, reads whose adapters were not identified were discarded. The output of 15 bp were used to analyze differential expression of sncRNAs (miRNAs, tRNAs, snRNAs, snoRNAs, and rRNAs) and differential expression of tRNAs, 5′ tRNA halves and 3′ tRNA halves by different strategies. First, differential expression of sncRNAs was analyzed on reads where the random nucleotides on 5′ (4 bp) and 3′ (6pb) were cut using FASTQ Trimmer. The obtained reads having <19 bp were discarded using Filter Fastq and the remaining reads were aligned against the medaka genome (Assembly ASM00223467v1) with RNA STAR (allowing multimapping reads, 1 mismatch, and not allowing introns). Expression of miRNAs, tRNAs, snRNAs, snoRNAs was analyzed using FeatureCounts (allowing multimapping reads to be counted, and assigning 1/n fractions to multimapping reads) with Ensembl annotation (Release v102). To analyze cytoplasmic and mitochondrial rRNA expression, reads were mapped to a custom fasta file containing 28s rRNA (RNA central: URS000215D18B_8090), 18s rRNA (refseq: XR_002874070.1), 5.8s rRNA (RNA central: URS0000671FD1_8090), 5s rRNA (refseq: XR_002875036.1), 16s mtRNA (RNA central: URS00003A7D46_8090) and 12s mtRNA (RNA central: URS000033338A_8090) sequences. Differential expression of all sncRNAs was calculated using DESEQ2. Second, to analyze differential expression of 5′ and 3′ tRNA halves, an additional 3 base pairs were removed with FASTQ Trimmer from the 3’end of all reads. Reads having less than 15 bp were discarded using Filter Fastq. The output was aligned to the reference genome and analyzed as mentioned before using custom GTF files with genomic coordinates for either 5′ or 3′ tRNA halves. To determine the sequence length of mapped tRNA and rRNA reads, BAM files were filtered (using GTFs files containing genomic coordinate for full length tRNAs o rRNAs), reads were extracted, converted to fasta and their length computed with in-house scripts. tRNA and rRNA read coverage was calculated with BamCoverage (bin size 1, no smoothing) with RPM values representing reads per million mapped to functional sncRNA categories (miRNAs, tRNAs, snRNAs, snoRNAs and rRNAs).
## UHPLC-MS-MS
The analysis of modified ribonucleotides from spermatic RNAs were performed by UHPLC-MS-MS. For that purpose, ~1.5 µg of total RNAs were isolated from two independent pools of stripped sperm from ~9 control and 10MTX-treated males and run in denaturant polyacrylamide gel ($15\%$, 7 M Urea). The gel was then stained with ethidium bromide and RNAs that have a ranged size from 20–50 nt and 50–90 nt were cut and recovered using the ZR small-RNATM PAGE Recovery Kit (Zymo Research) by following the manufacturer’s instruction. Approximately 100 ng of RNA was obtained on each fraction and utilized for UHPLC-MS-MS analysis. Then, 100 ng purified RNA samples were digested to nucleosides for 2 hr at 37 °C using the Nucleoside Digestion mix (NEB, M069S). Quantifications were performed as in89, briefly: digested RNA samples were diluted to 100 μl with ddH20 and filtered through 0.22 μm Millex Syringe Filters. 5 μl of the filtered solution was injected for LC-MS/MS analysis using the Agilent 1290 UHPLC-MS/MS system with a Hypersil Gold C18 reversed-phase column (2.1 × 150 mm, 3 μm). Mobile phase A consisted of water with $0.1\%$ (v/v) formic acid and mobile phase B consisted of acetonitrile with $0.1\%$ (v/v) formic acid. Mass spectrometry detection was performed using an Agilent 6470 triple quadrupole mass spectrometer in positive electrospray ionization mode and data were quantified in dynamic multiple reaction monitoring (dMRM) mode, by monitoring the mass transitions 268♋♊136 for Adenosine (A), 282♋♊150 for N6-methyladenosine (m1A), 244♋♊112 for Cytidine (C), 258♋♊126 for C5-methylcytidine (m5C), 284♋♊152 for Guanosine (G), 298♋♊166 for N7’-methyladenosine (m7G) and N2-methylguanosine (m2G), 312♋♊180 for N2,N2-dimethylguanosine (m2’2G), 282♋♊136 for 2’-O-methyladenosine (Am), 258♋♊112 for 2’-O-methylcytidine (Cm) and 298♋♊152 for 2’-O-methylguanosine (Gm). To quantify the concentrations of the various methylation modifications we used pure nucleosides of A, C, G, m1A, m5C, m7G, m2G m2’2G, Am, Cm, and Gm to generate calibration standard curves through serial dilution.
## RNA quantification by RT-PCR
Dissected testis from adult males (Control and 10MTX groups) were used for gene expression analysis. For this purpose, a complete testis from 8 males were individually grinded in 300 µl of TRIzol Reagent (Life Technologies) and total RNAs (~800 ng/testis) were isolated and retrotranscribed90. Expression of target genes were measured by qPCR using Fast Start Universal SYBR green Supermix (Roche Diagnostics, USA) on Thermal Cycler StepOne Plus (Applied Biosystem, USA), using ribosomal protein L7 (rpl7) and elongation factor 1 alpha (ef1α) as reference genes with the geometric mean calculation as described by Padilla et al. 90. Real-time PCR primers are listed in Supplementary Table 1. Each sample was run in duplicate and a PCR reaction without the addition of template, was used as negative control. The amplification protocol consisted of an initial cycle of 1 min at 95 °C, followed by 40 cycles: 15 s at 95 °C and 30 s at 60 °C. After the amplification, a melt curve was performed by 1 cycle: 15 s at 95 °C, 60 s at 60 °C, and 15 s at 95 °C enabling confirmation of amplification of single products. Gene expression levels were calculated by the 2−ΔΔCt comparative threshold cycle (Ct) method (where ΔΔCt = ΔCt sample - ΔCt reference). The efficiency of amplification ranged 95–$105\%$ for all genes studied. The expression level in each group was normalized to the control and was presented as a fold of change91.
## Zygotic RNA injection
RNAs were isolated from pools from stripped sperm from ~9 control, 10MTX, and 50MTX-treated males and run in denaturant polyacrylamide gel ($15\%$, 7 M Urea). RNAs that have a ranged size from 20–50 nt and 50–90 nt were cut and recovered using the ZR small-RNATM PAGE Recovery Kit (Zymo Research) by following the manufacturer’s instruction. Fertilized eggs at the stage of one cell were injected with 45.1 pg of purified RNAs from 20–50 nt, 50–90 nt or a combination of both (20–90 nt) by using the microinjector Nanoject II (Drummond Scientific Company). After this, the injected embryos were transferred into petri dish containing embryo media and let them grown up until 3 days post hatching. The embryos were monitored every day and alcian blue staining was performed for larvae as previously mentioned. Stained larvae were analyzed as previously mentioned.
## Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.
## Supplementary information
Supplementary Information Peer Review File Reporting Summary The online version contains supplementary material available at 10.1038/s41467-023-37427-7.
## Source data
Source Data
## Peer review information
Nature Communications thanks the anonymous reviewer(s) for their contribution to the peer review of this work. Peer reviewer reports are available.
## References
1. Sánchez-Vásquez E, Alata Jimenez N, Vázquez NA, Strobl-Mazzulla PH. **Emerging role of dynamic RNA modifications during animal development**. *Mech. Dev.* (2018.0) **154** 24-32. DOI: 10.1016/j.mod.2018.04.002
2. Greenberg MVC, Bourc’his D. **The diverse roles of DNA methylation in mammalian development and disease**. *Nat. Rev. Mol. Cell Biol.* (2019.0) **20** 590-607. DOI: 10.1038/s41580-019-0159-6
3. 3.Bernstein C., Bernstein H. Demethylation in early embryonic development and memory. In: DNA Methylation Mechanism. 10.5772/intechopen.90306 (2020).
4. O’Neill C. **The epigenetics of embryo development**. *Anim. Front.* (2015.0) **5** 42-49. DOI: 10.2527/af.2015-0007
5. Li J, Shi Y, Sun J, Zhang Y, Mao B. **Xenopus reduced folate carrier regulates neural crest development epigenetically**. *PLoS One* (2011.0) **6** e27198. DOI: 10.1371/journal.pone.0027198
6. Shaw GM, Schaffer D, Velie EM, Morland K, Harris JA. **Periconceptional vitamin use, dietary folate, and the occurrence of neural tube defects**. *Epidemiology* (1995.0) **6** 219-226. DOI: 10.1097/00001648-199505000-00005
7. Tolarova M. **Periconceptional supplementation with vitamins and folic acid to prevent recurrence of cleft lip**. *Lancet* (1982.0) **320** 217. DOI: 10.1016/S0140-6736(82)91063-7
8. McDonald SD, Ferguson S, Tam L, Lougheed J, Walker MC. **The prevention of congenital anomalies with periconceptional folic acid supplementation**. *J. Obstet. Gynaecol. Can.* (2003.0) **25** 115-121. DOI: 10.1016/S1701-2163(16)30207-9
9. Vergel RG, Sanchez LR, Heredero BL, Rodriguez PL, Martinez AJ. **Primary prevention of neural tube defects with folic acid supplementation: Cuban experience**. *Prenat. Diagn.* (1990.0) **10** 149-152. DOI: 10.1002/pd.1970100303
10. Blencowe H, Kancherla V, Moorthie S, Darlison MW, Modell B. **Estimates of global and regional prevalence of neural tube defects for 2015: a systematic analysis**. *Ann. N. Y. Acad. Sci.* (2018.0) **1414** 31-46. DOI: 10.1111/nyas.13548
11. Salari N. **Global prevalence of cleft palate, cleft lip and cleft palate and lip: A comprehensive systematic review and meta-analysis**. *J. Stomatol. Oral. Maxillofac. Surg.* (2022.0) **123** 110-120. DOI: 10.1016/j.jormas.2021.05.008
12. Impellizzeri A, Giannantoni I, Polimeni A, Barbato E, Galluccio G. **Epidemiological characteristic of Orofacial clefts and its associated congenital anomalies: Retrospective study**. *BMC Oral. Health* (2019.0) **19** 290. DOI: 10.1186/s12903-019-0980-5
13. Mahadevan U, McConnell RA, Chambers CD. **Drug safety and risk of adverse outcomes for pregnant patients with inflammatory bowel disease**. *Gastroenterology* (2017.0) **152** 451-462.e2. DOI: 10.1053/j.gastro.2016.10.013
14. Hyoun SC, Običan SG, Scialli AR. **Teratogen update: Methotrexate**. *Birth Defects Res A Clin. Mol. Teratol.* (2012.0) **94** 187-207. DOI: 10.1002/bdra.23003
15. Hernández-Díaz S, Werler MM, Walker AM, Mitchell AA. **Folic acid antagonists during pregnancy and the risk of birth defects**. *N. Engl. J. Med.* (2000.0) **343** 1608-1614. DOI: 10.1056/NEJM200011303432204
16. Nguyen GC. **The Toronto consensus statements for the management of inflammatory bowel disease in pregnancy**. *Gastroenterology* (2016.0) **150** 734-757.e1. DOI: 10.1053/j.gastro.2015.12.003
17. Beghin D, Cournot MP, Vauzelle C, Elefant E. **Paternal exposure to methotrexate and pregnancy outcomes**. *J. Rheumatol.* (2011.0) **38** 628-632. DOI: 10.3899/jrheum.100600
18. Sharma U. **Small RNAs are trafficked from the epididymis to developing mammalian sperm**. *Dev. Cell* (2018.0) **46** 481-494.e6. DOI: 10.1016/j.devcel.2018.06.023
19. Zhang Y. **Dnmt2 mediates intergenerational transmission of paternally acquired metabolic disorders through sperm small non-coding RNAs**. *Nat. Cell Biol.* (2018.0) **20** 535-540. DOI: 10.1038/s41556-018-0087-2
20. Chen X. **Early cleavage of preimplantation embryos is regulated by tRNAGln-TTG-derived small RNAs present in mature spermatozoa**. *J. Biol. Chem.* (2020.0) **295** 10885-10900. DOI: 10.1074/jbc.RA120.013003
21. Rodgers AB, Morgan CP, Leu NA, Bale TL. **Transgenerational epigenetic programming via sperm microRNA recapitulates effects of paternal stress**. *Proc. Natl Acad. Sci. USA.* (2015.0) **112** 13699-13704. DOI: 10.1073/pnas.1508347112
22. Fullston T. **Paternal obesity initiates metabolic disturbances in two generations of mice with incomplete penetrance to the F2 generation and alters the transcriptional profile of testis and sperm microRNA content**. *FASEB J.* (2013.0) **27** 4226-4243. DOI: 10.1096/fj.12-224048
23. Gapp K. **Implication of sperm RNAs in transgenerational inheritance of the effects of early trauma in mice**. *Nat. Neurosci.* (2014.0) **17** 667-669. DOI: 10.1038/nn.3695
24. 24.Chen, Q. et al. Sperm tsRNAs contribute to intergenerational inheritance of an acquired metabolic disorder. Science (1979).10.1126/science.aad7977 (2016).
25. Aarabi M. **High-dose folic acid supplementation alters the human sperm methylome and is influenced by the MTHFR C677T polymorphism**. *Hum. Mol. Genet.* (2015.0) **24** 6301-6313. DOI: 10.1093/hmg/ddv338
26. Boxmeer JC. **Low folate in seminal plasma is associated with increased sperm DNA damage**. *Fertil. Steril.* (2009.0) **92** 548-556. DOI: 10.1016/j.fertnstert.2008.06.010
27. Wang Y. **Reprogramming of meiotic chromatin architecture during spermatogenesis**. *Mol. Cell* (2019.0) **73** 547-561.e6. DOI: 10.1016/j.molcel.2018.11.019
28. Santos F, Hendrich B, Reik W, Dean W. **Dynamic reprogramming of DNA methylation in the early mouse embryo**. *Dev. Biol.* (2002.0) **241** 172-182. DOI: 10.1006/dbio.2001.0501
29. 29.Feng, S., Jacobsen, S. E. & Reik, W. Epigenetic reprogramming in plant and animal development. Science (1979).10.1126/science.1190614 (2010).
30. Chen Q, Yan W, Duan E. **Epigenetic inheritance of acquired traits through sperm RNAs and sperm RNA modifications**. *Nat. Rev. Genet.* (2016.0) **17** 733-743. DOI: 10.1038/nrg.2016.106
31. 31.Sharma U., et al. Biogenesis and function of tRNA fragments during sperm maturation and fertilization in mammals. Science (1979). 351 10.1126/science.aad6780 (2016)
32. 32.Grandjean, V. et al. RNA-mediated paternal heredity of diet-induced obesity and metabolic disorders. Sci. Rep.10.1038/srep18193 (2015).
33. Carone BR. **Paternally induced transgenerational environmental reprogramming of metabolic gene expression in mammals**. *Cell.* (2010.0) **143** 1084-1096. DOI: 10.1016/j.cell.2010.12.008
34. Wu L. **Paternal psychological stress reprograms hepatic gluconeogenesis in offspring**. *Cell Metab.* (2016.0) **23** 735-743. DOI: 10.1016/j.cmet.2016.01.014
35. 35.Blanco, S. et al. Aberrant methylation of t RNA s links cellular stress to neuro-developmental disorders. EMBO J.10.15252/embj.201489282 (2014).
36. Tuorto F. **RNA cytosine methylation by Dnmt2 and NSun2 promotes tRNA stability and protein synthesis**. *Nat. Struct. Mol. Biol.* (2012.0) **19** 900-905. DOI: 10.1038/nsmb.2357
37. Schaefer M. **RNA methylation by Dnmt2 protects transfer RNAs against stress-induced cleavage**. *Genes Dev.* (2010.0) **24** 1590-1595. DOI: 10.1101/gad.586710
38. Kao TT. **Methotrexate-induced decrease in embryonic 5-Methyl-Tetrahydrofolate is irreversible with Leucovorin Supplementation**. *Zebrafish* (2013.0) **10** 326-337. DOI: 10.1089/zeb.2013.0876
39. Cronstein BN, Aune TM. **Methotrexate and its mechanisms of action in inflammatory arthritis**. *Nat. Rev. Rheumatol.* (2020.0) **16** 145-154. DOI: 10.1038/s41584-020-0373-9
40. Dervieux T, Greenstein N, Kremer J. **Pharmacogenomic and metabolic biomarkers in the folate pathway and their association with methotrexate effects during dosage escalation in rheumatoid arthritis**. *Arthritis Rheum.* (2006.0) **54** 3095-3103. DOI: 10.1002/art.22129
41. Alata Jimenez N, Torres Pérez SA, Sánchez-Vásquez E, Fernandino JI, Strobl-Mazzulla PH. **Folate deficiency prevents neural crest fate by disturbing the epigenetic Sox2 repression on the dorsal neural tube**. *Dev. Biol.* (2018.0) **444** S193-S201. DOI: 10.1016/j.ydbio.2018.08.001
42. Antony AC, Hansen DK. **Hypothesis: Folate-responsive neural tube defects and neurocristopathies**. *Teratology* (2000.0) **62** 42-50. DOI: 10.1002/1096-9926(200007)62:1<42::AID-TERA9>3.0.CO;2-U
43. 43.van Rooij IALM, et al. Periconceptional folate intake by supplement and food reduces the risk of nonsyndromic cleft lip with or without cleft palate. Prev. Med. 39 10.1016/j.ypmed.2004.02.036 (2004).
44. Wilcox AJ. **Folic acid supplements and risk of facial clefts: National population-based case-control study**. *Br. Med J.* (2007.0) **334** 464. DOI: 10.1136/bmj.39079.618287.0B
45. Burgoon JM, Selhub J, Nadeau M, Sadler TW. **Investigation of the effects of folate deficiency on embryonic development through the establishment of a folate deficient mouse model**. *Teratology* (2002.0) **65** 219-227. DOI: 10.1002/tera.10040
46. 46.Prevention of neural tube defects: Results of the Medical Research Council Vitamin Study. Lancet338, 131–137 (1991).
47. Ord J, Heath PR, Fazeli A, Watt PJ. **Paternal effects in a wild-type zebrafish implicate a role of sperm-derived small RNAs**. *Mol. Ecol.* (2020.0) **29** 2722-2735. DOI: 10.1111/mec.15505
48. Rashad S. **The stress-specific impact of ALKBH1 on tRNA cleavage and tiRNA generation**. *RNA Biol.* (2020.0) **17** 1092-1103. DOI: 10.1080/15476286.2020.1779492
49. Kumar P, Anaya J, Mudunuri SB, Dutta A. **Meta-analysis of tRNA derived RNA fragments reveals that they are evolutionarily conserved and associate with AGO proteins to recognize specific RNA targets**. *BMC Med.* (2014.0) **12** 78. PMID: 24886626
50. Thompson DM, Lu C, Green PJ, Parker R. **tRNA cleavage is a conserved response to oxidative stress in eukaryotes**. *RNA* (2008.0) **14** 2095-2103. DOI: 10.1261/rna.1232808
51. Yamasaki S, Ivanov P, Hu GF, Anderson P. **Angiogenin cleaves tRNA and promotes stress-induced translational repression**. *J. Cell Biol.* (2009.0) **185** 35-42. DOI: 10.1083/jcb.200811106
52. Mishima E. **Conformational change in transfer RNA is an early indicator of acute cellular damage**. *J. Am. Soc. Nephrol.* (2014.0) **25** 2316-2326. DOI: 10.1681/ASN.2013091001
53. Huber SM, Leonardi A, Dedon PC, Begley TJ. **The versatile roles of the tRNA epitranscriptome during cellular responses to toxic exposures and environmental stress**. *Toxics* (2019.0) **7** 17. DOI: 10.3390/toxics7010017
54. 54.Hori H. Methylated nucleosides in tRNA and tRNA methyltransferases. Front Genet. 5. 10.3389/fgene.2014.00144(2014).
55. Smith SW, Nelson LS. **Case files of the New York City Poison Control Center: antidotal strategies for the management of methotrexate toxicity**. *J. Med. Toxicol.* (2008.0) **4** 132-140. DOI: 10.1007/BF03160968
56. LoVecchio F, Katz K, Watts D, Wood I. **Four-year experience with methotrexate exposures**. *J. Med Toxicol.* (2008.0) **4** 149-150. DOI: 10.1007/BF03161192
57. Jacabacci J, Guzzardi M, D’Aversa TG. **Exogenous folic acid is not able to inhibit the effects of methotrexate on the folate pathway in zebrafish embryos**. *Bios* (2015.0) **86** 125-133. DOI: 10.1893/BIOS-D-14-00027.1
58. Liu S, Narumi R, Ikeda N, Morita O, Tasaki J. **Chemical-induced craniofacial anomalies caused by disruption of neural crest cell development in a zebrafish model**. *Dev. Dyn.* (2020.0) **249** 794-815. DOI: 10.1002/dvdy.179
59. Sun S. **Effects of methotrexate on the developments of heart and vessel in zebrafish**. *Acta Biochim Biophys. Sin.* (2009.0) **41** 86-96. DOI: 10.1093/abbs/gmn010
60. Lee MS. **Disruption of the folate pathway in zebrafish causes developmental defects**. *BMC Dev. Biol.* (2012.0) **12** 12. DOI: 10.1186/1471-213X-12-12
61. Hirako A, Takeoka Y, Furukawa S, Sugiyama A. **Effects of methotrexate exposure on medaka (Oryzias latipes) testes**. *J. Toxicol. Pathol.* (2017.0) **30** 283-289. DOI: 10.1293/tox.2017-0029
62. Lambrot R. **Low paternal dietary folate alters the mouse sperm epigenome and is associated with negative pregnancy outcomes**. *Nat. Commun.* (2013.0) **4** 4. DOI: 10.1038/ncomms3889
63. Ly L. **Intergenerational impact of paternal lifetime exposures to both folic acid deficiency and supplementation on reproductive outcomes and imprinted gene methylation**. *Mol. Hum. Reprod.* (2017.0) **23** 461-477. DOI: 10.1093/molehr/gax029
64. Zhang Y, Shi J, Rassoulzadegan M, Tuorto F, Chen Q. **Sperm RNA code programmes the metabolic health of offspring**. *Nat. Rev. Endocrinol.* (2019.0) **15** 489-498. DOI: 10.1038/s41574-019-0226-2
65. Johnson GD, Mackie P, Jodar M, Moskovtsev S, Krawetz SA. **Chromatin and extracellular vesicle associated sperm RNAs**. *Nucleic Acids Res.* (2015.0) **43** 6847-6859. DOI: 10.1093/nar/gkv591
66. Peng H. **A novel class of tRNA-derived small RNAs extremely enriched in mature mouse sperm**. *Cell Res.* (2012.0) **22** 1609-1612. DOI: 10.1038/cr.2012.141
67. Kawano M, Kawaji H, Grandjean V, Kiani J, Rassoulzadegan M. **Novel small noncoding RNAs in Mouse Spermatozoa, Zygotes and Early Embryos**. *PLoS One* (2012.0) **7** e44542. DOI: 10.1371/journal.pone.0044542
68. Fu H. **Stress induces tRNA cleavage by angiogenin in mammalian cells**. *FEBS Lett.* (2009.0) **583** 437-442. DOI: 10.1016/j.febslet.2008.12.043
69. Su Z, Wilson B, Kumar P, Dutta A. **Noncanonical roles of tRNAs: TRNA fragments and beyond**. *Annu. Rev. Genet.* (2020.0) **54** 54-69. DOI: 10.1146/annurev-genet-022620-101840
70. 70.Schimmel P. RNA Processing and Modifications: The emerging complexity of the tRNA world: Mammalian tRNAs beyond protein synthesis. Nat. Rev. Mol. Cell Biol. 19 10.1038/nrm.2017.77 (2018).
71. Pan T. **Modifications and functional genomics of human transfer RNA**. *Cell Res.* (2018.0) **28** 395-404. DOI: 10.1038/s41422-018-0013-y
72. Suzuki T. **The expanding world of tRNA modifications and their disease relevance**. *Nat. Rev. Mol. Cell Biol.* (2021.0) **22** 375-392. DOI: 10.1038/s41580-021-00342-0
73. Pereira M. **Impact of tRNA modifications and tRNA-modifying enzymes on proteostasis and human disease**. *Int J. Mol. Sci.* (2018.0) **19** 3738. DOI: 10.3390/ijms19123738
74. 74.Goll M. G., et al. Methylation of tRNAAsp by the DNA methyltransferase homolog Dnmt2. Science (1979). 311 10.1126/science.1120976 (2006).
75. 75.Schulz N. K. E., et al. Paternal knockdown of tRNA(cytosine-5-)-methyltransferase (Dnmt2) increases offspring susceptibility to infection in red flour beetles. Insect Mol. Biol. 31 (2022).
76. Chen Q, Zhang X, Shi J, Yan M, Zhou T. **Origins and evolving functionalities of tRNA-derived small RNAs**. *Trends Biochem Sci.* (2021.0) **46** 790-804. DOI: 10.1016/j.tibs.2021.05.001
77. 77.Akiyama Y., et al. Multiple ribonuclease A family members cleave transfer RNAs in response to stress. bioRxiv. (2019).
78. Cho S, Zhang J. **Zebrafish ribonucleases are bactericidal: Implications for the origin of the vertebrate RNase a superfamily**. *Mol. Biol. Evol.* (2007.0) **24** 1259-1268. DOI: 10.1093/molbev/msm047
79. Pizzo E. **Ribonucleases and angiogenins from fish**. *J. Biol. Chem.* (2006.0) **281** 27454-27460. DOI: 10.1074/jbc.M605505200
80. Kazakou K, Holloway DE, Prior SH, Subramanian V, Acharya KR. **Ribonuclease a homologues of the Zebrafish: Polymorphism, crystal structures of two representatives and their evolutionary implications**. *J. Mol. Biol.* (2008.0) **380** 206-222. DOI: 10.1016/j.jmb.2008.04.070
81. Prats-Ejarque G, Lu L, Salazar VA, Moussaoui M, Boix E. **Evolutionary trends in RNA base selectivity within the RNase A superfamily**. *Front. Pharm.* (2019.0) **10** 10. DOI: 10.3389/fphar.2019.01170
82. Barraud P, Tisné C. **To be or not to be modified: Miscellaneous aspects influencing nucleotide modifications in tRNAs**. *IUBMB Life* (2019.0) **71** 1126-1140. PMID: 30932315
83. Fabik J, Psutkova V, Machon O. **The mandibular and hyoid arches—from molecular patterning to shaping bone and cartilage**. *Int J. Mol. Sci.* (2021.0) **22** 7529. DOI: 10.3390/ijms22147529
84. 84.Zhou C. F., Bhullar B. A. S., Neander A. I., Martin T., Luo Z. X. New Jurassic mammaliaform sheds light on early evolution of mammal-like hyoid bones. Science (1979). 36510.1126/science.aau9345 (2019).
85. Halsted CH. **Folate deficiency disturbs hepatic methionine metabolism and promotes liver injury in the ethanol-fed micropig**. *Proc. Natl Acad. Sci. USA* (2002.0) **99** 10072-10077. DOI: 10.1073/pnas.112336399
86. 86.Streissguth A. P., Landesman-Dwyer S., Martin J. C., Smith D. W. Teratogenic effects of alcohol in humans and laboratory animals. Science (1979). 209 10.1126/science.6992275 (1980).
87. 87.Kinoshita M., Murata K., Naruse K., Tanaka M. Medaka: Biology, Management, and Experimental Protocols. 10.1002/9780813818849 (2012).
88. Schneider CA, Rasband WS, Eliceiri KW. **NIH Image to ImageJ: 25 years of image analysis**. *Nat. Methods* (2012.0) **9** 671-675. DOI: 10.1038/nmeth.2089
89. 89.Liberman N., et al. Intergenerational hormesis is regulated by heritable 18S rRNA methylation. bioRxiv. (2021).
90. Padilla LFA. **Cystic proliferation of germline stem cells is necessary to reproductive success and normal mating behavior in Medaka**. *Elife* (2021.0) **10** 10
91. 91.Schmittgen T. D., Livak K. J. Analysis of relative gene expression data using real-time quantitative PCR and the 2(-Delta Delta C(T)) Method. Methods. 25 (4–8) (2001).
|
---
title: Comparative analysis of transferrin and IgG N-glycosylation in two human populations
authors:
- Irena Trbojević-Akmačić
- Frano Vučković
- Tea Pribić
- Marija Vilaj
- Urh Černigoj
- Jana Vidič
- Jelena Šimunović
- Agnieszka Kępka
- Ivana Kolčić
- Lucija Klarić
- Mislav Novokmet
- Maja Pučić-Baković
- Erdmann Rapp
- Aleš Štrancar
- Ozren Polašek
- James F. Wilson
- Gordan Lauc
journal: Communications Biology
year: 2023
pmcid: PMC10036557
doi: 10.1038/s42003-023-04685-6
license: CC BY 4.0
---
# Comparative analysis of transferrin and IgG N-glycosylation in two human populations
## Abstract
Human plasma transferrin (Tf) N-glycosylation has been mostly studied as a marker for congenital disorders of glycosylation, alcohol abuse, and hepatocellular carcinoma. However, inter-individual variability of Tf N-glycosylation is not known, mainly due to technical limitations of Tf isolation in large-scale studies. Here, we present a highly specific robust high-throughput approach for Tf purification from human blood plasma and detailed characterization of Tf N-glycosylation on the level of released glycans by ultra-high-performance liquid chromatography based on hydrophilic interactions and fluorescence detection (HILIC-UHPLC-FLD), exoglycosidase sequencing, and matrix-assisted laser desorption/ionization time-of-flight mass spectrometry (MALDI-TOF-MS). We perform a large-scale comparative study of Tf and immunoglobulin G (IgG) N-glycosylation analysis in two human populations and demonstrate that Tf N-glycosylation is associated with age and sex, along with multiple biochemical and physiological traits. Observed association patterns differ compared to the IgG N-glycome corroborating tissue-specific N-glycosylation and specific N-glycans’ role in their distinct physiological functions.
A large-scale comparative study in two human populations analyzes human plasma transferrin and immunoglobulin G (IgG) N-glycosylation profiles and their association with age, sex, and various biochemical and physiological traits.
## Introduction
Transferrin (Tf) is an iron-binding glycoprotein produced mainly by hepatocytes whose key function is the regulation of free iron levels in biological fluids. In humans, Tf has two N-glycosylation sites—at asparagine 432 and asparagine 630. Glycans determine glycoprotein stability and function and are involved in protein and cell recognition and interaction, signaling, trafficking, etc.1.
Glycan synthesis is a dynamic process that involves various enzymes involved in glycan attachment (glycosyltransferases), removal (glycosidases), glycan precursors synthesis and their amount, as well as sugar transporters delivering substrates for glycan chain synthesis2. In addition to enzymes directly involved in glycosylation, numerous other genetic loci have been implicated in the regulation of this complex process3,4.
Tf N-glycans have been reported to be mainly diantennary and triantennary sialylated complex-type glycans, with or without fucose, with diantennary disialylated afucosylated glycans being the most abundant5–8. Aberrant Tf N-glycosylation has been predominantly studied in the context of a biomarker for congenital disorders of glycosylation9–11, and chronic alcohol consumption12, and has been seen in e.g. hepatocellular carcinoma7, rheumatoid arthritis13, inflammation14. Bergström et al. in 2008 reported small and mostly non-statistically significant differences in carbohydrate-deficient Tf patterns in 1387 individuals of different ethnicity, age, sex, and body mass index (BMI)15, while observed differences in glycosylation profiles between smokers and non-smokers were attributed to higher alcohol intake in smokers. Besides that, Tf N-glycosylation has not been studied in much depth, and methods employed have usually been based on less sensitive and detailed techniques, e.g. lectins16, isoelectric focusing11, and high-performance liquid chromatography (HPLC) with absorbance detection (at 470 nm) of Tf glycoforms in an iron-Tf complex15,17. Recently, for Tf from a pooled healthy human serum, N-glycosylation has been described in more detail8. However, inter-individual variability or environmental factors influencing Tf N-glycosylation in a healthy human population remain largely unexplored, mainly because high-throughput methods enabling Tf purification and N-glycan analysis on a larger scale have been lacking.
One of the most studied glycoproteins is immunoglobulin G (IgG), predominantly because it is the most abundant antibody in human blood plasma, and one of the key molecules in the immune system response with N-glycosylation significantly impacting its function18. Moreover, there are efficient methods for its purification via binding to protein G or protein A. IgG N-glycans are predominantly of diantennary complex type, with a disialylated structure containing core fucose and bisecting N-acetylglucosamine (GlcNAc) being the most complex IgG N-glycan19. The variability of IgG N-glycosylation in the human population has been well studied20–22 as well as its changes in different diseases23.
Here, we present a highly specific robust high-throughput approach for purification and released N-glycan analysis of Tf from human blood plasma by ultra-high-performance liquid chromatography based on hydrophilic interactions and fluorescence detection (HILIC-UHPLC-FLD). We perform a detailed characterization of total Tf N-glycome by several complementary approaches—HILIC-UHPLC-FLD, exoglycosidase sequencing24,25, and matrix-assisted laser desorption/ionization time-of-flight mass spectrometry (MALDI-TOF-MS) analysis after linkage-specific sialic acid ethyl esterification26. We describe Tf N-glycosylation variability within two healthy human populations expanding on our parallel genetic study performed on the same set of samples27. We demonstrate that Tf N-glycosylation is associated with age and sex, as well as with a number of biochemical and physiological traits. In parallel, we analyze IgG N-glycosylation in the same two populations and demonstrate that Tf N-glycosylation correlates more with sex than age compared to IgG N-glycosylation. Additionally, regression analysis with biochemical and physiological traits reveals different patterns of associations for Tf and IgG N-glycomes, supporting N-glycan importance in distinct physiological functions of these two glycoproteins.
## Highly specific high-throughput isolation of human Tf
Tf was isolated in a high-throughput mode by immunoaffinity purification from blood plasma in two human populations, Korcula ($$n = 927$$, discovery cohort) and VIKING ($$n = 958$$, replication cohort) (Table 1), using our previously developed monolithic CIMac-@Tf 96-well plate with immobilized antibodies for human Tf28. Average purification capacity of the CIMac-@Tf 96-well plate was 300 μg of Tf per well (coefficient of variation, CV = $9.1\%$ for the plate)28. Tf purification repeatability within the cohorts was determined from the standard plasma samples and resulted in a CV of $18.9\%$ (average Tf mass = 156 ± 30 μg, $$n = 72$$) for Korcula and $23.4\%$ (average Tf mass = 85 ± 20 μg, $$n = 79$$) for VIKING. In parallel with Tf isolation, IgG isolation ($$n = 950$$ and $$n = 1087$$ for Korcula and VIKING, respectively) and subsequent N-glycosylation profiling of both glycoproteins were done for the same individuals to allow their comparative analysis. Both Tf and IgG were deglycosylated with PNGase F, and their total released N-glycans labeled with 2-aminobenzamide (2-AB), cleaned up, and analyzed by HILIC-UHPLC-FLD (Fig. 1a). Since IgG was shown in our early experiments to be the major glycosylated contaminant in Tf eluate, first IgG was isolated from plasma samples using protein G 96-well monolithic plate, and flowthrough was immediately applied on a CIMac-@Tf 96-well monolithic plate, thus bringing potential contamination from IgG in Tf eluate to a minimum (Supplementary Fig. 1a). Tf isolation process and purity of isolated Tf have been controlled throughout the study by sodium dodecyl sulfate–polyacrylamide electrophoresis (SDS–PAGE) analysis of a blank sample and four randomly taken Tf eluates from each plate, demonstrating successful Tf isolation and no visible presence of other protein contaminants (Supplementary Fig. 1b). Moreover, we wanted to ensure that quantified released N-glycans indeed originated from Tf, and not other potentially glycosylated contaminants present in the amount below SDS–PAGE detection. Therefore, isolated Tf purity was further assessed by proteomic analysis of Tf eluate (Supplementary Table 1). Proteomic analysis of Tf eluates resulted in the average relative intensity extracted for serotransferrin (UniProt P02787) of $99.36\%$, confirming the high purity of the Tf sample used for subsequent released N-glycan analysis. Moreover, other glycosylated contaminants were detected in only one replicate with a relative intensity of $0.22\%$ (Supplementary Table 1) additionally increasing the confidence that released N-glycans originate almost exclusively from Tf. Table 1Demographic characteristics of Korcula and VIKING cohorts. CharacteristicKorcula, $$n = 927$$aVIKING, $$n = 958$$aAge55 (41–67)52 (41–63)Sex F600 [65]578 [60] M327 [35]380 [40]BMI26.6 (23.7–29.4)26.8 (24.2–29.9)Smoking Never smoked274 [30]517 [54] Stopped434 [48]368 [38] Currently196 [22]72 (7.5)PerStill No334 [57]326 [56] Yes253 [43]251 [44]Cholesterol5.70 (5.00–6.70)5.30 (4.60–6.00)Triglycerides1.20 (0.90–1.60)0.90 (0.60–1.30)HDL1.50 (1.30–1.70)1.47 (1.24–1.76)LDL3.65 (2.90–4.40)3.32 (2.71–4.00)InsulinNA6.2 (4.3–9.0)HbA1c5.30 (5.00–5.60)5.30 (5.20–5.50)F females, M males, BMI body mass index, PerStill having a regular menstrual cycle, HDL high-density lipoprotein, LDL low-density lipoprotein, HbA1c hemoglobin A1c, IQR interquartile range.aMedian (IQR); N (%).Fig. 1Workflow for human plasma transferrin (Tf) isolation and N-glycan structures characterization.a Sequential isolation and N-glycan profiling of immunoglobulin G (IgG) and Tf from two human populations (Korcula and VIKING). HILIC-UHPLC-FLD ultra-high-performance liquid chromatography based on hydrophilic interactions with fluorescent detection. b Workflow for Tf purity assessment and N-glycan structures characterization by several complementary approaches. nanoLC-ESI-qTOF-MS nano-liquid chromatography coupled to electrospray ionization quadrupole time-of-flight mass spectrometry; MALDI-TOF-MS matrix-assisted laser desorption/ionization time-of-flight mass spectrometry.
## Characterization of human plasma Tf N-glycans
Human plasma Tf N-glycome was analyzed on the level of released glycans by HILIC-UHPLC-FLD and separated into 35 glycan peaks (Tf_GP1-Tf_GP35, Fig. 2, Supplementary Table 2). Several complementary approaches have been used to structurally characterize N-glycans eluting in each Tf_GP (Fig. 1b). Preliminary structures were assigned according to glucose unit (GU) values and GlycoStore database search (www.glycostore.org)29,30, and additionally confirmed by exoglycosidase sequencing24,25 (Supplementary Table 3, Supplementary Figs. 2 and 3) and MALDI-TOF-MS analysis of fractionated 2-AB labeled N-glycans eluting in individual Tf_GPs. MALDI-TOF-MS analysis was performed with or without ethyl esterification of collected N-glycan fractions, which enabled differentiation of α2,3- and α2,6-bound sialic acid26. N-glycan structures were successfully assigned to $\frac{34}{35}$ Tf_GPs accounting for approximately $99.8\%$ of Tf N-glycome (Fig. 2, Supplementary Table 2).Fig. 2N-glycosylation profile of human plasma transferrin (Tf) obtained by ultra-high-performance liquid chromatography based on hydrophilic interactions with fluorescent detection (HILIC-UHPLC-FLD).The most abundant glycan structure in each Tf glycan peak (Tf_GP) is shown. For full characterization see Supplementary Table 2. Sialic acid linkages are assigned based on matrix-assisted laser desorption/ionization time-of-flight mass spectrometry (MALDI-TOF-MS) analysis of fractionated 2-aminobenzamide (2-AB) labeled and ethyl-esterified N-glycans eluting in individual Tf_GPs. Structural schemes are given according to Consortium for Functional Glycomics (CFG) guidelines. EU emission unit, GU glucose unit. a The eight most abundant Tf N-glycan structures are shown. b Zoomed-in view showing less abundant Tf_GPs.
Predominantly, one glycan structure has been detected per chromatographic peak ranging in their complexity from high-mannose glycans, e.g. M5 in Tf_GP1, to complex tetraantennary sialylated and fucosylated glycans, e.g. F2A4G4S[3,3]2 in Tf_GP33. The most abundant glycans in Tf N-glycome are of diantennary digalactosylated type with two α2,6-sialic acids without (A2G2S[6,6]2, Tf_GP19) or with core fucose (FA2G2S[6,6]2, Tf_GP20), with both α2,3- and α2,6-sialic acid without core fucose (A2G2S[3,6]2, Tf_GP17), and α2,6-monosialylated structure without core fucose (A2G2S[6]1, Tf_GP13). In addition to directly measured N-glycans (Fig. 2, Supplementary Table 2, Supplementary Fig. 4), derived glycosylation traits that are biologically more related to activities of specific enzymes in the glycosylation pathway were calculated for both the Tf N-glycome (Supplementary Table 4) and IgG N-glycome (Supplementary Table 5).
The whole sample analysis procedure (from IgG and Tf isolation to HILIC-UHPLC-FLD analysis) was controlled by having one negative control (blank) and four to six internal standard samples per plate, according to which the method variability was assessed (Supplementary Table 6/Supplementary Figs. 5 and 6). Korcula cohort additionally contained nine to ten sample duplicates per each 96-well plate for which Tf N-glycan analysis showed a very good correlation (Supplementary Table 6/Supplementary Fig. 7).
## Age- and sex-dependent differences in the Tf N-glycome
To determine to which extent Tf N-glycome was associated with age and sex, regression analysis was performed in two independent cohorts (Table 1). After correction for multiple testing, statistically significant associations with both age and sex were observed for a majority of glycan structures (Supplementary Tables 7 and 8). These associations replicated well between the two populations, though some population-specific differences existed (Fig. 3). The most considerable associations with age were observed in the levels of glycan branching, galactosylation and sialylation. Triantennary (A3) and tetraantennary (A4) glycans, trigalactosylated (G3) and tetragalactosylated (G4) glycans, disialylated (S2) and trisialylated (S3) glycans steadily increased with age. On the other hand, diantennary (A2) glycans, agalactosylated (G0), monogalactosylated (G1), and digalactosylated (G2) glycans, asialylated (S0) and monosialylated (S1) glycans, high-mannose glycans (HM) and glycans with bisecting GlcNAc (B) or fucose (F) decreased with age. The observed changes were present both in men and women (Supplementary Table 9), with patterns of change through time being similar (Fig. 3).Fig. 3N-glycosylation profile of human plasma transferrin (Tf) is associated with age and sex. Derived N-glycosylation traits are shown for each cohort (KOR—Korcula, VIK—VIKING) and calculated according to formulas in Supplementary Table 4. Turquoise and red curves are fitted local regression models describing a sex-specific relationship between age and derived traits. The gray-shaded region is a pointwise $95\%$ confidence interval on the fitted values (there is $95\%$ confidence that the true regression curve lies within the shaded region). F females, M males, A2 diantennary glycans, A3 triantennary glycans, A4 tetraantennary glycans, HM high-mannose glycans, G0 agalactosylated glycans, G1 monogalactosylated glycans, G2 digalactosylated glycans, G3 trigalactosylated glycans, G4 tetragalactosylated glycans, F glycans containing fucose, B glycans containing bisecting N-acetylglucosamine, S0 asialylated glycans, S1 monosialylated glycans, S2 disialylated glycans, S3 trisialylated glycans.
Regression analysis revealed considerable differences in Tf N-glycome between females and males (Supplementary Tables 7 and 8). The largest differences were observed in the levels of bisecting GlcNAc, sialylation, and galactosylation. Statistically significant increases in S2, G2 and G4, A2 and A4 glycans were observed in males. Females showed higher levels of HM glycans, glycans with B or F, G0 and G1, S0 and S1 glycans (Supplementary Table 7). These differences replicated well between the two populations.
While a number of associations between Tf N-glycome and age were observed, with 30 out of 35 Tf GPs being significantly associated with age (Supplementary Table 8), the strongest associations were not as strong as associations observed between IgG N-glycome and age (Supplementary Fig. 8). The strongest associations between Tf N-glycome and age were observed for A4 ($R = 0.278$, p.adj = 6.87E−39), G4 ($R = 0.278$, p.adj = 6.87E−39), and S1 (R = −0.262, p.adj = 9.30E−34) glycans for which <$10\%$ of the variance was explained by age (Supplementary Table 10), compared to the strongest associations between IgG N-glycome and age, for G0 ($R = 0.666$, p.adj < 1.00E−300), G2 (R = −0.667, p.adj < 1.00E−300) and S1 (R = −0.595, p.adj < 1.00E−300) glycans for which more than $40\%$ of the variance was explained by age (Supplementary Table 11). In contrast to age-related associations, associations between sex and Tf N-glycome were stronger than those with IgG N-glycome. For the strongest observed association between Tf N-glycome and sex, B glycans (R = −0.331, p.adj = 3.63E−57), more than $10\%$ of the variation was explained by sex (Supplementary Table 10); while for the strongest observed association between IgG N-glycome and sex, G2 glycans (R = −0.209, p.adj = 1.84E−22), <$5\%$ of the variation was explained by sex (Supplementary Table 11). Patterns of observed correlations revealed that age and sex have quite different influences on Tf and IgG N-glycomes. For the IgG N-glycome, the strongest observed age effects were several times stronger than the strongest observed sex effects. Unlike for IgG N-glycome, the strongest observed age and sex associations for Tf N-glycome were of similar magnitude.
## Tf N-glycome and biochemical and physiological traits
To identify factors that may be responsible for the remaining variability in the Tf N-glycome, we performed regression analysis with available biochemical and physiological traits in our databases (Supplementary Table 12). Since the majority of available biochemical and physiological traits correlate with age and sex and consequently univariate regression analysis showed significant associations for virtually all traits (Supplementary Fig. 9), age and sex were included as covariates in all further analyses (Fig. 4). Regression analysis revealed strong associations between multiple Tf N-glycome and clinical (weight, BMI, blood pressure, etc.) as well as biochemical (HDL, insulin, triglycerides, etc.) parameters (Supplementary Tables 13 and 14). These associations replicated well between the two populations. The strongest associations were observed between Tf N-glycome and weight, with 22 out of 35 Tf GPs being significantly associated with weight (Supplementary Table 14). A2, S2, and G2 glycans were positively correlated with weight. A3, G1, and G3 glycans, S0 and S1 glycans, B glycans, and HM glycans were negatively correlated with weight (Supplementary Table 13). The majority of clinical and biochemical traits that are known to be associated with an unhealthy lifestyle (insulin, uric acid, glucose, triglycerides, HbA1c, blood pressure) revealed patterns of correlations with Tf N-glycome similar to those obtained for weight. A similar pattern was observed for several medical conditions (hypertension, gout, diabetes), although associations were not as strong as for biochemical traits (Supplementary Table 13). Since the majority of biochemical and physiological traits correlate with weight, the possibility remained that some of the observed associations were a reflection of weight. To further investigate the impact biochemical and physiological traits have on Tf N-glycome, regression analysis was repeated with BMI included as an additional covariate (Table 2, Supplementary Tables 15 and 16). While the number of significant associations in the repeated analysis was lower, with obtained effects not as strong as effects from previous analysis with only age and sex as covariates, the pattern of correlations with Tf N-glycome remained similar (Supplementary Fig. 10). Most associations between clinical parameters, medical conditions, and Tf N-glycome were reduced below the level of statistical significance, however, most of the associations with biochemical parameters remained statistically significant (Supplementary Tables 15 and 16).Fig. 4Age- and sex-adjusted correlations between directly measured transferrin (Tf)/immunoglobulin G (IgG) N-glycan traits (GPs) and biochemical and physiological traits. Correlation analysis was performed on Korcula and VIKING cohorts separately and then combined using an inverse-variance weighted meta-analysis approach. Prior to correlation analysis, glycan measurements were adjusted for age and sex. Only statistically significant (adjusted meta p-value < 0.01, method = Benjamini–Hochberg) meta-correlations are shown, where the blue color indicates a positive correlation, and the red color indicates a negative correlation. Description of biochemical and physiological traits is given in Supplementary Table 12.Table 2Age-, sex- and BMI-adjusted associations of derived transferrin (Tf) N-glycome traits with biochemical and physiological traits. GlycanTraitKorculaVIKINGMETADFeffectSEp.valp.adjDFeffectSEp.valp.adjeffectSEp.valp.adjS2 totalinsulinNANANANANA9480.2480.0347.02E−134.00E−100.2480.0344.53E−132.86E−10S1 totalinsulinNANANANANA948−0.2340.0347.34E−122.09E-09−0.2340.0345.22E−121.64E−09G2 totalhdl865−0.2030.0354.63E−092.03E−06945−0.1140.0371.82E−034.31E-02−0.1620.0251.38E−102.91E−08A2 totalhdl865−0.2040.0356.59E−092.03E−06945−0.1120.0372.27E−034.61E−02−0.1600.0252.60E−104.10E−08A3 totalhdl8650.1810.0364.59E−077.05E−059450.1270.0364.80E−041.83E−020.1550.0261.51E−091.58E−07G3 totalhdl8650.1810.0364.59E−077.05E−059450.1270.0364.80E−041.83E−020.1550.0261.51E−091.58E−07S1 totalfbrinogen846−0.1030.0331.57E−032.42E−02945−0.1510.0322.01E−062.87E−04−0.1280.0232.04E-081.84E−06S2 totalhdl865−0.1060.0352.20E−032.98E−02945−0.1500.0337.20E−066.84E−04−0.1290.0248.92E−087.03E−06S2 totalhba1c8600.1580.0355.70E−063.90E−049460.0860.0341.16E−021.27E−010.1210.0246.53E−074.57E−05S0 totalinsulinNANANANANA948−0.1780.0369.24E−071.76E−04−0.1780.0369.00E−075.67E−05S3 totaltot_chol8650.1090.0362.68E−033.50E−029480.1240.0353.39E−041.76E−020.1170.0253.25E−061.86E−04F totalfibrinogen8460.1090.0351.84E−032.76E−029450.1200.0356.94E−042.24E−020.1140.0254.48E−062.35E−04S3 totalhdl8650.1420.0371.22E−045.34E−039450.0940.0371.06E−021.23E−010.1180.0266.38E−063.09E−04HM totalinsulinNANANANANA948−0.1620.0366.99E−066.84E−04−0.1620.0366.99E−063.15E−04S0 totaltriglyc865−0.0940.0356.57E−036.12E−02945−0.1180.0344.61E−041.83E−02−0.1060.0241.13E-054.73E-04A4 totalfibrinogen8460.1140.0334.84E−041.24E−029450.0870.0351.21E−021.28E−010.1020.0242.14E-057.95E-04G4 totalfibrinogen8460.1140.0334.84E−041.24E−029450.0870.0351.21E−021.28E−010.1020.0242.14E-057.95E-04G2 totaluric8650.1920.0401.65E−062.03E−049480.0450.0412.68E−016.74E−010.1200.0292.65E-059.27E-04S1 totalhba1c860−0.1290.0363.41E−041.05E−02946−0.0770.0342.15E−021.81E−01−0.1020.0253.92E-051.30E-03G1 totaltriglyc865−0.0930.0345.94E−035.89E−02945−0.1050.0342.21E−034.61E−02−0.0990.0244.21E-051.33E-03G1 totalarth_sr8450.3800.1303.44E−034.25E−029470.3160.1155.76E−039.12E−020.3440.0866.54E-051.96E-03A2 totaluric8650.1870.0413.98E−063.18E−049480.0400.0413.25E−017.28E−010.1150.0297.04E-052.02E-03S0 totalwaist867−0.1200.0604.52E−022.53E−01948−0.2500.0713.99E−041.83E−02−0.1750.0461.40E-043.76E-03All associations are reported in Supplementary Table 15. Description of biochemical and physiological traits is given in Supplementary Table 12.DF degrees of freedom, SE standard error, p.adj false discovery rate was controlled using Benjamini–Hochberg method at the specified level of 0.05, S2 disialylated glycans, S1 monosialylated glycans, G2 digalactosylated glycans, A2 diantennary glycans, A3 triantennary glycans, G3 trigalactosylated glycans, S0 asialylated glycans, S3 trisialylated glycans, HM high-mannose glycans, A4 tetraantennary glycans, G4 tetragalactosylated glycans, G1 monogalactosylated glycans.
Regression analysis with biochemical and physiological traits revealed different patterns of associations for Tf and IgG N-glycomes (Supplementary Tables 17–20). For most clinical parameters (BMI, weight, blood pressure), medical conditions (hypertension, gout, diabetes, arthritis), and biochemical parameters (insulin, HDL, uric acid, glucose, triglycerides, HbA1c, fibrinogen) considerably stronger associations were observed with Tf N-glycome in comparison with IgG N-glycome. For IgG N-glycome, the analysis showed comparatively stronger associations only for traits related to fecundity (regular menstrual cycle) and smoking, although significant associations between Tf N-glycome and smoking were also observed.
## Discussion
Glycosylation is one of the most common co- and posttranslational modifications affecting not only protein structure but also its functionality1. Variability of IgG glycosylation has been extensively studied in different populations20–22, diseases23, and functional studies18,31. However, glycosylation changes of other glycoproteins, including Tf, have been mostly studied in the context of changes happening in different disease states vs. healthy controls, while their variability in a healthy human population remains generally unknown and is considered stable. Moreover, while high-throughput purification (and glycosylation analysis) strategies for large-scale study applications do exist for IgG, this is largely not the case for other glycoproteins. For only a few other glycoproteins, e.g. IgA32, apolipoprotein CIII33, α-1-acid glycoprotein34,35, high-throughput purification and/or glycosylation analysis strategies with application in population studies have been reported.
Here, we have developed a highly specific robust high-throughput approach for purification and glycosylation analysis of Tf from human blood plasma by HILIC-UHPLC-FLD. Developed reusable monolithic CIMac-@Tf 96-well plate with immobilized antibodies for human Tf enables fast large-scale and cost-effective purification of Tf without the use of packed columns or tips used in previously reported approaches8,36. Combining the developed Tf purification workflow with N-glycan quantification by HILIC-UHPLC-FLD enabled a detailed characterization of total released Tf N-glycome, including information on sialic acid linkages using ethyl esterification and MALDI-TOF-MS analysis26 of fractionated chromatographic peaks. The repertoire of identified Tf N-glycan structures expands on the previous work8,37, especially in terms of sialic acid linkages for disialylated and trisialylated structures, making this study the most extensive one in terms of Tf N-glycome characterization on the level of total released N-glycans.
Previously, Bergström et al. studied differences in Tf glycosylation patterns related to ethnicity, age, sex, BMI, and smoking as potential confounders for carbohydrate-deficient Tf testing in the context of a biomarker for heavy alcohol use15. Based on the relative quantification of seven iron-saturated Tf glycoforms (asialo-, monosialo-, disialo-, trisialo-, tetrasialo-, pentasialo-, and hexasialotransferrin) in serum using an HPLC17 they concluded that there are no significant differences in disialotransferrin levels (the main carbohydrate-deficient Tf glycoform) between individuals of different ethnic origin, age, sex, BMI or smoking status15. They did observe some statistically significant differences in other Tf glycoforms between individuals of different sex, BMI, and smoking statuses, although they concluded that the latter mostly reflected higher alcohol consumption by smokers.
We have applied the developed state-of-the-art analytical approach to explore in-depth the variability of Tf N-glycome in a healthy human population and to compare it with the variability of IgG N-glycosylation in the same set of samples. By relative quantification of 35 directly measured and 15 calculated derived Tf N-glycosylation traits, we performed the most extensive analysis of Tf N-glycome variability in a healthy human population demonstrating that Tf N-glycome associates with the age and sex of individuals, as well as with a number of biochemical and physiological traits. Compared to the IgG glycome, Tf N-glycome correlates more with sex than age, which could be a reflection of the different roles of glycans on these two proteins. As the main function of *Tf is* the binding and transport of iron it is somewhat expected that its glycosylation is well conserved during aging, while on the other hand, it is plausible that its synthesis and glycosylation are differentially regulated in females compared to males due to, e.g. menstrual cycles and hormonal fluctuations13. It is known that both Tf concentration and glycosylation change during pregnancy13,38,39 as well as during the use of oral contraceptives13. Although the biological meaning of different Tf glycosylation patterns is still not completely understood, early research studies have demonstrated that the absence of Tf glycosylation doesn’t affect iron binding or Tf binding to a Tf receptor40–42, while it was shown that it affects cellular iron uptake in vitro and in vivo40,42. However, changes in Tf N-glycome can alter its circulation half-life and affect iron transport43, which might be an important element of the response to acute inflammation44,45. On the other hand, the function of IgG is recognition of foreign pathogens and activation of an appropriate immune response, so it is conceivable IgG glycosylation correlates more with age than with sex, since it was proposed that it somewhat reflects previous antigen encounters during past infections46–48.
The role of IgG N-glycans in the modulation of its effector functions has been well recognized49–52. One of the most studied examples is the addition of core fucose to the Fc fragment Asn297 N-glycosylation site that decreases antibody-dependent cell-mediated cytotoxicity (ADCC) up to 100-fold through lower IgG affinity toward Fcγ receptor IIIa53,54. Moreover, both galactosylation and sialylation have been shown to modulate the inflammatory activity of IgG. While galactosylated IgG has an anti-inflammatory effect by inhibiting the complement pathway55, agalactosylated IgG glycans act pro-inflammatory by activating the alternative pathway56, as well as lectin complement pathway via interaction with mannose-binding lectin57. Sialylation also changes IgG activity from pro-inflammatory to anti-inflammatory31,51, although the exact mechanisms remain to be elucidated. Pro-inflammatory glycosylation profile of IgG, predominantly characterized by decreased galactosylation and sialylation, has been seen with aging21,58, as well as with different diseases23. It was hypothesized that the lifelong antigenic challenges and increasing antigenic burden cause immune system remodeling leading to the state of a low-grade chronic inflammation that characterizes aging, i.e. inflammaging46. In this context, pro-inflammatory IgG glycans are considered both biomarkers and functional effectors of aging47.
Next, we demonstrated that Tf N-glycome correlates with a number of biochemical and physiological traits and with different patterns of association compared to the IgG N-glycome validating distinctive protein-specific roles of N-glycans in Tf and IgG physiological functions. Strong associations of Tf N-glycome with weight and other clinical and biochemical parameters potentially implicate systemic changes in the iron transport by Tf within the organism, e.g. after intestinal iron uptake, by iron recycling from iron-containing proteins or iron stores59, reflected by changes in these physiological and metabolic factors. During iron transport, Tf binds to two plasma membrane receptors—Tf receptor and asialoglycoprotein receptor, with early studies demonstrating that the affinity of asialo Tf glycoforms for asialoglycoprotein receptor differs depending on the Tf glycan chain complexity60. Even though the functional relevance of Tf glycosylation remains underexplored, it is possible that specific Tf glycoforms influence Tf affinity and interactions with its receptors, fine-tuning the iron transport between sites of absorption and delivery to requiring cells59.
In conclusion, by developing a robust high-throughput approach for the analysis of total Tf N-glycome we have expanded the current knowledge of human plasma Tf N-glycan repertoire. Moreover, we have performed an extensive analysis of Tf N-glycome variability in a healthy human population demonstrating its changes with age, sex, biochemical, and physiological status of individuals, providing the basis for future population and clinical studies. By performing IgG glycosylation analysis in the same samples, we demonstrated that glycosylation of these two proteins has independent regulation, which confirms a similar observation from a previous genetic study27.
## Korcula cohort
This study was performed in the adult population of the island of Korčula, Croatia61. The fieldwork was performed in 2007 in the eastern parts of the island, focusing on the town of Korčula and villages Lumbarda, Žrnovo, and Račišće. The sampling approach was convenient, with population-wide invites to non-institutionalized individuals living in the island. All subjects were aged 18 and over, and had signed informed consent before entering the study, which was approved by the Ethical Committee of the Medical School, University of Zagreb, and Multi-Centre Research Ethics Committee for Scotland. All relevant ethical regulations were followed. Fasting plasma samples were collected, processed, and stored at −80 °C on-site immediately upon processing, ensuring the highest possible sample quality62.
## VIKING cohort
The Viking Health Study—Shetland (VIKING) is a family-based, cross-sectional study that seeks to identify genetic factors affecting cardiovascular and other disease risks in the isolated population of the Shetland Islands in northern Scotland63. Compared to Mainland Scotland, genetic diversity in this population is decreased, consistent with the high levels of endogamy historically. Between 2013 and 2015, 2105 participants were recruited, most having at least three grandparents from Shetland. Numerous health-related phenotypes and environmental exposures were measured and fasting blood samples were collected from each individual. The study was approved by the South East Scotland Research Ethics Committee, NHS Lothian (reference: 12/SS/0151), all participants gave informed consent, and all relevant ethical regulations were followed.
## Experimental design
This is an observational study and samples were collected as samples of convenience. No statistical calculation of sample size was performed; the sample size was determined based on availability. All samples were processed in batches (96-well in the case of Korcula samples and 60-well in the case of VIKING samples), following a predetermined experimental design, which was blocked on sex and age information. All plates included several technical replicates of a standard plasma sample and blank samples for quality control and batch correction. Standard plasma sample used for *Korcula analysis* was prepared as a pool of cohort samples. Standard plasma sample used for VIKING cohort analysis was obtained from the Croatian National Institute of Transfusion Medicine after approval by the Ethical Committee of the institute. Additionally, Korcula samples contained around $10\%$ of duplicate samples per plate to further assess method variability. The sample analysis was done blindly, and all samples were analyzed with a uniform set of techniques.
## Immunoglobulin G and transferrin isolation
The whole procedure was done in a 96-well plate manner according to the previously prepared randomization plan and ultrapure water (≥18.2 MΩ at 25 °C) was used throughout. IgG was isolated from blood plasma samples by CIM® Protein G 96-well monolithic plate (BIA Separations, a Sartorius company, Ajdovščina, Slovenia) with individual column volume of 200 μL using a vacuum manifold (Pall Corporation, Ann Arbor, MI, USA)20. All steps during the isolation procedure were performed at around 380 mmHg, except for plasma sample application and glycoprotein elution (around 200 mmHg). The protein G plate was preconditioned by washing with 2 mL of ultrapure water, 2 mL of 1× PBS, pH 7.4, 1 mL of 0.1 mol L−1 formic acid (Merck, Darmstadt, Germany), 2 mL of 10× PBS, and 4 mL of 1× PBS, pH 7.4. Then, 120 µL of plasma sample was centrifuged for 3 min at 12,100×g, diluted with 1× PBS, pH 7.4 (1:7), and filtered through a 0.45 μm AcroPrep hydrophilic polypropylene (GH Polypro, GHP) filter plate (Pall Corporation) using a vacuum manifold (around 380 mmHg). After plasma filtration, samples were applied to the preconditioned protein G plate and flowthrough was collected for immediate subsequent Tf isolation by previously developed CIMac-@Tf 96-well monolithic plate (BIA Separations)28 with individual column volume of 200 μL. The CIMac-@Tf 96-well monolithic plate was preconditioned by washing with 2 mL of ultrapure water, 1 mL of 0.1 mol L−1 formic acid pH 3.0 (pH adjusted with $25\%$ ammonia solution, Merck), and 4 mL of 1× PBS, pH 7.4. Unbound proteins during IgG and Tf isolation were washed away with 6 mL of 1× PBS, pH 7.4, and 1× PBS (0.25 mol L−1 NaCl), pH 7.4, respectively. Bound IgG was eluted with 1 mL of 0.1 mol L−1 formic acid and bound Tf with 0.7 mL of 0.1 mol L−1 formic acid pH 3.0 and immediately neutralized with 1 mol L−1 ammonium hydrogencarbonate (Sigma-Aldrich, St. Louis, MO, USA) to pH 7.0. The protein G plate was regenerated by washing with 2 mL of 0.1 mol L−1 formic acid, 2 mL of 10× PBS, 4 mL of 1× PBS, pH 7.4, and 1 mL of $20\%$ (v/v) ethanol in 20 mmol L−1 Tris + 0.1 mol L−1 NaCl, while the CIMac-@Tf plate was washed with 2 mL of 0.1 mol L−1 formic acid pH 3.0, 4 mL of 1× PBS, pH 7.4, and 1 mL of 1× PBS + $0.02\%$ NaN3 (Sigma-Aldrich), pH 7.4. Monolithic plates were stored in $20\%$ (v/v) ethanol in 20 mmol L−1 Tris + 0.1 mol L−1 NaCl (protein G plate) and 1× PBS + $0.02\%$ NaN3, pH 7.4 (CIMac-@Tf plate) at 4 °C until the next isolation. IgG and Tf concentrations were measured at 280 nm using a NanoDrop 8000 spectrophotometer (Thermo Fisher Scientific, Waltham, ME, USA). Each elution fraction (300 μL) was dried in a vacuum centrifuge (Thermo Fisher Scientific) and stored at −20 °C until subsequent N-glycan analysis.
## Sodium dodecyl sulfate–polyacrylamide gel electrophoresis
Tf elution fractions were analyzed by sodium dodecyl sulfate–polyacrylamide gel electrophoresis (SDS–PAGE) using NuPAGE 4–$12\%$ Bis-Tris gradient protein gels (1.0 mm thickness) under reducing conditions according to the manufacturer’s instructions (Invitrogen, Waltham, MA, USA). The gels were run at 200 V for 35 min using a NuPAGE MES SDS buffering system (Invitrogen). Protein bands were visualized by GelCode Blue staining reagent (Thermo Fisher Scientific).
## Transferrin purity assessment by proteomic analysis
Tf purity after isolation was assessed by performing trypsin digestion and LC–MS analysis of obtained (glyco)peptides. Pooled isolated Tf sample (≈20 µg of protein in 150 µL of the eluate) was reduced with 16 µL of 20 mmol L−1 dithiothreitol (Sigma-Aldrich), incubated at 60 °C for 30 min and then alkylated with the addition of 30 µL of 40 mmol L−1 iodoacetamide (IAA, Sigma-Aldrich) and incubation in the dark for 30 min. Before overnight trypsin digestion (0.2 µg per sample), samples were left in bright light at room temperature for 20 min to stop IAA activity. Tryptic glycopeptides and peptides were separated and analyzed by nano-liquid chromatography (Waters, Milford, MA, USA) coupled to electrospray ionization quadrupole time-of-flight mass spectrometry (Bruker Daltonics, Bremen, Germany) (nanoLC-ESI-qTOF-MS). Samples were loaded on a PepMap 100 C18 trap column (5 mm × 300 µm i.d.; Thermo Fisher Scientific) for 3 min of trapping at a 40 µL min−1 flow rate. After that, the trap column was switched in-line with the analytical SunShell C18 column (150 mm × 100 µm i.d., 90 Å, ChromaNik Technologies Inc., Osaka, Japan), and tryptic (glyco)peptides were separated with $0.1\%$ formic acid as solvent A and $80\%$ acetonitrile (ACN, Honeywell, Charlotte, NC, USA), $0.02\%$ formic acid as solvent B at a flow rate of 1 µL min−1 with a linear gradient of 0–$80\%$ solvent B in 80 min, in a 90 min analytical run. MS was set to the following parameters: positive ion mode, captive spray with nanoBooster technology, capillary voltage 1 kV, dry gas temperature 150 °C, and flow 8 L h−1. Mass spectra were recorded in the range from m/z 50 to 3000 with a frequency of 0.5 Hz. Three precursors with the highest intensities were automatically selected for CID fragmentation. A search for specific tryptic peptides with a maximum of two miscleavages was done in MaxQuant version 1.6.10.43 software64 against Homo sapiens protein sequences (UniProt fasta file) with the methionine oxidation and asparagine carrying N-acetylhexosamine as variable modifications, and carbamidomethyl on cysteine as the fixed modification. Analysis was performed in triplicates.
## N-glycan release and fluorescent labeling of Tf and IgG glycans
Dried Tf and IgG eluates after isolation from plasma samples were denatured with 30 μl of 13.3 g L−1 sodium dodecyl sulfate (SDS, Invitrogen) and by incubation at 65 °C for 10 min. After cooling down to room temperature for 30 min, 10 μl of $4\%$ (v/v) Igepal CA-630 (Sigma-Aldrich) was added and the mixture was shaken for 15 min on a plate shaker. N-glycans were released after the addition of 10 μL of 5× PBS and 1.2 U of PNGase F (10 U/μL, Promega, Madison, WI, USA) by incubation at 37 °C for 18 h. Released N-glycans were labeled with 2-AB (Sigma-Aldrich). The labeling mixture was freshly prepared by dissolving 2-AB and 2-methylpyridine borane complex (final concentrations of 19.2 and 44.8 mg mL−1, respectively) in the mixture of dimethyl sulfoxide (Sigma-Aldrich) and glacial acetic acid (Merck) (7:3, v/v). The labeling mixture (25 μL) was added to each sample and the plate was sealed using an adhesive seal. After 10 min of shaking, samples were incubated for 2 h at 65 °C.
## Cleanup of 2-AB-labeled glycans
Excess reagents from previous steps were removed from the samples using hydrophilic interaction liquid chromatography solid phase extraction (HILIC-SPE). After free N-glycan labeling samples were cooled down to room temperature for 30 min and 700 μL of ACN (previously cooled down to 4 °C) was added to each sample. The cleanup procedure was performed on a hydrophilic 0.2 μm AcroPrep GHP filter plate (Pall Corporation) using a vacuum manifold at around 25 mmHg. All wells of a GHP filter plate were prewashed with 200 μL of $70\%$ (v/v) ethanol in water, 200 μL of ultrapure water, and 200 μL of $96\%$ (v/v) ACN in water (previously cooled down to 4 °C). Diluted samples were loaded to the GHP filter plate wells, and after short incubation subsequently washed with 5 × 200 μL of $96\%$ (v/v) ACN in water. The last washing step was followed by centrifugation at 164×g for 5 min. Glycans were eluted from the plate with 2 × 90 μL of ultrapure water after 15 min shaking at room temperature and centrifugation at 164×g for 5 min in each step. Combined eluates of 2-AB labeled Tf and IgG N-glycans were stored at −20 °C until ultra-high-performance liquid chromatography (UHPLC) analysis.
## Glycan analysis by ultra-high-performance liquid chromatography
Fluorescently labeled and purified Tf and IgG N-glycans were analyzed by UHPLC based on hydrophilic interactions (HILIC-UHPLC) and detected using excitation and emission wavelengths of 250 and 428 nm, respectively. Acquity UHPLC instrument (Waters) was under the control of Empower 3 software, build 3471 (Waters). Mobile phases were 100 mmol L−1 ammonium formate, pH 4.4 (solvent A), and ACN (solvent B) and samples were maintained at 10 °C before injection. Tf 2-AB labeled N-glycans prepared in $75\%$ ACN were separated on a Waters BEH Glycan column, 150 × 2.1 mm i.d., 1.7 μm BEH particles at 25 °C in a 23-min linear gradient of 30–$47\%$ solvent A at a flow rate of 0.56 mL min−1. IgG 2-AB-labeled N-glycans prepared in $80\%$ ACN were separated on a Waters BEH Glycan column, 100 × 2.1 mm i.d., 1.7 μm BEH particles at 60 °C in a 29-min linear gradient of 25–$38\%$ solvent A at a flow rate of 0.4 mL min−1.
The HILIC-UHPLC system was calibrated using a dextran ladder (external standard of hydrolyzed and 2-AB labeled glucose oligomers) according to which the retention times for the individual chromatographic peaks (representing the 2-AB-labeled glycan) were converted to glucose units (GU). Data was processed using an automatic processing method with a traditional integration algorithm. Each Tf N-glycans chromatogram integrated into 35 peaks was manually corrected to maintain the same intervals of integration for all the samples, while IgG N-glycans chromatograms were integrated into 24 peaks20 using automatic integration65. The amount of glycans in each chromatographic peak was expressed as a percentage of the total integrated area (% area).
## Exoglycosidase digestions
A pool of Tf 2-AB-labeled N-glycans (volume equivalent to glycans from 10 μg of Tf) was dried in PCR tubes and processed using the N-Glycan Sequencing Kit according to the manufacturer’s instructions (New England Biolabs, Ipswich, MA, USA) (Supplementary Table 3). Each of the eight reactions was done in triplicates. After 18-h digestion at 37 °C reaction mixtures were cleaned up on AcroPrep Omega 96-well filter plate MWCO 10 K NTRL, 350 μL well volume (Pall) using a vacuum manifold at a maximum vacuum. The filter plate was washed 2 × 100 μL of ultrapure water, reaction mixtures (10 μL) were transferred to prewashed wells of the filter plate, and eluates were collected in a clean collection plate. Reaction tubes were rinsed 2 × 20 μL of ultrapure water and transferred in each step to the Omega filter plate. The second and third fractions were collected in the same collection plate resulting in a total volume of 50 μL. Digested Tf N-glycan samples were stored at −20 °C until HILIC-UHPLC analysis, as described.
## Ethyl esterification of Tf 2-AB-labeled N-glycans
Before mass spectrometry analysis of Tf glycans eluting in each chromatographic peak, the Tf 2-AB labeled N-glycan pool was fractionated by HILIC-UHPLC using the chromatographic method described above. Four equivalent HILIC-UHPLC fractions were combined, dried in a vacuum centrifuge, and reconstituted in 1 μL of ultrapure water. Ethyl esterification reaction and cotton HILIC glycan enrichment26 were performed as follows. The volume of 20 µL of 1-ethyl-3-(3-(dimethylamino)propyl)carbodiimide (EDC, AcrosOrganics, Thermo Fisher Scientific)/1-hydroxy-benzotriazole (HOBt, Sigma-Aldrich) reagent in ethanol (0.25 mol L−1 each) was added and samples incubated for 1 h at 37 °C on a heated platform without shaking (Heidolph, Schwabach, Germany). ACN (20 µL) was added to each sample, resuspended, incubated at −20 °C for 15 min, and then at room temperature for 15 min. Each cotton pipette tip used for glycan enrichment was prewashed 3× 20 µL of ultrapure water and 3× 20 µL of $85\%$ (v/v) ACN in water. The sample was aspirated and resuspended 20× in the tip. Cotton tip was then washed 3× 20 µL of $85\%$ (v/v) ACN, $1\%$ (v/v) trifluoroacetic acid (TFA, Sigma-Aldrich) and 3× 20 µL $85\%$ (v/v) ACN. The sample was eluted to a clean tube by pipetting 10 µL of ultrapure water and resuspending 20×. Additional elution fraction was collected in the same way, both fractions were combined and dried in a vacuum centrifuge (Martin Christ, Osterode, Germany).
## MALDI-TOF-MS Tf N-glycan analysis
Chromatographic peak fractions containing 2-AB-labeled ethyl-esterified Tf N-glycans were analyzed by MALDI-TOF-MS. Each sample was reconstituted in 1.5 μL of ultrapure water to which 1.5 μL of matrix solution, containing 5 g L−1 2,5-dihydroxybenzoic acid (2,5-DHB, Bruker Daltonics), 1 mmol L−1 sodium hydroxide in $75\%$ (v/v) ACN in ultrapure water, was added. Droplets were mixed by pipetting and 1 μL of the mixture was spotted to MALDI AnchorChip 384 BC (Bruker). Spots were allowed to air dry for 30 min and recrystallized by briefly tapping them with a pipette tip containing 0.2 μL of ethanol.
Analyses were done on an ultrafleXtreme MALDI-TOF-MS (Bruker Daltonics) using positive-ion reflectron mode under the control of Flexcontrol 3.3 software (Bruker Daltonics). Dextran calibration standard (0.8 g L−1) was used for instrument calibration before measuring the surrounding eight spotted samples. A mass window of m/z 1000–5000 with suppression up to m/z 900 was used for Tf N-glycans detection. For each spectrum, 3 × 10,000 laser shots were accumulated at a laser frequency of 2000 Hz, using a complete sample random walk with 200 shots per raster spot. Spectra were recorded using $65\%$ laser intensity. Tandem mass spectrometry (MALDI-TOF/TOF-MS/MS) was performed via LIFTTM technology (Bruker Daltonics). Compass DataAnalysis 4.1, build 362.7 (Bruker Daltonics) was used for spectra analysis.
## Tf N-glycan structure assignation
N-glycan structures were proposed using the GlycoStore database (www.glycostore.org)29,30 according to UHPLC-HILIC data, GlycoMod Tool (https://web.expasy.org/glycomod/)66 according to experimentally determined monoisotopic masses (0.5 Da mass tolerance, 143 Da adduct mass that includes 2-AB and Na+ as adducts), results of exoglycosidase array reactions, literature search and established biosynthetic pathways. GlycoWorkbench version 2.1, build 146 was used for MS/MS spectra annotation and depiction of glycan structures on the figures. Figures have been annotated with glycan cartoons following the recommendations of the Consortium for Functional Glycomics67.
## Statistics and reproducibility
High-throughput UHPLC data was normalized and batch-corrected to remove experimental variation from glycan measurements. Normalization by total area resulted in % area values expressing the amount of N-glycans in each chromatographic peak. Normalized glycan measurements were then log-transformed due to the right-skewness of their distributions and the multiplicative nature of batch effects. ComBat method (R package sva) was used to perform batch correction on log-transformed measurements, where the technical source of variation (which sample was analyzed on which plate) was modeled as a batch covariate. Estimated batch effects were subtracted from log-transformed measurements to get measurements corrected for experimental noise. In addition to directly measured glycan structures, derived glycosylation traits were calculated based on the shared glycan structural features (e.g. sialylation, galactosylation, fucosylation, etc.) according to Supplementary Tables 4 and 5 for Tf and IgG, respectively.
Method variability was assessed by estimating the measurement error of each Tf chromatographic peak. Estimations are based on the internal standard samples in Korcula ($$n = 72$$) and VIKING ($$n = 71$$) cohorts, and based on the sample replicates in Korcula ($$n = 2$$ × 107 = 214) cohort. Both internal standards and replicates represent a set of samples that are biologically identical and differences among their measurements are a result of experimental variation. Estimations of the measurement error based on standard samples were calculated as a ratio of the variation of standard samples and variation of cohort samples, multiplied by 100 (100*Var(Stand)/Var(Cohort)). Estimations of the measurement error based on sample replicates were calculated using the formula (1−Correlation(Replicates))*100.
Associations analyses between glycan traits and biochemical and physiological traits were performed by implementing a general linear regression model or Pearson correlation test (where stated). In analyses where the general regression model was applied, age and sex (and BMI where stated) were included as additional covariates except in those models where age and sex were variables of primary interest. Before regression analyses, glycan variables as well as quantitative biochemical and physiological traits were all transformed to standard Normal distribution (mean = 0, sd = 1) by an inverse transformation of ranks to Normality (R package “GenABEL”, function rntransform). Transformed glycan variables have the same standardized variance so using rank transformed variables in analyses makes estimated effects of different glycans in different cohorts comparable. In analyses where the Pearson correlation test was used, glycan measurements were adjusted for age, sex, and BMI (where stated) prior to correlation analysis. Analyses were firstly performed for each cohort separately (N(Korcula) = 927; N(VIKING) = 958 for Tf and N(Korcula) = 950; N(VIKING) = 1087 for IgG) and then combined using fixed-effects meta-analysis approach (R package meta, metagen(method = “FE”)). The false discovery rate was controlled using the Benjamini–Hochberg procedure (function p.adjust(method = “BH”)). Data were analyzed and visualized using the R programming language (version 4.0.2).
## Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.
## Supplementary information
Peer Review File Supplementary Information Description of Additional Supplementary Files Supplementary Data 1 Reporting Summary The online version contains supplementary material available at 10.1038/s42003-023-04685-6.
## Peer review information
Communications Biology thanks Mary Ann Comunale, Angelo Palmigiano and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Primary Handling Editor: Gene Chong. Peer reviewer reports are available.
## References
1. Varki A. **Biological roles of oligosaccharides: all of the theories are correct**. *Glycobiology* (1993.0) **3** 97-130. DOI: 10.1093/glycob/3.2.97
2. Ohtsubo K, Marth JD. **Glycosylation in cellular mechanisms of health and disease**. *Cell* (2006.0) **126** 855-867. DOI: 10.1016/j.cell.2006.08.019
3. Krištić J. **Profiling and genetic control of the murine immunoglobulin G glycome**. *Nat. Chem. Biol.* (2018.0) **14** 516-524. DOI: 10.1038/s41589-018-0034-3
4. Klarić L. **Glycosylation of immunoglobulin G is regulated by a large network of genes pleiotropic with inflammatory diseases**. *Sci. Adv.* (2020.0) **6** eaax0301. DOI: 10.1126/sciadv.aax0301
5. Spik G. **Studies on glycoconjugates. LXIV. Complete structure of two carbohydrate units of human serotransferrin**. *FEBS Lett.* (1975.0) **50** 296-299. DOI: 10.1016/0014-5793(75)80513-8
6. Spik G, Debruyne V, Montreuil J, van Halbeek H, Vliegenthart JFG. **Primary structure of two sialylated triantennary glycans from human serotransferrin**. *FEBS Lett.* (1985.0) **183** 65-69. DOI: 10.1016/0014-5793(85)80955-8
7. Yamashita K, Koide N, Endo T, Iwaki Y, Kobata A. **Altered glycosylation of serum transferrin of patients with hepatocellular carcinoma**. *J. Biol. Chem.* (1989.0) **264** 2415-2423. DOI: 10.1016/S0021-9258(19)81629-0
8. O’Flaherty R. **A robust and versatile automated glycoanalytical technology for serum antibodies and acute phase proteins: ovarian cancer case study**. *Mol. Cell. Proteom.* (2019.0) **18** 2191-2206. DOI: 10.1074/mcp.RA119.001531
9. Guillard M. **Transferrin mutations at the glycosylation site complicate diagnosis of congenital disorders of glycosylation type I**. *J. Inherit. Metab. Dis.* (2011.0) **34** 901-906. DOI: 10.1007/s10545-011-9311-y
10. Barroso A, Giménez E, Benavente F, Barbosa J, Sanz-Nebot V. **Classification of congenital disorders of glycosylation based on analysis of transferrin glycopeptides by capillary liquid chromatography–mass spectrometry**. *Talanta* (2016.0) **160** 614-623. DOI: 10.1016/j.talanta.2016.07.055
11. Wolking AB. **Transferrin glycosylation analysis from dried blood spot cards and capillary blood samples**. *J. Chromatogr. B Anal. Technol. Biomed. Life Sci.* (2019.0) **1106–1107** 64-70. DOI: 10.1016/j.jchromb.2019.01.004
12. Stibler H. **Carbohydrate-deficient transferrin in serum: a new marker of potentially harmful alcohol consumption reviewed**. *Clin. Chem.* (1991.0) **37** 2029-2037. DOI: 10.1093/clinchem/37.12.2029
13. de Jong G, van Noort WL, Feelders RA, de Jeu-Jaspars CMH, van Eijk HG. **Adaptation of transferrin protein and glycan synthesis**. *Clin. Chim. Acta* (1992.0) **212** 27-45. DOI: 10.1016/0009-8981(92)90135-D
14. Gornik O, Lauc G. **Glycosylation of serum proteins in inflammatory diseases**. *Dis. Mark.* (2008.0) **25** 267-278. DOI: 10.1155/2008/493289
15. Bergström JP, Helander A. **Influence of alcohol use, ethnicity, age, gender, BMI and smoking on the serum transferrin glycoform pattern: implications for use of carbohydrate-deficient transferrin (CDT) as alcohol biomarker**. *Clin. Chim. Acta* (2008.0) **388** 59-67. DOI: 10.1016/j.cca.2007.10.011
16. Yoshikawa K. **Determination of carbohydrate-deficient transferrin separated by lectin affinity chromatography for detecting chronic alcohol abuse**. *FEBS Lett.* (1999.0) **458** 112-116. DOI: 10.1016/S0014-5793(99)01137-0
17. Helander A, Husa A, Jeppsson JO. **Improved HPLC method for carbohydrate-deficient transferrin in serum**. *Clin. Chem.* (2003.0) **49** 1881-1890. DOI: 10.1373/clinchem.2003.023341
18. Li T. **Modulating IgG effector function by Fc glycan engineering**. *Proc. Natl Acad. Sci. USA* (2017.0) **114** 3485-3490. DOI: 10.1073/pnas.1702173114
19. Parekh RB. **Association of rheumatoid arthritis and primary osteoarthritis with changes in the glycosylation pattern of total serum IgG**. *Nature* (1985.0) **316** 452-457. DOI: 10.1038/316452a0
20. Pučić M. **High throughput isolation and glycosylation analysis of IgG-variability and heritability of the IgG glycome in three isolated human populations**. *Mol. Cell. Proteom.* (2011.0) **10** M111.010090-M111.010090. DOI: 10.1074/mcp.M111.010090
21. Krištić J. **Glycans are a novel biomarker of chronological and biological ages**. *J. Gerontol.—Ser. A Biol. Sci. Med. Sci.* (2014.0) **69** 779-789. DOI: 10.1093/gerona/glt190
22. Štambuk J. **Global variability of the human IgG glycome**. *Aging (Albany, NY)* (2020.0) **12** 1-13. PMID: 31905168
23. Gudelj I, Lauc G, Pezer M. **Immunoglobulin G glycosylation in aging and diseases**. *Cell. Immunol.* (2018.0) **333** 65-79. DOI: 10.1016/j.cellimm.2018.07.009
24. Küster B, Wheeler SF, Hunter AP, Dwek RA, Harvey DJ. **Sequencing of N-linked oligosaccharides directly from protein gels: In-gel deglycosylation followed by matrix-assisted laser desorption/ionization mass spectrometry and normal-phase high-performance liquid chromatography**. *Anal. Biochem.* (1997.0) **250** 82-101. DOI: 10.1006/abio.1997.2199
25. Rudd PM. **Oligosaccharide sequencing technology**. *Nature* (1997.0) **388** 205-207. DOI: 10.1038/40677
26. Reiding KR, Blank D, Kuijper DM, Deelder AM, Wuhrer M. **High-throughput profiling of protein N-glycosylation by MALDI-TOF-MS employing linkage-specific sialic acid esterification**. *Anal. Chem.* (2014.0) **86** 5784-5793. DOI: 10.1021/ac500335t
27. Landini A. **Genetic regulation of post-translational modification of two distinct proteins**. *Nat. Commun.* (2022.0) **13** 1586. DOI: 10.1038/s41467-022-29189-5
28. Trbojević-Akmačić I. **Chromatographic monoliths for high-throughput immunoaffinity isolation of transferrin from human plasma**. *Croat. Chem. Acta* (2016.0) **89** 203-211. DOI: 10.5562/cca2815
29. Campbell MP, Royle L, Radcliffe CM, Dwek RA, Rudd PM. **GlycoBase and autoGU: tools for HPLC-based glycan analysis**. *Bioinformatics* (2008.0) **24** 1214-1216. DOI: 10.1093/bioinformatics/btn090
30. Abrahams JL, Campbell MP, Packer NH. **Building a PGC-LC-MS N-glycan retention library and elution mapping resource**. *Glycoconj. J.* (2018.0) **35** 15-29. DOI: 10.1007/s10719-017-9793-4
31. Kaneko Y, Nimmerjahn F, Ravetch JV. **Anti-inflammatory activity of immunoglobulin G resulting from Fc sialylation**. *Science* (2006.0) **313** 670-673. DOI: 10.1126/science.1129594
32. Bondt A. **IgA N- and O-glycosylation profiling reveals no association with the pregnancy-related improvement in rheumatoid arthritis**. *Arthritis Res. Ther.* (2017.0) **19** 1-8. DOI: 10.1186/s13075-017-1367-0
33. Demus D. **Large-scale analysis of apolipoprotein ciii glycosylation by ultrahigh resolution mass spectrometry**. *Front. Chem.* (2021.0) **9** 1-9. DOI: 10.3389/fchem.2021.678883
34. Keser T. **High-throughput and site-specific N-glycosylation analysis of human alpha-1-acid glycoprotein offers a great potential for new biomarker discovery**. *Mol. Cell. Proteom.* (2021.0) **20** 0-15. DOI: 10.1074/mcp.RA120.002433
35. Tijardović M. **Fucosylated AGP glycopeptides as biomarkers of HNF1A-Maturity onset diabetes of the young**. *Diabetes Res. Clin. Pract.* (2022.0) **185** 109226. DOI: 10.1016/j.diabres.2022.109226
36. Quaranta A, Sroka-Bartnicka A, Tengstrand E, Thorsén G. **N-Glycan profile analysis of transferrin using a microfluidic compact disc and MALDI-MS**. *Anal. Bioanal. Chem.* (2016.0) **408** 4765-4776. DOI: 10.1007/s00216-016-9570-4
37. Sarrats A. **Glycosylation of liver acute-phase proteins in pancreatic cancer and chronic pancreatitis**. *Proteom.-Clin. Appl.* (2010.0) **4** 432-448. DOI: 10.1002/prca.200900150
38. Leger D, Campion B, Decottignies JP, Montreuil J, Spik G. **Physiological significance of the marked increased branching of the glycans of human serotransferrin during pregnancy**. *Biochem. J.* (1989.0) **257** 231-238. DOI: 10.1042/bj2570231
39. Kenan N, Larsson A, Axelsson O, Helander A. **Changes in transferrin glycosylation during pregnancy may lead to false-positive carbohydrate-deficient transferrin (CDT) results in testing for riskful alcohol consumption**. *Clin. Chim. Acta* (2011.0) **412** 129-133. DOI: 10.1016/j.cca.2010.09.022
40. Hu WL, Chindemi PA, Regoeczi E. **Reduced hepatic iron uptake from rat aglycotransferrin**. *Biol. Met.* (1991.0) **4** 90-94. DOI: 10.1007/BF01135384
41. Mason AB. **Expression of glycosylated and nonglycosylated human transferrin in mammalian cells. Characterization of the recombinant proteins with comparison to three commercially available transferrins**. *Biochemistry* (1993.0) **32** 5472-5479. DOI: 10.1021/bi00071a025
42. Hoefkens P, Huijskes-Heins MIE, De Jeu-Jaspars CMH, Van Noort WL, Van Eijk HG. **Influence of transferrin glycans on receptor binding and iron-donation**. *Glycoconj. J.* (1997.0) **14** 289-295. DOI: 10.1023/A:1018510309524
43. Hu WL, Chindemi PA, Regoeczi E. **In vivo behaviour of rat transferrin bearing a hybrid glycan and its interaction with macrophages**. *Biochem. Cell Biol.* (1992.0) **70** 636-642. DOI: 10.1139/o92-098
44. Gornik O, Gornik I, Gašparović V, Lauc G. **Change in transferrin sialylation is a potential prognostic marker for severity of acute pancreatitis**. *Clin. Biochem.* (2008.0) **41** 504-510. DOI: 10.1016/j.clinbiochem.2008.01.026
45. Gornik O, Gornik I, Kolednjak IZ, Lauc G. **Change of transferrin sialylation differs between mild sepsis and severe sepsis and septic shock**. *Intern. Med.* (2011.0) **50** 861-869. DOI: 10.2169/internalmedicine.50.4704
46. De Martinis M, Franceschi C, Monti D, Ginaldi L. **Inflamm-ageing and lifelong antigenic load as major determinants of ageing rate and longevity**. *FEBS Lett.* (2005.0) **579** 2035-2039. DOI: 10.1016/j.febslet.2005.02.055
47. Dall’Olio F. **N-glycomic biomarkers of biological aging and longevity: a link with inflammaging**. *Ageing Res. Rev.* (2013.0) **12** 685-698. DOI: 10.1016/j.arr.2012.02.002
48. De Jong SE. **IgG1 Fc N-glycan galactosylation as a biomarker for immune activation**. *Sci. Rep.* (2016.0) **6** 1-9. PMID: 28442746
49. Shade K-TC, Anthony RM. **Antibody glycosylation and inflammation**. *Antibodies* (2013.0) **2** 392-414. DOI: 10.3390/antib2030392
50. Dekkers G. **Decoding the human immunoglobulin G-glycan repertoire reveals a spectrum of Fc-receptor- and complement-mediated-effector activities**. *Front. Immunol.* (2017.0) **8** 877. DOI: 10.3389/fimmu.2017.00877
51. Vattepu R, Sneed SL, Anthony RM. **Sialylation as an important regulator of antibody function**. *Front. Immunol.* (2022.0) **13** 818736. DOI: 10.3389/fimmu.2022.818736
52. 52.Kavur, M. M., Lauc, G. & Pezer, M. Systems glycobiology: immunoglobulin G glycans as biomarkers and functional effectors in aging and diseases. (ed Joseph Barchi) In Comprehensive Glycoscience 2nd edn. 439–478 (Elsevier 2021).
53. Shields RL. **Lack of fucose on human IgG1 N-linked oligosaccharide improves binding to human FcγRIII and antibody-dependent cellular toxicity**. *J. Biol. Chem.* (2002.0) **277** 26733-26740. DOI: 10.1074/jbc.M202069200
54. Shinkawa T. **The absence of fucose but not the presence of galactose or bisecting N-acetylglucosamine of human IgG1 complex-type oligosaccharides shows the critical role of enhancing antibody-dependent cellular cytotoxicity**. *J. Biol. Chem.* (2003.0) **278** 3466-3473. DOI: 10.1074/jbc.M210665200
55. Karsten CM. **Anti-inflammatory activity of IgG1 mediated by Fc galactosylation and association of FcγRIIB and dectin-1**. *Nat. Med.* (2012.0) **18** 1401-1406. DOI: 10.1038/nm.2862
56. Banda NK. **Initiation of the alternative pathway of murine complement by immune complexes is dependent on N-glycans in IgG antibodies**. *Arthritis Rheum.* (2008.0) **58** 3081-3089. DOI: 10.1002/art.23865
57. Malhotra R. **Glycosylation changes of IgG associated with rheumatoid arthritis can activate complement via the mannose-binding protein**. *Nat. Med.* (1995.0) **1** 237-243. DOI: 10.1038/nm0395-237
58. Parekh R, Roitt I, Isenberg D, Dwek R, Rademacher T. **Age-related galactosylation of the N-linked oligosaccharides of human serum IgG**. *J. Exp. Med.* (1988.0) **167** 1731-1736. DOI: 10.1084/jem.167.5.1731
59. de Jong G, van Dijk JP, van Eijk HG. **The biology of transferrin**. *Clin. Chim. Acta* (1990.0) **190** 1-46. DOI: 10.1016/0009-8981(90)90278-Z
60. Debanne MT, Chindemi PA, Regoeczi E. **Binding of asialotransferrins by purified rat liver plasma membranes**. *J. Biol. Chem.* (1981.0) **256** 4929-4933. DOI: 10.1016/S0021-9258(19)69346-4
61. Rudan I. **10 001 Dalmatians:’ Croatia launches Its National Biobank**. *Croat. Med. J.* (2009.0) **50** 4-6. DOI: 10.3325/cmj.2009.50.4
62. Campbell H. **Effects of genome-wide heterozygosity on a range of biomedically relevant human quantitative traits**. *Hum. Mol. Genet.* (2007.0) **16** 233-241. DOI: 10.1093/hmg/ddl473
63. Kerr SM. **An actionable KCNH2 Long QT Syndrome variant detected by sequence and haplotype analysis in a population research cohort**. *Sci. Rep.* (2019.0) **9** 1-11. DOI: 10.1038/s41598-019-47436-6
64. Tyanova S, Temu T, Cox J. **The MaxQuant computational platform for mass spectrometry-based shotgun proteomics**. *Nat. Protoc.* (2016.0) **11** 2301-2319. DOI: 10.1038/nprot.2016.136
65. 65.Agakova, A., Vučković, F., Klarić, L., Lauc, G. & Agakov, F. Automated Integration of a UPLC Glycomic Profile. in High-Throughput Glycomics and Glycoproteomics. Methods in Molecular Biology (eds Lauc, G. & Wuhrer, M.) 217–233 (Humana Press, 2017).
66. Cooper CA, Gasteiger E, Packer NH. **GlycoMod—a software tool for determining glycosylation compositions from mass spectrometric data**. *Proteomics* (2001.0) **1** 340-349. DOI: 10.1002/1615-9861(200102)1:2<340::AID-PROT340>3.0.CO;2-B
67. Varki A. **Symbol nomenclature for graphical representations of glycans**. *Glycobiology* (2015.0) **25** 1323-1324. DOI: 10.1093/glycob/cwv091
|
---
title: Combining machine learning and structure-based approaches to develop oncogene
PIM kinase inhibitors
authors:
- Haifa Almukadi
- Gada Ali Jadkarim
- Arif Mohammed
- Majid Almansouri
- Nasreen Sultana
- Noor Ahmad Shaik
- Babajan Banaganapalli
journal: Frontiers in Chemistry
year: 2023
pmcid: PMC10036574
doi: 10.3389/fchem.2023.1137444
license: CC BY 4.0
---
# Combining machine learning and structure-based approaches to develop oncogene PIM kinase inhibitors
## Abstract
Introduction: PIM kinases are targets for therapeutic intervention since they are associated with a number of malignancies by boosting cell survival and proliferation. Over the past years, the rate of new PIM inhibitors discovery has increased significantly, however, new generation of potent molecules with the right pharmacologic profiles were in demand that can probably lead to the development of Pim kinase inhibitors that are effective against human cancer.
Method: In the current study, a machine learning and structure based approaches were used to generate novel and effective chemical therapeutics for PIM-1 kinase. Four different machine learning methods, namely, support vector machine, random forest, k-nearest neighbour and XGBoost have been used for the development of models. Total, 54 Descriptors have been selected using the Boruta method.
Results: SVM, Random Forest and XGBoost shows better performance as compared to k-NN. An ensemble approach was implemented and, finally, four potential molecules (CHEMBL303779, CHEMBL690270, MHC07198, and CHEMBL748285) were found to be effective for the modulation of PIM-1 activity. Molecular docking and molecular dynamic simulation corroborated the potentiality of the selected molecules. The molecular dynamics (MD) simulation study indicated the stability between protein and ligands.
Discussion: Our findings suggest that the selected models are robust and can be potentially useful for facilitating the discovery against PIM kinase.
## Introduction
Proto-oncogene PIM-1 kinase is a member of the serine/threonine protein kinase family (Narlik-Grassow et al., 2014). PIM kinases are involved in cancer cell survival, proliferation, and tumor growth and are overexpressed in a number of hematological malignancies, in addition to solid cancers such as pancreatic, prostate, and colon cancers (Amson et al., 1989; Li et al., 2006; Nawijn et al., 2011). PIM-1, PIM-2, and PIM-3 are the three highly homologous genes that make up the PIM family. This kinase family is highly homologous with the kinase domains, especially in the linker region and the ATP-binding sites (Warfel and Kraft, 2015). These enzymes are constitutively expressed in tumors and are becoming more widely acknowledged as crucial survival signal mediators in malignancies, stress responses, and neurological development. PIM-1 kinase is a genuine oncogene that is the focus of drug development research initiatives since it has been linked to the emergence of leukemias, lymphomas, and prostate cancer (Li et al., 2011; Le et al., 2015; Huang et al., 2022). PIM kinases regulate the network of signaling pathways that are critical for tumorigenesis and development, making them attractive drug targets (Drygin et al., 2012; Tursynbay et al., 2016).
The crystal structure of PIM-1 has been published by numerous independent groups in both the presence and the absence of its inhibitors (Wang et al., 2013; Nonga et al., 2021). Structural research on PIM-1 has found a number of distinctive characteristics that set it apart from other kinases with known structures. The catalytic domain of PIM-1 kinase spans amino acid positions 38 to 290 and includes a conserved glycine loop motif at positions 45 to 50, phosphate-binding sites at positions 44 to 52 and 67, and a proton acceptor site at position 167. The hunt for small-molecule ATP-competitive inhibitors with the potential to develop into novel targeted oncology treatments has been sparked by the involvement of the PIM kinases in important cancer hallmarks. The majority of PIM-1 inhibitors have failed to evolve into a new anticancer medication despite having excellent biochemical potency, largely because they were found to have subpar pharmacological qualities (Dakin et al., 2012; Drygin et al., 2012; Ogawa et al., 2012; Vivek et al., 2017; Zhao et al., 2017; Park et al., 2021). Due to their therapeutic value in cancer, the discovery of PIM-1 inhibitors has increasingly attracted much attention in past few years. The rate of new PIM inhibitor discovery has increased significantly, and there has been demand for a new generation of potent molecules with the right pharmacologic profiles that can probably lead to the development of PIM kinase inhibitors that are effective against human cancer.
This work was undertaken to develop machine learning-based classification models to identify a new class of PIM-1 inhibitors. Under this approach, four different machine learning methods were applied to develop the classification models. These models were further used to screen chemical libraries to retrieve novel potent PIM-1 inhibitors. In addition, we also carried out molecular docking and molecular dynamics simulations to investigate the interaction and stability within the catalytic site of PIM-1 kinase. This multistage approach allows us to screen large chemical libraries efficiently and effectively in a reasonable time. Moreover, it can also help us identify novel chemical scaffolds for potent PIM-1 inhibitors.
## Data collection and model building
All chemical compounds with activity against PIM-1 were collected from the literature and the ChEMBL database (Gaulton et al., 2012). Inorganic and duplicate compounds were removed from the list. Generally, compounds with IC50 ≤ 10 μM will likely be “active,” predicting a large number of active molecules. However, such a high fraction of active compounds cannot be expected from any experimental platform. Therefore, in order to make the most efficient use of costly experimental validation, the optimal model should identify compounds with affinity higher than 10 μM. The higher the value, the higher the drug dose needed to achieve the required potency and, thus, the higher the chance of “off-target” activity. To address this issue, we chose to set the decision boundary at IC50 ≤ 1 μM for active molecules. Molecular descriptors were calculated using the PaDEL software (Yap, 2011). A two-tier selection procedure was applied to select the best descriptors. First, we randomly selected one descriptor from a pair showing >0.85 correlation. Second, descriptors were reduced using the Boruta method (Kursa et al., 2010). We used four different machine learning methods, namely, Support Vector Machine (SVM) (Mitchell, 1997), random forest (Breiman, 2001), Extreme Gradient Boosting (XGBoost) (Chen and Guestrin, 2016), and kappa nearest neighbor (kNN) (Voulgaris and Magoulas, 2008), to build the classification models. All the classification experiments and calculations were conducted using the R.3.0.2 environment (http://www.R-project.org/) and Python (http://www.python.org/) platform. The compounds used in training and test sets are given in Supplementary Tables S1 and S2, respectively.
## Model validation
A receiver operating characteristic (ROC) plot and area under the curve (AUC) were used to assess the performance of the model (Hanley and McNeil, 1983; Park et al., 2004). In Table 1, the terms precision (Eq. 1), recall (Eq. 2), accuracy (Eq. 3), and F1 score (Eq. 4) are defined along with their relationships to the statistical performance calculations used to assess the quality of the model. Precision=True positiveTrue positive+False Negative, [1] Recall=True positiveTrue positive+False Negative, [2] Accuracy=TP+TNTP+TN+FP+FN, [3] F1=1.Precision X RecallPrecision+Recall. [ 4]
**TABLE 1**
| Method | Descriptors | Precision | Recall | Accuracy (Q) | F1 score | AUC |
| --- | --- | --- | --- | --- | --- | --- |
| XGBoost | All descriptors | 0.82 | 0.81 | 0.83 | 0.97 | 0.89 |
| XGBoost | Boruta | 0.81 | 0.79 | 0.85 | 0.8 | 0.88 |
| XGBoost | MACCS | 0.8 | 0.76 | 0.81 | 0.77 | 0.92 |
| Random forest | All descriptors | 0.85 | 0.81 | 0.86 | 0.98 | 0.91 |
| Random forest | Boruta | 0.86 | 0.81 | 0.87 | 0.83 | 0.92 |
| Random forest | MACCS | 0.8 | 0.76 | 0.82 | 0.78 | 0.9 |
| SVM | All descriptors | 0.74 | 0.73 | 0.78 | 0.86 | 0.83 |
| SVM | Boruta | 0.75 | 0.72 | 0.78 | 0.71 | 0.82 |
| SVM | MACCS | 0.7 | 0.73 | 0.7 | 0.69 | 0.82 |
| kNN | All descriptors | 0.77 | 0.75 | 0.8 | 0.75 | 0.84 |
| kNN | Boruta | 0.72 | 0.67 | 0.75 | 0.68 | 0.78 |
| kNN | MACCS | 0.81 | 0.76 | 0.82 | 0.77 | 0.82 |
## Applicability domain
In order to highlight the region of the chemical space that contains the chemicals for which the model is expected to make accurate predictions, a well-validated predictive model needs to have a defined applicability domain (AD) (Rakhimbekova et al., 2020). Any predictive model must verify its constraints in terms of its structural domain and response space. As a result, determining a model’s AD and evaluating the accuracy of its predictions are both challenging tasks. These QSAR models typically use the training set to cover a certain chemical space. The model’s predictions are accurate if any query compound falls within this definition of AD. If not, the prediction might not conform to the model’s presumptions. Principal component analysis (PCA) (Sushko et al., 2010) has been employed in our work to define the AD of the compounds used in this study.
## Y-randomization
To test the robustness of the proposed models, y-randomization was applied. This technique involves randomly mixing up the values of the target variable in the training set (Rücker et al., 2007; Lipiński and Szurmak, 2017). The same parameters used in the initial model are then applied to a new prediction generated with the scrambled data. Every estimate of the model’s accuracy was recorded. In total, $50\%$ of the compounds in the training set were resampled and used in a 500-run y-randomization test.
## Similarity calculations
The Tanimoto coefficient (Tc) (Eq. 5) was computed using MACCS-166 fingerprints to quantify chemical similarity. The active and inactive chemicals in the training set were compared against false and true positive compounds in systematic pairwise similarity computations. Tc=CA+B−C. [5]
## Substructure analyses
Molecular substructures related to PIM activity were analyzed using the distribution of MACSS fingerprints in active and inactive compounds (Eq. 6). Frequency=∑iNFP1|0NX100. [ 6]
## Analysis of probability scores
Additionally, the probability scores of the developed classification models were examined. *In* general, a molecule is defined as inactive if its probability score is lower than 0.5, while a compound with a probability score of 0.5 is considered active (Ponzoni et al., 2019). The more this score approaches 1, the more confident we are in our prediction. Here, we examined the probability score distributions for TP (true positive), TN (true negative), FP (false positive), and FN (false negative) results.
## Chemical database screening
The developed models were used to screen the hits against PIM-1. The NCI library and Maybridge databases were used for virtual screening. The National Cancer Institute maintains a repository of compounds that have been evaluated as potential anticancer agents. These compounds represent unique structural diversity based on synthetic and natural products. The Maybridge library consists of a highly diverse set of over 53,000 lead-like compounds. Maybridge Hit-to-Lead was designed for medicinal chemistry, allowing SAR development and hit-to-lead optimization. The following filters were used to select the hits: Filter 1: compounds predicted to be active by all the validated models; Filter 2: compounds having a probability score; and Filter 3: compounds falling within the chemical space of the training set. These compounds were further processed for molecular docking, followed by molecular dynamics simulations. Finally, compounds with the best affinity and conformance within the active site were selected and analyzed.
## Molecular docking
Molecular docking was implemented to identify the best physical confirmation of inhibitor binding within the active site of PIM-1 kinase. The PIM kinase enzyme structure was taken from the Protein Data Bank (PDB ID: 5KZI). All of the docking simulations for this work were performed using AutoDock Vina (Trott and OlsonAutoDock, 2009) with a 1 spacing, default exhaustiveness, and full ligand flexibility. The grid resolution was internally set to 1Å. We set the number of binding modes to 10 and exhaustiveness to 8. A cubical grid of size 60 × 60 × 60 size with 0.375 Å spacing was used around the active sites of the protein. To acquire the structure in the PDBQT format, polar hydrogen atoms were added using AutoDock Tools 92.
## Molecular dynamics simulations
Selected best compounds were further subjected to molecular dynamics (MD) simulations using Groningen Machine for Chemical Simulations (GROMACS v5.1.5) (Pronk et al., 2013). The parameters and coordinate files for PIM-1 kinase and selected potential hit compounds were generated using the CHARMM27 forcefield in GROMACS and PRODRG, respectively. The TIP3P water model was used for each simulation system, which was neutralized by the addition of Na+ ions in a dodecahedron periodic box. Energy minimization was performed for 50,000 nstep using the steepest descent algorithm to avoid steric clashes. Equilibration of each system was performed in two stages: the first phase was carried out with a constant number of particles, volume, and temperature (NVT) ensemble for 500 ps at 300 K, using the V-rescale thermostat (Bussi et al., 2007); and in the second phase, the pressure of each system was equilibrated for 500 ps at a constant number of particles, pressure, and temperature (NPT) at 1 bar using a Parrinello–Rahman barostat (Parrinello and Rahman, 1981). Each equilibrated system was simulated for 30 ns under periodic boundary conditions to avoid edge effects. Electrostatic interactions were handled by the particle mesh Ewald (PME) method, while the heavy-atom bonds were restrained using the LINCS algorithm.
## Model development and evaluation
In total, 54 descriptors from the set of 240 were eventually selected using the Boruta method (Supplementary Table S3). All these descriptors belonged to 12 different classes. The descriptors include autocorrelation, information content, atom-type electrotopological state, Burden modified eigenvalues, molecular distance edge, carbon type, and molecular linear free energy relation. The models were trained using four machine learning methods (SVM, random forest, XGBoost, and kNN). Evaluation metrics for the developed models are given in Table 1, including accuracy, recall, precision, F1 Score (a measure of a model’s accuracy, which takes into account both precision and recall), and Area Under the Curve (AUC) values. SVM, random forest, and XGBoost performed than kNN according to these metrics in combination with the selected descriptor set. Among the three, random forest achieved the best accuracy, at 0.87 for the test set (with selected descriptors), as compared to SVM (0.78) and XGBoost (0.84). In addition, these models also had significant AUC values (Figure 1).
**FIGURE 1:** *ROC curves of the models based on four machine learning approaches for (A) all descriptors; (B) selected descriptors (Boruta method); (C) MACCS fingerprints.*
## Applicability domain and y-randomization
An applicability domain (AD) analysis was performed to check the reliability of the generated classification models. Figure 2 shows a scatter plot of the PC1 and PC2 coordinates derived from the set of selected PIM-1 compound descriptors. The training and test compounds share similar PC1 and PC2 coordinates, suggesting that predictions were within the applicability domain (AD) of both the training and test sets. To check the robustness of the developed models, y-randomization tests were performed (Rücker et al., 2007). Y-randomization test accuracies were found to be lower, and none of the random trials achieved higher scores than our main models (Figure 3). The average accuracy across all randomly generated models were found to be less than 0.58. This confirms that the selected models are robust and reliable and were not generated by chance correlations. A pairwise comparison of the compounds in each cluster was found to reflect reasonable Tanimoto coefficient similarities between them.
**FIGURE 2:** *Applicability domain plot based on principal component analysis (PCA) for (A) training set and (B) test set.* **FIGURE 3:** *Y-randomization models. (A) Accuracy; (B) AUC values. A total of 500 y-randomization runs were performed.*
## Probability analyses
Probability scores of the selected models, reflecting the probability of belonging to each class, were also analyzed. It is known that a compound with a probability score of ≥0.5 is classified as active, whereas a molecule with a probability below <0.5 is classified as inactive. As this score approaches 1, the higher the value, the higher the model’s confidence in the prediction is (Minerali et al., 2020; Esposito et al., 2021). In our study, we analyzed the distribution of probability scores among TN (true negative), FP (false positive), TP (true positive), and FN (false negative) results. For the SVM model, compounds with a probability score of more than 0.80 (an average value) were more likely to be active, whereas compounds with a probability score of 0.36 were more likely to be inactive. In the case of the random forest model, a compound with a probability score of more than 0.87 was more likely to be active, whereas a compound with a probability score of 0.24 was more likely to be inactive. Random forest achieved values of 0.95 and 0.11 for active and inactive compounds, respectively, indicating greater success in predicting compound activity with the desired probability score (Supplementary Figure S1). False positive compounds were predicted with probability scores of 0.63, 0.65, and 0.69 for the random forest, XGBoost, and SVM models, respectively. In contrast, false negative compounds were found to have probability scores of 0.31, 0.42, and 0.14 for the random forest, SVM, and XGBoost models, respectively. Each predictive model’s effectiveness in the early recognition of hits was visually evaluated using a cumulative gain plot (Table 2). The cumulative gain curve is an evaluation curve that evaluates the model’s performance and contrasts the outcomes with a random selection. It displays the percentage of targets identified when taking into account a particular portion of the population that has the highest likelihood of being a target based on the model. The comparison showed that the XGBoost and random forest methods performed better than SVM and kNN in terms of early recognition of hits (Figure 4).
## MACCS fingerprint analyses
Molecular substructures related to the PIM-1 activity of the compounds can be identified by analyzing the bits in the MACCS fingerprints. We analyzed the MACCS fingerprints showing a reasonable difference between active and inactive compounds (Supplementary Table S4). The occurrence of MACCS fingerprints differed significantly between active and inactive compounds in the training dataset, suggesting that the substructures represented by these features may be closely related to PIM-1 activity. Descriptions and the number of occurrences of these substructures are listed in Supplementary Table S4. It was found that MACCS38, MACCS52, MACCS92, MACCS98, MACCS107, MACCSFP142, etc. are prevalent in active molecules. This is consistent with previous studies, which shows that compounds with such functional groups have therapeutic potential against PIM kinase (Tsuganezawa et al., 2012; El-Hawary et al., 2018; Park et al., 2021).
## Database screening and molecular interaction analyses
The NCI and Maybridge databases were used to screen the potential hits from validated models. Commonly predicted active compounds with high probability scores were selected and further filtered out within the applicability domain (AD) of the training set. These compounds were further subjected to molecular docking simulation (Table 2). Finally, four compounds (CHEMBL303779, CHEMBL690270, CHEMBL748285, and N-[(1-ethylbenzimidazol-2-yl)methyl]-3-(4-methoxyphenyl)-1H-pyrazole-4-carboxamide (EBM-MPC)) were observed to have reasonable binding affinity and stable interaction with the catalytic residues in the active site (Table.3 and Figure 5). A literature survey revealed that Leu44, Lys67, Glu121, and Asp186 are crucial for the interaction of inhibitors (Tsuganezawa et al., 2012; El-Hawary et al., 2018; Park et al., 2021). It can be observed in Figure 4 that CHEMBL690270, CHEMBL303779, and EBM-MPC form hydrogen bond interactions with Lys67 and hydrophobic interactions with Asp186 (Figure 6). In contrast, CHEMBL748285 forms hydrogen bonds with Asp186 (Figure 6). The quinazoline ring of compounds was involved in multiple p–alkyl interactions. In addition, a number of hydrophobic contacts, particularly residues Leu44, Gly47, Phe49, Ile104, and Leu120, stabilize interaction with hits. PIM inhibitors fall into two broad categories: ATP mimetics, which form hydrogen bonds with the glutamate residue that serves as the hinge (Glu121), and non-ATP mimetics, which bind far from the hinge or interact with the hinge through hydrophobic interactions with a number of residues in the specific hydrophobic pocket that serves as the hinge environment (El-Hawary et al., 2018; Park et al., 2021). The Tanimoto coefficient (Tc) similarity score of these selected hits was found to be ≤ 0.5 with high-activity compounds (Figure 5B).
## MD simulation analyses
By analyzing 100-ns MD trajectories, the structural changes to PIM-1 upon inhibitor binding were studied. We examined the RMSD of the protein backbone and the RMSF of the protein’s alpha-carbon atoms. As shown in Figure 6, all the systems exhibited stability throughout the 100-ns simulation. The average RMSD value for all four systems was observed to be below 0.31 nm, which indicated that simulated complexes displayed RMSD values below the threshold. The average RMSD values further showed that the CHEMBL690270 PIM-1 complex displayed less deviation (0.26 nm), whereas CHEMBL303779 and CHEMBL748285 demonstrated similar average values of 0.34 nm (Figure 7A). RMSF is a significant value, used to characterize each residue’s fluctuation rate upon ligand binding. It was observed that the inhibitor binding residues (Leu44, Phe49, Lys67, Glu121, and Asp186) did not fluctuate significantly (Figure 7B).
**FIGURE 7:** *Molecular dynamics simulation analyses. (A) RMSD plot; (B) RMSF plot; (C) hydrogen plot for selected compounds to illustrate protein–ligand stability during a 100-ns simulation.*
## Discussion
This study was designed with the aim of building a classification model to predict potential hits for PIM-1 kinase. Four different machine learning approaches were used to build the models. Our proposed models performed well in terms of accuracy, F1 score, precision, and recall. We used the area under the receiver operating characteristic curve approach to compare classifiers. The ROC curve is a graphical representation that contrasts a classifier’s true positive rate and false positive rate at various threshold levels. The area under this curve, or AUC, is thus a useful metric for assessing machine learning algorithms, since it shows the degree of separability (Parrinello and Rahman, 1981). A ROC curve with a higher AUC value implies greater sensitivity in identifying active molecules and specificity in rejecting inactive compounds (Figure 1). In addition, our study also distinguished and ranked the top 18 variables, including 2D autocorrelation, Burden modified eigenvalues, and topological charge. These descriptors have the capacity to distinguish between active and inactive compounds.
QSAR Classification models must undergo an extensive validation process, and the reliability of those models must be objectively determined. The OECD guidelines state that a model must have a clearly defined domain of applicability (Dwyer et al., 2013). Additionally, the dataset for such models with a defined AD should cover a broad chemical space and a diverse range of structural types. The AD of PIM kinase inhibitors has been defined using a principal component analysis-based approach for model development. A sufficient level of assurance in the produced models can be seen in the 2D plot obtained from the first two PCs, which represents the training and test set compounds, illustrating their structural variety and similar chemical space (Figure 2). To assess the likelihood of a random correlation for a chosen descriptor, y-randomization was utilized. This technique is used to assess the reliability or robustness of QSAR models and is recognized as one of the most effective validation processes (Rücker et al., 2007). By comparing a developed model’s performance to the average measure of 500 random models, which are obtained by using the same parameters as those used to construct the original model along with a randomly scrambled target variable class, the statistical significance of the developed model can be examined. The results of the y-randomization tests demonstrated that the models created for this study did not exhibit these connections by chance and that a true structure–activity relationship existed (Figure 3).
Fingerprints describe the molecular makeup of a compound. The description of each molecule is given as a string of binary substructures called a fingerprint. The corresponding fingerprint bit is set to 1 if the specified substructure is present in the given molecule; otherwise, it is set to 0. In our study, we used MACCS fingerprints to represent the presence of structures and their representative substructures in active and inactive compounds. These molecules contained MACCS65, MACCS128, and MACCS90. Compounds having such substructures were found to exhibit reasonable levels of activity toward PIM-1 kinase (Akué-Gédu et al., 2010; Dwyer et al., 2013; Hu et al., 2015; Wurz et al., 2015; Li et al., 2016).
To identify potent PIM-1 inhibitors, virtual screening of the NCI and Maybridge databases was performed using the validated models. To gain structural insight relevant to the inhibitory activities of the newly identified inhibitors, their binding modes in the binding site of PIM-1 were examined. Figure 6 shows the most stable binding configurations of selected four compounds derived via docking simulations with potent inhibitors. These compounds appear to be accommodated in a similar way in the binding site of PIM1 (Xia et al., 2009; Abdelaziz et al., 2018; Ibrahim et al., 2022). The necessity of the interactions with the hinge region and Gly-loop residues (Qian et al., 2005; Pogacic et al., 2007; Tsuganezawa et al., 2012; Casuscelli et al., 2013; Fan et al., 2016; Abdelaziz et al., 2018; Bima et al., 2022; Ibrahim et al., 2022; Shaik et al., 2022) for tight binding to PIM-1 was also implicated with potent inhibitors (Xia et al., 2009; Ibrahim et al., 2022). Moreover, these four compounds can also interact with the activation loop including the Asp186 residue. A hydrophobic cavity is formed among the Ala65, Ile104, Phe187, Val52, Lys67, and Leu120 residues, and this maintains molecular stability through various hydrophobic forces. Similar interactions have also been noted in earlier published investigations, highlighting the significance of these amino acids for the assembly of PIM-1 inhibitor complexes (Tsuganezawa et al., 2012; El-Hawary et al., 2018; Park et al., 2021). Residue Lys67 is known to be significant in stabilizing the interaction with the compound and to play an important role in the catalytic activity of PIM-1 (Pogacic et al., 2007; Fan et al., 2016). In our study, we found that all four compounds interacted with Lys67, either with hydrogen bonds or through hydrophobic contact. Compared to the currently available PIM-1 inhibitors, the four selected compounds exhibit low Tanimoto coefficient (Tc) similarities, highlighting their structural novelty and druggability. Moreover, all these compounds were found to have a similar chemical boundary (Figure 5). Therefore, models constructed using these selected descriptors have good interpretability and reliability.
Molecular docking studies were conducted to analyze the binding mode of inhibitors at the PIM-1 catalytic domain. Notably, these inhibitors are positioned in the active site, between the residues Leu44, Gly45, Phe49, Lys67, Ile104, Lys67, Leu172, Leu174, and Asp186 (Table 3). These inhibitors were found to have stabilized the complex with hydrogen and hydrophobic interactions with residues, namely, Lys67 and Asp186. This is consistent with earlier research that revealed that these amino acid residues were essential for the catalytic activity of PIM-1 kinase (Qian et al., 2005; Banaganapalli et al., 2016; Shaik et al., 2021; Bima et al., 2022; Shaik et al., 2022).
Although molecular docking has strong computational capabilities, its predictions of the shape of the protein–ligand binding are frequently inaccurate. Thus, in this study, we performed 100-ns MD simulations to test the stability of the chosen compounds in the PIM-1 binding pocket. It was determined that selected compounds remained stable in the binding pocket, as analyzed through the RMSD, RMSF, and hydrogen bonds. Most notably, stable hydrogen bonds with the residues Lys67 and Asp186 were observed in the complexes with the compounds (namely, CHEMBL748285, and CHEMBL690270).
## Conclusion
The PIM kinase family has become a focus of attention in drug discovery. In particular, the search for inhibitors simultaneously targeting PIM-1 isoforms is of great interest because it opens new horizons toward the discovery of new chemicals capable of therapeutically modulating many biochemical pathways involved in the emergence and development of various cancers. In the present study, ensemble learning based on four different machine learning approaches, together with molecular docking and molecular dynamics simulation, was successfully utilized to identify novel scaffold inhibitors against PIM kinase. By combining machine learning and structure-based approaches, it was possible to evaluate the quantitative contributions of the molecules to the activity. This permitted the guided design of four new molecules, predicted to be potential PIM-1 inhibitors. The molecular docking analyses showed that the active inhibitors were able to interact with the amino acids (Lys67, Asp186, Leu44, Glu171, etc.) crucial for catalytic activity of PIM kinase. The interactions were found to be stable, as investigated through 100-ns molecular dynamics simulation.
## Data availability statement
The original contributions presented in the study are included in the article/Supplementary Material; further inquiries can be directed to the corresponding authors.
## Author contributions
HA, NS, and BB: conceptualization; HA, GJ, and BB: data curation; BB and GJ: formal analysis; HA: funding acquisition; HA, GJ, and BB: methodology; HA: project administration; BB, NS, and HA: resources; BB: software; NS: supervision; HA and NS: validation; BB: visualization; HA, BB, AM, MA, NS, and GJ: writing—original draft and review.
## Conflict of interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Supplementary material
The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fchem.2023.1137444/full#supplementary-material
## References
1. Abdelaziz M. E., M El-Miligy M., Fahmy S. M., Mahran M. A., A Hazzaa A.. **Design, synthesis and docking study of pyridine and thieno[2, 3-b] pyridine derivatives as anticancer PIM-1 kinase inhibitors**. *Bioorg Chem.* (2018) **80** 674-692. DOI: 10.1016/j.bioorg.2018.07.024
2. Akué-Gédu R., Nauton L., Théry V., Bain J., Cohen P., Anizon F.. **Synthesis, Pim kinase inhibitory potencies and**. *Bioorg Med. Chem.* (2010) **18** 6865-6873. DOI: 10.1016/j.bmc.2010.07.036
3. Amson R., Sigaux F., Przedborski S., Flandrin G., Givol D., Telerman A.. **The human protooncogene product p33pim is expressed during fetal hematopoiesis and in diverse leukemias**. *Proc. Natl. Acad. Sci. U. S. A.* (1989) **86** 8857-8861. PMID: 2682662
4. Banaganapalli B., Mohammed K., Khan I. A., Al-Aama J. Y., Elango R., Shaik N. A.. **A computational protein phenotype prediction approach to analyze the deleterious mutations of human MED12 gene**. *J. Cell Biochem.* (2016) **117** 2023-2035. DOI: 10.1002/jcb.25499
5. Bima A. I. H., Elsamanoudy A. Z., Albaqami W. F., Khan Z., Parambath S. V., Al-Rayes N.. **Integrative system biology and mathematical modeling of genetic networks identifies shared biomarkers for obesity and diabetes**. *Math. Biosci. Eng.* (2022) **19** 2310-2329. DOI: 10.3934/mbe.2022107
6. Breiman L.. **Random forests**. *Mach. Learn.* (2001) **45** 5-32. DOI: 10.1023/A:1010933404324
7. Bussi G., Donadio D., Parrinell M.. **Canonical sampling through velocity rescaling**. *J. Chem. Phys.* (2007) **126** 014101. DOI: 10.1063/1.2408420
8. Casuscelli F., Ardini E., Avanzi N., Casale E., Cervi G., D'Anello M.. **Discovery and optimization of pyrrolo[1, 2-a]pyrazinones leads to novel and selective inhibitors of PIM kinases**. *Bioorg Med. Chem.* (2013) **21** 7364-7380. DOI: 10.1016/j.bmc.2013.09.054
9. Chen T., Guestrin C.. **XGBoost: A scalable tree boosting system**. (2016)
10. Dakin L. A., Block M. H., Chen H., Code E., Dowling J. E., Feng X.. **Discovery of novel benzylidene-1, 3-thiazolidine-2, 4-diones as potent and selective inhibitors of the PIM-1, PIM-2, and PIM-3 protein kinases**. *Bioorg Med. Chem. Lett.* (2012) **22** 4599-4604. DOI: 10.1016/j.bmcl.2012.05.098
11. Drygin D., Haddach M., Pierre F., Ryckman D. M.. **Potential use of selective and nonselective pim kinase inhibitors for cancer therapy**. *J. Med. Chem.* (2012) **55** 8199-8208. DOI: 10.1021/jm3009234
12. Dwyer M. P., Keertika K., Paruch K., Alvarez C., Labroli M., Poker C.. **Discovery of pyrazolo[1, 5-a]pyrimidine-based pim inhibitors: A template-based approach**. *Bioorg Med. Chem. Lett.* (2013) **23** 6178-6182. DOI: 10.1016/j.bmcl.2013.08.110
13. El-Hawary S. S., Sayed A. M., Mohammed R., Khanfar M. A., Rateb M. E., Mohammed T. A.. **New pim-1 kinase inhibitor from the Co-culture of two sponge-associated actinomycetes**. *Front. Chem.* (2018) **6** 538. DOI: 10.3389/fchem.2018.00538
14. Esposito C., Landrum G. A., Schneider N., Stiefl N., Riniker S.. **GHOST: Adjusting the decision threshold to handle imbalanced data in machine learning**. *J. Chem. Inf. Model* (2021) **61** 2623-2640. DOI: 10.1021/acs.jcim.1c00160
15. Fan Y. B., Li K., Huang M., Cao Y., Li Y., Jin S. Y.. **Design and synthesis of substituted pyrido[3, 2-d]-1, 2, 3-triazines as potential Pim-1 inhibitors**. *Bioorg Med. Chem. Lett.* (2016) **26** 1224-1228. DOI: 10.1016/j.bmcl.2016.01.032
16. Gaulton A., Bellis L. J., Bento A. P., Chambers J., Davies M., Hersey A.. **ChEMBL: A large-scale bioactivity database for drug discovery**. *Nucleic Acids Res.* (2012) **40** D1100-D1107. DOI: 10.1093/nar/gkr777
17. Hanley J. A., McNeil B. J.. **A method of comparing the areas under receiver operating characteristic curves derived from the same cases**. *Radiology* (1983) **148** 839-843. DOI: 10.1148/radiology.148.3.6878708
18. Hu H., Wang X., Chan G. K., Chang J. H., Do S., Drummond J.. **Discovery of 3, 5-substituted 6-azaindazoles as potent pan-Pim inhibitors**. *Bioorg Med. Chem. Lett.* (2015) **25** 5258-5264. DOI: 10.1016/j.bmcl.2015.09.052
19. Huang J., Yuan Y., Zhu X., Li G., Xu Y., Chen W.. **Identification of pim-1 kinase inhibitors by pharmacophore model, molecular docking-based virtual screening, and biological evaluation**. *Curr. Comput. Aided. Drug. Des.* (2022) **18** 240-246. DOI: 10.2174/1573409918666220427120524
20. Ibrahim M. H., Harras M. F., Mostafa S. K., Mohyeldin S. M., Kamaly O., Altwaijry N.. **Development of novel cyanopyridines as PIM-1 kinase inhibitors with potent anti-prostate cancer activity: Synthesis, biological evaluation, nanoparticles formulation and molecular dynamics simulation**. *Bioorg Chem.* (2022) **129** 106122. DOI: 10.1016/j.bioorg.2022.106122
21. Kursa M. B., Jankowski A., Rudnicki W. R.. **Boruta—a system for feature selection**. *Fundam. Inf.* (2010) **101** 271-285. DOI: 10.3233/fi-2010-288
22. Le B. T., Kumarasiri M., Adams J. R., Yu M., Milne R., Sykes M. J.. **Targeting pim kinases for cancer treatment: Opportunities and challenges**. *Future. Med. Chem.* (2015) **7** 35-53. DOI: 10.4155/fmc.14.145
23. Li J., Loveland B. E., Xing P. X.. **Anti-Pim-1 mAb inhibits activation and proliferation of T lymphocytes and prolongs mouse skin allograft survival**. *Cell Immunol.* (2011) **272** 87-93. DOI: 10.1016/j.cellimm.2011.09.002
24. Li K., Li Y., Zhou D., Fan Y., Guo H., Ma T.. **Synthesis and biological evaluation of quinoline derivatives as potential anti-prostate cancer agents and Pim-1 kinase inhibitors**. *Bioorg Med. Chem.* (2016) **24** 1889-1897. DOI: 10.1016/j.bmc.2016.03.016
25. Li Y. Y., Popivanova B. K., Nagai Y., Ishikura H., Fujii C., Mukaida N.. **Pim-3 a proto-oncogene with serine/threonine kinase activity, is aberrantly expressed in human pancreatic cancer and phosphorylates bad to block bad-mediated apoptosis in human pancreatic cancer cell lines**. *Cancer Res.* (2006) **66** 6741-6747. DOI: 10.1158/0008-5472.can-05-4272
26. Lipiński P. F. J., Szurmak P.. **SCRAMBLE’N’GAMBLE: A tool for fast and facile generation of random data for statistical evaluation of QSAR models**. *Chem. Pap.* (2017) **71** 2217-2232. DOI: 10.1007/s11696-017-0215-7
27. Minerali E., Foil D. H., Zorn K. M., Lane T. R., Ekins S.. **Comparing machine learning algorithms for predicting drug-induced liver injury (DILI)**. *Mol. Pharm.* (2020) **17** 2628-2637. DOI: 10.1021/acs.molpharmaceut.0c00326
28. Mitchell T.. *Machine learning* (1997)
29. Narlik-Grassow M., Blanco-Aparicio C., Carnero A.. **The PIM family of serine/threonine kinases in cancer**. *Med. Res. Rev.* (2014) **34** 136-159. PMID: 23576269
30. Nawijn M., Alendar A., Berns A.. **For better or for worse: The role of pim oncogenes in tumorigenesis**. *Nat. Rev. Cancer* (2011) **11** 23-34. DOI: 10.1038/nrc2986
31. Nonga O. E., Lavogina D., Enkvist E., Kestav K., Chaikuad A., Dixon-Clarke S. E.. **Crystal structure-guided design of bisubstrate inhibitors and photoluminescent probes for protein kinases of the PIM family**. *Molecules* (2021) **26** 4353. DOI: 10.3390/molecules26144353
32. Ogawa N., Yuki H., Tanaka A.. **Insights from Pim1 structure for anti-cancer drug design**. *Expert Opin. Drug Discov.* (2012) **7** 1177-1192. DOI: 10.1517/17460441.2012.727394
33. Park H., Jeon J., Kim K., Choi S., Hong S.. **Structure-based virtual screening and de novo design of PIM1 inhibitors with anticancer activity from natural products**. *Pharm. (Basel)* (2021) **14** 275. DOI: 10.3390/ph14030275
34. Park S. H., Goo J. M., Jo C. H.. **Receiver operating characteristic (ROC) curve: Practical review for radiologists**. *Korean J. Radiol.* (2004) **5** 11-18. DOI: 10.3348/kjr.2004.5.1.11
35. Parrinello M., Rahman A.. **Polymorphic transitions in single crystals: A new molecular dynamics method**. *J. Appl. Phys.* (1981) **52** 7182-7190. DOI: 10.1063/1.328693
36. Pogacic V., Bullock A. N., Fedorov O., Filippakopoulos P., Gasser C., Biondi A.. **Structural analysis identifies imidazo[1, 2-b]pyridazines as PIM kinase inhibitors with**. *Cancer Res.* (2007) **67** 6916-6924. DOI: 10.1158/0008-5472.can-07-0320
37. Ponzoni I., Sebastián-Pérez V., Martínez M. J., Roca C., Cruz Pérez C. D., Cravero F.. **QSAR classification models for predicting the activity of inhibitors of beta-secretase (BACE1) associated with alzheimer's disease**. *Sci. Rep.* (2019) **9** 9102. DOI: 10.1038/s41598-019-45522-3
38. Pronk S., Páll S., Schulz R., Larsson P., Bjelkmar P., Apostolov R.. **Gromacs 4.5: A high-throughput and highly parallel open source molecular simulation toolkit**. *Bioinformatics* (2013) **29** 845-854. DOI: 10.1093/bioinformatics/btt055
39. Qian K. C., Wang L., Hickey E. R., Studts J., Barringer K., Peng C.. **Structural basis of constitutive activity and a unique nucleotide binding mode of human Pim-1 kinase**. *J. Biol. Chem.* (2005) **280** 6130-6137. DOI: 10.1074/jbc.m409123200
40. Rakhimbekova A., Madzhidov T. I., Nugmanov R. I., Gimadiev T. R., Baskin I. I., Varnek A.. **Comprehensive analysis of applicability domains of QSPR models for chemical reactions**. *Int. J. Mol. Sci.* (2020) **21** 5542. DOI: 10.3390/ijms21155542
41. Rücker C., Rücker G., Meringer M.. **y-Randomization and its variants in QSPR/QSAR**. *J. Chem. Inf. Model* (2007) **47** 2345-2357. DOI: 10.1021/ci700157b
42. Shaik N. A., Nasser K. K., Alruwaili M. M., Alallasi S. R., Elango R., Banaganapalli B.. **Molecular modelling and dynamic simulations of sequestosome 1 (SQSTM1) missense mutations linked to Paget disease of bone**. *J. Biomol. Struct. Dyn.* (2021) **39** 2873-2884. DOI: 10.1080/07391102.2020.1758212
43. Shaik N. A., Saud Al-Saud N. B., Abdulhamid Aljuhani T., Jamil K., Alnuman H., Aljeaid D.. **Structural characterization and conformational dynamics of alpha-1 antitrypsin pathogenic variants causing alpha-1-antitrypsin deficiency**. *Front. Mol. Biosci.* (2022) **9** 1051511. DOI: 10.3389/fmolb.2022.1051511
44. Sushko I., Novotarskyi S., Körner R., Pandey A. K., Cherkasov A., Li J.. **Applicability domains for classification problems: Benchmarking of distance to models for ames mutagenicity set**. *J. Chem. Inf. Model.* (2010) **50** 2094-2111. DOI: 10.1021/ci100253r
45. Trott O., OlsonAutoDock A. J.. **AutoDock Vina: Improving the speed and accuracy of docking with a new scoring function, efficient optimization, and multithreading**. *J. Comput. Chem.* (2009) **31** 455-461. DOI: 10.1002/jcc.21334
46. Tsuganezawa K., Watanabe H., Parker L., Yuki H., Taruya S., Nakagawa Y.. **A novel Pim-1 kinase inhibitor targeting residues that bind the substrate peptide**. *J. Mol. Biol.* (2012) **417** 240-252. DOI: 10.1016/j.jmb.2012.01.036
47. Tursynbay Y., Zhang J., Li Z., Tokay T., Zhumadilov Z., Wu D.. **Pim-1 kinase as cancer drug target: An update**. *Biomed. Rep.* (2016) **4** 140-146. DOI: 10.3892/br.2015.561
48. Vivek A., Bharti S. K., Kumar Budhwani A.. **3D-QSAR and virtual screening studies of thiazolidine-2, 4-dione analogs: Validation of experimental inhibitory potencies towards PIM-1 kinase**. *J. Mol. Struct.* (2017) **1133** 278-293. DOI: 10.1016/j.molstruc.2016.12.006
49. Voulgaris Z., Magoulas G. D.. **Extensions of the k nearest neighbour methods for classification problems**. (2008) 23-28
50. Wang X., Magnuson S., Pastor R., Fan E., Hu H., Tsui V.. **Discovery of novel pyrazolo[1, 5-a]pyrimidines as potent pan-Pim inhibitors by structure- and property-based drug design**. *Bioorg Med. Chem. Lett.* (2013) **23** 3149-3153. DOI: 10.1016/j.bmcl.2013.04.020
51. Warfel N. A., Kraft A. S.. **PIM kinase (and Akt) biology and signaling in tumors**. *Pharmacol. Ther.* (2015) **151** 41-49. DOI: 10.1016/j.pharmthera.2015.03.001
52. Wurz R. P., Pettus L. H., Jackson C., Wu B., Wang H. L., Herberich B.. **The discovery and optimization of aminooxadiazoles as potent Pim kinase inhibitors**. *Bioorg Med. Chem. Lett.* (2015) **25** 847-855. DOI: 10.1016/j.bmcl.2014.12.067
53. Xia Z., Knaak C., Ma J., Beharry Z. M., McInnes C., Wang W.. **Synthesis and evaluation of novel inhibitors of Pim-1 and Pim-2 protein kinases**. *J. Med. Chem.* (2009) **52** 74-86. DOI: 10.1021/jm800937p
54. Yap C. W.. **PaDEL-descriptor: An open source software to calculate molecular descriptors and f ingerprints**. *J. Comput. Chem.* (2011) **32** 1466-1474. DOI: 10.1002/jcc.21707
55. Zhao W., Qiu R., Li P., Yang J.. **PIM1: A promising target in patients with triple-negative breast cancer**. *Med. Oncol.* (2017) **34** 142. DOI: 10.1007/s12032-017-0998-y
|
---
title: The impact of circulating IGF-1 and IGFBP-2 on cardiovascular prognosis in
patients with acute coronary syndrome
authors:
- Wei Wang
- Kang Yu
- Shou-Yong Zhao
- De-Gang Mo
- Jia-Hui Liu
- Li-Jinn Han
- Tai Li
- Heng-Chen Yao
journal: Frontiers in Cardiovascular Medicine
year: 2023
pmcid: PMC10036580
doi: 10.3389/fcvm.2023.1126093
license: CC BY 4.0
---
# The impact of circulating IGF-1 and IGFBP-2 on cardiovascular prognosis in patients with acute coronary syndrome
## Abstract
### Background
While insulin-like growth factor 1 (IGF-1) exerts a cardioprotective effect in the setting of atherosclerosis, insulin-like growth factor binding protein 2 (IGFBP-2) is involved in metabolic syndrome. Although IGF-1 and IGFBP-2 are known to be predictors for mortality in patients with heart failure, their use in clinic as prognostic biomarkers for acute coronary syndrome (ACS) requires investigation. We evaluated the relationship between IGF-1 and IGFBP-2 levels at admission and the risk of major adverse cardiovascular events (MACEs) in patients with ACS.
### Methods
A total of 277 ACS patients and 42 healthy controls were included in this prospective cohort study. Plasma samples were obtained and analyzed at admission. Patients were followed for MACEs after hospitalization.
### Results
Among patients who suffered acute myocardial infarction, plasma levels of IGF-1 and IGFBP-2 were lower and higher, respectively, as compared to healthy controls (both $p \leq 0.05$). The mean follow-up period was 5.22 (1.0–6.0) months and MACEs incidence was $22.4\%$ (62 of 277 patients). Kaplan–Meier survival analysis revealed that patients with low IGFBP-2 levels had a greater event-free survival rate than patients with high IGFBP-2 levels ($p \leq 0.001$). Multivariate Cox proportional hazards analysis revealed IGFBP-2, but not IGF-1, to be a positive predictor of MACEs (hazard ratio 2.412, $95\%$ CI 1.360–4.277; $$p \leq 0.003$$).
### Conclusion
Our findings suggest that high IGFBP-2 levels are associated with the development of MACEs following ACS. Moreover, IGFBP-2 is likely an independent predictive marker of clinical outcomes in ACS.
## Introduction
Acute coronary syndrome (ACS) is characterized by a sudden decrease in blood flow to the heart. Worldwide, more than seven million people are annually diagnosed with ACS; approximately $5\%$ of this patient population was reported to die prior to hospital discharge [1, 2]. Although scientific advances have greatly facilitated implementation of effective secondary cardiovascular prevention strategies, previously unrecognized mediators of cardiovascular disease (CVD) continue to be discovered.
Insulin-like growth factors (IGFs) are conserved peptide hormones structurally similar to insulin that are expressed universally in multiple tissues [3]. Interestingly, IGF-1 is not only found in the circulation but also in arteries, with studies having reported IGF-1 to exert cardioprotective effects in the setting of atherosclerosis [4, 5]. Preclinical model studies reported that administration of IGF-1 suppresses cardiac fibrosis induced by angiotensin II [6], and treatment of sheep fetuses with IGF-1 was reported to stimulate growth of the coronary vasculature and myocardium [7]. Furthermore, bone marrow mesenchymal stem cells overexpressing IGF-1 were reported to better resist apoptosis in myocardial infarction [8]. In clinical practice, serum IGF-1 levels are decreased in heart failure (HF) patients [9] and serve as a predictor of cardiovascular mortality in this condition [10].
All six members of the IGF-binding protein (IGFBP) family regulate IGF bioavailability [11, 12]. As the second most abundant protein of this family [13], IGFBP-2 plays critical roles in several pathological processes including carcinogenesis [14], pulmonary arterial hypertension (PAH) [15], obesity and insulin resistance [13]. In addition to strongly predicting mortality in HF patients [16], IGFBP-2 has recently emerged as a novel candidate biomarker for cardiovascular risk assessment in patients with aortic stenosis who undergo transcatheter aortic valve implantation [17] and elderly men [18].
To date, relevant research has been limited to animal experiments, preclinical analyses or patients with HF, PAH or aortic stenosis. As such, the prognostic influences of circulating IGF-1 and IGFBP-2 in ACS patients remain unclear. Here, we evaluated the relationship of circulating IGF-1 and IGFBP-2 with major adverse cardiovascular events (MACEs) in ACS patients who underwent coronary angiography (CAG).
## Study design and population
This study was a prospective cohort study conducted according to regulations set forth by the Declaration of Helsinki. This study was approved by the Ethics Committee of Liaocheng People's Hospital and all subjects enrolled provided informed consent. From June 1 2021 to October 1 2021, 304 ACS patients who underwent CAG and 42 site-matched controls free of clinical heart disease were initially enrolled in this study. All ACS patients enrolled met relevant diagnostic criteria for either ST-elevation myocardial infarction (STEMI) or non-ST-elevation ACS [19, 20]. Patients with severe liver disease or renal failure, neoplasms of any kind or infectious or inflammatory conditions were excluded from this study. Patients lost to follow-up were also excluded from analyses. The treatment therapy, including intensive treatment with medicine, percutaneous coronary intervention (PCI) or coronary artery bypass grafting (CABG) was decided by two experienced cardiologists according to the results of CAG and international standards and guidelines [21].
## Laboratory measurement
Venous blood samples were draw from study subjects for evaluation prior to administration of any medications. Blood samples were collected using tubes containing ethylenediaminetetraacetic acid and centrifuged at 3,000 rpm for 15 min immediately after collection. Samples were aliquoted and stored at −80°C until use. Plasma levels of IGF-1 and IGFBP-2 were assayed using the enzyme linked immunosorbent assays (ELISA) IGF-1 (DG100B) and IGFBP-2 (DGB 200) (both R&D Systems, USA), according to manufacturer's protocol. No significant cross-reactivity or interference of IGFBP/IGF-1 with IGFBP-2 was found in the immunoassay according to product description. ELISA IGFBP-2 kit did not measure IGFBP in complex with IGF-1.
## Definitions
Participants with one major coronary artery ≥$50\%$ stenosis were considered as single-vessel disease, whereas multi-vessel disease (MVD) was defined in cases of stenoses ≥$50\%$ in 2 or 3 major epicardial coronary arteries. Stenosis of the left main coronary artery ≥$50\%$ was regarded as the left main disease [22]. Incomplete revascularization (IR) was defined as one or more vessels stenosis ≥$50\%$ being left untreated after revascularization [23].
## Follow-up and outcomes
The entire cohort was followed up for 6 months starting from date of hospitalization. Data were systematically obtained via phone interviews and review of medical records. A total of 27 ACS patients were lost of follow-up and excluded from this study. A total of 277 ACS patients who completed follow-up were finally evaluated and included 93 unstable angina (UA), 89 non-ST-elevation myocardial infarction (NSTEMI) and 95 STEMI patients. Cardiovascular death, angina, new-onset HF, recurrent myocardial infarction (MI) or any revascularization were all defined as MACEs.
## Statistical analyses
The Shapiro–Wilk test was used to determine whether continuous data were normally distributed. Normally distributed continuous variables were expressed as mean ± standard deviation; data not normally distributed were expressed as median (interquartile range). Comparisons of continuous variables between two groups were performed using the Mann–Whitney U test or t-test. Comparisons of continuous variables among four groups were performed using one-way analysis of variance or the Kruskal–Wallis H test. The pairwise test for multiple comparisons was used to analyze intergroup differences for IGF-1 and IGFBP-2 levels after Kruskal–Wallis H analysis. Categorical variables were presented as frequency and percentage and compared using the chi-squared test. Spearman correlation analysis was performed to analyze the correlation of plasma IGF-1 and IGFBP-2 levels.
A receiver operator characteristics (ROC) curve was generated and the area under the curve (AUC) was calculated, and Z test were used to compare AUC values. Optimal IGF-1 and IGFBP-2 cutoff points for MACEs prediction were determined based on maximal Youden's index. Kaplan–Meier survival curves were constructed to analyze the short-term event-free survival (EFS) rate and comparisons were performed using the log-rank test. Cox proportional hazards regression analysis was performed to determine independent factors predictive for MACEs; confounders with unadjusted p-values <0.05 in univariate analysis were included in a multivariate regression model. A two-tailed p-value of <0.05 was considered as statistically significant. Statistical analyses were performed using SPSS 23.0 (IBM, USA).
## Clinical characteristics
Median values (interquartile ranges) of IGF-1 and IGFBP-2 concentrations in healthy controls ($$n = 42$$) were 170.39 (106.14) ng/ml and 199.1 (255.01) ng/ml, respectively. While IGF-1 levels in healthy controls were higher than in patients who suffered NSTEMI or STEMI ($p \leq 0.05$), IGFBP-2 levels showed the opposite pattern. Although IGFBP-2 levels in STEMI patients were higher than those in UA patients ($p \leq 0.05$), no significant difference between healthy controls and UA patients was found ($p \leq 0.05$; Table 1).
**Table 1**
| Variables | HC (n = 42) | UA (n = 93) | NSTEMI (n = 89) | STEMI (n = 95) |
| --- | --- | --- | --- | --- |
| IGF-1 (ng/ml) | 170.39 (106.14) | 138.89 (86.53) | 117.53 (76.31)a | 117.95 (96.6)a |
| IGFBP-2 (ng/ml) | 199.1 (255.01) | 258.74 (245.03) | 318.55 (394.59)a | 364.75 (400.79)a,b |
In a total of 277 ACS patients, 195 ($70.4\%$) presented with MVD, 37 ($13.4\%$) presented with left main vessel disease, and the number of patients with IR treatment was 32 ($11.6\%$). 89 ($32.1\%$) patients were treated with intensive medication, and 188 ($67.9\%$) patients were treated with PCI or CABG. Patients were divided into two groups based on median IGF-1 concentrations: a high IGF-1 group (IGF-1 levels ≥126.92 ng/ml; $$n = 139$$) and a low IGF-1 group (IGF-1 levels <126.92 ng/ml; $$n = 138$$). High IGF-1 group patients were found to have a higher ejection fraction (EF), a lower rate of acute myocardial infarction (AMI) and a higher rate of multi-vessel lesion as compared to those in the low IGF-1 group. No other statistically significant differences in demographic characteristics or laboratory evaluations were noted between the two groups (Table 2).
**Table 2**
| Characteristics | IGF-1 ≧ 126.92 ng/ml | IGF-1 < 126.92 ng/ml | p value |
| --- | --- | --- | --- |
| Characteristics | n = 139 | n = 138 | p value |
| Age (years) | 59 ± 12 | 62 ± 12 | 0.491 |
| Male, n (%) | 111 (79.9) | 102 (73.9) | 0.241 |
| Smoking, n (%) | 77 (55.4) | 73 (52.9) | 0.677 |
| Hypertension, n (%) | 85 (61.2) | 74 (53.6) | 0.205 |
| DM, n (%) | 37 (26.6) | 38 (27.5) | 0.864 |
| Heart rate (bpm) | 72 (17) | 72 (17) | 0.584 |
| BMI (kg/m2) | 25.78 ± 3.75 | 25.25 (4.3) | 0.137 |
| Hemoglobin (g/L) | 137 ± 16 | 134 ± 16 | 0.786 |
| WBC count (×109/L) | 7.5 (3.6) | 7.7 (3.7) | 0.671 |
| D-dimer (mg/L) | 0.35 (0.45) | 0.35 (0.37) | 0.912 |
| Creatinine (umol/L) | 67.3 (19.5) | 69.6 (18.9) | 0.576 |
| TG (mmol/L) | 1.35 (1.06) | 1.28 (0.80) | 0.115 |
| LDL (mmol/L) | 2.62 ± 0.79 | 2.63 (0.94) | 0.539 |
| TC (mmol/L) | 4.27 ± 1.08 | 4.29 (1.33) | 0.761 |
| CRP (mg/L) | 3.55 (6.02) | 4.26 (7.48) | 0.227 |
| EF (%) | 58 (14) | 54 (14) | 0.025 |
| LVEDD (mm) | 45 (5) | 46 (6) | 0.198 |
| Diagnosis, n (%) | Diagnosis, n (%) | Diagnosis, n (%) | Diagnosis, n (%) |
| AMI | 83 (59.7) | 101 (73.2) | 0.018 |
| UA | 56 (40.1) | 37 (26.8) | 0.018 |
| Angiography, n (%) | Angiography, n (%) | Angiography, n (%) | Angiography, n (%) |
| One-vessel lesion | 33 (23.7) | 49 (35.5) | 0.032 |
| Multi-vessel lesion | 106 (76.3) | 89 (64.5) | 0.032 |
| Left main vessel lesion | 22 (15.8) | 15 (10.8) | 0.225 |
| IR | 18 (12.9) | 14 (10.1) | 0.465 |
| Treatment strategies, n (%) | Treatment strategies, n (%) | Treatment strategies, n (%) | Treatment strategies, n (%) |
| Intensive medication | 39 (28.1) | 50 (36.2) | 0.145 |
| PCI/CABG | 100 (71.9) | 88 (63.8) | 0.145 |
| IGF-1 (ng/ml) | 173.6 (78.6) | 88.9 (37.3) | <0.001 |
| IGFBP-2 (ng/ml) | 259.3 (317.9) | 326.53 (394.5) | 0.112 |
Patients were divided into two groups based on median IGFBP-2 concentrations: a high IGFBP-2 group (IGFBP-2 ≥ 308.3 ng/ml; $$n = 139$$) and a low IGFBP-2 group (IGFBP-2 < 308.3 ng/ml; $$n = 138$$). As shown in Table 3, high IGFBP-2 group patients were older, had lower levels of hemoglobin, triglycerides and EF, higher D-dimer levels, and lower body mass indices (BMI). Moreover, high IGFBP-2 group patients were found to have a higher rate of AMI as compared to low IGFBP-2 group patients.
**Table 3**
| Characteristics | IGFBP-2 ≧ 308.3 ng/ml | IGFBP-2 < 308.3 ng/ml | p value |
| --- | --- | --- | --- |
| Characteristics | n = 139 | n = 138 | p value |
| Age (years) | 62 ± 12 | 58 ± 12 | 0.006 |
| Male, n (%) | 108 (77.7) | 105 (76.1) | 0.750 |
| Smoking, n (%) | 81 (58.3) | 69 (50.0) | 0.167 |
| Hypertension, n (%) | 79 (56.8) | 80 (58.0) | 0.848 |
| DM, n (%) | 38 (27.3) | 37 (26.8) | 0.921 |
| Heart rate (bpm) | 74 (18) | 70 (16) | 0.552 |
| BMI (kg/m2) | 25.1 ± 3.46 | 25.9 ± 3.49 | 0.049 |
| Hemoglobin (g/L) | 134 ± 16 | 138 ± 17 | 0.034 |
| WBC count (×109/L) | 7.59 (3.59) | 7.66 (3.45) | 0.561 |
| D-dimer (mg/L) | 0.4 (0.57) | 0.33 (0.34) | 0.014 |
| Creatinine (umol/L) | 67 (19) | 68 (20) | 0.143 |
| TG (mmol/L) | 1.24 (0.73) | 1.45 (0.96) | 0.015 |
| LDL (mmol/L) | 2.63 ± 0.74 | 2.67 ± 0.81 | 0.682 |
| TC (mmol/L) | 4.24 ± 1.01 | 4.26 ± 1.09 | 0.871 |
| CRP (mg/L) | 3.88 (7.41) | 3.84 (6.46) | 0.421 |
| EF (%) | 54 (15) | 59 (14) | 0.017 |
| LVEDD (mm) | 46 (7) | 45 (5) | 0.174 |
| Diagnosis, n (%) | Diagnosis, n (%) | Diagnosis, n (%) | Diagnosis, n (%) |
| AMI | 103 (74.1) | 81 (58.7) | 0.007 |
| UA | 36 (25.9) | 57 (41.3) | 0.007 |
| Angiography, n (%) | Angiography, n (%) | Angiography, n (%) | Angiography, n (%) |
| One-vessel lesion | 40 (28.9) | 42 (30.4) | 0.762 |
| Multi-vessel lesion | 99 (71.1) | 96 (69.6) | 0.762 |
| Left main vessel lesion | 17 (12.2) | 20 (14.4) | 0.580 |
| IR | 18 (12.9) | 14 (10.1) | 0.465 |
| Treatment strategies, n (%) | Treatment strategies, n (%) | Treatment strategies, n (%) | Treatment strategies, n (%) |
| Intensive medication | 38 (27.3) | 51 (37) | 0.087 |
| PCI/CABG | 101 (72.7) | 87 (63) | 0.087 |
| IGF-1 (ng/ml) | 121.9 (82.4) | 129.0 (88.4) | 0.328 |
| IGFBP-2 (ng/ml) | 518.25 (312.63) | 187.89 (123.35) | <0.001 |
The correlation of plasma IGF-1 and IGFBP-2 levels was evaluated by Spearman correlation analysis. A significant negative correlation was found among IGF-1 and IGFBP-2 (r = −0.172, $$p \leq 0.002$$; Figure 1).
**Figure 1:** *Correlation graph of IGF-1 and IGFBP-2. IGF-1: Insulin like growth factor 1; IGFBP-2: Insulin like growth factor binding protein 2.*
## Clinical outcomes of adverse cardiovascular events
The mean follow-up period was 5.22 (1.0–6.0) months. The incidence of MACEs was $22.4\%$ (62 of 277 patients) and included instances of cardiovascular death ($$n = 8$$), angina ($$n = 21$$), HF ($$n = 13$$) and reinfarction or revascularization ($$n = 20$$). No patient died of non-cardiovascular causes. No statistical differences in MACEs rates between high and low IGF-1 group patients were found ($$p \leq 0.141$$). Total MACEs incidence was higher among high IGFBP-2 group patients as compared to low IGFBP-2 group patients ($p \leq 0.001$). Furthermore, incidences of angina ($$p \leq 0.003$$) and HF ($$p \leq 0.011$$) were higher among high IGFBP-2 group patients as compared to low IGFBP-2 group patients (Table 4).
**Table 4**
| Complications | Plasma IGF-1 levels | Plasma IGF-1 levels.1 | Plasma IGF-1 levels.2 | Plasma IGFBP-2 levels | Plasma IGFBP-2 levels.1 | Plasma IGFBP-2 levels.2 |
| --- | --- | --- | --- | --- | --- | --- |
| Complications | Low (n = 138) | High (n = 139) | p value | Low (n = 138) | High (n = 139) | p value |
| Death, n (%) | 5 (3.6) | 3 (2.2) | 0.467 | 3 (2.2) | 5 (3.6) | 0.728 |
| Angina, n (%) | 11 (8.0) | 10 (7.2) | 0.807 | 4 (2.9) | 17 (12.2) | 0.003 |
| Heart failure, n (%) | 9 (6.5) | 4 (2.9) | 0.152 | 2 (1.4) | 11 (7.9) | 0.011 |
| Reinfarction or Revascularization, n (%) | 11 (6.0) | 9 (6.5) | 0.630 | 8 (5.8) | 12 (8.6) | 0.362 |
| Total MACEs, n (%) | 36 (25.9) | 26 (18.8) | 0.141 | 17 (12.3) | 45 (32.4) | <0.001 |
## Kaplan-Meier survival curves of circulating IGF-1 and IGFBP-2 in ACS patients during follow-up
Kaplan-Meier survival analysis revealed no statistically significant differences in event-free survival (EFS) between high and low IGF-1 group patients ($$p \leq 0.145$$; Figure 2A). Low IGFBP-2 group patients were found to have had a higher EFS as compared to high IGFBP-2 group patients ($p \leq 0.001$; Figure 2B). As such, patients with low levels of IGFBP-2 were found to have had a more favorable prognosis as compared to those with high levels of IGFBP-2.
**Figure 2:** *Kaplan-Meier curves in patients with ACS with individual levels of IGF-1 (A) and IGFBP-2 (B) during follow-up (p = 0.145 and <0.001, respectively). MACEs, Major adverse cardiovascular events; ACS, Acute coronary syndrome; IGF-1, Insulin like growth factor 1; IGFBP-2, Insulin like growth factor binding protein 2.*
## Independent predictors for MACEs in ACS patients
As shown in Table 5, we analyzed potential confounders for association with short-term MACEs using univariate analysis. Confounders with p-values of <0.05 in univariate analysis were included in multivariate Cox regression analysis. After correcting for age, diagnosis of AMI, creatinine level, EF, intensive medication therapy and left main vessel lesion, high IGFBP-2 level was confirmed to have positively predicted value for MACEs [adjusted hazard ratio: 2.412, $95\%$ confidential interval (CI) 1.360–4.277; $$p \leq 0.003$$]. Furthermore, EF and left main vessel lesion were found to be independent predictors for MACEs, while IGF-1 level was not.
**Table 5**
| Variables | HR | 95% CI | p value | Adjusted HR | 95% CI.1 | p value.1 |
| --- | --- | --- | --- | --- | --- | --- |
| Male | 1.295 | 0.741–2.264 | 0.363 | | | |
| Age | 1.037 | 1.014–1.061 | 0.001 | 1.016 | 0.993–1.040 | 0.175 |
| Smoking status | 1.044 | 0.634–1.720 | 0.865 | | | |
| Hypertension | 1.062 | 0.643–1.874 | 0.816 | | | |
| DM | 1.563 | 0.929–2.629 | 0.093 | | | |
| Diagnosis of AMI | 3.319 | 1.637–6.731 | 0.001 | 1.76 | 0.787–3.936 | 0.169 |
| Heart rate | 1.012 | 0.996–1.029 | 0.145 | | | |
| BMI | 0.971 | 0.957–0.986 | 0.052 | | | |
| WBC count | 1.031 | 0.955–1.113 | 0.440 | | | |
| Creatinine | 1.012 | 1.005–1.019 | 0.001 | 1.006 | 0.999–1.014 | 0.11 |
| TG | 0.79 | 0.578–1.079 | 0.138 | | | |
| LDL | 0.906 | 0.658–1.247 | 0.543 | | | |
| TC | 0.869 | 0.684–1.105 | 0.252 | | | |
| CRP | 1.01 | 1.000–1.021 | 0.054 | | | |
| EF | 0.935 | 0.911–0.960 | <0.001 | 0.956 | 0.927–0.985 | 0.004 |
| Intensive medication | 1.762 | 1.066–2.911 | 0.027 | 0.609 | 0.356–1.043 | 0.071 |
| Multi-vessel lesion | 1.243 | 0.704–2.196 | 0.454 | | | |
| Left main Vessel lesion | 2.134 | 1.193–3.816 | 0.011 | 1.868 | 1.014–3.440 | 0.045 |
| IR | 1.525 | 0.752–3.091 | 0.242 | | | |
| High IGF-1 | 0.694 | 0.419–1.150 | 0.157 | | | |
| High IGFBP-2 | 2.894 | 1.656–5.057 | <0.001 | 2.412 | 1.360–4.277 | 0.003 |
## MACEs prediction in ACS patients using receiver operator characteristics curves of circulating IGFBP-2 and EF
To evaluate the potential prognostic power of circulating IGFBP-2 and EF for MACEs prediction, ROC curves were generated. Analysis revealed that AUC values of plasma IGFBP-2 (Figure 3A) and EF (Figure 3B) for MACEs prediction in ACS patients were 0.722 ($95\%$ CI 0.640–0.804; $p \leq 0.001$) and 0.659 ($95\%$ CI 0.570–0.748; $p \leq 0.001$), respectively. No statistical differences in AUC values of IGFBP-2 and EF for MACEs prediction were found ($Z = 1.02$; $p \leq 0.05$).
**Figure 3:** *Receiver operating characteristic curves of circulating IGFBP-2 (A) and EF (B) for predicting MACEs in patients with ACS. IGFBP-2, Insulin like growth factor binding protein 2; EF, Ejection fraction; MACEs, Major adverse cardiovascular events; ACS, Acute coronary syndrome.*
## Discussion
We evaluated a cohort of 277 ACS patients and 42 healthy controls to investigate the relationship between IGF-1 and IGFBP-2 levels and prognosis for short-term outcomes. Our findings revealed that [1] circulating IGF-1 levels in healthy controls were higher than in NSTEMI or STEMI patients, and IGFBP-2 levels showed an opposite pattern; [2] EFS was poor in patients with high levels of IGFBP-2; and [3] high IGFBP-2 levels, but not low IGF-1 levels, independently predicted for MACEs in ACS patients who underwent CAG.
IGF-1 levels were previously found to be lower in acute MI patients compared to healthy controls [24, 25]. To date, however, studies evaluating the influence of IGFBP-2 levels on cardiovascular disease remain scarce. In this study, we not only confirmed IGF-1 levels in healthy controls to have been significantly higher than those in acute MI patients, but also found IGFBP-2 levels to have been higher in acute MI patients than those in healthy controls. Levels of IGF-1 and IGFBP-2 were previously reported to associate with EF and potentially serve as biomarkers for HF (9, 26–28). In agreement with previous studies, we found that IGF-1 levels positively associated with EF, whereas IGFBP-2 negatively associated with EF. Although HF incidence was greater in high IGFBP-2 group patients, no differences between groups were found for IGF-1 levels. We noted that high IGFBP-2 group patients had lower triglyceride levels and BMI. Our findings are in agreement with prior literature that reported IGFBP-2 to be a marker of metabolic syndrome and inversely correlate with BMI and triglyceride levels (29–31).
No differences in incidences of MACEs or death were found between low and high IGF-1 group patients in this study. Although no difference in mortality was noted between high and low IGFBP-2 group patients, the incidence of MACEs in high IGFBP-2 group patients was found significantly higher as compared to low IGFBP-2 group patients. Moreover, after adjusting for age and other variables, we found that IGFBP-2 was an independent predictive factor for MACEs, while IGF-1 was not. Iswandi et al. [ 32] reported that IGF-1 was not an independent predictor of cardiovascular mortality or morbidity in ACS patients over a 5-year follow-up period. Furthermore, Wallander et al. [ 33] found that IGF-1 levels at hospital admission were not related to cardiovascular death over a three-year follow-up period in patients with type 2 diabetes who suffered acute MI. Although our findings were in agreement with most previously reported, a study by Bourron and et al. [ 34] reported that low IGF-1 levels not only associate with increased mortality risk but also with risk of any MACEs in acute MI patients over 2 years of follow-up. As various populations may yield disparate findings, future studies should confirm whether IGF-1 is predictive for adverse outcomes in diverse groups.
Increasing evidence has highlighted the cardioprotective effects of IGF-1 in the setting of cardiovascular disease. Numerous in vivo and in vitro studies reported that IGF-1 facilitates resistance to apoptosis in hypoxic conditions [8], stimulates fetal cardiac growth [7], increases production of circulating angiogenic cytokines [35] and exerts positive inotropic and antioxidant effects [36]. Translational research has further revealed that treatment with IGF-1 significantly reduces left ventricular volume, attenuates left ventricular mass and improves stroke volume in STEMI patients [37], and improves EF in HF patients [38]. However, Conover et al. [ 39] found that transgenic overexpression of pregnancy-associated plasma protein-A increases IGF-1 activity and results in accelerated atherosclerotic lesion development. Hirai et al. [ 40] reported that IGF-1 promotes atherosclerosis by affecting endothelial function and increasing aging in rabbits fed a cholesterol-rich diet. As such, the roles of IGF-1 in cardiovascular disease remain unclear and warrant further study.
Previously, IGFBP-2 was reported to inversely correlate with BMI [41]. Indeed, studies reported that low IGFBP-2 independently associates with an increased risk of metabolic syndrome as well as elevated fasting glucose levels [30]. Higher circulating IGFBP-2 concentrations were also longitudinally associated with lower type 2 diabetes risk [42, 43]. Although IGFBP-2 is considered to protect against cardiovascular risk factors, high levels of IGFBP-2 were reported to associate with poor prognoses in several diseases. Prior studies reported that IGFBP-2 independently predicts for adverse clinical outcomes in patients with HF [16, 44], severe aortic stenosis [17], dilated cardiomyopathy [45] and PAH [46]. To date, studies about IGFBP-2 in CVD is seldom, and the present study first uncovered that IGFBP-2 is an independent predictor of MACEs in ACS patients. Additionally, although no significant differences in prognosis power among IGFBP-2 and EF were noted, the AUC value of circulating IGFBP-2 was found to be greater than that of EF. Studies enrolling more eligible patients may suggest the prognostic power of IGFBP-2 was superior to EF in ACS patients. The seemingly paradoxical influences of IGFBP-2 on cardiovascular risk factors and pathological processes warrant detailed study.
The potential mechanism of IGFBP-2 and MACE is thought to be multifactorial. IGFBP-2 plays a crucial role in regulating mitogen-activated protein kinase (MAPK) pathway, which is a driver of atherosclerosis and involved in inflammatory signaling and oxidative stress [47, 48]. Moreover, IGFBP-2 regulates the phosphatidylinositol 3-kinase (PI3K)/alpha serine/threonine-protein kinase (Akt) signaling pathway, which has a fundamental role in the pathological processes of atherosclerosis [49, 50]. In addition, IGFBP-2 enhances the migration and proliferation of vascular smooth muscle cells (VSMC), this process is associated with the development of atherosclerosis [51]. Further studies about the underlying mechanisms of IGFBP-2 in CVD are certainly warranted.
Our study, although well-designed, was not without limitations. First, the participants in this study were predominantly male, our sample size was relatively small and the follow-up period was short. As such, sex differences were not investigated, and the low incidence of mortality limited the hard endpoint analysis of the study. In addition, a lack of significant findings concerning IGF-1 and MACEs prediction may have occurred due to a type II statistical error. Second, the mechanisms behind the association between circulating levels of IGF-1 and IGFBP-2 and the incidence of MACE were not elucidated. Third, healthy controls were not well-matched with ACS patients for gender or age. Finally, blood samples were collected at admission and lacked data concerning serial fluctuations in IGF-1 and IGFBP-2 levels during follow-up. Therefore, multi-center studies with a greater sample size and animal studies about potential biological mechanisms of IGF-1 and IGFBP-2 are needed to confirm these results.
## Conclusion
Plasma IGFBP-2 levels were higher in patients who suffered acute MI compared to healthy controls. A high IGFBP-2 level, but not a low IGF-1 level, likely has clinical use as a prognostic biomarker for MACEs in patients with ACS. Although the underlying mechanisms for our findings remain unclear, we provide a foundation for further study of IGFBP-2 and improvements in clinical management of patients suffering ACS.
## Data availability statement
The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.
## Ethics statement
The studies involving human participants were reviewed and approved by Ethics Committee of Liaocheng People's Hospital. The patients/participants provided their written informed consent to participate in this study.
## Author contributions
HY put forward conception and study design. KY, SZ and LH researched data, tested biomarkers. WW wrote the manuscript and contributed to statistical analysis. TL, DM and JL edited and contributed to the manuscript, data interpretation, and discussion. All authors contributed to the article and approved the submitted version.
## Conflict of interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher's note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## References
1. Bhatt DL, Lopes RD, Harrington RA. **Diagnosis and treatment of acute coronary syndromes: a review**. *JAMA* (2022) **327** 662-75. DOI: 10.1001/jama.2022.0358
2. Collet JP, Thiele H. **Management of acute coronary syndromes in patients presenting without persistent ST-segment elevation and coexistent atrial fibrillation—dual versus triple antithrombotic therapy**. *Eur Heart J* (2021) **42** 2020-1. DOI: 10.1093/eurheartj/ehaa909
3. Haywood NJ, Slater TA, Matthews CJ, Wheatcroft SB. **The insulin like growth factor and binding protein family: novel therapeutic targets in obesity & diabetes**. *Mol Metab* (2019) **19** 86-96. DOI: 10.1016/j.molmet.2018.10.008
4. Higashi Y, Gautam S, Delafontaine P, Sukhanov S. **IGF-1 and cardiovascular disease**. *Growth Horm IGF Res* (2019) **45** 6-16. DOI: 10.1016/j.ghir.2019.01.002
5. Higashi Y, Sukhanov S, Shai SY, Danchuk S, Tang R, Snarski P. **Insulin-Like growth factor-1 receptor deficiency in macrophages accelerates atherosclerosis and induces an unstable plaque phenotype in apolipoprotein E-deficient mice**. *Circulation* (2016) **133** 2263-78. DOI: 10.1161/CIRCULATIONAHA.116.021805
6. Ock S, Ham W, Kang CW, Kang H, Lee WS, Kim J. **IGF-1 protects against angiotensin II-induced cardiac fibrosis by targeting alphaSMA**. *Cell Death Dis* (2021) **12** 688. DOI: 10.1038/s41419-021-03965-5
7. Jonker SS, Giraud GD, Chang EI, Elman MR, Louey S. **Coronary vascular growth matches IGF-1-stimulated cardiac growth in fetal sheep**. *FASEB J* (2020) **34** 10041-55. DOI: 10.1096/fj.202000215R
8. Lin M, Liu X, Zheng H, Huang X, Wu Y, Huang A. **IGF-1 enhances BMSC viability, migration, and anti-apoptosis in myocardial infarction via secreted frizzled-related protein 2 pathway**. *Stem Cell Res Ther* (2020) **11** 22. DOI: 10.1186/s13287-019-1544-y
9. Barroso MC, Kramer F, Greene SJ, Scheyer D, Kohler T, Karoff M. **Serum insulin-like growth factor-1 and its binding protein-7: potential novel biomarkers for heart failure with preserved ejection fraction**. *BMC Cardiovasc Disord* (2016) **16** 199. DOI: 10.1186/s12872-016-0376-2
10. De Giorgi A, Marra AM, Iacoviello M, Triggiani V, Rengo G, Cacciatore F. **Insulin-like growth factor-1 (IGF-1) as predictor of cardiovascular mortality in heart failure patients: data from the T.O.S.CA. Registry**. *Intern Emerg Med* (2022) **17**. DOI: 10.1007/s11739-022-02980-4
11. Clemmons DR. **Role of IGF binding proteins in regulating metabolism**. *Trends Endocrinol Metab* (2016) **27** 375-91. DOI: 10.1016/j.tem.2016.03.019
12. Firth SM, Baxter RC. **Cellular actions of the insulin-like growth factor binding proteins**. *Endocr Rev* (2002) **23** 824-54. DOI: 10.1210/er.2001-0033
13. Wheatcroft SB, Kearney MT, Shah AM, Ezzat VA, Miell JR, Modo M. **IGF-binding protein-2 protects against the development of obesity and insulin resistance**. *Diabetes* (2007) **56** 285-94. DOI: 10.2337/db06-0436
14. Sun L, Zhang X, Song Q, Liu L, Forbes E, Tian W. **IGFBP2 Promotes tumor progression by inducing alternative polarization of macrophages in pancreatic ductal adenocarcinoma through the STAT3 pathway**. *Cancer Lett* (2021) **500** 132-46. DOI: 10.1016/j.canlet.2020.12.008
15. Yang J, Griffiths M, Nies MK, Brandal S, Damico R, Vaidya D. **Insulin-like growth factor binding protein-2: a new circulating indicator of pulmonary arterial hypertension severity and survival**. *BMC Med* (2020) **18** 268. DOI: 10.1186/s12916-020-01734-3
16. Barutaut M, Fournier P, Peacock WF, Evaristi MF, Caubere C, Turkieh A. **Insulin-like growth factor binding protein 2 predicts mortality risk in heart failure**. *Int J Cardiol* (2020) **300** 245-51. DOI: 10.1016/j.ijcard.2019.09.032
17. Muessig JM, Lichtenauer M, Wernly B, Kelm M, Franz M, Baz L. **Insulin like growth factor binding protein 2 (IGFBP-2) for risk prediction in patients with severe aortic stenosis undergoing transcatheter aortic valve implantation (TAVI)**. *Int J Cardiol* (2019) **277** 54-9. DOI: 10.1016/j.ijcard.2018.09.091
18. van den Beld AW, Blum WF, Brugts MP, Janssen JA, Grobbee DE, Lamberts SW. **High IGFBP2 levels are not only associated with a better metabolic risk profile but also with increased mortality in elderly men**. *Eur J Endocrinol* (2012) **167** 111-7. DOI: 10.1530/EJE-12-0160
19. Amsterdam EA, Wenger NK, Brindis RG, Casey DE, Ganiats TG, Holmes DR. **2014 AHA/ACC guideline for the management of patients with non-ST-elevation acute coronary syndromes: a report of the American college of cardiology/American heart association task force on practice guidelines**. *J Am Coll Cardiol* (2014) **64** e139-228. DOI: 10.1016/j.jacc.2014.09.017
20. Jneid H, Addison D, Bhatt DL, Fonarow GC, Gokak S, Grady KL. **2017 AHA/ACC clinical performance and quality measures for adults with ST-elevation and non-ST-elevation myocardial infarction: a report of the American college of cardiology/American heart association task force on performance measures**. *J Am Coll Cardiol* (2017) **70** 2048-90. DOI: 10.1016/j.jacc.2017.06.032
21. Neumann FJ, Sousa-Uva M, Ahlsson A, Alfonso F, Banning AP, Benedetto U. **2018 ESC/EACTS guidelines on myocardial revascularization**. *Eur Heart J* (2019) **40** 87-165. DOI: 10.1093/eurheartj/ehy394
22. Su J, Li Z, Huang M, Wang Y, Yang T, Ma M. **Triglyceride glucose index for the detection of the severity of coronary artery disease in different glucose metabolic states in patients with coronary heart disease: a RCSCD-TCM study in China**. *Cardiovasc Diabetol* (2022) **21** 96. DOI: 10.1186/s12933-022-01523-7
23. Gaba P, Gersh BJ, Ali ZA, Moses JW, Stone GW. **Complete versus incomplete coronary revascularization: definitions, assessment and outcomes**. *Nat Rev Cardiol* (2021) **18** 155-68. DOI: 10.1038/s41569-020-00457-5
24. Conti E, Andreotti F, Sciahbasi A, Riccardi P, Marra G, Menini E. **Markedly reduced insulin-like growth factor-1 in the acute phase of myocardial infarction**. *J Am Coll Cardiol* (2001) **38** 26-32. DOI: 10.1016/S0735-1097(01)01367-5
25. Hajsadeghi S, Mohseni H, Moradi M, Rahmani E, Kordshakeri K, Manteghi MJ. **Evaluating the association between insulin-like growth factor-1 values and short-term survival rates following acute myocardial infarction**. *Clin Med Insights Cardiol* (2011) **5** 7-11. DOI: 10.4137/CMC.S6629
26. Bruno C, Silvestrini A, Calarco R, Favuzzi AMR, Vergani E, Nicolazzi MA. **Anabolic hormones deficiencies in heart failure with preserved ejection fraction: prevalence and impact on antioxidants levels and myocardial dysfunction**. *Front Endocrinol (Lausanne)* (2020) **11** 281. DOI: 10.3389/fendo.2020.00281
27. D'Assante R, Arcopinto M, Rengo G, Salzano A, Walser M, Gambino G. **Myocardial expression of somatotropic axis, adrenergic signalling, and calcium handling genes in heart failure with preserved ejection fraction and heart failure with reduced ejection fraction**. *ESC Heart Fail* (2021) **8** 1681-6. DOI: 10.1002/ehf2.13067
28. Prelevic V, Antunovic T, Radunovic D, Gligorovic Barhanovic N, Gledovic B, Bulatovic N. **Decreased insulin-like growth factor-1 (IGF-1) concentration correlates with reduced left-ventricle ejection fraction (LVEF) in hemodialysis patients**. *Int Urol Nephrol* (2020) **52** 2385-91. DOI: 10.1007/s11255-020-02595-8
29. Carter S, Li Z, Lemieux I, Almeras N, Tremblay A, Bergeron J. **Circulating IGFBP-2 levels are incrementally linked to correlates of the metabolic syndrome and independently associated with VLDL triglycerides**. *Atherosclerosis* (2014) **237** 645-51. DOI: 10.1016/j.atherosclerosis.2014.09.022
30. Heald AH, Kaushal K, Siddals KW, Rudenski AS, Anderson SG, Gibson JM. **Insulin-like growth factor binding protein-2 (IGFBP-2) is a marker for the metabolic syndrome**. *Exp Clin Endocrinol Diabetes* (2006) **114** 371-6. DOI: 10.1055/s-2006-924320
31. van den Beld AW, Carlson OD, Doyle ME, Rizopoulos D, Ferrucci L, van der Lely AJ. **IGFBP-2 and aging: a 20-year longitudinal study on IGFBP-2, IGF-I, BMI, insulin sensitivity and mortality in an aging population**. *Eur J Endocrinol* (2019) **180** 109-16. DOI: 10.1530/EJE-18-0422
32. Iswandi CP, van den Berg VJ, Simsek S, Velzen DV, Boekel ET, Cornel JH. **IGF-1 is not related to long-term outcome in hyperglycemic acute coronary syndrome patients**. *Diab Vasc Dis Res* (2021) **18** 14791641211047436. DOI: 10.1177/14791641211047436
33. Wallander M, Norhammar A, Malmberg K, Ohrvik J, Ryden L, Brismar K. **IGF Binding protein 1 predicts cardiovascular morbidity and mortality in patients with acute myocardial infarction and type 2 diabetes**. *Diabetes Care* (2007) **30** 2343-8. DOI: 10.2337/dc07-0825
34. Bourron O, Le Bouc Y, Berard L, Kotti S, Brunel N, Ritz B. **Impact of age-adjusted insulin-like growth factor 1 on major cardiovascular events after acute myocardial infarction: results from the fast-MI registry**. *J Clin Endocrinol Metab* (2015) **100** 1879-86. DOI: 10.1210/jc.2014-3968
35. Haleagrahara N, Chakravarthi S, Mathews L. **Insulin like growth factor-1 (IGF-1) causes overproduction of IL-8, an angiogenic cytokine and stimulates neovascularization in isoproterenol-induced myocardial infarction in rats**. *Int J Mol Sci* (2011) **12** 8562-74. DOI: 10.3390/ijms12128562
36. Yeves AM, Burgos JI, Medina AJ, Villa-Abrille MC, Ennis IL. **Cardioprotective role of IGF-1 in the hypertrophied myocardium of the spontaneously hypertensive rats: a key effect on NHE-1 activity**. *Acta Physiol (Oxf)* (2018) **224** e13092. DOI: 10.1111/apha.13092
37. Caplice NM, DeVoe MC, Choi J, Dahly D, Murphy T, Spitzer E. **Randomized placebo controlled trial evaluating the safety and efficacy of single low-dose intracoronary insulin-like growth factor following percutaneous coronary intervention in acute myocardial infarction (RESUS-AMI)**. *Am Heart J* (2018) **200** 110-7. DOI: 10.1016/j.ahj.2018.03.018
38. Osterziel KJ, Blum WF, Strohm O, Dietz R. **The severity of chronic heart failure due to coronary artery disease predicts the endocrine effects of short-term growth hormone administration**. *J Clin Endocrinol Metab* (2000) **85** 1533-9. DOI: 10.1210/jcem.85.4.6575
39. Conover CA, Bale LK, Harrington SC, Resch ZT, Overgaard MT, Oxvig C. **Cytokine stimulation of pregnancy-associated plasma protein A expression in human coronary artery smooth muscle cells: inhibition by resveratrol**. *Am J Physiol Cell Physiol* (2006) **290** C183-8. DOI: 10.1152/ajpcell.00199.2005
40. Hirai H, Kanaya R, Maeda M, Qungfang D, Ina K, Hayashi T. **The role of insulin growth factor on atherosclerosis and endothelial function: the effect on hyperlipidemia and aging**. *Life Sci* (2011) **88** 425-31. DOI: 10.1016/j.lfs.2010.12.021
41. Crowe FL, Key TJ, Allen NE, Appleby PN, Overvad K, Gronbaek H. **A cross-sectional analysis of the associations between adult height, BMI and serum concentrations of IGF-I and IGFBP-1 -2 and -3 in the European prospective investigation into cancer and nutrition (EPIC)**. *Ann Hum Biol* (2011) **38** 194-202. DOI: 10.3109/03014460.2010.507221
42. Lappas M, Jinks D, Shub A, Willcox JC, Georgiou HM, Permezel M. **Postpartum IGF-I and IGFBP-2 levels are prospectively associated with the development of type 2 diabetes in women with previous gestational diabetes mellitus**. *Diabetes Metab* (2016) **42** 442-7. DOI: 10.1016/j.diabet.2016.06.004
43. Wittenbecher C, Ouni M, Kuxhaus O, Jahnert M, Gottmann P, Teichmann A. **Insulin-like growth factor binding protein 2 (IGFBP-2) and the risk of developing type 2 diabetes**. *Diabetes* (2019) **68** 188-97. DOI: 10.2337/db18-0620
44. Brankovic M, Akkerhuis KM, Mouthaan H, Brugts JJ, Manintveld OC, van Ramshorst J. **Cardiometabolic biomarkers and their temporal patterns predict poor outcome in chronic heart failure (bio-SHiFT study)**. *J Clin Endocrinol Metab* (2018) **103** 3954-64. DOI: 10.1210/jc.2018-01241
45. Hassfeld S, Eichhorn C, Stehr K, Naegele H, Geier C, Steeg M. **Insulin-like growth factor-binding proteins 2 and 3 are independent predictors of a poor prognosis in patients with dilated cardiomyopathy**. *Heart* (2007) **93** 359-60. DOI: 10.1136/hrt.2006.090092
46. Griffiths M, Yang J, Nies M, Vaidya D, Brandal S, Williams M. **Pediatric pulmonary hypertension: insulin-like growth factor-binding protein 2 is a novel marker associated with disease severity and survival**. *Pediatr Res* (2020) **88** 850-6. DOI: 10.1038/s41390-020-01113-x
47. Chakrabarty S, Kondratick L. **Insulin-like growth factor binding protein-2 stimulates proliferation and activates multiple cascades of the mitogen-activated protein kinase pathways in NIH-OVCAR3 human epithelial ovarian cancer cells**. *Cancer Biol Ther* (2006) **5** 189-97. DOI: 10.4161/cbt.5.2.2333
48. Reustle A, Torzewski M. **Role of p38 MAPK in atherosclerosis and aortic valve sclerosis**. *Int J Mol Sci* (2018) **19**. DOI: 10.3390/ijms19123761
49. Mehrian-Shai R, Chen CD, Shi T, Horvath S, Nelson SF, Reichardt JK. **Insulin growth factor-binding protein 2 is a candidate biomarker for PTEN status and PI3K/Akt pathway activation in glioblastoma and prostate cancer**. *Proc Natl Acad Sci U S A* (2007) **104** 5563-8. DOI: 10.1073/pnas.0609139104
50. Ghafouri-Fard S, Khanbabapour Sasi A, Hussen BM, Shoorei H, Siddiq A, Taheri M. **Interplay between PI3K/AKT pathway and heart disorders**. *Mol Biol Rep* (2022) **49** 9767-81. DOI: 10.1007/s11033-022-07468-0
51. Shen X, Xi G, Maile LA, Wai C, Rosen CJ, Clemmons DR. **Insulin-like growth factor (IGF) binding protein 2 functions coordinately with receptor protein tyrosine phosphatase beta and the IGF-I receptor to regulate IGF-I-stimulated signaling**. *Mol Cell Biol* (2012) **32** 4116-30. DOI: 10.1128/MCB.01011-12
|
---
title: Integrating RNA-seq and scRNA-seq to explore the mechanism of macrophage ferroptosis
associated with COPD
authors:
- Pengbei Fan
- Yige Zhang
- Shenao Ding
- Zhixin Du
- Chunyu Zhou
- Xiaodan Du
journal: Frontiers in Pharmacology
year: 2023
pmcid: PMC10036582
doi: 10.3389/fphar.2023.1139137
license: CC BY 4.0
---
# Integrating RNA-seq and scRNA-seq to explore the mechanism of macrophage ferroptosis associated with COPD
## Abstract
Aims: Our study focused on whether macrophages ferroptosis is associated with the pathogenesis of chronic obstructive pulmonary disease (COPD) or not.
Main methods: We first identified macrophage module genes by weighted gene co-expression network analysis (WGCNA) in RNA sequencing (RNA-seq) date from COPD, and then identified macrophage marker genes by comprehensive analysis of single-cell RNA sequencing (scRNA-seq) data from COPD macrophages. There were 126 macrophage marker genes identified, and functional enrichment analyses indicated that ferroptosis pathway genes were significantly enriched. Secondly, we identified eight macrophage ferroptosis related genes and based on these eight genes, we performed co-expression analysis and drug prediction. Thirdly, two biomarkers (SOCS1 and HSPB1) were screened by the least absolute shrinkage and selection operator (LASSO), random forest (RF), and support vector machine-recursive feature elimination (SVM-RFE) and established an artificial neural network (ANN) for diagnosis. Subsequently, the biomarkers were validated in the dataset and validation set. These two biomarkers were then subjected to single gene-gene set enrichment analysis (GSEA) and gene set variation analysis (GSVA) analysis, and the ceRNA network was constructed. Finally, we carried out molecular validation with COPD models in vitro for cell counting kit-8 (CCK8) experiments, Western blot and quantitative real-time PCR (qRT-PCR) analysis and transmission electron microscopy (TEM).
Key findings: This study revealed the vital role of macrophage ferroptosis in COPD, and novel biomarkers (SOCS1 and HSPB1) may be involved in the pathogenesis of COPD by regulating macrophage ferroptosis.
Significance: Taken together, our results suggest that targeting SOCS1 and HSPB1 could treat COPD by inhibiting macrophage ferroptosis.
## 1 Introduction
As a major cause of morbidity, mortality, and healthcare use globally, the pathogenesis of chronic obstructive pulmonary disease (COPD) includes inflammation, oxidative stress, airway remodeling, and accelerated lung aging (Christenson et al., 2022; Yang et al., 2022). Characterized by chronic respiratory symptoms, COPD is an inflammatory lung condition (Wang et al., 2020). As the most abundant type of immune cell in the lung, macrophages are crucial to the inflammatory response associated with COPD (Higham et al., 2020). Several studies have demonstrated that macrophages contribute to the development of acute and chronic inflammatory responses through the secretion of inflammatory cytokines/chemokines and the activation of inflammation associated transcription factors (Singh et al., 2021; Li et al., 2022a). Furthermore, COPD patients have decreased macrophage clearance of apoptotic cells, resulting in persistent inflammation in the lungs (Grabiec and Hussell, 2016).
Death is the common fate for all life forms, from cells to organisms. In a healthy organism, cell death is an essential component of the development of the cell and the maintenance of internal environmental homeostasis, but when it is dysregulated, it can result in a variety of pathological consequences. Recently, researchers have proposed new types of death, including necroptosis, autophagic death, scorch death, and ferroptosis, based on different morphological characteristics of cell death (Han et al., 2020; Peng et al., 2022). Among these, ferroptosis, a form of regulated cell death discovered by Dixon et al., in 2012, has a significant influence on various diseases (Dixon et al., 2012). Several diseases have been linked to ferroptosis, such as tumors, neurodegenerative diseases, brain injuries, atherosclerosis, diabetes mellitus, and pulmonary inflammatory (Han et al., 2020; Sepand et al., 2021; Wang et al., 2021). Emerging evidence has revealed that pulmonary macrophage death plays important roles in the progression of inflammatory, while study shows ferroptosis strongly related to smoking and COPD progression (Liao et al., 2022). Moreover, pulmonary macrophage has a key role in the initiation and progression of the chronic inflammatory process in COPD lung tissue, a systematic understanding of macrophage ferroptosis involved in COPD is essential to explore key targets in COPD.
In recent years, single-cell RNA sequencing (scRNA-seq) has become a popular way to identify genes in various types of cells. In view of this advantage, we analyzed scRNA-seq and RNA sequencing (RNA-Seq) data from COPD macrophages and found that the pathogenesis of COPD is closely related to ferroptosis in macrophages. Afterwards, we identified ferroptosis related differential genes (DEGs) and pathways, screened for biomarkers, performed drug prediction, and constructed a ceRNA network. In addition, we studied further cellular experiments to verify the related gene expression, which will provide potential biomarkers and therapeutic targets for COPD.
## 2.1 Acquisition of scRNA-seq and RNA-Seq data
The scRNA-seq data of COPD macrophages and RNA-*Seq data* used in this study were screened from the public database GEO database platform (http://www.ncbi.nlm.nih.gov/geo). The scRNA-seq dataset (GSE183974) used is based on the Illumina NovaSeq assay platform and contains 9 samples. And the GSE13896 and GSE130928 were used as the dataset and validation set, respectively. GSE13896 included 12 COPD patients and 24 controls, while GSE130928 included 22 COPD patients and 24 controls.
## 2.2 Immune infiltration analysis
CIBERSORT inverse convolution estimates 22 immune cell types within tissues using linear support vector regression. Initially, CIBERSORT was used to calculate the relative abundance of immune cell types in samples. Using the Wilcoxon test, immune cells were compared among COPD samples and control samples. Lastly, R software was used to calculate correlation coefficients between different immune cells based on the results obtained through the CIBERSORT inverse convolution method.
## 2.3 Identifying macrophage-related modules
The weighted gene co-expression network analysis (WGCNA) technology is a high-throughput algorithm for analyzing gene expression data. Initially, we selected a soft threshold for constructing an adjacency matrix. The adjacency matrix was then converted to the topological overlap matrix (TOM) and the corresponding dissimilarity (1-TOM). In the following steps, a hierarchical clustering dendrogram was established to group genes with similar expression profiles into modules. Using the merged modules, associations were identified with immune cells, the module-immune cell correlation heat map was plotted and the module with the highest correlation with macrophages were screened.
## 2.4 Analysis of scRNA-seq data
The scRNA-seq data was converted into Seurat objects with the help of the R software “Seurat package”. Following quality control and data filtering, the top 1500 highly variable genes were analyzed using principal component analysis (PCA) to identify the 12 most significant principal components (PC). Then, using t-distributed stochastic neighbor embedding (t-SNE), we were able to achieve unsupervised clustering and unbiased visualization of cell subpopulations. *The* gene expression differences between clusters were compared using the “FindAllMarkers” package. To identify the marker genes for each cluster, |log2 FC (fold change)| > 1 and adjusted $p \leq 0.05$ were used. Additionally, subpopulations within each cluster were annotated using the “SingleR” package.
We drew a venn diagram by intersecting macrophage marker genes with macrophage module genes. Afterwards, the intersecting genes were analyzed for functional enrichment. Gene ontology (GO) and *Kyoto encyclopedia* of genes and genomes (KEGG) pathway enrichment analysis was done using the R package “clusterProfiler”, and the q-value and p-value cutoffs were set to 0.05. The results of the enrichment analysis were visualized using R software.
## 2.5 Identification and enrichment analysis of macrophages ferroptosis-related genes
To explore potential ferroptosis-related genes in macrophages, we intersected macrophage marker genes with ferroptosis-related genes obtained from FerrDb (http://www.zhounan.org/ferrdb/) and drew a Venn diagram, after which the obtained macrophage ferroptosis-related genes were subjected to functional enrichment analysis by R software.
## 2.6 Drug prediction of macrophage ferroptosis-related genes
To identify drugs that target genes, the Drug-Gene Interaction Database (DGIdb, http://dgidb.genome.wustl.edu/) be used. Based on the DGIdb database, we were able to identify potential drug candidates targeting macrophage ferroptosis-related genes. And the network of drug-genes was plotted using Cytoscape software.
## 2.7 Determination of macrophage ferroptosis-related DEGs
To begin with, DEGs were identified using the “limma” package and were defined according to the screening criteria (|log2FC| > 1, $p \leq 0.05$). The heatmap and volcano plot of DEGs were constructed using the R packages “Pheatmap” and “ggplot2”. We then conducted a Pearson’s correlation analysis on DEGs and macrophage ferroptosis-related genes to identify DEGs that may be associated with macrophage ferroptosis. Our thresholds were as follows: correlation coefficient (|r|) > 0.5 and $p \leq 0.001.$ Functional enrichment analysis was conducted on these macrophage ferroptosis-related DEGs.
## 2.8 Screening of biomarkers using machine learning techniques
Machine learning is the new tool for analyzing algorithms. The least absolute shrinkage and selection operator (LASSO) is a regularized regression algorithm using the “glmnet” package. Using Support vector machine-recursive feature elimination (SVM-RFE), features can be ranked based on recursion. Random forest (RF) analysis allows feature selection, calculates the mean decrease gini (MDG) for each gene and ranks them, and determines the optimal number of features by adding differential genes one by one from the largest to the smallest MDG value to determine the accuracy of the classification results. The three algorithms were used to better screen for biomarkers.
## 2.9 Artificial neural network construction
Using the packages “neuralnet” and “neuralnettools” in R software, we constructed an artificial neural network (ANN) model for the biomarkers based on the gene score. The number of hidden nodes was set to five. Based on the derived “gene score” information, a classification model for COPD was constructed. To evaluate the predictive performance of the ANN, the receiver operating characteristic curves (ROC) were used both on the dataset and on the validation set.
## 2.10 ROC analysis and validation of the biomarkers
In addition, the gene expression and diagnostic value of biomarkers was verified in both GSE13896 and GSE130928. We examined the diagnostic effectiveness of the biomarkers with the ROC using the “pROC” package. The expression levels of the biomarkers were also compared between COPD and control samples using an independent t-test, with a $p \leq 0.05$ considered statistically significant.
## 2.11 GSVA and GSEA analyses
According to the median expression levels of biomarkers in the GSE13896 dataset, COPD samples were divided into two groups (high and low-expression group). And single gene-gene set variation analysis (GSVA) and gene set enrichment analysis (GSEA) analyses were used to clarify the enriched KEGG pathways, using the gene set “c2. cp.kegg.v7.4. symbols.gmt” as a reference. Gene sets with $p \leq 0.05$ were treated as significantly enriched gene sets.
## 2.12 ceRNA network construction
To explore the ceRNA network, the miRNet (https://www.mirnet.ca/miRNet/home.xhtml) were performed to identify possible miRNA targeting biomarkers. Only the intersection of miRNAs targeting biomarkers from databases were selected for ceRNA network construction. Based on data from the ENCORI (http://starbase.sysu.edu.cn/index.php), we identified lncRNAs that target co-miRNAs. The ceRNA regulation network was then constructed. Cytoscape was used to visualize the network.
## 2.13 Cell lines and agents
Human monocyte leukemia cell line THP-1 were obtained from the Procell Life Science and Technology Collection (Procell, Wuhan, China) and cultured with RPMI1640 medium (Solarbio, Beijing, China) supplemented by $10\%$ fetal bovine serum (Gibco, Australia) in 37°C, $5\%$ CO2 humid incubator. 12-O-tetradecanoyl phorbol-13-acetate (PMA) was purchased from Glpbio, storage at 4°. THP-1 was induced by 100 ng/mL PMA for 24 h prior to the experiment to adhere and differentiate into macrophages.
Smoking is one of the major causes of COPD. In our experiment, we used cigarette smoke extract (CSE) to stimulate macrophages in vitro, to simulate macrophages of COPD patients. CSE was prepared as described. Briefly, three cigarettes (Yuxi, Hongta Tobacco Co., China) were ignition and combustion and the smoke was collected using a vacuum pump through a container containing the phosphate-buffered saline (PBS, 15 mL) maker as $100\%$ CSE. This $100\%$ CSE was adjusted to pH 7.4 and aseptically filtered through a 0.22-µm filter. CSE was prepared fresh for each experiment and diluted with medium containing $10\%$ FBS immediately prior to use.
## 2.14 Cell viability
THP-1 were planted with 6000 cells per well in 96-well plates and PMA induced macrophages for 24 h, then treated with several concentration of $2.5\%$, $5\%$, $10\%$ CSE for 12 h, 24 h, 48 h. After that, we used the cell counting kit-8 (CCK8; APExBIO, America) to measure cell viability. Refer to the manufacture’s instruction and add 10 μL/well CCK8 reagent to the cells. After incubated at 37°C for 3 h, incubated for 30min at 37°C. The Opticaldensity (OD) value was measured at 450 nm using a Microplate Reader (Bio-Tek, Vermont, United States).
## 2.15 Western blotting
THP-1 were planted in 6-well plates with 2*105 cells per well and PMA induced macrophages for 24 h, then treated with specified concentration of $5\%$ CSE or PBS for 12 h, 24 h, 48 h. The cells were washed with PBS for three times. Collecting cells in each group by trypsin digestion, adding RIPA (Servicebio, G2002-100 ML) lysis buffer containing PMSF (Servicebio, G2008-1 Ml), collecting that supernatant to obtain total protein solution, then protein denaturation. After samples were separated by $12\%$ SDS-PAGE gel electrophoresis, transferred to PVDF membranes (Servicebio, WGPVDF45), blocked with $5\%$ defatted milk for 1 h, the membrane was incubated overnight with primary antibodys (1:1000 diluted GPX4, Servicebio, GB113091; 1:5000 diluted GAPDH, Proteintech, 6004–1) at 4°C, followed by TBST washing three times the next day, incubation with secondary antibody (1:20,000 diluted) for 1 h. Following incubation and washing, uniformly dropwise added that high-sensitivity ECL luminescent liquid (Servicebio, G2014-50 ML). The bands were detected using the Bio-RadChemiDoc MP luminescence imaging system. The images in the film were quantified for the gray value using ImageJ.
## 2.16 Determination of GPX4, SOCS1 and HSPB1 secretion in macrophages
THP-1 were planted in 6-well plates with 2*105 cells per well and PMA induced macrophages for 24 h, then treated with specified concentration of $5\%$ CSE or PBS for 12 h, 24 h, 48 h. These total RNA was extracted from $5\%$ CSE-induced macrophages as a COPD model (CSE12h, CSE24h, CSE48h) and $5\%$ PBS-induced macrophages as a control group (Con), using RNA-Quick Purification Kit (EScience, Shanghai, China) and quantified by Nanodrop spectrophotometer (NanoDrop 2000, Thermo, United States). Complementary cDNA was synthesized from total RNA using the Servicebio®RT First Strand cDNA Synthesis Kit (Servicebio, Wuhan, China). The qRT-PCR assay was performed using the SYBR Green qPCR Master Mix (None ROX) (Servicebio, Wuhan, China) on Quant Studio Real-Time PCR System (Thermo Fisher Scientific, United States). Determination of GPX4, SOCS1 and HSPB1 secretion were normalized to GAPDH and calculated with the 2−ΔΔCT method. Three biotechnical replicates were set up for each sample. The primer sequences used in this study see in Table 1.
**TABLE 1**
| Human | Forward primer (5′-3′) | Reverse primer | Fragment length (bp) |
| --- | --- | --- | --- |
| GPX4 | TGAAGATCCAACCCAAGGGC | GACGGTGTCCAAACTTGGTG | 75 |
| SOCS1 | GACACGCACTTCCGCACATT | TAGAATCCGCAGGCGTCCA | 86 |
| HSPB1 | AGCATGGCTACATCTCCCGG | GACTCGAAGGTGACTGGGATG | 172 |
| GAPDH | GGAAGCTTGTCATCAATGGAAATC | TGATGACCCTTTTGGCTCCC | 168 |
## 2.17 Transmission electron microscopy
THP-1 were planted in 6-well plates with 2*105 cells per well and PMA induced macrophages for 24 h, then treated with the concentration of $5\%$ CSE or PBS for 48 h. Macrophages were stimulated with $5\%$ CSE or PBS for 48 h. Macrophages were gently scraped from the culture dish using a cell scraper, and the cells were collected by centrifugation in the 1.5 ml Eppendorf tubes. And medium was removed and pre-chilled $2.5\%$ glutaraldehyde (pH 7.4) solution was added for 2 h at 4°C, resuspended and fixed at 4 °C. Then macrophages were fixed and stored at 4°C. Cells or bacteria were centrifuged by centrifuge, supernatant was discarded and 0.1 M phosphate buffer (pH 7.4) was added, mixed and rinsed for 3 min before centrifugation, and the washing was repeated 3 times. Prepare $1\%$ agarose solution by heating and dissolving in advance, add to EP tubes after cooling slightly, and pick up the precipitate with forceps before the agarose solidifies and wrap it in agarose in suspension. The specimens were sequentially dehydrated using $30\%$ alcohol −$50\%$ alcohol −$70\%$ alcohol −$80\%$ alcohol −$95\%$ alcohol −$100\%$ alcohol −$100\%$ alcohol for every 20 min then $100\%$ acetone twice for every 15 min. Finally, after permeabilization, embedding, and sectioning, the samples were observed under transmission electron microscopy and images were collected for analysis.
## 3.1 Identification of macrophage module genes
The results of CIBERSORT showed that both M0 macrophage and M1 macrophages were altered in the COPD group (Figure 1A). Figure 1B illustrates the immune cell heat map of COPD, and Figure 1C illustrates the correlation analysis, which showed a remarkable correlation in the 22 types of immune cells. Following this, we performed the WGCNA analysis. As a first step, we set the threshold at 0.25 to merge similar modules (Figure 1D). Then, we chose β = 16 as a suitable soft threshold for the construction of a scale-free network (Figure 1E). After merging similar gene modules, a dynamic cut tree was generated (Figure 1G). Of the 23 gene modules, the midnight blue module shows a close relationship with the features of M1 macrophages and M0 macrophages (Figure 1H). In the end, we obtained 763 macrophage module genes.
**FIGURE 1:** *Immune infiltration analysis and identification of macrophage-related modular genes. (A) The CIBERSORT algorithm was used to analyze the content of immune cells. (B) A holistic view of the distribution of immune cells between different tissues. (C) Correlation analysis was performed on all immune cells in the CIBERSORT algorithm. (D) Correlation diagram between modules obtained by clustering according to inter-gene expression levels. (E) Screening for suitable soft thresholds and scale-free network validation. The soft threshold is selected as 16. (F) The cluster dendrogram with the gene modules and module merging. (G) The correlation between gene modules and immune cell fraction.*
## 3.2 Identification of macrophage marker genes and functional enrichment analysis
Figure 2A shows the range of gene numbers detected, the depth of sequencing, and the percentage of mitochondrial content in each sample. Following quality control and normalization of the data, we selected the top 1500 genes with high variability (Figure 2B). The PCA method was used to reduce dimensionality (Figure 2C), and 12 PCs with $p \leq 0.05$ were selected for further analysis (Figure 2D). We identified 5 clusters using the t-SNE algorithm. The heatmap illustrating the relative expression of five cluster marker genes (Figure 2E). Following this, we visualized the five clusters (Figure 2F). The “singleR” algorithm was used to annotate cell subpopulations, and clusters 0, 1, 2, 3 and 4 were identified as macrophage subpopulations (Figure 2G). Finally, we identified 184 macrophage marker genes.
**FIGURE 2:** *Identification of macrophage cell marker genes by scRNA-seq analysis. (A) Quality control of scRNA-seq data. (B) The variance plot showed 14,175 genes in all cells, red dots represent the top 1500 highly variable genes. (C) PCA was used for dimensionality reduction. (D) 12 PCs were identified based on p < 0.05. (E) The heatmap showed the relative expression of genes in 5 clusters. Yellow represents high expressed genes and purple represents low expressed genes. (F) 5 clusters were visualized based on the t-SNE algorithm. (G) Cell subpopulations identified by marker genes. (H) Venn diagram illustrating macrophage module genes and macrophage marker genes. (I,J) GO and KEGG analysis of macrophage marker genes.*
Interestingly, we found that the macrophage marker genes completely overlapped with the macrophage module genes (Figure 2H). After that, GO analysis showed that the biological processes (BP) were mainly focused on neutrophil activation, neutrophil-mediated immunity and neutrophil activation involved in immune response. Cell composition (CC) was mainly focused on endoplasmic membrane, vacuolar membrane, vesicular membrane and lysosomal membrane and molecular functions (MF) were mainly focused on amide binding, peptide binding, amyloid-beta binding and proteoglycan binding. The KEGG signaling pathway was significantly enriched in cell adhesion molecules, NF-κB signaling pathway, apoptosis and ferroptosis. Ferroptosis, as a pathway enriched by COPD macrophage marker genes, is likely to play a great role in the pathology of COPD, therefore, we selected ferroptosis-related genes for the next study.
## 3.3 Identification and functional enrichment analysis of macrophage ferroptosis-related genes
A total of 8 macrophage ferroptosis-related genes were identified (Figure 3A), and further functional enrichment analysis revealed that GO analysis (Figure 3B) showed that the biological processes were mainly focused on the lipoxygenase pathway and vascular endothelial growth factor receptor signaling pathway. The cellular composition was mainly focused on autophagosomes, NADPH oxidase complex, secondary cell bodies. The molecular functions were mainly focused on iron binding, and KEGG analysis (Figure 3C) showed that Leukocyte trans endothelial migration, necroptosis and ferroptosis signaling pathways were significantly enriched.
**FIGURE 3:** *Macrophage ferroptosis-related genes identification and drug prediction. (A) Venn plot exhibiting the macrophage marker genes and ferroptosis-related genes. (B,C) GO and KEGG analysis of Macrophage ferroptosis-related genes. (D) The drug and macrophage ferroptosis-related genes interaction network.*
## 3.4 Identification of the potential drugs
A total 61 compounds or drugs corresponding to genes were identified. As shown in Figure 3D, two drugs target SOCS1, 31 drugs target HSPB1, 25 drugs target ALOX5, and three drugs target CYBB. Among them, insulin (targeting SOCS1), quercetin (targeting HSPB1), zileuton (targeting ALOX5) and apigenin (targeting CYBB) could be the potential effective drugs in the future.
## 3.5 Identification and functional enrichment analysis of macrophage ferroptosis‐related DEGs
Next, we aimed to identify macrophage ferroptosis-elated DEGs. First, 845 DEGs were identified, and heat maps and volcano maps are shown in Figures 4A,B. In total, 121 macrophage ferroptosis-related DEGs were identified by correlation analysis (Figure 4C). According to functional enrichment analysis (Figure 4D), macrophage ferroptosis-related DEGs are functionally enriched in the regulation of phosphatidylinositol 3-kinase signaling, phosphatidylinositol 3-kinase signaling and MAPK signaling pathway.
**FIGURE 4:** *Identification and enrichment analysis of macrophage ferroptosis-related DEGs. (A,B) heat maps and volcano maps of GSE13896. (C) Macrophage ferroptosis related genes and DEGs interaction network. (D) GO and KEGG pathways enriched by the macrophage ferroptosis-related DEGs.*
## 3.6 Construction of biomarkers
Four diagnostic genes were identified by the RF algorithm (Figures 5A, B). By SVM-RFE algorithm, six genes were extracted as candidate biomarkers (Figure 5C). Using the LASSO regression algorithm, two genes were identified as potential diagnostic biomarkers (Figure 5D). Using a Venn diagram, two genes (SOCS1 and HSPB1) were then overlapping and served as robust diagnostic biomarkers.
**FIGURE 5:** *Identification and validation of the biomarkers. (A,B) RF algorithm. (C) SVM-RFE algorithm. (D) LASSO regression analysis. (E) Venn plot exhibiting the reliable biomarkers among LASSO, SVM-RFE, and RF. (F) The visualization of the ANN diagnostic model. (G) The assessment result of the date set. (H) The testification result of the validation set. (I) Validation of the key crosstalk genes in GSE13896. (J) Validation of the key crosstalk genes in GSE130928.*
## 3.7 Development and validation of an ANN model
Based on gene weight, an ANN diagnostic model was developed (Figure 5F). During testing of the model, the AUC was 0.816 for the date set and 0.777 for the validation set, indicating that the model performed very well in diagnosing COPD. Results showed that we had developed a good diagnostic model between COPD samples and controls.
## 3.8 ROC analysis and validation of the biomarkers
We validated the biomarkers in GSE13896 (Figure 5I) and GSE130928 (Figure 5J) to further verify their efficacy. According to the ROC results in the GSE13896 dataset, HSPB1 and SOCS1 were effective at discriminating between COPD and control samples. Furthermore, the expression of the biomarkers was significantly lower in COPD samples, suggesting that HSPB1 and SOCS1 are strongly associated with COPD. In addition, the ROC results showed good diagnostic efficacy in the validation dataset, and the levels of mRNA expression of the biomarkers were lower in case groups.
## 3.9 GSVA and GSEA analysis
GSEA analysis showed significant differences between different expression groups of HSPB1 in autophagy, asthma, and glutathione metabolism signaling pathways (Figure 6A), and different expression groups of SOCS1 in autophagy and folate biosynthesis signaling pathways, etc. ( Figure 6C). After that, GSVA was used to analyze the relevant signaling pathways regulated by biomarkers. The results showed that the Hedgehog signaling pathway was highly enriched in the high expression group of HSPB1, and the low expression of HSPB1 was mainly enriched in asthma and glutathione metabolic signaling pathway (Figure 6B); the low expression of SOCS1 was mainly enriched in folate biosynthesis signaling pathway and so on (Figure 6D).
**FIGURE 6:** *Single gene GSEA and GSVA analysis. (A,B) GSEA and GSVA analysis of HSPB1. (C,D) GSEA and GSVA analysis of SOCS1.*
## 3.10 Construction of ceRNA network
The miRNet database was used to extract the target miRNAs of the 2 biomarkers of macrophage ferroptosis, and a total of 84 miRNAs were obtained, among which hsa-mir-146a-5p, hsa-mir-155–5p and hsa-mir-34a-5p were the co-miRNAs of the 2 biomarkers. After that, the 3 shared target miRNAs were entered into the ENCORI database, 114 target lncRNAs were obtained (Figure 7), among which, only XIST could target the 3 co-miRNAs simultaneously.
**FIGURE 7:** *The ceRNA network. The ceRNA network was constructed through Cytoscape software. Triangles represent miRNAs, circles represent biomarkers, and lozenges represent lncRNAs.*
## 3.11 Cellular experimental validation
Cellular experimental validation revealed no significant effect of CSE at $2.5\%$, $5\%$ on macrophage viability at 12 h, with large changes in cell viability, and at 24 h and 48 h, CSE ($5\%$ concentration) significantly reduced macrophage viability, but the effect on viability was <$50\%$, compared to $10\%$ CSE which had >$50\%$ effect on cell viability (Figure 8A). Therefore, we used $5\%$ CSE in the subsequent experiments. GPX4 played a vital role in ferroptosis. Though cellular experimental validation, we detected that GPX4 decreased in COPD macrophage (Figures 8B,C). This validates the relevance of macrophage ferroptosis to COPD. Furthermore, our results showed that the biomarkers HSPB1 and SOCS1 were significantly reduced in macrophages with increasing duration of CSE stimulation (Figures 8D,E). One of the most prominent morphological features of ferroptosis is mitochondrial changes. Compared with the control group, mitochondria are wrinkled, cristae are reduced, and membrane density is increased in CSE-induced 48 h group (CSE group), showing a typical expression of ferroptosis (Figure 8F).
**FIGURE 8:** *Cell viability and molecular validation. (A) Cell viability of macrophages stimulated by different concentrations of CSE. (B) Western blot for probing the expression of GPX4. (C) GPX4 expression after CSE stimulation at different time. (D) HSPB1 expression after CSE stimulation at different time. (E) SOCS1 expression after CSE stimulation at different time. *p < 0.05, **p < 0.01, ***p < 0.001, the CSE-induced group (CSE group) vs. the control group (Con). (F) Observation of mitochondrial morphology (orange arrows) using TEM.*
## 4 Discussion
The high prevalence and mortality of COPD impose a heavy economic burden on society and families, with high annual direct medical costs, direct non-medical costs and indirect costs, and there are no effective measures to reverse the continuous decline of lung function in COPD patients (Adeloye et al., 2022). Due to the heterogeneity of COPD pathological mechanisms, its treatment is very challenging. Research has shown that macrophages play a significant role in the remodeling of small airways and chronic inflammation associated with this disease (Kaku et al., 2014; Lee et al., 2021). Genes associated with macrophages and their biological functions may provide insight into the molecular mechanisms underlying COPD. Therefore, the novel biomarkers identification of macrophages based on macrophage RNA-seq and scRNA-seq data is one of the key tasks to refine personalized and targeted drug use in COPD in the future.
In this study, we integrated scRNA-seq data of macrophages in COPD alveolar lavage fluid as well as RNA-*Seq data* and identified a total of 126 macrophage marker genes. Functional enrichment suggests that macrophage marker genes are mainly enriched in ferroptosis-related pathways. Macrophage ferroptosis has been found to play a vital role in lung disease, but no study has yet explored the role of macrophage ferroptosis in COPD. A total of eight COPD macrophage ferroptosis-related genes were identified in this study, including ALOX5, NCF2, CYBB, HSPB1, GPX4, FTH1, FTL, SOCS1. Researchers found that ALOX5 was significantly upregulated in macrophages of COPD mice, which led to lipid peroxidation and ferroptosis in alveolar macrophages, thereby exacerbating COPD (Gunes Gunsel et al., 2022). NCF2 is a key transcription factor in the antioxidant response, and can active the nicotinamide adenine dinucleotide phosphate oxidase (NOXs), then ferroptosis could be triggered (Li et al., 2022a). CYBB, also referred to as NOX2, is mainly expressed in lung macrophages (Seimetz et al., 2020). And NOX2 protein expression is found to be upregulated in the lungs of COPD mice and is involved in the oxidative and inflammatory response in the early stages of COPD (Zhang et al., 2022). Heat shock protein (HSP) is a potent inducer of natural and adaptive immunity, and the association between increased expression levels of HSPB1 and the risk of COPD has been reported and verified (Cui et al., 2015). The lipid repair enzyme GPX4, a member of the selenoprotein family, has been shown to act as a negative regulator of ferroptosis (Ji et al., 2022). Ferroptosis has an important role in the pathogenesis of COPD, and GPX4 expression levels were found to be significantly decreased in bronchial epithelial cells of COPD patients (Yoshida et al., 2019). FTL encodes ferritin light chain, which is involved in iron metabolism and utilization, and FTL protein can regulate ferroptosis by controlling free iron levels. FTH1 is responsible for intracellular iron storage and iron metabolism and altered expression of FTH1 is closely associated with the occurrence of ferroptosis. SOCS1 is a signal transducer of Janus kinase (JAK) and a negative regulator of the activator of transcription (STAT) pathway. Recent studies have found that SOCS1 is associated with the pathogenesis of COPD and is related to the time of COPD onset (Liao et al., 2022).
Based on these eight COPD macrophage ferroptosis-related genes, we performed co-expression analysis as well as drug prediction, and we identified a total of 121 COPD macrophage ferroptosis related DEGs that were significantly enriched in the MAPK signaling pathway. MAPK is a group of protein kinases that phosphorylate serine and threonine. Three signaling pathways are known, JNK$\frac{1}{2}$/3, REK$\frac{1}{2}$ and p38, which can regulate stress, apoptosis, immune defense and other biological processes. Recent studies have found that ferroptosis is closely related to the MAPK signaling pathway (Guo et al., 2022). In addition, MAPK signaling pathways has been shown to induce ferroptosis and to regulate the polarization of macrophages (Li et al., 2018). During the present study, significant reductions in M1 macrophages were observed, resulting in an imbalance in the polarization between M1 and M2 macrophages, which further exacerbates the pathology of the disease (Finicelli et al., 2022).
To better guide COPD treatment, we made drug predictions for these eight macrophage ferroptosis-related genes. In our study, insulin was predicted to target SOCS1 and treat COPD. As a common comorbidity of COPD, insulin resistance is mainly associated with smoking, obesity, genetics, lack of exercise, and long-term application of glucocorticoids, which can aggravate airway inflammation in COPD patients (Anker et al., 2022; Park et al., 2022). In addition, quercetin, which targets HSPB1, is a phenolic compound belonging to the flavonoid family and is found in various foods or agricultural products, such as onions, apples and broccoli, with high antioxidant activity (Li et al., 2022b). Quercetin can alleviate the course of COPD through various mechanisms, such as reducing lung inflammation and macrophage infiltration, and is expected to be one of the drugs for COPD (Araujo et al., 2022). Leukotrienes play an important role in the pathogenesis of COPD, and 5-LO can regulate the production of leukotrienes, while zileuton, an inhibitor of 5-LO, can significantly improve the exercise ability and quality of life in COPD exacerbation patients with a better safety profile (Garland et al., 2022). As a natural flavonoid present in a variety of herbs, lignan can reduce the expression of inflammatory factors and eliminate oxygen free radicals in LPS-induced mouse macrophage models, which may have therapeutic effects on COPD (Li et al., 2021). Since the efficacy of these drugs and targets is based solely on theoretical predictions, further animal experiments and clinical trials are required to provide more evidence.
Then, we further screened two biomarkers (HSPB1 and SOCS1) by multiple machine learning and built an ANN diagnostic model to evaluate their performance. The performance was assessed by AUC (0.816). Also, we testified the diagnostic ability in the validation set and the AUC was 0.777. Together, the developed diagnostic model could offer a novel perspective on our research of the mechanism of COPD.
In addition, we performed ROC analysis and external dataset validation, and performed single gene GSEA and GSVA analysis for biomarkers. HSPB1 expression was found to be closely related to the Hedgehog signaling pathway as well as glutathione metabolism. Recent studies suggest that Hedgehog signaling could be altered in COPD, promotes airway inflammatory processes and induces apoptosis in bronchial epithelial cells (Lahmar et al., 2022). Therefore, inhibition of Hedgehog signaling activation may serve as a new therapeutic. Glutathione is a major intracellular antioxidant that correlates with the severity of COPD. Glutathione levels are increased in bronchoalveolar lavage fluid of patients with stable COPD, whereas glutathione levels are reduced in patients with deteriorating COPD, suggesting that increasing intrapulmonary glutathione levels may be a new therapeutic strategy (Engelen et al., 2004). Further, SOCS1 is primarily involved in the folate biosynthesis signaling pathway, and folate levels were significantly decreased in patients with COPD (Kim et al., 2020).
We also identified upstream miRNAs and lncRNAs of those two biomarkers. We found that three of miRNAs (miR-34a-5p, miR-155–5p and miR-146a-5p) can target two biomarkers at the same time, so they are considered as key miRNAs (Figure 7). The miR-34a-5p has been found to be elevated in the airway epithelium of COPD patients (Zeng et al., 2022). In addition, miR-146a-5p in lung fibroblasts can inhibit IL-8 secretion, while miR-146a-5p expression is reduced in COPD patients, and the inhibitory effect on IL-8 secretion is diminished, thus exacerbating the development of chronic inflammation and COPD (Osei et al., 2017). Moreover, miR-155–5p is a classical multifunctional miRNA involved in the development, differentiation, activation, and maintenance of homeostasis of macrophages. For example, miR-155–5p expression was increased in alveolar macrophages of COPD mice (De Smet et al., 2020). Subsequently, we found that XIST (a key regulator of inflammatory response, that is significantly upregulated in lung tissue of COPD patients (Chen et al., 2021)) can simultaneously target three key miRNAs. And XIST is reported to negatively regulate the expression of miR-155–5p (Zhang et al., 2021; Wang and Cao, 2022). Moreover, SOCS1 and miR-155 are believed to be key regulators of the inflammatory response. The expression of miR-155 has also been demonstrated to be induced in animals by smoke, and miR-155 plays a role in the inflammatory response to lung injury by inhibiting the expression of SOCS1 (Zhang et al., 2020). Thus, we propose that the XIST/miR-155–5p/SOCS1 axis plays a significant role in the development and progression of COPD.
Despite the contribution of this study to the understanding of COPD macrophages, our study has several limitations due to small sample size and lack of information-rich sample data: first, the accuracy of COPD assessment and prediction could be improved by increasing the sample size; second, the macrophage biomarkers identified in this study with potential drugs and associated pathways need to be further validated to provide practical evidence for clinically targeted therapies; Third, analysis of protein expression levels of marker macrophages genes could provide substantial evidence. And the ferroptosis-induced cell death assay is necessary, some ferroptosis inhibitors or agonists may be used in experimental research to perfect this scientific problem. Despite recent advances in understanding, the relationship between macrophage ferroptosis and the HSPB1 and SOCS1 level needs to be further explored. In addition, the relationship between macrophage ferroptosis and chronic inflammation of COPD deserves to be investigated. Therefore, more extensive in vivo and in vitro experiments are necessary for further validation in the future.
## 5 Conclusion
In conclusion, we explored the two biomarkers (HSPB1 and SOCS1) of macrophage ferroptosis in COPD by combining scRNA-seq and RNA-seq with a view to treating COPD. Our study provides new theoretical insights into the role of macrophage ferroptosis biomarkers in COPD.
## Data availability statement
The original contributions presented in the study are included in the article/supplementary material, further inquiries can be directed to the corresponding authors.
## Author contributions
PF and ZD designed, wrote and polished the main manuscript. ZD, YZ, and XD performed the differential analysis. SD, CZ, and XD performed the molecular validation. All authors listed have made a substantial, direct, and intellectual contribution to the work and approved the final version of the manuscript for publication.
## Conflict of interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher’s note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## References
1. Adeloye D., Song P., Zhu Y., Campbell H., Sheikh A., Rudan I.. **Global, regional, and national prevalence of, and risk factors for, chronic obstructive pulmonary disease (COPD) in 2019: A systematic review and modelling analysis**. *Lancet Respir. Med.* (2022) **10** 447-458. DOI: 10.1016/S2213-2600(21)00511-7
2. Anker S. D., Sander L. E., Fitchett D. H., Zinman B., Pernille Ofstad A., Wanner C.. **Empagliflozin in patients with type 2 diabetes mellitus and chronic obstructive pulmonary disease**. *Diabetes Res. Clin. Pract.* (2022) **186** 109837. DOI: 10.1016/j.diabres.2022.109837
3. Araujo N., de Matos N. A., Oliveira M., de Souza A. B. F., Castro T. F., Machado-Junior P. A.. **Quercetin improves pulmonary function and prevents emphysema caused by exposure to cigarette smoke in male mice**. *Antioxidants (Basel)* (2022) **11** 181. DOI: 10.3390/antiox11020181
4. Chen P., Jiang P., Chen J., Yang Y., Guo X.. **XIST promotes apoptosis and the inflammatory response in CSE-stimulated cells via the miR-200c-3p/EGR3 axis**. *BMC Pulm. Med.* (2021) **21** 215. DOI: 10.1186/s12890-021-01582-8
5. Christenson S. A., Smith B. M., Bafadhel M., Putcha N.. **Chronic obstructive pulmonary disease**. *Lancet (London), Engl.* (2022) **399** 2227-2242. DOI: 10.1016/S0140-6736(22)00470-6
6. Cui X., Xing J., Liu Y., Zhou Y., Luo X., Zhang Z.. **COPD and levels of Hsp70 (HSPA1A) and Hsp27 (HSPB1) in plasma and lymphocytes among coal workers: A case-control study**. *Cell Stress Chaperones* (2015) **20** 473-481. DOI: 10.1007/s12192-015-0572-5
7. De Smet E. G., Van Eeckhoutte H. P., Avila Cobos F., Blomme E., Verhamme F. M., Provoost S.. **The role of miR-155 in cigarette smoke-induced pulmonary inflammation and COPD**. *Mucosal Immunol.* (2020) **13** 423-436. DOI: 10.1038/s41385-019-0241-6
8. Dixon S. J., Lemberg K. M., Lamprecht M. R., Skouta R., Zaitsev E. M., Gleason C. E.. **Ferroptosis: An iron-dependent form of nonapoptotic cell death**. *Cell* (2012) **149** 1060-1072. DOI: 10.1016/j.cell.2012.03.042
9. Engelen M. P., Orozco-Levi M., Deutz N. E., Barreiro E., Hernandez N., Wouters E. F.. **Glutathione and glutamate levels in the diaphragm of patients with chronic obstructive pulmonary disease**. *Eur. Respir. J.* (2004) **23** 545-551. DOI: 10.1183/09031936.04.00022204
10. Finicelli M., Digilio F. A., Galderisi U., Peluso G.. **The emerging role of macrophages in chronic obstructive pulmonary disease: The potential impact of oxidative stress and extracellular vesicle on macrophage polarization and function**. *Antioxidants (Basel)* (2022) **11** 464. DOI: 10.3390/antiox11030464
11. Garland L. L., Guillen-Rodriguez J., Hsu C. H., Davis L. E., Szabo E., Husted C. R.. **Clinical study of aspirin and zileuton on biomarkers of tobacco-related carcinogenesis in current smokers**. *Cancers (Basel)* (2022) **14** 2893. DOI: 10.3390/cancers14122893
12. Grabiec A. M., Hussell T.. **The role of airway macrophages in apoptotic cell clearance following acute and chronic lung inflammation**. *Seminars Immunopathol.* (2016) **38** 409-423. DOI: 10.1007/s00281-016-0555-3
13. Gunes Gunsel G., Conlon T. M., Jeridi A., Kim R., Ertuz Z., Lang N. J.. **The arginine methyltransferase PRMT7 promotes extravasation of monocytes resulting in tissue injury in COPD**. *Nat. Commun.* (2022) **13** 1303. DOI: 10.1038/s41467-022-28809-4
14. Guo B., Zuo Z., Di X., Huang Y., Gong G., Xu B.. **Salidroside attenuates HALI via IL-17A-mediated ferroptosis of alveolar epithelial cells by regulating Act1-TRAF6-p38 MAPK pathway**. *Cell Commun. Signal* (2022) **20** 183. DOI: 10.1186/s12964-022-00994-1
15. Han C., Liu Y., Dai R., Ismail N., Su W., Li B.. **Ferroptosis and its potential role in human diseases**. *Front. Pharmacol.* (2020) **11** 239. DOI: 10.3389/fphar.2020.00239
16. Higham A., Scott T., Li J., Gaskell R., Dikwa A. B., Shah R.. **Effects of corticosteroids on COPD lung macrophage phenotype and function**. *Clin. Sci. Lond. Engl.* (2020) **134** 751-763. DOI: 10.1042/CS20191202
17. Ji Q. X., Zeng F. Y., Zhou J., Wu W. B., Wang X. J., Zhang Z.. **Ferroptotic stress facilitates smooth muscle cell dedifferentiation in arterial remodelling by disrupting mitochondrial homeostasis**. *Cell Death Differ.* (2022) **30** 457-474. DOI: 10.1038/s41418-022-01099-5
18. Kaku Y., Imaoka H., Morimatsu Y., Komohara Y., Ohnishi K., Oda H.. **Overexpression of CD163, CD204 and CD206 on alveolar macrophages in the lungs of patients with severe chronic obstructive pulmonary disease**. *PLoS One* (2014) **9** e87400. DOI: 10.1371/journal.pone.0087400
19. Kim T., Oak C. H., Jung M. H., Jang T. W., Kim J.. **High serum folate concentration is associated with better lung function in male chronic obstructive pulmonary disease patients who are current smokers: Analysis of nationwide population-based survey**. *Nutrients* (2020) **12** 2219. DOI: 10.3390/nu12082219
20. Lahmar Z., Ahmed E., Fort A., Vachier I., Bourdin A., Bergougnoux A.. **Hedgehog pathway and its inhibitors in chronic obstructive pulmonary disease (COPD)**. *Pharmacol. Ther.* (2022) **240** 108295. DOI: 10.1016/j.pharmthera.2022.108295
21. Lee J-W., Chun W., Lee H. J., Min J-H., Kim S-M., Seo J-Y.. **The role of macrophages in the development of acute and chronic inflammatory lung diseases**. *Cells* (2021) **10** 897. DOI: 10.3390/cells10040897
22. Li B. S., Zhu R. Z., Lim S. H., Seo J. H., Choi B. M.. **Apigenin alleviates oxidative stress-induced cellular senescence via modulation of the SIRT1-NAD[formula: See text]-CD38 Axis**. *Am. J. Chin. Med.* (2021) **49** 1235-1250. DOI: 10.1142/S0192415X21500592
23. Li L., Hao Y., Zhao Y., Wang H., Zhao X., Jiang Y.. **Ferroptosis is associated with oxygen-glucose deprivation/reoxygenation-induced Sertoli cell death**. *Int. J. Mol. Med.* (2018) **41** 3051-3062. DOI: 10.3892/ijmm.2018.3469
24. Li L. Y., Zhang C. T., Zhu F. Y., Zheng G., Liu Y. F., Liu K.. **Potential natural small molecular compounds for the treatment of chronic obstructive pulmonary disease: An overview**. *Front. Pharmacol.* (2022) **13** 821941. DOI: 10.3389/fphar.2022.821941
25. Li M., Xin S., Gu R., Zheng L., Hu J., Zhang R.. **Novel diagnostic biomarkers related to oxidative stress and macrophage ferroptosis in atherosclerosis**. *Oxid. Med. Cell Longev.* (2022) **2022** 8917947. DOI: 10.1155/2022/8917947
26. Liao S. X., Chen J., Zhang L. Y., Zhang J., Sun P. P., Ou-Yang Y.. **Effects of SOCS1-overexpressing dendritic cells on Th17- and Treg-related cytokines in COPD mice**. *BMC Pulm. Med.* (2022) **22** 145. DOI: 10.1186/s12890-022-01931-1
27. Osei E. T., Florez-Sampedro L., Tasena H., Faiz A., Noordhoek J. A., Timens W.. **miR-146a-5p plays an essential role in the aberrant epithelial-fibroblast cross-talk in COPD**. *Eur. Respir. J.* (2017) **49** 1602538. DOI: 10.1183/13993003.02538-2016
28. Park S. S., Perez Perez J. L., Perez Gandara B., Agudelo C. W., Rodriguez Ortega R., Ahmed H.. **Mechanisms linking COPD to type 1 and 2 diabetes mellitus: Is there a relationship between diabetes and COPD?**. *Med. Kaunas.* (2022) **58** 1030. DOI: 10.3390/medicina58081030
29. Peng F., Liao M., Qin R., Zhu S., Peng C., Fu L.. **Regulated cell death (RCD) in cancer: Key pathways and targeted therapies**. *Signal Transduct. Target. Ther.* (2022) **7** 286-366. DOI: 10.1038/s41392-022-01110-y
30. Seimetz M., Sommer N., Bednorz M., Pak O., Veith C., Hadzic S.. **NADPH oxidase subunit NOXO1 is a target for emphysema treatment in COPD**. *Nat. Metab.* (2020) **2** 532-546. DOI: 10.1038/s42255-020-0215-8
31. Sepand M. R., Maghsoudi A. S., Shadboorestan A., Mirnia K., Aghsami M., Raoufi M.. **Cigarette smoke-induced toxicity consequences of intracellular iron dysregulation and ferroptosis**. *Life Sci.* (2021) **281** 119799. DOI: 10.1016/j.lfs.2021.119799
32. Singh R., Belchamber K. B. R., Fenwick P. S., Chana K., Donaldson G., Wedzicha J. A.. **Defective monocyte-derived macrophage phagocytosis is associated with exacerbation frequency in COPD**. *Respir. Res.* (2021) **22** 113. DOI: 10.1186/s12931-021-01718-8
33. Wang C., Zhou J., Wang J., Li S., Fukunaga A., Yodoi J.. **Progress in the mechanism and targeted drug therapy for COPD**. *Signal Transduct. Target. Ther.* (2020) **5** 248. DOI: 10.1038/s41392-020-00345-x
34. Wang L., Cao Q. M.. **Long non-coding RNA XIST alleviates sepsis-induced acute kidney injury through inhibiting inflammation and cell apoptosis via regulating miR-155-5p/WWC1 axis**. *Kaohsiung J. Med. Sci.* (2022) **38** 6-17. DOI: 10.1002/kjm2.12442
35. Wang Y., Quan F., Cao Q., Lin Y., Yue C., Bi R.. **Quercetin alleviates acute kidney injury by inhibiting ferroptosis**. *J. Adv. Res.* (2021) **28** 231-243. DOI: 10.1016/j.jare.2020.07.007
36. Yang I. A., Jenkins C. R., Salvi S. S.. **Chronic obstructive pulmonary disease in never-smokers: Risk factors, pathogenesis, and implications for prevention and treatment**. *Lancet Respir. Med.* (2022) **10** 497-511. DOI: 10.1016/S2213-2600(21)00506-3
37. Yoshida M., Minagawa S., Araya J., Sakamoto T., Hara H., Tsubouchi K.. **Involvement of cigarette smoke-induced epithelial cell ferroptosis in COPD pathogenesis**. *Nat. Commun.* (2019) **10** 3145-3214. DOI: 10.1038/s41467-019-10991-7
38. Zeng X. L., Yang X. N., Liu X. J.. **Resveratrol attenuates cigarette smoke extract induced cellular senescence in human airway epithelial cells by regulating the miR-34a/SIRT1/NF-κB pathway**. *Med. Baltim.* (2022) **101** e31944. DOI: 10.1097/MD.0000000000031944
39. Zhang H., Ma J., Liu F., Zhang J.. **Long non-coding RNA XIST promotes the proliferation of cardiac fibroblasts and the accumulation of extracellular matrix by sponging microRNA-155-5p**. *Exp. Ther. Med.* (2021) **21** 477. DOI: 10.3892/etm.2021.9908
40. Zhang P., Zhang Y., Wang L., Wang X., Xu S., Zhai Z.. **Reversal of NADPH oxidase-dependent early oxidative and inflammatory responses in chronic obstructive pulmonary disease by puerarin**. *Oxid. Med. Cell Longev.* (2022) **2022** 5595781. DOI: 10.1155/2022/5595781
41. Zhang Y., Xie Y., Zhang L., Zhao H.. **MicroRNA-155 participates in smoke-inhalation-induced acute lung injury through inhibition of SOCS-1**. *Molecules* (2020) **25** 1022. DOI: 10.3390/molecules25051022
|
---
title: Pharmacological activities and effective substances of the component-based
Chinese medicine of Ginkgo biloba leaves based on serum pharmacochemistry, metabonomics
and network pharmacology
authors:
- Hongbao Liang
- Jingchun Yao
- Yu Miao
- Ying Sun
- Yanbing Gao
- Chenghong Sun
- Rui Li
- He Xiao
- Qun Feng
- Guofei Qin
- Xiaoyan Lu
- Zhong Liu
- Guimin Zhang
- Feng Li
- Mingguo Shao
journal: Frontiers in Pharmacology
year: 2023
pmcid: PMC10036596
doi: 10.3389/fphar.2023.1151447
license: CC BY 4.0
---
# Pharmacological activities and effective substances of the component-based Chinese medicine of Ginkgo biloba leaves based on serum pharmacochemistry, metabonomics and network pharmacology
## Abstract
As a potential drug candidate for the treatment of hypertension and complications, it is speculated that the component-based Chinese medicine of *Ginkgo biloba* leaves (GBCCM) which mainly composed of flavonoid aglycones (FAs) and terpene lactones (TLs) may have different pharmacological effects at different doses or ratios. Taking the normal mice as the study object, metabonomics was conducted by giving different doses of GBCCM. Based on the components of GBCCM absorbed into the blood, the network pharmacological prediction was carried out. By integrating the results of metabonomics and network pharmacology, predict the possible pharmacological effects of GBCCM and conduct experimental verification. It was found that eight of the 19 compounds in GBCCM could be absorbed into the blood. GBCCM mainly affected the signal pathways of unsaturated fatty acid, pyruvate, bile acid, melanin and stem cells. It was speculated that GBCCM might have activities such as lowering blood pressure, regulating stem cell proliferation and melanogenesis. By establishing the models of mushroom tyrosinase, rat bone marrow mesenchymal stem cells (BMSCs) and spontaneously hypertensive rats (SHRs), we found that FAs and TLs showed synergistic effect in hypertension and tyrosinase models, and the optimal ratio was 3:2 (4.4 mg/kg) and 1:1 (0.4 mg/ml), respectively. As effective substances, FAs significantly promoted the proliferation of rat BMSCs on the third and fifth days at the concentration of 0.2 μg/ml ($p \leq 0.05$). GBCCM showed a variety of pharmacological effects at different doses and ratios, which provided an important reference for the druggability of GBCCM.
## 1 Introduction
GBE has a variety of pharmacological effects, such as vasodilation, blood lipid regulation, platelet activating factor antagonism, ischemic injury protection, anti-inflammatory and anti-tumor (Huang et al., 2013; Abdel-Zaher et al., 2017). Ginkgo biloba preparations with GBE as the main raw material are used in the treatment of cardiovascular and cerebrovascular diseases in clinical practice, such as coronary heart disease, cerebral infarction and memory loss, but there still exist some problems as follows: 1) Flavonoids and TLs are the main active components of GBE, but the total content of them only about $30\%$ (Liu et al., 2021). Whether or not other components of GBE play a role in reducing toxicity and improving efficacy is unknown. 2) The activities of flavonoids in regard to preventing reperfusion injury, lowering blood lipid, enhancing memory and immune regulation are all based on their antioxidant effects (Crascì et al., 2018; Shen et al., 2022), and the antioxidant activity of FAs is stronger than flavonol glycosides (Odontuya et al., 2005). However, flavonoids of GBE mainly exist in form of the flavonol glycosides (Crespy et al., 1999). 3) The ratio between flavonoids and TLs of GBE is basically fixed, but it may not be the best.
Our research group prepared a component-based Chinese medicine of *Ginkgo biloba* leaves (GBCCM) mainly composed of flavonoid aglycones (FAs) and terpene lactones (TLs) from *Ginkgo biloba* leaves. FAs were obtained by converting flavonol glycosides into aglycones through hydrolysis in vitro and purification. However, TCM or natural products have the characteristics of multi-targets, multi-pathways and multi-pharmacological activities (Jiang et al., 2022; Zhao et al., 2022). GBCCM may exhibit a variety of pharmacological effects and complex dose-effect relationship at different drug concentrations and composition ratios. Therefore, solving the doubts is conducive to evaluating the potential value of GBCCM.
As a newly developed technology in recent years, metabonomics can be used to characterize the changes of endogenous substances in organisms caused by drug effects (Yang et al., 2005; Lyu et al., 2022). Although the drug may show different effects under physiological and pathological conditions, the potential action characteristics and pharmacological activities of the drug may be inferred from the analysis of the changes of differential metabolites and metabolic pathways at different doses under physiological conditions. Network pharmacology is a new cross discipline based on the theory of system biology, which uses bioinformatics and network analysis methods to analyze biological systems, study the mechanism of drug action from the system level, and carry out multi target drug molecular design (Zhou et al., 2020; Wang et al., 2021). It can help us find the complex action rules of traditional Chinese medicine by combing the relationship among components, targets and pathways.
In this study, integrating network pharmacology and serum metabolomics to reveal the relationship between blood components in vivo, drug concentration and signal pathways, so as to predict the possible pharmacological activities of GBCCM. The potential activities will be verified by using molecular, cellular and animal models. The optimal concentrations and composition ratios of GBCCM will also be explored. It is hoped that this study can provide data support for the clinical development of GBCCM.
## 2.1 Chemicals
Tween-80 was purchased from Nanjing Well Pharmaceutical co., LTD. ( Nanjing, China). Ethanol (AR grade), phosphate acid (AR grade), ethyl acetate (AR grade), hydrochloric acid (AR grade), n-hexane (AR grade) were all purchased from Sinopharm Chemical Reagent Co. LTD. ( Shanghai, China). Standards of ginkgolide A ≥ $95\%$, ginkgolide B ≥ $95\%$, ginkgolide C ≥ $95\%$, ginkgolide J ≥ $95\%$, bilobalide ≥ $98\%$, quercetin ≥ $98\%$, isorhamnetin ≥ $98\%$, kaempferol ≥ $98\%$ were purchased from National Institutes for Food and Drug Control (Beijing, China). Formic acid (MS grade), acetonitrile (HPLC grade), triffuoroacetic acid (HPLC grade), tetrahydrofuran (HPLC grade) and methanol (HPLC grade) were acquired from Merck (Darmstadt, Germany). Fetal Bovine Serum (FBS) was obtained from GIBCO (Australia Origin). DME/F12, PBS buffer, Trypsin, and Penicillin-streptomycin were purchased from Hyclone (Logan, Utah, United States). Anti-CD90-PerCP, anti-CD45-FITC, anti-CD29-PE and anti-CD34-Alexa Fluor 647 were acquired from BD Biosciences (San Diego, United States).
## 2.2 Animals
Mice (Weight: 18 ± 2 g), Wistar rats and SHRs (Weight: 200 ± 20 g) were purchased from Beijing Weitong Lihua Experimental Animal Technology Co., Ltd. (Beijing, China; animal license number: SCXK [Jing] 2016-0006). All experimental animal procedures were carried out according to the Guide and Use of Laboratory Animals and approved by the Ethics Committee for Experimental Animals at State Key Laboratory of Generic Manufacture Technology of Chinese Traditional Medicine (Approved on 13 November 2019; No. NH-IACUC-2019-38) for minimizing animal suffering.
## 2.3 Preparation of the component-based Chinese medicine of Ginkgo biloba leaves (GBCCM)
Ginkgo biloba leaves were collected from Shandong, China in August 2021 and authenticated by the botanist Feng Li, Shandong University of Traditional Chinese Medicine, Shandong, China. The samples were dried, and stored without light (Figure 1). The preparation method for GBCCM was based on a previously published article (Liang et al., 2022b) with some modifications. GBCCM can be obtained by mixing FAs and TLs in different proportions as required. The specific method was as follows: *Ginkgo biloba* leaves (1 kg) were extracted two times with $50\%$ ethanol, concentrated and ethanol precipitated. The supernatant was concentrated and purified with macroporous resin, and $50\%$ of the eluent was collected. Then, the concentrated eluent was extracted twice with equal volume of ethyl acetate. Among them, flavonol glycosides were mainly enriched in the aqueous phase, and TLs were mainly enriched in the ethyl acetate phase. After secondary purification with macroporous resin, the aqueous phase was acid hydrolyzed at 80°C for 3.5 h. The hydrolysate was adjusted to neutral with NaOH and concentrated to dry. Performing reflux extraction with ethyl acetate, and concentrating the extract until to dry. FAs (2.8 g) would be obtained through crystallization in ethanol/water (1:3) system. The ethyl acetate phase was deacidified and decolorized twice with activated carbon. TLs (5.0 g) would be obtained through crystallization in ethanol/water (1:1.5) system.
**FIGURE 1:** *Ginkgo biloba plants (left), fresh leaves (middle), dried leaves (right).*
## 2.4 Preparation of samples for analysis in vivo
Twenty-four mice were randomly divided into control group, GBCCM low-dose group and GBCCM high-dose group according to body weight with eight mice in each group. The mice in low-dose group and high-dose group were orally administered with GBCCM at the dose of 4.4 mg/kg and 44 mg/kg (FAs: TLs = 3: 2) respectively. The control group was given equal volume of $0.5\%$ CMC-Na, once a day. Before the last administration, the mice were fasting for 12 h (Free drinking water). 30 min after drug supplementation on the day 5, all animals were anesthetized with ketamine (100 mg/kg, i.p.) and xylazine (7.5 mg/kg, i.m.).
Blood was collected from the abdominal inferior vena cava. After placing at room temperature for 30 min, the blood samples were centrifuged (Z326K, Hermle, Germany) at 3,500 rpm for 10 min. 1,000 μL methanol-acetonitrile (V: $V = 2$: 1) was added to 300 μL serum sample, and the mixture was agitated for 2 min using a vortex agitator followed by centrifugation at 18,000 rpm for 10 min. The supernatant was blown to dry with N2at 35°C and the residue was dissolved again in 200 μL methanol-water (V: $V = 8$: 2). Subsequently, the samples were centrifuged at 12,000 rpm for 15 min at 4°C. The supernatant was used for LC-MS/MS analysis. The QC samples were prepared by pooling the same volume of supernatant from each of the samples.
## 2.5.1 Chromatographic and mass spectrometric conditions
The qualitative analysis of the serum samples from the control group and high-dose group was carried out by UPLC-Q-Exactive-MS/MS (Thermo Fisher Scientiffc, United States). The liquid chromatographic separation of all analytes was carried out on a Waters-ACQUITY UPLC HSS analytical column (2.1 × 100 mm, 1.8 µm) at 30°C with $0.1\%$ formic acid-water as phase A and acetonitrile as phase B. The flow rate was maintained at 0.2 mL/min and the injection volume was 2 µL. The column temperature was 30°C. The gradient elution was as follows: 0 min, $15\%$B; 30 min, $40\%$B; 33 min, $55\%$B; 35 min, $15\%$B.
The Q-Exactive-Orbitrap-MS was coupled to the LC system via an electrospray ionization interface. Ultrahigh-purity helium (He) was used as collision gas and high-purity nitrogen (N2), as nebulizing gas. Mass spectrum was recorded in the range m/z 150-1,500 in the positive and negative ion modes. The conditions were as follows: ion spray voltage was 3.5 kV, capillary temperature was 350°C, collision voltage was 40 eV, the sheath and auxiliary gases flow rate were 35 and 10 (arbitrary units), respectively. The contents of GBCCM were analyzed by HPLC-CAD (Liang et al., 2022a).
## 2.5.2 Data analysis
The chemical components of GBCCM were identified using the PeakView 2.1 software that was supplied with the instrument. The data of blank serum and GBCCM containing serum were imported into the software. Blank serum was taken as the normal control group, the information such as retention time, accurate molecular weight, accurate mass charge ratio and secondary ion fragments were compared with the in vitro components, and if they were consistent, they were identified as prototype components. Previous studies (Pietta et al., 1997; Li et al., 2012) had shown that flavonoids and terpene lactones of GBE are metabolized by the liver via phase II reactions after absorption into blood, and mainly forming glucuronic acid, sulfate or glutathione conjugates. The conjugated compounds were not found to have biological activity, so this study focused on the components absorbed into the blood rather than phase I and phase II metabolites.
## 2.6.1 Chromatographic and mass spectrometric conditions
The liquid chromatographic separation of all analytes was carried out on a Waters-HSS T3 C18 analytical column (2.1 × 100 mm, 1.7 μm) with $0.01\%$ formic acid-water as phase A and acetonitrile as phase B. The flow rate was maintained at 0.20 mL/min. The column temperature was 40°C and the injection volume was 2.0 μL. The gradient elution conditions were as follows: 0 min, $5\%$B; 2 min, $5\%$B; 4 min, $45\%$B; 23 min, $60\%$B; 27 min, $100\%$B; 32 min, $5\%$B; 34 min $5\%$B.
The UPLC-LTQ-Oribitrap-MS (Thermo Scientific, Santa Clara, United States) was coupled to the LC system via an electrospray ionization interface. Mass spectrum was recorded in the range m/z 50-1800 in positive and negative ion mode. The spray and capillary voltages were 4.0 kV and 35.0 V, respectively. The capillary temperature was 350°C and the tube lens voltage was set to 110V. N2 (purity > $99.99\%$) was used as both the sheath gas (40 arb) and auxiliary gas (20 arb). Data-dependent acquisition (ddms3) of high-resolution Fourier transform (TF, full scan, resolution 30,000) and CID fragmentation were used for positive and negative ion data acquisition.
## 2.6.2 Data pre-processing and multivariate pattern recognition
The collected LC–MS/MS raw data files were imported into Compound Discoverer 3.1 software (Thermo, MA, United States) to obtain matched peak data. The parameters were set as follows: quality range, 100-1,500; quality deviation, 5 × 10−6; retention time deviation, 0.05 min; SNR threshold, 3. Peak area normalization. Normalized data were imported into SIMCA-P 13.0 software for principal component analysis (PCA), and orthogonal partial least squares discriminant analysis (OPLS-DA). A 200-iteration permutation test was used to verify the robustness of the supervised OPLS-DA model and to assess the degree of overfitting. The differential metabolites were selected on the basis of the combination of a statistically significant threshold of variable influence on projection (VIP) values obtained from the OPLS-DA model and p values from a two-tailed Student’s t-test on the normalized peak areas, where metabolites with VIP > 1.0 and $p \leq 0.05$ were considered as differential metabolites.
The HMDB database and the mass spectrometry ion fragments were used to identify the selected compounds. The positive and negative ion data were combined into a data matrix table containing all the information extracted from the original data and used for subsequent analysis. SPASS 23.0 software was used to conduct one-way analysis of variance (ANOVA) for differential metabolite data. Pathway enrichment analysis was performed on the Metaboanalyst platform (https://www.metaboanalyst.ca/).
## 2.7 Network pharmacology analysis
The molecular targets of the constituents absorbed into the blood were searched from STITCH 5.0 (http://stitch.embl.de/) and bioinformatics analysis tool for molecular mechanism of traditional Chinese medicine (BATMAN-TCM, http://bionet.ncpsb.org/batman-tcm/). After removing duplicates, the related targets of GBCCM were obtained. All the intersected targets were normalized to their official symbols by the UniProt data-base (https://www.uniprot.org/). The Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analyses were performed based on the online platform KOBAS (http://kobas.cbi.pku.edu.cn/index.php) with $p \leq 0.01.$ The pathways were presented as bubble plots using the “pathview” package in the R software, and the “active ingredient-target” network was visualized using the Cytoscape software to visualize.
## 2.8.1 Inhibitory effect of GBCCM on tyrosinase activity in vitro
The tyrosinase inhibitory activity of FAs and TLs was evaluated detected according to the instruction of the reagent kit (Sigma-Aldrich, St. Louis, MO, United States). Weigh appropriate amounts of FAs, TLs and kojic acid to prepare a stock solution of 40 mg/mL with $20\%$ tween-80 ethanol solution. Dilute them to the final concentration with ultrapure water before testing (Table 1). Adding 20 μL test samples and tyrosinase analysis buffer to the sample hole (S) and control hole (EC) to be tested respectively. Adding 50 μL tyrosinase solution, incubating at 25°C for 10 min, and then adding 30 μL tyrosinase substrate solution to each well, incubating at 25°C for 30–60 min. Two time points (T1 and T2) were selected within 30–60 min and a full wavelength microplate reader was used for acquiring absorbance values of AbT1 and AbT2 at 510 nm. The tyrosinase activity inhibition rate (A%) was calculated according to the following formula [1]. A%=AbT2−AbT1 EC−AbT2−AbT1 S/ AbT2−AbT1 EC×$100\%$ [1]
**TABLE 1**
| Unnamed: 0 | A (mg/ml) | B (mg/ml) | C (mg/ml) | D (mg/ml) | E (mg/ml) |
| --- | --- | --- | --- | --- | --- |
| FAs | 0.002 | 0.02 | 0.2 | 0.5 | 1.0 |
| TLs | 0.002 | 0.02 | 0.2 | 0.5 | 1.0 |
| Kojic acid | 0.002 | 0.02 | 0.2 | 0.5 | 1.0 |
| FAs + TLs | 0.2 + 0 (2:0) | 0.2 + 0.1 (2:1) | 1.2 + 0.2 (2:2) | 0.1 + 0.2 (1:2) | 0 + 0.2 (0:2) |
## 2.8.2 Molecular docking
The molecular structure of mushroom tyrosinase (AbTYR; PDB code: 2Y9X) was used for the docking studies. The structures of the ligand compound were drawn by ChemDraw ultra and saved in mol format. After being imported into the software for pretreatment, flexible docking was carried out. DS visualizer 3.5 software was used for analysis. The LibDock score, the CDocker energy, the CDocker interaction energy and the hydrogen bond formation of the ligand-receptor complex were comprehensively considered so as to determine the final steady conformation.
## 2.9.1 Isolation, cultivation and passage of rat BMSCs
The healthy Wistar rats of 3–4 weeks were killed after cervical spondylectomy, and the femur and tibia were separated under sterile conditions. Wash the bone marrow cavity repeatedly with serum-free DME/F12 medium until white. Centrifuge the cleaning solution at 1,200 rpm for 5 min and discard the supernatant. The cells were resuspended with DME/F12 culture medium containing $10\%$ FBS and inoculated into the culture bottle. The cells were cultured in a 37°C, $5\%$ CO2 incubator. Culture medium was changed every 48 h. When the cells adherent growth density reached about $85\%$, they were digested with $0.25\%$ trypsin. The cell morphology was observed with an inverted phase contrast microscope. When most of the cell body retracted and became round, stop the digestion. The cells were subcultured at a ratio of 1:2.
## 2.9.2 Detection of surface markers of rat BMSCs by flow cytometry
The third generation BMSCs were collected and resuspended after digestion with $0.25\%$ trypsin. The cells were resuspended with 50 μl of pre-cooled PBS buffer, and an appropriate amount antibodies and corresponding homotypic control antibodies were added respectively. After incubating at room temperature in the dark for 30 min, add 1 ml PBS and shake it well. Centrifuge for 5 min at 1,200 rpm and discard the supernatant. Add an appropriate amount of PBS buffer, mix well, and then being detected by flow cytometry (CytoFLEX, Beckman Coulter, United States).
## 2.9.3 Effect of GBCCM on rat BMSCs
CCK-8 method was used to detect the effects of FAs and TLs on the proliferation of BMSCs (Kong and Chen, 2020). The experiment was divided into nine groups, including control group (C), FAs low-dose group (GFL), medium-dose group (GFM), high-dose group (GFH), TLs low-dose group (GLL), medium-dose group (GLM), high-dose group (GLH), FAs + TLs (1:1) low-dose group (GFLL), medium-dose group (GFLM) and high-dose group (GFLH). The drug concentrations were 0.02 μg/ml (L), 0.2 μg/ml (M), 2 μg/ml (H) respectively.
The third generation BMSCs were inoculated into 96-well plates at the concentration of 2 × 104 cells/well. 100 μL of test samples under different concentrations (five wells in each concentration) were added. The culture was terminated on 1, 3, 5 days. After 100 μL of basic medium and 20 μL of CCK-8 solution were added to each well, 96-well plates were kept in the incubator at 37°C for 2 h. The microplate reader was used for acquiring absorbance values (OD) at 450 nm.
## 2.10 Effect of GBCCM on blood pressure
Wistar rats were used as the normal group consisting of 10 rats. SHRs were divided into six groups with 10 rats in each group, including model group, FAs + TLs (5:0), FAs + TLs (4:1), FAs + TLs (3:2), FAs + TLs (2:3), FAs + TLs (1:4) and FAs + TLs (0:5) groups, and amlodipine besylate group, respectively. The normal group and the model group were given water (containing $0.5\%$ sodium carboxymethyl cellulose) by gavage. FAs + TLs (4.4 mg/kg) and amlodipine besylate (0.5 mg/kg) were suspended in $0.5\%$ sodium carboxymethyl cellulose and orally administered to SHRs once daily for 60 days as administration groups.
Blood pressure was measured at day 0, and 60 after the start of treatment. The rats were allowed to rest for at least 15 min at 30°C before systolic blood pressure (SBP) was measured by a tail-cuff method (BP-2000, Visitech Systems, Inc., Apex, NC, United States). Blood pressure was measured four times for each rat, and the mean value was recorded.
## 2.11 Statistical analysis
Statistical analysis was performed using SPSS software (Version 23.0; IBM SPSS Statistics Inc., Chicago, IL, United States). The results were expressed as mean ± SD, and three repeats were performed for each experiment. One-way analysis of variance (ANOVA) was performed to evaluate the differences in mean values. Significant differences were verified by the Tukey-Kramer honestly significant difference test ($p \leq 0.05$).
## 3.1 Identification of chemical constituents of GBCCM before and after blood transfusion
A total of 19 components were identified from GBCCM by LC-MS/MS in vitro (Figure 2), including eight flavonoid glycosides, five flavonoid aglycones (FAs) and six terpene lactones (TLs) (Supplementary Table S1). FAs and TLs were the main components. The content of GBCCM (FAs + TLs) was reached $92.3\%$. Less flavonoid glycosides also had been identified, indicating that acid hydrolysis was incomplete.
**FIGURE 2:** *Total ion chromatograms of GBCCM before and after blood transfusion. Test solution ((A), ESI-; (B), ESI+), mixed standards ((C), ESI-; (D), ESI+), blank serum ((E), ESI-; (F), ESI+), medicated serum ((G), ESI-; (H), ESI+). The corresponding compounds of peaks 1-19 are shown in Supplementary Table S1.*
According to the method described in “2.5.1”, data were collected in both positive and negative ion modes. The total ion chromatograms (TICs) were shown in Figure 3. Using the analysis method described in “2.5.2”, a total of eight components absorbed into blood were identified, including three flavonoids (quercetin, kaempferol and isorhamnetin) and five terpene lactones (bilobalide, ginkgolide A, B, C and J). The structural formulas of these compounds were shown in Figure 4. Other compounds of GBCCM were not be identified, it may be because the concentrations of them were too low or they had been converted into phase I and phase II metabolites.
**FIGURE 3:** *Chemical structures of components absorbed into the blood.* **FIGURE 4:** *The PCA (A), OPLS-DA (B, D) and (F) score plots of C, L and H and permutation test of OPLS-DA model (C, E) and (G). (C) Control group samples, L: Low-dose group samples, H: High-dose group samples, QC: Quality control (QC) samples, R2: Goodness of fit, and Q2: Predictability of the models.*
## 3.2.1 UHPLC-MS/MS analysis of serum samples
The representative LC–MS/MS TICs of the serum samples were shown in Supplementary Figure S1. In the positive and negative ion mode, the chromatographic peaks of the control group are different from those of the low-dose group and high-dose group, indicating that there were obvious differences in endogenous metabolites among the three groups of serum samples.
## 3.2.2 Statistical analysis of metabolic data by PCA and OPLS-DA
The unsupervised PCA is performed on all sample data to show theiroriginal classification status. The PCA scoring plots showed a tight aggregation of quality control (QC) samples, showing a good reproducibility of the instrument throughout the analysis period (Figure 4A). As expected, we observed a clear separation and clustering between the control group, low-dose group and high-dose group, with no extreme outliers to exclude, suggesting that there were significant differences at the metabolic level of mice. The supervised OPLS-DA was used to identify more specific metabolites among the groups. Such distributions and changing trends in aggregation and separation among control group, low-dose group and high-dose group became more apparent with OPLS-DA analysis (Figures 4B,D,F). To further verify the above observations, we performed a permutation procedure test using the OPLS-DA model with the same number of components. In total, 200 rounds of random permutations of the y variable were performed, and the results showed that while the R2 values (R2 represents the validity of the model and indicates the goodness of fit) were largely steady, the Q2 values which represents the accuracy of the model prediction were substantially decreased with increasing cycles of interaction validation (Figures 4C,E,G). We found that both goodness-of-fit parameters (R2 and Q2) calculated for the ranked data were lower than the corresponding original points on the right-hand side (1 on X-axis), and the intercepts of Q2 regression lines were all less than zero, indicating little overfitting in the original prediction model. Therefore, these analyses show that the separation model is statistically valid, and that the high value of predictability is not caused by overfitting. Therefore, we used these data for subsequent analyses.
## 3.2.3 Identification and analysis of sifferential metabolites
We identified the potential metabolites among the groups combined with S-plot obtained from OPLS-DA analysis. S-plot analysis (Figures 5A–C) represented the farther away metabolite ions from origin represent the higher VIP value of the ions, and the higher VIP value represents the greater contribute to the difference between the two sample groups. Based on VIP >1.0 in the OPLS-DA model and $p \leq 0.05$ in Student’s t-test, a total of 31 differential metabolites between the low-dose group and control group were identified, including 17 up-regulated and 14 down-regulated; A total of 38 differential metabolites between the high-dose group and control group were identified, including 29 up-regulated and nine down-regulated; *And a* total of 44 differential metabolites between the high-dose group and low-dose group were identified, including 20 up-regulated and 24 down-regulated. Through the venn diagram (Figure 5D), we found 12 common differential metabolites, which were linoleic acid, docosahexaenoic acid, palmitoleic acid, oleic acid, palmitic acid, myristic acid, stearic acid, 3,4-dihydroxyhydrocinnamic acid, 3-(3,4-dihydroxy-5-methoxy)-2-propenoic acid, 15H-11,12-EETA, l-lactic acid and 2,3-dihydro-2-S-glutathionyl-3-hydroxy bromobenzene. It was speculated that GBCCM might mainly affect unsaturated fatty acids and fatty acid related metabolic pathways in a dose-dependent manner.
**FIGURE 5:** *OPLS-DA S- Plot (A, B) and (C) and venn diagram of differential metabolites (D). (A) Low-dose group vs. Control group, (B) High-dose group vs. Control group, (C) Low-dose group vs. High-dose group. (C) Control group, L: Low-dose group, H: High-dose group.*
## 3.2.4 Enrichment analysis of serum metabolite pathway
By setting adjusted $p \leq 0.01$ as the screening criteria, the KEGG pathway analysis of differentially abundant metabolites was performed by MetaboAnalyst 4.0 to identify the disturbed metabolic pathways caused by GBCCM. 10, eight and eight metabolic pathways were enriched among C VS H, C VS L, and L VS H, respectively (Figure 6). It can be seen from the above that at different doses, the active ingredients of GBCCM may affect the same or different metabolic pathways, showing the characteristics of multi-component and multi target effects. We found the biosynthesis of unsaturated fatty acids, fatty acid biosynthesis, pyruvate metabolish, phenylalanine, tyrosine and tryptophan biosynthesis, phenylalanine metabolish and linoleic acid metabolish were common metabolic pathways of the three groups. It indicated that the effective components absorbed into blood of GBCCM could affect these metabolic pathways in a dose dependent manner. Changes of metabolites of these common metabolic pathways have a strong correlation with cardiovascular and cerebrovascular diseases, cardiac hypertrophy, biological characteristics of mesenchymal stem cells (Smith et al., 2012; Abdelhamid et al., 2018; Borges et al., 2022), which also suggests that GBCCM may have the potential to treat hypertension, hyperlipidemia, and promote mesenchymal stem cell differentiation and proliferation.
**FIGURE 6:** *Bubble diagram of KEGG metabolic pathway enrichment analysis of significantly different metabolites (A, B) and (C), and signal pathway enrichment analysis of components absorbed into the blood (D). (A) Low-dose group vs. Control group, (B) High-dose group vs. Control group, (C) High-dose group vs. Low-dose group. The horizontal axis represents the rich factor, and the vertical axis represents the pathways. The bubble size represents the number of targets in the pathway. The bubble color indicates the magnitude of the -log10(p) values.*
## 3.3 Network pharmacology analysis
According to the prediction of the two online platforms, 303 targets of eight blood components were obtained and a total of 72 pathways were enriched through the online software KOBAS ($p \leq 0.01$). After removing irrelevant signal paths, the KEGG analysis showed that the target genes corresponding to the constituents absorbed into the blood were enriched in the signaling pathways related to linoleic acid metabolism, pyruvate metabolism, arachidonic acid metabolism, bile secretion, regulating pluripotency of stem cells and melanogenesis (Figure 6). Such a high consistency validated the accuracy of pathway analysis in the metabolomics. As an unsaturated fatty acid, Linoleic acid can prevent or reduce the occurrence of cardiovascular and cerebrovascular diseases, especially hypertension, hyperlipidemia, angina pectoris, coronary heart disease, arteriosclerosis and elderly obesity (den Hartigh, 2019). Pyruvate metabolism was closely related to myocardial hypertrophy (Cluntun et al., 2021). Integrating the results of network pharmacology and metabonomics, we speculated that GBCCM might have the activities of regulating blood pressure, blood lipid, proliferation of stem cells and inhibiting melanin synthesis.
## 3.4.1 Inhibitory effect of GBCCM on tyrosinase activity in vitro
As shown in Figure 7A, FAs and TLs exhibited potent inhibitory activities on tyrosinase dose-dependently with IC50 values of 0.02 ± 0.01 and 0.05 ± 0.01 mg/ml, respectively. Both of them were higher than that of kojic acid (0.006 ± 0.001 mg/ml).
**FIGURE 7:** *Tyrosinase inhibitory activity of FAs and TLs at different concentrations (A) and ratios (B). *p < 0.05, ***p < 0.001 vs. FAs + TLs (1:1) group, ##
p < 0.01, vs. FAs + TLs (0:2) group.*
The inhibition rates of FAs and TLs increased significantly with the increase of concentration in the range of 0–0.2 mg/ml. When the concentration reached 0.5 mg/ml, the tyrosinase inhibition rates of FAs and TLs reached $79.71\%$ and $68.00\%$ respectively. It could be seen from Figure 7B, FAs had stronger tyrosinase inhibitory activity than TLs at the same concentration ($p \leq 0.01$). When FAs and TLs were used together at a ratio of 1:1, the inhibition rate reached $86.55\%$, which was higher than the inhibition rate when used alone ($p \leq 0.05$, $p \leq 0.001$).
## 3.4.2 The result of molecular docking
The docking results of eight active ingredients and mushroom tyrosinase protein active sites were shown in Figure 8. It was generally believed that the higher the docking score, the stronger the binding force between the compound and the target, the more stable the conformation and the stronger the potential role (Liang et al., 2021). Molecular docking results showed that the docking scores of the quercetin, kaempferol and isorhamnetin were close to or higher than those of the original ligand, indicating that these three components might have similar effects with the original ligand (Figures 8A–E). It could be seen from the molecular docking diagram that the flavone aglycone occupied the active center near two copper ions in tyrosinase. Through the hydrogen bond and hydrophobic force formed with multiple sites of the receptor, the affinity with the target protein was enhanced, and the stability of the conformation was also improved.
**FIGURE 8:** *Molecular docking diagram of active components and key targets. The compound is presented in the form of stick, the mushroom tyrosinase is presented in the form of ribbon, and the yellow dotted line represents hydrogen bonding. (A–J) represent the docking of quercetin, kaempferol, isorhamnetin, original ligand, all compounds, ginkgolide A, ginkgolide J, ginkgolide B, bilobalide, ginkgolide C with mushroom tyrosinase, respectively.*
Although the docking scores of bilobalide, ginkgolide A, B, C and J were lower than those of the original ligands, the conformations match the pockets well, and they could also form hydrogen bonds and hydrophobic forces (Figures 8F,G). The collision fraction and polarity were within a reasonable range, indicating that they also had a certain tyrosinase inhibitory effect. The results of molecular docking were basically consistent with those of activity detection, which further proved the accuracy of docking results.
Molecular docking results showed that FAs could form more hydrogen bonds and hydrophobicity near the active center of tyrosinase than TLs due to their multiple phenolic hydroxyl groups, which also indicated that FAs may have stronger inhibitory activity than TLs. Among the three flavonoid aglycones, quercetin had the highest docking score, suggesting that its activity may be the strongest. This was mainly due to the fact that quercetin had the most phenolic hydroxyl groups on its C-ring and the strongest ability to chelate copper from the tyrosinase active center (Fan et al., 2017; Roulier et al., 2020).
## 3.5 Effects of GBCCM on the proliferation of rat BMSCs
Flow cytometry was used to detect the third generation BMSCs. As seen in Supplementary Figure S2, the positive expression of CD29 and CD90 was detected while CD45 and CD34 were negative, indicating that the BMSCs were successfully isolated.
The effects of FAs and TLs on proliferation of BMSCs were shown in Figure 9. Compared with the normal control group, FAs significantly promoted BMSCs proliferation on the third and fifth days ($p \leq 0.05$, $p \leq 0.001$) at the concentration of 0.2 μg/ml. In each dose group, TLs had no effect on the proliferation of BMSCs ($p \leq 0.05$). When FAs and TLs (0.2 μg/ml) was used together at a ratio of 1:1, the effect was basically the same with FAs ($p \leq 0.05$), indicating that there was no synergistic effect between them. It was reported (Wu et al., 2016) that GBE did not show the activity of promoting the proliferation of rat BMSCs. It was speculated that the main reason may be that the content of free FAs in GBE was very little, and it was difficult to reach the effective concentration.
**FIGURE 9:** *The effects of FAs and TLs on proliferation of rat BMSCs. *p < 0.05, ***p < 0.001 vs. Control group.*
## 3.6 Effects of GBCCM on blood pressure
Previous study (Liang et al., 2022b) found that FAs and TLs had better antihypertensive activity at 4.4 mg/kg (3:2), but the proportion was not necessarily the best. Therefore, under the condition that the total dose remained unchanged, different ratio of FAs and TLs were designed based on the method of increasing proportion to study the optimal proportion.
The therapeutic effect of each group on SHRs was shown in Figure 10. The mean SBP of model group was significantly higher than that of normal group ($p \leq 0.01$) (Figure 10A). After 60 days of treatment, FAs + TLs (5:0) group, FAs + TLs (1:4) group and FAs + TLs (0:5) group had no significant improvement on SBP ($p \leq 0.05$), while FAs + TLs (4:1) group ($p \leq 0.01$), FAs + TLs (3:2) group ($p \leq 0.001$) and FAs + TLs (2:3) group ($p \leq 0.05$) significantly improved the SBP of SHRs (Figure 10B). FAs + TLs (3:2) group and amlodipine besylate group showed significant hypotensive effect ($p \leq 0.001$), indicating that 3:2 was the optimal ratio between FAs and TLs.
**FIGURE 10:** *Effects on SBP of each group after repeated administration for 0 days (A) and 60 days (B). ***p < 0.001 vs. Normal group, #
p < 0.05, ##
p < 0.01, ###
p < 0.001 vs. Model group.*
## 4 Discussion
Many researches proved that the network pharmacology integrated with metabolomics strategy was effective method and can be used to study the effect and mechanism of TCM (Hua et al., 2019; Li et al., 2021). In this study, we innovatively predicted the possible pharmacological effects of GBCCM by integrating serum pharmacochemistry, network pharmacology and metabolomics. The potential activities were verified through experiments.
In metabonomics research, we found that GBCCM regulates six metabolic pathways including biosynthesis of unsaturated fatty acids, fatty acid biosynthesis, pyruvate metabolish, phenylalanine, tyrosine and tryptophan biosynthesis, phenylalanine metabolish and linoleic acid metabolish in a dose-dependent manner at two doses (low and high), which showed the characteristics of multi-components, multi-targets and multi-pathways of GBCCM. Changes of metabolites of these common metabolic pathways have a strong correlation with cardiovascular and cerebrovascular diseases, cardiac hypertrophy, biological characteristics of mesenchymal stem cells (Smith et al., 2012; Abdelhamid et al., 2018; Borges et al., 2022). Combined with the primary excimer ion peak and secondary fragment information provided by high resolution mass spectrometry, nineteen components, including eight flavonoid glycosides, five flavonoid aglycones (FAs) and six terpene lactones (TLs) were identified in vitro by referring to relevant literatures and comparing with reference standards. As the main active components, the content of FAs and TLs in GBCCM reached $92.3\%$. Compared with GBE, the chemical composition of GBCCM was relatively clear. According to pharmaceutical theory, the premise for the efficacy of traditional Chinese medicine is that the ingredients are absorbed into the blood (Yan et al., 2015). Therefore, it is particularly important to investigate the components of GBCCM that are really absorbed into the blood through serum pharmacochemistry. Only three FAs and five TLs of GBCCM were identified as the main blood components by LC-MS/MS. Other ingredients were not be identified, it may be because the concentrations of them were too low or they had been converted into phase I and phase II metabolites. Network pharmacological analysis showed that a total of 72 signal pathways were enriched, mainly involving linoleic acid metabolism, pyruvate metabolism, arachidonic acid metabolism, bile secretion, regulating pluripotency of stem cells and melanogenesis. Integrating the results of network pharmacology and metabonomics, we speculated that GBCCM might have the activities of regulating blood pressure, blood lipid, proliferation of stem cells and inhibiting melanin synthesis.
Tyrosinase is a copper containing polyphenol oxidase, which is the key enzyme involved in melanogenesis. The occurrence and treatment of pigment disorders, malignant melanoma, albinism and senile dementia are directly related to tyrosinase (El-Nashar et al., 2021). It has been reported (Xue et al., 2011; Solimine et al., 2016; Klomsakul et al., 2022) that flavonoid aglycones (FAs) in *Ginkgo biloba* extract (GBE) have certain tyrosinase inhibitory activity. In the tyrosinase inhibitory activity experiment, we found that FAs had a good inhibitory effect on tyrosinase, which was consistent with the literature reports. At the same time, TLs were also found to have significant inhibitory activity for the first time. The inhibitory activity of FAs showed a good dose dependence at low concentrations, and the inhibition rate was about $79.71\%$ at 0.5 mg/ml. However, when the concentration continued to increase to 1.0 mg/ml, the inhibition rate decreased. The reason might be that FAs were easy to precipitate at high concentration due to their poor solubility. When FAs and TLs of GBCCM were used at a ratio of 1:1, the inhibition rate reached $86.55\%$, which was higher than the inhibition rate when used alone, indicating that they played a synergistic role when used together. Through molecular docking, it was speculated that the active components might play a role in the tyrosinase inhibitory through competitively bind to the active site in the enzyme with the substrate.
Bone marrow mesenchymal stem cells (BMSCs) are a kind of multipotent stem cells derived from bone marrow. They can not only provide hematopoietic support, but also have the potential of self-renewal, high proliferation and multi-directional differentiation (Liang et al., 2019). In recent years, many TCM extracts have been found to have certain effects on promoting the proliferation of BMSCs. Treatment of BMSCs with curcumin after 48 h, increased cell survival and proliferation in a dose-dependent manner (Attari et al., 2015). It was also found that Polygonatum, Plastrum testudinis, Panax notoginseng could promote the proliferation of stem cells (Li et al., 2011; Zong et al., 2015; Shen et al., 2018). GBE can promote the osteogenic differentiation of BMSCs, but had no significant effect on their proliferation (Wu et al., 2016). Through co-culture of BMSCs with GBCCM of different concentrations and proportions, we found that only FAs had a certain activity of promoting BMSCs proliferation at the concentration of 0.4 mg/ml. However, the activity was not strong enough to be further developed.
Hypertension is a chronic cardiovascular disease characterized by the rise of systemic arterial pressure, which can cause damage to heart, brain, kidney and other organs (Boesen and Kakalij, 2021). With the accelerated aging of the population, the prevalence of hypertension and its related diseases is increasing. Hypertension prevention and treatment has become a severe challenge facing the world’s public health (Frohlich, 2017). In recent years, the clinical application of western medicine combined with *Ginkgo biloba* preparation in the treatment of hypertensive patients shows that the antihypertensive effect of these two drugs is better than that of antihypertensive drugs taken alone (Mansour et al., 2011), but the antihypertensive active components and mechanism of GBE are still unclear. Our research group found that FAs and TLs mixed in a certain ratio had a better effect on lowering blood pressure when used alone (Liang et al., 2022b), but the composition ratios need to be further optimized. In this study, we found that the activity of FAs and TLs was not strong when used alone ($p \leq 0.05$) in the spontaneously hypertensive rat model. When the ratio of FAs to TLs was 3:2, GBCCM showed the best antihypertensive effect ($p \leq 0.001$), and there was no significant difference compared with amlodipine besylate, indicating that it had the potential clinical value.
Based on the above experiments, we found that GBCCM had significant tyrosinase inhibitory activity and blood pressure lowering activity, and there were complex network connections among components, targets, and pharmacological activities (Figure 11). Compared with GBE, GBCCM had the advantages of basically clear effective substances, relatively clear mechanism of action and convenience of prescription adjustment. We predict that GBCCM may become a new raw material for *Ginkgo biloba* preparations in the future. Strengthen the research on prescription ratio, dose-effect relationship and mechanism of TCM will help to improve the quality control level, enhance the efficacy and reduce the toxic and side effects, and it will also be the key point and breakthrough in the research of the modernization of TCM (Wang et al., 2016; Chen et al., 2019). As a novel mode for exploring the pharmacological activity of component-based Chinese medicine, we hope this study can bring some inspiration to other researchers.
**FIGURE 11:** *Network diagram of components-targets-pharmacological activities. QCT: Quercetin, KMF: Kaempferol, ISR: Isorhamnetin, BB: Bilobalide, GA: Ginkgolide A, GB: Ginkgolide B, GC: Ginkgolide C, GJ: Ginkgolide J, TYR: Tyrosinase.*
## 5 Conclusion
In this study, the serum differential metabolites of mice after intragastric administration of different GBCCM doses were investigated, and the metabolic pathway enrichment was analyzed. Based on the components absorbed into the blood, the action network of active components in GBCCM was predicted by network pharmacology technology, and the enrichment analysis of signal pathway was also carried out. By integrating serum pharmacochemistry, metabonomics and network pharmacology, it was found that GBCCM mainly affected the signal pathways of unsaturated fatty acid, pyruvate, bile acid, melanin and stem cells. It was speculated that GBCCM might have activities such as lowering blood pressure, regulating stem cell proliferation and melanogenesis. The pharmacological activities of GBCCM were verified by molecular, cellular and animal models, and the effective substances of GBCCM in different models were also confirmed.
## Data availability statement
The original contributions presented in the study are included in the article/Supplementary Materials, further inquiries can be directed to the corresponding authors.
## Ethics statement
The animal study was reviewed and approved by the Ethics Committee for Experimental Animals at State Key Laboratory of Generic Manufacture Technology of Chinese Traditional Medicine.
## Author contributions
GZ, MS, and FL conceived of the project and provided guidance; HL, YM, XL, and YG created the models and edited the draft; RL, JY, and HX carried out the mathematical analysis and validation; YS made the charts; QF, GQ, ZL, and CS collected the samples and edited the draft. All authors read and approved the final manuscript.
## Conflict of interest
HL, JY, YM, YS, YG, CS, RL, HX, QF, GQ, XL, ZL, GZ, and MS, were Shandong New Time Pharmaceutical Co., Ltd., Lunan Pharmaceutical Group Co., Ltd.
The remaining author declares that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher’s note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## Supplementary material
The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fphar.2023.1151447/full#supplementary-material
## Abbreviations
TCM, Traditional Chinese medicine; GBCCM, Component-based Chinese medicine of *Ginkgo biloba* leaves; BMSCs, Bone marrow mesenchymal stem cells; SHRs, Spontaneously hypertensive rats; GBE, *Ginkgo biloba* extract; FAs, Flavonoid aglycones; TLs, Terpene lactones; PCA, Principal component analysis; OPLS-DA, Orthogonal partial least squares discriminant analysis; VIP, Variable influence on projection.
## References
1. Abdelhamid A. S., Martin N., Bridges C., Brainard J. S., Wang X., Brown T. J.. **Polyunsaturated fatty acids for the primary and secondary prevention of cardiovascular disease**. *Cochrane Database Syst. Rev.* (2018) **11** Cd012345. DOI: 10.1002/14651858.CD012345.pub3
2. Abdel-Zaher A. O., Farghaly H. S. M., El-Refaiy A. E. M., Abd-Eldayem A. M.. **Protective effect of the standardized extract of ginkgo biloba (EGb761) against hypertension with hypercholesterolemia-induced renal injury in rats: Insights in the underlying mechanisms**. *Biomed. Pharmacother.* (2017) **95** 944-955. DOI: 10.1016/j.biopha.2017.08.078
3. Attari F., Zahmatkesh M., Aligholi H., Mehr S. E., Sharifzadeh M., Gorji A.. **Curcumin as a double-edged sword for stem cells: Dose, time and cell type-specific responses to curcumin**. *Daru* (2015) **23** 33. DOI: 10.1186/s40199-015-0115-8
4. Boesen E. I., Kakalij R. M.. **Autoimmune-mediated renal disease and hypertension**. *Clin. Sci. (Lond)* (2021) **135** 2165-2196. DOI: 10.1042/cs20200955
5. Borges M. C., Haycock P. C., Zheng J., Hemani G., Holmes M. V., Davey Smith G.. **Role of circulating polyunsaturated fatty acids on cardiovascular diseases risk: Analysis using mendelian randomization and fatty acid genetic association data from over 114,000 UK biobank participants**. *BMC Med.* (2022) **20** 210. DOI: 10.1186/s12916-022-02399-w
6. Chen C. C., Qin X. M., Guan-Hua D. U., Zhou Y. Z.. **The "compatible art" of effective components from traditional Chinese medicine: Research on the compatibility and proportion of effective components**. *Acta Pharm. Sin* (2019) **54** 808-817. DOI: 10.16438/j.0513-4870.2018-1119
7. Cluntun A. A., Badolia R., Lettlova S., Parnell K. M., Shankar T. S., Diakos N. A.. **The pyruvate-lactate axis modulates cardiac hypertrophy and heart failure**. *Cell Metab.* (2021) **33** 629-648.e10. DOI: 10.1016/j.cmet.2020.12.003
8. Crascì L., Lauro M. R., Puglisi G., Panico A.. **Natural antioxidant polyphenols on inflammation management: Anti-glycation activity vs metalloproteinases inhibition**. *Crit. Rev. Food Sci. Nutr.* (2018) **58** 893-904. DOI: 10.1080/10408398.2016.1229657
9. Crespy V., Morand C., Manach C., Besson C., Demigne C., Remesy C.. **Part of quercetin absorbed in the small intestine is conjugated and further secreted in the intestinal lumen**. *Am. J. Physiol.* (1999) **277** 370-126. DOI: 10.1152/ajpgi.1999.277.1.G120
10. den Hartigh L. J.. **Conjugated linoleic acid effects on cancer, obesity, and atherosclerosis: A review of pre-clinical and human trials with current perspectives**. *Nutrients* (2019) **11**. DOI: 10.3390/nu11020370
11. El-Nashar H. A. S., El-Din M. I. G., Hritcu L., Eldahshan O. A.. **Insights on the inhibitory power of flavonoids on tyrosinase activity: A survey from 2016 to 2021**. *Molecules* (2021) **26** 7546. DOI: 10.3390/molecules26247546
12. Fan M., Zhang G., Hu X., Xu X., Gong D.. **Quercetin as a tyrosinase inhibitor: Inhibitory activity, conformational change and mechanism**. *Food Res. Int.* (2017) **100** 226-233. DOI: 10.1016/j.foodres.2017.07.010
13. Frohlich E. D.. **Hypertension: New and future challenges**. *Med. Clin. North Am.* (2017) **101** 1-6. DOI: 10.1016/j.mcna.2016.08.018
14. Hua Y. L., Ma Q., Yuan Z. W., Zhang X. S., Yao W. L., Ji P.. **A novel approach based on metabolomics coupled with network pharmacology to explain the effect mechanisms of Danggui Buxue Tang in anaemia**. *Chin. J. Nat. Med.* (2019) **17** 275-290. DOI: 10.1016/s1875-5364(19)30031-7
15. Huang C. H., Yang M. L., Tsai C. H., Li Y. C., Lin Y. J., Kuan Y. H.. **Ginkgo biloba leaves extract (EGb 761) attenuates lipopolysaccharide-induced acute lung injury via inhibition of oxidative stress and NF-κB-dependent matrix metalloproteinase-9 pathway**. *Phytomedicine* (2013) **20** 303-309. DOI: 10.1016/j.phymed.2012.11.004
16. Jiang W., Qi J., Li X., Chen G., Zhou D., Xiao W.. **Post-infectious cough of different syndromes treated by traditional Chinese medicines: A review**. *Chin. Herb. Med.* (2022) **14** 494-510. DOI: 10.1016/j.chmed.2022.09.002
17. Klomsakul P., Aiumsubtub A., Chalopagorn P.. **Evaluation of antioxidant activities and tyrosinase inhibitory effects of ginkgo biloba tea extract**. *ScientificWorldJournal* (2022) **2022** 4806889. DOI: 10.1155/2022/4806889
18. Kong Y., Chen Z. T.. **MiR-146a regulates osteogenic differentiation and proliferation of bone marrow stromal cells in traumatic femoral head necrosis**. *Eur. Rev. Med. Pharmacol. Sci.* (2020) **24** 11465. DOI: 10.26355/eurrev_202011_23753
19. Li X. D., Wang J. S., Chang B., Chen B., Guo C., Hou G. Q.. **Panax notoginseng saponins promotes proliferation and osteogenic differentiation of rat bone marrow stromal cells**. *J. Ethnopharmacol.* (2011) **134** 268-274. DOI: 10.1016/j.jep.2010.11.075
20. Li L., Zhao Y., Du F., Yang J., Xu F., Niu W.. **Intestinal absorption and presystemic elimination of various chemical constituents present in GBE50 extract, a standardized extract of Ginkgo biloba leaves**. *Curr. Drug Metab.* (2012) **13** 494-509. DOI: 10.2174/1389200211209050494
21. Li J. X., Li R. Z., Sun A., Zhou H., Neher E., Yang J. S.. **Metabolomics and integrated network pharmacology analysis reveal Tricin as the active anti-cancer component of Weijing decoction by suppression of PRKCA and sphingolipid signaling**. *Pharmacol. Res.* (2021) **171** 105574. DOI: 10.1016/j.phrs.2021.105574
22. Liang J. Q., Lu F., Gan B., Wen Y. Y., Chen J., Wang H. G.. **Low-dose tubacin promotes BMSCs proliferation and morphological changes through the ERK pathway**. *Am. J. Transl. Res.* (2019) **11** 1446-1459. PMID: 30972173
23. Liang H. B., Li R., Yao J. C., Qin G. F., Zhang H., Zhang G. M.. **Mechanism of Shouhui Tongbian Capsules in treating constipation based on network pharmacology and molecular docking**. *Zhongguo Zhong Yao Za Zhi* (2021) **46** 511-519. DOI: 10.19540/j.cnki.cjcmm.20201117.406
24. Liang H., Sun C., Feng Z., Wang X., Kong L., Zhu F.. **Study on integrated pharmacokinetics of the component-based Chinese medicine of ginkgo biloba leaves based on nanocrystalline solid dispersion technology**. *Int. J. Nanomed.* (2022a) **17** 4039-4057. DOI: 10.2147/ijn.S379736
25. Liang H., Yuan X., Sun C., Sun Y., Yang M., Feng S.. **Preparation of a new component group of Ginkgo biloba leaves and investigation of the antihypertensive effects in spontaneously hypertensive rats**. *Biomed. Pharmacother.* (2022b) **149** 112805. DOI: 10.1016/j.biopha.2022.112805
26. Liu L., Wang Y., Zhang J., Wang S.. **Advances in the chemical constituents and chemical analysis of Ginkgo biloba leaf, extract, and phytopharmaceuticals**. *J. Pharm. Biomed. Anal.* (2021) **193** 113704. DOI: 10.1016/j.jpba.2020.113704
27. Lyu Y. H., Xie L., Chen W., Wang J., Wei X. T., Wei Y. P.. **Application of metabonomics in study of traditional Chinese medicine syndrome:a review**. *Zhongguo Zhong Yao Za Zhi* (2022) **47** 367-375. DOI: 10.19540/j.cnki.cjcmm.20210817.602
28. Mansour S. M., Bahgat A. K., El-Khatib A. S., Khayyal M. T.. **Ginkgo biloba extract (EGb 761) normalizes hypertension in 2K, 1C hypertensive rats: Role of antioxidant mechanisms, ACE inhibiting activity and improvement of endothelial dysfunction**. *Phytomedicine* (2011) **18** 641-647. DOI: 10.1016/j.phymed.2011.01.014
29. Odontuya G., Hoult J. R., Houghton P. J.. **Structure-activity relationship for antiinflammatory effect of luteolin and its derived glycosides**. *Phytother. Res.* (2005) **19** 782-786. DOI: 10.1002/ptr.1723
30. Pietta P. G., Gardana C., Mauri P. L.. **Identification of Gingko biloba flavonol metabolites after oral administration to humans**. *J. Chromatogr. B Biomed. Sci. Appl.* (1997) **693** 249-255. DOI: 10.1016/s0378-4347(96)00513-0
31. Roulier B., Pérès B., Haudecoeur R.. **Advances in the design of genuine human tyrosinase inhibitors for targeting melanogenesis and related pigmentations**. *J. Med. Chem.* (2020) **63** 13428-13443. DOI: 10.1021/acs.jmedchem.0c00994
32. Shen G. Y., Ren H., Huang J. J., Zhang Z. D., Zhao W. H., Yu X.. **Plastrum testudinis extracts promote BMSC proliferation and osteogenic differentiation by regulating let-7f-5p and the TNFR2/PI3K/AKT signaling pathway**. *Cell Physiol. Biochem.* (2018) **47** 2307-2318. DOI: 10.1159/000491541
33. Shen N., Wang T., Gan Q., Liu S., Wang L., Jin B.. **Plant flavonoids: Classification, distribution, biosynthesis, and antioxidant activity**. *Food Chem.* (2022) **383** 132531. DOI: 10.1016/j.foodchem.2022.132531
34. Smith A. N., Muffley L. A., Bell A. N., Numhom S., Hocking A. M.. **Unsaturated fatty acids induce mesenchymal stem cells to increase secretion of angiogenic mediators**. *J. Cell Physiol.* (2012) **227** 3225-3233. DOI: 10.1002/jcp.24013
35. Solimine J., Garo E., Wedler J., Rusanov K., Fertig O., Hamburger M.. **Tyrosinase inhibitory constituents from a polyphenol enriched fraction of rose oil distillation wastewater**. *Fitoterapia* (2016) **108** 13-19. DOI: 10.1016/j.fitote.2015.11.012
36. Wang J. B., Cui H. R., Bai Z. F., Xiao X. H.. **Precision medicine-oriented safety assessment strategy for traditional Chinese medicines: Disease-syndrome-based toxicology**. *Yao Xue Xue Bao* (2016) **51** 1681-1688. DOI: 10.16438/j.0513-4870.2016-0812
37. Wang X., Wang Z. Y., Zheng J. H., Li S.. **TCM network pharmacology: A new trend towards combining computational, experimental and clinical approaches**. *Chin. J. Nat. Med.* (2021) **19** 1-11. DOI: 10.1016/s1875-5364(21)60001-8
38. Wu Z., Zhang J., Gu X., Zhang X., Shi S., Liu C.. **Effects of the extract of Ginkgo biloba on the differentiation of bone marrow mesenchymal stem cells**. *Am. J. Transl. Res.* (2016) **8** 3032-3040. PMID: 27508023
39. Xue Y. L., Miyakawa T., Hayashi Y., Okamoto K., Hu F., Mitani N.. **Isolation and tyrosinase inhibitory effects of polyphenols from the leaves of persimmon, Diospyros kaki**. *J. Agric. Food Chem.* (2011) **59** 6011-6017. DOI: 10.1021/jf200940h
40. Yan G. L., Sun H., Zhang A. H., Han Y., Wang P., Wu X. H.. **Progress of serum pharmacochemistry of traditional Chinese medicine and further development of its theory and method**. *Zhongguo Zhong Yao Za Zhi* (2015) **40** 3406-3412. PMID: 26978981
41. Yang J., Song S. L., Castro-Perez J., Plumb R. S., Xu G. W.. **Metabonomics and its applications**. *Sheng Wu Gong Cheng Xue Bao* (2005) **21** 1-5. PMID: 15859320
42. Zhao Y., Zhong X., Yan J., Sun C., Zhao X., Wang X.. **Potential roles of gut microbes in biotransformation of natural products: An overview**. *Front. Microbiol.* (2022) **13** 956378. DOI: 10.3389/fmicb.2022.956378
43. Zhou Z., Chen B., Chen S., Lin M., Chen Y., Jin S.. **Applications of network pharmacology in traditional Chinese medicine research**. *Evid. Based Complement. Altern. Med.* (2020) **2020** 1646905. DOI: 10.1155/2020/1646905
44. Zong S., Zeng G., Zou B., Li K., Fang Y., Lu L.. **Effects of Polygonatum sibiricum polysaccharide on the osteogenic differentiation of bone mesenchymal stem cells in mice**. *Int. J. Clin. Exp. Pathol.* (2015) **8** 6169-6180. PMID: 26261494
|
---
title: Antigen discrimination by T cells relies on size-constrained microvillar contact
authors:
- Edward Jenkins
- Markus Körbel
- Caitlin O’Brien-Ball
- James McColl
- Kevin Y. Chen
- Mateusz Kotowski
- Jane Humphrey
- Anna H. Lippert
- Heather Brouwer
- Ana Mafalda Santos
- Steven F. Lee
- Simon J. Davis
- David Klenerman
journal: Nature Communications
year: 2023
pmcid: PMC10036606
doi: 10.1038/s41467-023-36855-9
license: CC BY 4.0
---
# Antigen discrimination by T cells relies on size-constrained microvillar contact
## Abstract
T cells use finger-like protrusions called ‘microvilli’ to interrogate their targets, but why they do so is unknown. To form contacts, T cells must overcome the highly charged, barrier-like layer of large molecules forming a target cell’s glycocalyx. Here, T cells are observed to use microvilli to breach a model glycocalyx barrier, forming numerous small (<0.5 μm diameter) contacts each of which is stabilized by the small adhesive protein CD2 expressed by the T cell, and excludes large proteins including CD45, allowing sensitive, antigen dependent TCR signaling. In the absence of the glycocalyx or when microvillar contact-size is increased by enhancing CD2 expression, strong signaling occurs that is no longer antigen dependent. Our observations suggest that, modulated by the opposing effects of the target cell glycocalyx and small adhesive proteins, the use of microvilli equips T cells with the ability to effect discriminatory receptor signaling.
T cells can use TCR on microvilli to interact with peptide-MHC (pMHC) complexes on antigen presenting cells. Here the authors characterise how T cells use microvilli to interrogate reconstituted membranes for pMHC complexes and how this is regulated by a balance between glycoproteins/glycocalyces that reduce detection, and the small adhesion protein CD2, which enhances detection.
## Introduction
The initiation of adaptive immune responses relies on T cells forming contacts with other cells. This enables, especially, T-cell receptors (TCRs) to engage ‘foreign’ peptide fragments complexed with major histocompatibility complex proteins (pMHC) on target and antigen-presenting cells (APCs), leading to the initiation of signaling and eventual clearance of pathogens and tumors. From the earliest days of the application of scanning electron microscopy, lymphocytes were distinguished by the presence of numerous finger-like surface projections called “microvilli”, in contrast, e.g., to monocytes that mainly displayed ruffled membranes and ridge-like protrusions1. It is now known that microvilli are used by T cells to efficiently scan target cells for antigen. Using lattice light-sheet imaging, Cai et al. showed that $98\%$ of the T cell/APC interface could be visited by microvilli within one minute2. But why T cells use microvilli instead of larger structures to scan targets is largely unknown. One explanation is that the use of microvilli represents the preferred energetic solution for penetrating the glycocalyx of apposing cells and detecting the relatively short ligands of the TCR2. A second, non-exclusive possibility, supported by in silico simulations based on a “dwell-time” model of TCR signaling3, is that individual contacts must be small in order for ligand discrimination to be possible.
The glycocalyx is a negatively charged, dense array of glycoconjugates and/or glycoproteins that extends 50–500 nm from most cell membranes, comprising an important barrier to cell contact and adhesion4–8. In the case of nucleated cells of hemopoietic origin, the two key glycoprotein elements of the glycocalyx are CD43 and CD459–12, with the former also frequently expressed by (non-hemopoietic cell-derived) tumors13,14. In the absence of active processes, the CD43/CD45-based glycocalyx will likely prevent interactions between T cells and APCs at distances shorter than 50–100 nm. Accordingly, CD43 has been shown to antagonize cell contact15–17 and, whereas the impact of CD45 on cell adhesion is less well-studied, it has been estimated that the spontaneous local exclusion of CD45 from a 100 nm diameter region of cell surface, allowing passive microvillar contact, would take 109 s (31.7 years)18. This suggests that there will be a requirement for T cells to physically exclude CD45 on opposing surfaces to establish contacts19. Since T cells spend just 1–5 min searching their targets for antigen20, active processes will need to drive close membrane approximation long enough for adhesive proteins to stabilize T cell/target contact, allowing efficient pMHC scanning and engagement by the TCR.
The two most important adhesive proteins on the surface of T cells that are likely to mediate contact are CD2 and LFA-121. CD2 is a relatively small (7.5 nm) single-pass transmembrane protein composed of immunoglobulin superfamily domains that binds a similarly small, evolutionarily related protein, CD58. In contrast, LFA-1 is a much larger (20–25 nm) heterodimer, that binds intercellular adhesion molecules (ICAMs) 1-5 and junctional adhesion molecules 1-2, with the strongest-binding and best-characterized ligand being ICAM-1 (19 nm)22–24. CD58 and ICAM-1 are ubiquitously expressed on hemopoietic and non-hemopoietic cells in humans, underscoring the importance of CD2 and LFA-1 to T-cell function14–16. These proteins enhance T-cell activation through their adhesive and costimulatory properties and are considered important therapeutic targets and risk factors for a variety of pathologies25–34. In particular, the CD2/CD58 pair has recently been identified as having an important function in the etiology of cancer35–38.
Here, we show that microvilli are required to penetrate a model glycocalyx, forming numerous small, uniform contacts that are stabilized by CD2/ligand interactions and facilitate ligand engagement by TCRs. Importantly, altering the size of the individual contacts allowed us to also show that TCR discrimination requires each contact to be size-limited. These data suggest that the use of microvilli equips T cells with a capacity for discriminatory receptor signaling.
## A supported lipid-bilayer mimic of the APC surface
T-cell interactions are typically studied with supported lipid-bilayers (SLBs) presenting pMHC and ICAM-1 extracellular domains (ECDs) only (SLB1s; Fig. 1a)39. To study the likely interplay between microvilli, a glycocalyx, and adhesion proteins, we created more complex SLBs that mimic the APC cell surface by presenting physiological densities of ‘null’ pMHC, agonist pMHC, ICAM-1, CD58, and a model glycocalyx comprising the two major glycoprotein elements of the glycocalyx formed by APCs, i.e., CD43 and CD45 (for the present experiments, CD45RABC), utilizing histidine tags to anchor the ECDs of each protein to the SLB (see Supplementary Fig. 1a–e for quantification of the proteins versus human monocyte-derived dendritic cells, and SLB configurations). The complex of the ‘9V’ variant of the cancer/testis NY-ESO-1 peptide (157–165; SLLMWITQV) and HLA-A2 (pMHC9V), which binds the 1G4 TCR40 with a solution KD of 7.2 μM41, was used as the agonistic ligand. HLA-A2 complexed with the non-cognate gp100 peptide (YLEPGPVTV) served as the null ligand (pMHCnull). To study T-cell engagement in a ligand-dependent manner, we generated a human CD4-CD8αβ+ 1G4 TCR-expressing Jurkat T cell line42, transduced with additional LFA-1 to match the surface phenotype of primary CD8+ effector T cells (Supplementary Fig. 1f), and a genetically encoded calcium indicator (GECI) to monitor intracellular calcium release43 as a marker of TCR triggering/early T-cell activation. GECI levels were kept low to prevent it behaving as a calcium sink44. We refer to this as the ‘J8-GECI’ cell line. Fig. 1SLB2s balance antigen sensitivity and specificity.a–d Calcium response curves for J8-GECI cells on the indicated SLBs presenting pMHCnull ± pMHC9V. A cartoon indicating the SLB compositions is shown (top panel), along with the fraction of cells exhibiting a Ca2+ signal (middle panel) and the time taken for $50\%$ of the cells to signal (bottom panel). Each datapoint corresponds to a separate SLB. Data were fitted with a four-point dose-response curve, constrained to a minimum using responses to pMHCnull. a “*First* generation” SLB (‘SLB1’; black); $$n = 44$$ SLBs with ≥116 cells analyzed per SLB. b SLB1 + CD58 (cyan); $$n = 28$$ SLBs with ≥127 cells analyzed per SLB. c SLB1 + glycocalyx (magenta); $$n = 23$$ SLBs with ≥140 cells analyzed per SLB. d “*Second* generation” SLB (‘SLB2’; blue); $$n = 42$$ SLBs with ≥109 cells analyzed per SLB. e, f Spreading, synapse formation, and calcium release for J8-GECI cells interacting with non-adhesive (ΔICAM-1ΔCD58) to highly adhesive (ΔCD43ΔCD45) SLBs. All the SLBs presented pMHCnull, except for the right-most SLB2, which presented pMHCnull plus pMHC9V at ~100 molecules/μm2. J8-GECI cells were tracked for calcium signals for 10 min and immediately imaged afterward to allow IRM-based contact area measurement and synapse formation frequency. e Images of cells spreading (dark areas in IRM image) and forming synapses. Colored boxes denote the same SLB composition shown in (a–d). Images are representative of J8-GECI cells on $$n = 4$$ independent SLBs for each SLB composition. f Left plot: fraction of cells that exhibit a calcium signal. Shown is the mean (±S.D.) of $$n = 4$$ independent SLBs with ≥158 cells analyzed per SLB. ΔCD43ΔCD45, ΔCD58, and SLB2s presenting pMHCnull correspond to the zero density values in (b), (c), and (d), respectively. f Middle plot (log scale): quantification of cell spreading; $$n = 115$$ (ΔICAM-1ΔCD58), 133 (ΔCD58), 174 (ΔICAM-1), 154 (SLB2 pMHC9V−), 132 (ΔCD43), 108 (ΔCD45), 118 (ΔCD43ΔCD45), and 122 SLB2 (SLB2 pMHC9V+) cells pooled from four independent SLBs for each SLB composition. The red line indicates median. f Right plot: quantification of synapse formation. Shown is the mean (±S.D.) fraction of cells forming synapses from $$n = 4$$ independent SLBs with ≥17 cells analyzed per SLB. In f, means were compared to an SLB2 presenting pMHCnull via one-way ANOVA with Tukey correction. Comparison between the ΔCD43ΔCD45 SLB and SLB2 (with pMHC9V) was also included. Only statistics for comparisons of interest are shown. Source data are provided in the *Source data* file.
We first examined the impact of CD58 and the model glycocalyx on pMHC sensing. On SLB1s presenting pMHCnull with and without increasing densities of pMHC9V (0.01–100 molecules/μm2; see Supplementary Fig. 1g for pMHC9V density measurements), the fraction of J8-GECI cells that produced a calcium response increased from ~$10\%$ to ~$85\%$, taking between ~1000 s to ~100 s to reach a $50\%$ response level (Fig. 1a). Adding CD58 (SLB1 + CD58) caused a striking increase in the extent and tempo of the response, at the cost of a large reduction in pMHC specificity, since ≥$60\%$ of cells responded to pMHCnull and pMHC9V at all densities within 200 s (Fig. 1b). The increase in responsiveness was dependent on the TCR but not CD2 signaling (considered below). Adding the model glycocalyx to the SLB1 substantially reduced calcium responses at all pMHC9V levels (by >$30\%$) and introduced delays in the response (up to 2–3-fold; Fig. 1c). These effects are likely due to reduced contact formation, reflecting the barrier effects of the glycocalyx5, due in part to its heavy glycosylation (with, e.g., sialic acid; Supplementary Fig. 1h). Adding both CD58 and the glycocalyx balanced sensitivity and specificity, returning the response curve to near SLB1 levels, albeit with an enhanced response to pMHCnull ($25\%$ vs $10\%$), and an approximately 2-fold faster response at low pMHC9V levels (Fig. 1d). We refer to this configuration, i.e., the presentation of CD58 and a glycocalyx alongside pMHC and ICAM-1, as the ‘second generation’ SLB (SLB2). Overall, responses to SLB1 and SLB2 bilayers were comparable; however, the SLB2 allowed the interplay between a glycocalyx and adhesion proteins to be studied.
We further investigated the countervailing effects of adhesion and the model glycocalyx on responses to pMHCnull by analyzing, alongside calcium signaling, cell spreading as a measure of activation using interference reflection microscopy (IRM), and synapse formation (Supplementary Fig. 1i)19,45. Removal of either glycocalyx element (CD43 or CD45) from SLB2s presenting pMHCnull, had a modest impact on calcium signaling, contact area, and synapse formation (Fig. 1e, f). However, removal of both dramatically increased signaling, cell spreading, and synapse formation, producing responses comparable to those with pMHC9V. In contrast, removing CD58, but not ICAM-1, from SLB2s reduced the fraction of cells that responded to pMHCnull. Strikingly, removing either or both adhesive proteins blocked spreading and synapse formation, confirming the importance of adhesive protein-mediated contact formation in enhancing T-cell responsiveness. Primary T cells responded similarly to the J8-GECI cell line on null- versus agonist-presenting SLB2s in terms of both calcium signaling (Supplementary Fig. 1j) and synapse formation (Supplementary Fig. 1k). These results confirmed the important functions of glycocalyces and adhesive proteins in securing the specificity and sensitivity of early T-cell signaling, respectively. As described in what follows, we explored contact formation on SLB2s and how this affects T-cell antigen recognition and discrimination.
## Overcoming a glycocalyx barrier
T cells are believed to use microvilli to search for pMHC on apposing cells2,46. Having shown that the presence of a glycocalyx substantially suppresses early signaling events, we sought to show that microvilli are instrumental in overcoming glycocalyx barriers. Electron microscopy confirmed that the surfaces of J8-GECI cells were populated with microvilli (Supplementary Fig. 2a), that were enriched for L-selectin at their tips (Supplementary Fig. 2b)46. We labeled CD43 and CD45 inserted into SLB2s presenting pMHC9V at ~100 molecules/μm2, and the cell membrane of J8-GECI cells, and then imaged the cells as they formed contacts with the SLBs using total internal reflection fluorescence microscopy (TIRFM). Prior to the cell settling on the surface, we observed small puncta of membrane fluorescence appearing and disappearing, suggestive of active surface sampling by the cell (Supplementary Fig. 2c). The puncta persisted for ~8 s (Supplementary Fig. 2d); similar observations were made previously for primary T cells47. Once a cell had settled, small and distinct ‘holes’ in the fluorescent CD$\frac{43}{45}$ layer began to form within the area bounded by the membrane fluorescence (Supplementary Movie 1). Given the similarity in [1] the sizes of the membrane puncta prior to contact, [2] the size of the holes subsequently created in the bilayer fluorescence, and [3] the dimensions of the microvillar protrusions of J8-GECI cells (~0.45 μm diameter; Supplementary Fig. 2e), we conclude that, in these experiments, the model glycocalyx is being breached by microvilli, creating the holes in the fluorescence of the SLB2. To analyze protein localization at these sites, we developed a bespoke segmentation and analysis pipeline that utilized gradient-based filtering and hysteresis thresholding to automatically segment the images (see “Methods” for details). Formation of the holes in the CD$\frac{43}{45}$ fluorescence correlated with heightened TCR (Supplementary Fig. 2f), pMHC (Supplementary Fig. 2g), and ZAP70 fluorescence (Supplementary Fig. 2h). L-selectin was initially observable but was later excluded (Supplementary Fig. 2i; Supplementary Movie 2). We also confirmed that the holes in the fluorescence were formed by exclusion of the glycocalyx, rather than other photophysical effects, by showing that fluorescent CD$\frac{43}{45}$ immobilized directly onto a glass surface exhibited uniform fluorescence during cell-contact formation (Supplementary Fig. 2j). We refer to the holes in the fluorescence observed on SLB2s, created by microvilli and marked by the local exclusion of glycocalyx elements, which correspond to regions of TCR engagement and signaling, as ‘close contacts’.
To further confirm that microvilli are needed to actively penetrate a glycocalyx barrier and establish close contacts, we used a suite of actin-modifying drugs to alter or ‘paralyze’ J8-GECI microvillus formation (Fig. 2a, b). Using our segmentation and analysis pipeline, we found that the loss of microvilli, or their paralysis, reduced the number of cells overcoming the glycocalyx barrier to form close contacts (Fig. 2c, d), and slowed contact formation (Fig. 2e; Supplementary Movie 3). Cells unable to form or to mobilize microvilli produced substantially smaller calcium responses on SLB2s (Fig. 2f). The treated cells were nevertheless capable of calcium signaling in the absence of the glycocalyx, albeit with slightly reduced capacity (Fig. 2f). These results confirmed the need for actin remodeling to overcome a glycocalyx barrier and efficiently detect pMHC, via the formation of microvilli48.Fig. 2Microvilli allow T cells to overcome a glycocalyx barrier.a Cartoon depiction of experiments testing whether dynamic actin remodeling of microvilli is needed to establish close-contacts in the presence of a glycocalyx barrier. Jasplakinolide (Jasplak) enhances the nucleation and stabilization of actin filaments, resulting in ‘paralysis’ of the actin cytoskeleton and microvillar activity. Latrunculin B (Lat B) and cytochalasin D (Cyto D) prevent the formation of actin filaments by sequestering actin monomers and by blocking their recruitment to pre-existing filaments, respectively, resulting in filament breakdown and subsequent loss of membrane topography. Cells in the cartoon are shown interacting with an SLB2. b Confocal fluorescence images of fixed J8-GECI cells imaged at the midplane after treatment with DMSO or an actin-modifying drug. A digitally-magnified region (white box) is shown below. Images are representative of J8-GECI cells for $$n = 3$$ independent experiments. c Examples of contact formation in the presence of DMSO or the actin-modifying drugs, for J8-GECI cells interacting with an SLB2 presenting pMHCnull plus pMHC9V-lo, i.e., ~1 pMHC9V molecule/μm2. Close contacts are indicated by black holes in the SLB2 glycocalyx fluorescence (with point of initiation indicated by white arrows). See Supplementary Movie 3. Images are representative of drug-treated J8-GECI cells for $$n = 4$$ independent experiments. d Fraction of cells that formed detectable close contacts. Data are from the same experiment as in (c). Shown is the mean (±S.D.) of $$n = 4$$ independent SLBs with 13–113 cells imaged per SLB. e Time taken for cells to form a close contact versus first appearance of membrane fluorescence. Data are from the same experiment as in (c). Data were pooled from four independent SLBs with $$n = 93$$ (DMSO), 88 (Cyto D), 18 (Jasplak), and 64 (Lat B) total contact-forming cells analyzed. The red line indicates the median. In d and e, means were tested using one-way ANOVA with Dunnett correction using DMSO as the control group. f Fraction of DMSO- or actin-modifying drug-treated J8-GECI cells that exhibit calcium release on an SLB2 or SLB2Δglycocalyx (i.e., SLB1 + CD58) presenting pMHCnull plus pMHC9V-lo. Shown is the mean (±S.D.) of $$n = 4$$ independent SLBs with 138-785 cells analyzed per SLB. Two-way ANOVA with Šidák correction was used to compare means between glycocalyx-positive and -negative SLBs for each treatment. Source data are provided in the *Source data* file.
## The four stages of close-contact formation
To explore close-contact formation and its relationship to pMHC sensing, we used three-color TIRF-based imaging to simultaneously analyze the triggering state (calcium signaling), interaction footprint (cell membrane staining), and close contacts (fluorescent glycocalyx components of the SLB) for cells interacting with SLB2s presenting pMHCnull and two levels of pMHC9V ligands (pMHC9V-lo, i.e., ~1 molecule/μm2 and pMHC9V-hi, i.e., ~100 molecules/μm2). Supplementary Movie 1, and Supplementary Movies 4, and 5, show J8-GECI cells interacting with SLB2s presenting both pMHC9V-hi and pMHCnull, and just pMHCnull (signaling cells and non-signaling cells), respectively. For cells interacting with SLB2s presenting pMHC9V-hi, we identified four distinct stages of contact: ‘searching’, ‘scanning’, ‘spreading’, and ‘synapsing’ (Fig. 3a). Searching (stage I) was marked by microvillar tips moving in and out of the evanescent field with no observable close contacts being formed. The scanning stage (stage II) was initiated by close-contact formation, marked by holes in the glycocalyx fluorescence, which, importantly, persisted throughout the following stages. The formation of close contacts presumably allows the TCR to scan for cognate pMHC, which we explore further below. As dynamic actin remodeling was key to efficient contact formation (Fig. 2d), the transition between searching and scanning is likely a stochastic process, dependent on intrinsic pathways governing protrusion dynamics. Calcium signaling marked the onset of spreading (stage III), which led to an increase in both the cell footprint and number of close contacts. Stage III ended with the beginning of the synapsing stage (IV), characterized by contraction of the footprint and centripetal movement of the close contacts. Fig. 3The four stages of close-contact formation.a Key stages of close-contact formation. Images show a J8-GECI cell interacting with an SLB2 presenting pMHCnull plus pMHC9V-hi (i.e., ~100 molecules of pMHC9V/μm2). The different stages were identified by simultaneously imaging close contacts (black holes in the SLB glycocalyx fluorescence), cell footprint (cell membrane area), and triggering state (calcium signal) of a cell. A cartoon of each stage is indicated below. See main text for stage descriptions. b–e Image-based analysis of J8-GECI cells on an SLB2 presenting pMHCnull ± pMHC9V-lo or pMHC9V-hi. The baseline response is given by non-signaling cells on pMHCnull. See Supplementary Movies 1, 4, and 5. b Cumulative distribution of the searching to scanning stage transition for J8-GECI cells on SLB2s. The analysis uses both signaling and non-signaling cells; $$n = 26$$ (pMHC9V-hi), 26 (pMHC9V-lo), and 34 (pMHCnull) cells. Plotted is the cumulative distribution function of the Kaplan–Meier estimator with the exponential Greenwood confidence interval. A pairwise log-rank test indicated that there were no significant differences. c Cell footprint versus time, plotted relative to the first appearance of a close contact (timepoint 0 s). Plotted is the mean (±S.D.); $$n = 18$$ (pMHC9V-hi, 5 SLBs), 17 (pMHC9V-lo, 5 SLBs), and 11 (pMHCnull, 6 SLBs) signaling cells, and 23 (pMHCnull, 6 SLBs) non-signaling cells. d Number of close contacts versus time. Plotted is the mean (±S.D.). The analysis uses the same cells as in (c). e Area and diameter of individual close contacts within the first 10 s after their formation. Plotted is the mean (±S.D.). The analysis uses the same cells as in (c). f–i *Same analysis* as in (b–e) for primary CD8+ T cells. UCHT-1 Fab-HaloTag was used as an agonist. See Supplementary Movies 6 and 7. f Cumulative distribution of the searching to scanning stage transition for primary cells on SLB2s. The analysis uses data for both signaling and non-signaling cells [$$n = 19$$ (UCHT-1 Fab-HaloTag) and 28 (pMHCnull) cells]. g Cell footprint versus time, plotted relative to the first appearance of a close contact (timepoint 0 s). Plotted is the mean (±S.D.); $$n = 8$$ signaling cells (UCHT-1 Fab-HaloTag, 3 SLBs) and 14 non-signaling cells (pMHCnull, 6 SLBs). h Number of close contacts versus time. Plotted is the mean (±S.D.). The analysis uses the same cells as in (g). i Area and diameter of individual close contacts within the first 10 s after their formation. Plotted is the mean (±S.D.). The analysis uses the same cells as in (g). j, k J8-GECI cells interacting with SLB2s presenting pMHCnull plus pMHC9V-hi. j Detection of pMHC from a single close contact at calcium release. See Supplementary Movie 8. Images are representative of J8-GECI cells for $$n = 3$$ independent SLBs. k Accumulation of ZAP70 at close contacts prior to calcium release. See Supplementary Movie 9. Images are representative of J8-GECI cells for $$n = 3$$ independent SLBs. l *Total area* of close contacts versus total cell membrane area at calcium release. ‘ UCHT-1’ refers to UCHT-1 Fab-HaloTag. m Number of close contacts at calcium release. In l, m, boxplots indicate the quartiles with a line at the median. Whiskers extend to points that lie within 1.5 IQRs of the lower and upper quartile. Conditions were compared using the Kruskal–Wallis H test, and, if $p \leq 0.05$, further compared using the pairwise Mann–Whitney U test. The p values are shown. The analysis uses the same cells as in (c) for J8-GECI cells and (g) for primary cells. n Detection of single bound pMHC9V at close contacts prior to (t = −4 s) and at calcium release ($t = 0$ s). Images show a J8-GECI cell interacting with an SLB2 presenting pMHCnull plus pMHC9V-lo. The arrow indicates a single pMHC9V molecule at a close contact. o Maximum number of pMHC9V bound per close contact per cell for signaling and non-signaling cells; $$n = 13$$ FOVs, with 8 calcium signaling cells. Source data are provided in the *Source data* file.
The transition from searching to scanning occurred independently of the nature of the pMHC (Fig. 3b), whereas the progression to spreading was pMHC sensitive (Fig. 3c). For example, non-signaling cells on pMHCnull-presenting SLB2s could penetrate the glycocalyx but then exhibited only small increases in cell footprint (Fig. 3c) or close-contact number (Fig. 3d). In contrast, signaling cells on pMHC9V-hi- and pMHC9V-lo-presenting SLB2s exhibited large increases in cell footprint and number of close contacts formed, although spreading was accelerated on the pMHC9V-hi-presenting SLB2s. Cells responding to pMHCnull exhibited smaller increases in cell footprint and contact formation than with pMHC9V-lo, indicative of there being a relationship between signal strength and spreading19. A striking finding was that, whereas the cell footprint and close-contact number each increased with the level of stimulation up to and during the spreading stage (i.e., up to 150 s for cells on pMHC9V-hi SLB2s, and >300 s for other surfaces), the initial close-contacts formed (i.e., during the first 10 s of observation) were uniformly small (0.2–0.4 μm in diameter), even after calcium signaling (Fig. 3e). During the synapsing stage close-contact size increased, owing to the consolidation of contacts at the center of the cell. We address the importance of close-contact size below.
We also examined close-contact formation for primary CD8+ T cells on SLB2s presenting pMHCnull ± ~20 molecules/μm2 of a UCHT-1 Fab-HaloTag construct used as a strong agonist (SLB-bound UCHT-1 Fab has been used previously as a ‘pan-specific’ TCR ligand for human T cells29,49,50). Responding primary cells loaded with Fluo-4 exhibited all four stages of contact formation (Supplementary Movie 6; see Supplementary Movie 7 for a non-signaling cell). Although the main features of their interactions with agonist- and non-agonist-presenting SLB2s were broadly similar for primary cells (Fig. 3f–i), compared to J8-GECI cells, primary cells had a smaller cellular footprint (maximum of 127 versus 229 μm2; Fig. 3g), and formed more close contacts (82 versus 65 contacts/cell; Fig. 3h) of comparable size over time (Fig. 3i). Collectively, these results indicate that antigen detection during the early stages of T-cell activation occurs exclusively at small, microvillus-generated close contacts.
## Impact of glycocalyx density on close-contact formation
Although the glycocalyx of the SLB2 was designed to mimic that of an APC, we sought to understand how contact formation is affected by glycocalyx density, which is known to vary13,14,30. We therefore increased the density of the SLB2 glycocalyx three-fold, whilst keeping other protein densities constant (e.g., CD58; Supplementary Fig. 2k). This reduced, to some extent, the fraction of J8-GECI cells producing calcium responses for all conditions but had a modest, if any, effect on signaling times (Supplementary Fig. 2l). Increasing the density of the glycocalyx slightly delayed the transition from searching to scanning (22 s vs. 4 s, $$p \leq 0.003$$; Supplementary Fig. 2m). For cells that produced calcium signals we observed a reduction in the cellular footprint (maximum 90 versus 175 μm2), the number of close contacts formed (maximum 40 versus 60 contacts), and the extent of exclusion of the glycocalyx at each close contact (a proxy for contact ‘tightness’; $27\%$ versus $35\%$; Supplementary Fig. 2n). Importantly, the areas of the initially formed close contacts (~0.3 μm in diameter) remained constant (Supplementary Fig. 2n), further indicating that close-contact size is set, at least in part, by the dimensions of microvilli as they penetrate glycocalyx barriers. These results suggest that T cells can penetrate glycocalyx barriers of varying density to form close contacts, albeit with differing efficiency.
## Individual microvilli are capable of efficient antigen detection
We observed pMHC (Fig. 3j; Supplementary Movie 8) and ZAP70 (Fig. 3k; Supplementary Movie 9) accumulation at close contacts formed by J8-GECI cells interacting with pMHC9V-hi-presenting SLB2s during the scanning stage, prior to calcium release, confirming that close contacts are sites of TCR engagement and receptor triggering. Measurements of contact area at calcium release indicated that pMHC sensing was highly efficient, requiring <$0.2\%$ of the total cellular surface area (Fig. 3l) shared across fewer than 10 close contacts, irrespective of pMHC9V density (Fig. 3m); primary T cells used slightly more surface area and extra close-contacts, on average (Fig. 3m). In some instances, J8-GECI cells (Fig. 3m; Supplementary Movie 8) and primary cells (Fig. 3m) initiated signaling after forming 1–2 close contacts with the agonist-presenting SLB2. We therefore tested whether single pMHCs engaged at close contacts could mobilize signaling. By detecting bound pMHC on SLB2s presenting very low densities of pMHC9V, measured over the lifetime of the contact, we found that very small numbers of pMHC9V (i.e., 1–2 molecules) were engaged per close contact in cells that signaled, leading to cell spreading (Supplementary Movie 10), whereas none were bound by cells that did not signal (Fig. 3n, o). These results suggest that the use of microvilli allows for highly sensitive antigen detection by T cells.
## CD2, but not LFA-1, stabilizes close-contact formation
Previous work implied that TCR occupancy is required for microvillar contact stabilization2. We speculated that, in the presence of a glycocalyx, and in the more physiological setting of low agonist pMHC levels, adhesion molecules would be key to stabilizing contacts. To test this, we generated TCR-deficient J8-GECI T cells (TCRKO cells; Supplementary Fig. 3a). The TCRKO cells did not produce calcium responses on OKT3-coated glass or pMHCnull-presenting SLB2s (Supplementary Fig. 3b, c), and neither did they spread on the SLB2s, despite forming contacts detectable using IRM (Supplementary Fig. 3d). TIRFM confirmed that these cells entered the scanning stage but did not transition to the spreading stage (Supplementary Fig. 3e; Supplementary Movie 11). The close contacts were persistent (Supplementary Fig. 3f) and similar in number and size to those formed by non-signaling TCR-expressing cells (Supplementary Fig. 3g). For these cells, the maximum CD58 fluorescence intensity and track-length for each contact were correlated ($r = 0.75$), suggesting that close-contact persistence depended on the level of CD2/CD58 complex accumulation (Supplementary Fig. 3h). These observations indicate that the TCR is dispensable for close-contact formation and stabilization.
To understand how close contacts are stabilized, we imaged CD58 and ICAM-1 accumulation on pMHC9V-lo-presenting SLB2s, to mimic rare agonist presentation in vivo. For J8-GECI cells, we observed that CD58 accumulated exclusively within, and stably tracked with, close contacts during the scanning to synapsing stages (Fig. 4a; Supplementary Movie 12). The TCR colocalized in regions of CD58 accumulation (Supplementary Movie 13), suggesting that CD2 would be well placed to enhance pMHC detection at close contacts, by controlling membrane separation at the contact (since CD2/CD58 and TCR/pMHC complexes have similar dimensions51,52). In contrast, ICAM-1 was stably excluded from the close contacts (Fig. 4b; Supplementary Movie 14)2 and regions of CD58 accumulation (Supplementary Movie 15), as observed elsewhere53,54. The levels of glycocalyx exclusion on the SLB, measured as reductions in fluorescence, were greater in regions of CD58 versus ICAM-1 accumulation ($45\%$ versus $20\%$), indicating that CD2 and CD58 engagement formed tighter contacts (Supplementary Fig. 4a), correlating with the different dimensions of each complex. During the scanning stage, the sequence of events involving the adhesive proteins was: [1] close-contact formation/CD58 accumulation, [2] ICAM-1 exclusion, [3] ICAM-1 accumulation in rings around regions of CD58 accumulation, i.e., ‘micro adhesion rings’, and [4] calcium release (Supplementary Fig. 4b). Similar temporal patterns of CD58 and ICAM-1 accumulation occurred for primary CD8+ T cells (Supplementary Movie 16).Fig. 4CD2 stabilizes close contacts, enhancing antigen detection.a, b Image-based analysis of J8-GECI cells on an SLB2 presenting pMHCnull ± pMHC9V-lo. a CD58 accumulation occurs at close contacts formed by J8-GECI cells. The min/max normalized intensity line profile was taken along the direction of the white arrow and indicates CD58 accumulation at close contacts. The kymograph indicates persistence of CD58 accumulation spots/close contacts. See Supplementary Movie 12. Images are representative of J8-GECI cells on $$n = 3$$ independent SLBs. b Engaged ICAM-1 is excluded from close contacts formed by J8-GECI cells. The min/max normalized intensity line profile was taken along the direction of the white arrow and indicates ICAM-1 exclusion from close contacts. The kymograph indicates persistent exclusion of ICAM-1 from close contacts. See Supplementary Movie 14. Images are representative of J8-GECI cells on $$n = 3$$ independent SLBs. c–e Close-contact analysis for J8-GECI cells interacting with an SLB2, SLB2ΔCD58, or SLB2ΔICAM-1 presenting pMHCnull; $$n = 34$$ (SLB2, the same SLBs as in Fig. 3b–e), 14 (ΔCD58, from 4 SLBs), and 8 (ΔICAM-1, from 4 SLBs) cells. See Supplementary Movies 5, 17, and 19. c Cumulative distribution of the searching to scanning stage transition for cells on the indicated SLBs. The analysis uses data for both signaling and non-signaling cells. Plotted is the cumulative distribution function of the Kaplan–Meier estimator with the exponential Greenwood confidence interval. Data were examined using a pairwise log-rank test, which indicated that there were no significant differences. d Mean glycocalyx exclusion (i.e. contact tightness) at close contacts during the scanning stage for each cell. e Mean close-contact area during the scanning stage for each cell. The analysis uses the same cells as in (d). In d and e, conditions were compared using the pairwise Mann–Whitney U test. The p values are shown. The boxplots indicate the quartiles with a line at the median. Whiskers extend to points that lie within 1.5 IQRs of the lower and upper quartile. f–h *Same analysis* as in (c–e) using human primary CD8+ T cells; $$n = 28$$ (SLB2, the same SLBs as in Fig. 3f–i), 8 (ΔCD58, from 3 SLBs), and 18 (ΔICAM-1, from 3 SLBs) cells. See Supplementary Movies 7, 18, and 20. In f, a pairwise log-rank test indicated that there were no significant differences. i Fraction of J8-GECI cells that signal. Shown is the mean ± S.D. of $$n = 4$$ SLBs per condition with ≥257 cells analyzed per SLB. SLB2 data (blue bars) and SLB2Δglycocalyx (gray bars) were compared with a one-way ANOVA with Dunnett correction using data for the 'intact' SLBs, i.e., SLB2s with and without glycocalyces, as the control groups. j Displacement tracks. Each dot/line represents a single J8-GECI cell tracked for up to 10 min, exploiting the currents generated by adding cells to the SLBs to probe the adhesiveness of the different SLBs. Data are from the same experiment as in (i). k The time taken for $50\%$ of J8-GECI cells to adhere to the SLBs. SLB2 data (blue bars) and SLB2Δglycocalyx (gray bars) were compared separately with a one-way Kruskal–Wallis test with Dunn’s correction using data for the intact SLBs as the control groups. Data are from the same experiment as in (i). Source data are provided in the *Source data* file.
To distinguish between the effects of CD58 and ICAM-1 on the searching and scanning stages, we removed each protein from pMHCnull-presenting SLB2s. For J8-GECI cells and in the absence of CD58 on the bilayer, the glycocalyx was penetrated by microvilli (Fig. 4c), but the contacts were less stable (Supplementary Movie 5 versus Supplementary Movie 17), less tight (measured as glycocalyx exclusion: $32\%$ versus $41\%$; Fig. 4d), and smaller (mean diameter of 0.33 versus 0.44 μm; Fig. 4e). Primary CD8+ T cells exhibited similar behavior (Fig. 4f–h; Supplementary Movie 7 versus Supplementary Movie 18). We observed similar trends with CD58 removal for events beyond the scanning stage for J8-GECI cells activated on pMHC9V-lo-presenting SLB2s (Supplementary Fig. 4c), and for primary T cells (Supplementary Fig. 4d). In contrast, the removal of ICAM-1 from pMHCnull-presenting SLB2s had little impact on the scanning stage of close-contact formation (see Supplementary Movies 19 and 20 for J8-GECI and primary cells, respectively), but profoundly impacted the spreading and synapsing behavior of cells that produced signaling on pMHC9V-lo- and pMHCnull-presenting SLB2s: following calcium release, the cell footprint was greatly reduced (Supplementary Fig. 4c, d).
To relate differences in contact formation and stability during the scanning stage to functional outcomes, we monitored J8-GECI cell calcium responses to pMHC9V-lo-presenting SLB2s from which CD58 or ICAM-1 were removed. Consistent with a need for more stable and perhaps larger contacts to better detect pMHC, removal of CD58 caused a substantial reduction in the fraction of responding cells ($34\%$ versus $76\%$; Fig. 4i). In the absence of CD58 the cells exhibited greater displacement, i.e., the cells ‘drifted’ across the SLB, consistent with CD2 enhancing SLB adhesion generally (Fig. 4j, k). A much smaller reduction in signaling occurred upon ICAM-1 removal ($60\%$ versus $76\%$), likely reflecting a small overall decrease in adhesion to the SLB, since the properties of the close contacts were unchanged. As in the case of pMHCnull-presenting SLB2s (Fig. 1f), the absence of both adhesion molecules completely blocked signaling and adhesion to the SLB. Notably, the impact on pMHC detection and adhesion, of the loss of CD58 or ICAM-1, was only apparent in the presence of the model glycocalyx, highlighting the important interplay between adhesion proteins and glycocalyx barriers. Collectively, these data indicate that CD2 enhances close-contact formation and stabilization, facilitating antigen detection, whereas ICAM-1/LFA-1 interactions influence the cellular footprint, i.e., degree of spreading, following calcium signaling.
## CD2 enhances CD45 exclusion at close contacts, increasing antigen sensitivity
We recently proposed that sensitive, specific TCR signaling depends on the time the receptor spends in contacts that exclude cellular CD45 (the dwell-time model of T-cell signaling)3,54. It seemed likely that CD2 would make important contributions to signaling in this context as CD45 on J8-GECI cells was excluded from areas of CD58 accumulation (Supplementary Fig. 5a)55. To explore the function of CD2 in TCR signaling, we created a CD2-deficient J8-GECI cell line (J8-GECI-CD2KO) and re-introduced at endogenous levels either wild-type CD2 (J8-GECI-CD2WT), or CD2 whose cytosolic domain was deleted (J8-GECI-CD2ΔCYT). J8-GECI-CD2WT, J8-GECI-CD2ΔCYT, and J8-GECI-CD2KO cells were incubated with fluorescently labeled anti-CD45 antibody and tested for close-contact formation, CD45 exclusion, and calcium release (Supplementary Movies 21–23). To focus on the events leading to T-cell signaling, we restricted our analysis to the scanning stage of close-contact formation by using pMHCnull-presenting SLB2s.
J8-GECI-CD2WT-expressing cells formed stable close contacts that locally excluded CD45 prior to, and in the absence of, calcium release (Fig. 5a). The numbers of close contacts formed, their mean area, and average glycocalyx exclusion levels were similar for J8-GECI-CD2WT- and J8-GECI-CD2ΔCYT-expressing cells (Fig. 5b–d). J8-GECI-CD2WT- and J8-GECI-CD2ΔCYT-expressing cells excluded ~$40\%$ and ~$32\%$ of CD45 from close contacts, respectively (Fig. 5e). In marked contrast, J8-GECI-CD2KO cells formed smaller close contacts (Fig. 5c) that were characterized by significantly reduced exclusion of the glycocalyx (Fig. 5d), consistent with the effects of removing CD58 from SLB2s (Fig. 4c–h). Surprisingly, for J8-GECI-CD2KO cells, CD45 exclusion at each contact was similar to that for J8-GECI-CD2ΔCYT-expressing cells (~$33\%$; Fig. 5e), despite the contacts being smaller. EC50 values for calcium signaling by J8-GECI-CD2WT-, J8-GECI-CD2ΔCYT-expressing cells, and for J8-GECI-CD2KO cells on SLB2s presenting pMHCnull and increasing amounts of pMHC9V, were 0.7, 3.2, and 64.5 pMHC9V molecules/μm2, respectively (Fig. 5f). The differences were CD2 dependent, as the cell lines expressed other key surface proteins comparably (Supplementary Fig. 5b) and responded equally well to direct TCR stimulation with an activating antibody (OKT3; Supplementary Fig. 5c). The slight increase in the sensitivity of J8-GECI-CD2WT- versus J8-GECI-CD2ΔCYT-expressing cells likely reflects the slightly lower CD45 exclusion by J8-GECI-CD2ΔCYT-expressing cells and/or the enhanced recruitment of Lck to the contact by full-length CD2 (Fig. 5g; Supplementary Fig. 5d). These results indicate that CD2 has a pivotal role in stabilizing close contacts and altering the kinase/phosphatase balance at these sites. Fig. 5CD2 enhances T-cell sensitivity.a CD45 exclusion at close contacts formed by J8-GECI-CD2WT cells. The images show a J8-GECI-CD2WT cell during the scanning stage on an SLB2 presenting pMHCnull; inset shows GECI fluorescence, indicating a lack of signaling. A min/max normalized intensity line profile was taken along the direction of the white arrow and indicates CD45 exclusion at close contacts. The kymograph shows contact stabilization and persistent CD45 exclusion from contacts. See Supplementary Movies 21–23. b–e Close contact feature analysis for both signaling and non-signaling J8-GECI-CD2WT and variant cells during the scanning stage on an SLB2 presenting pMHCnull; $$n = 9$$ (J8-GECI-CD2WT, from 2 SLBs), 11 (J8-GECI-CD2ΔCYT, from 2 SLBs), and 17 (J8-GECI-CD2KO, from 3 SLBs) cells. The boxplots indicate the quartiles with a line at the median. Whiskers extend to points that lie within 1.5 IQRs of the lower and upper quartile. Conditions were compared using the Kruskal–Wallis H test, and, if $p \leq 0.05$, further compared using the pairwise Mann–Whitney U test. The p values are shown. b Mean number of close contacts per cell. c Mean single close-contact area. The analysis uses the same cells as in (b). d Mean glycocalyx exclusion at close contacts. The analysis uses the same cells as in (b). e Mean exclusion from close contacts of CD45 on the J8-GECI cell surface. The analysis uses the same cells as in (b). f Calcium response curves for J8-GECI-CD2WT, J8-GECI-CD2ΔCYT, and J8-GECI-CD2KO cell lines interacting with SLB2s presenting pMHCnull ± pMHC9V at the indicated densities. Data were fitted using a four-point dose-response curve, constrained to a minimum based on responses to pMHCnull; $$n = 27$$ (J8-GECI-CD2WT), 25 (J8-GECI-CD2ΔCYT), and 25 (J8-GECI-CD2KO) SLBs per curve with ≥139 cells analyzed per SLB. The EC50 values for J8-GECI-CD2WT, J8-GECI-CD2ΔCYT, and J8-GECI-CD2KO cells were 0.7, 3.2, and 64.5 molecules pMHC9V/μm2, respectively. g Increased Lck recruitment to close contacts for cells expressing full-length CD2 versus CD2ΔCYT on an SLB2 presenting pMHCnull + pMHC9V-lo; $$n = 1080$$ (J8-GECI-CD2WT) and 983 (J8-GECI-CD2ΔCYT) contacts analyzed from 10 FOVs. Means were compared using a two-sided Student’s t-test. Source data are provided in the *Source data* file.
## Dependence of TCR discrimination on constrained close-contact size
Finally, we examined the relationship between close contact formation and TCR ligand discrimination. J8-GECI cells were capable of efficient ligand discrimination, with EC50 values for responses to pMHC with affinities ranging from 7 μM to >2000 μM varying by three orders of magnitude (Fig. 6a), all within a narrow time window of 100–200 s (Fig. 6b). Using low densities of agonist pMHC (1 molecule/μm2), the transition to the scanning stage was found to be independent of pMHC affinity (Supplementary Fig. 6a), whereas transitioning from scanning to spreading was affinity dependent, reflecting discrimination (Supplementary Fig. 6a). Once again, irrespective of ligand affinity, or the delay before signaling, close-contact diameter was constrained to the small range of 0.32–0.43 μm at calcium release (Fig. 6c), with no difference in the degree of glycocalyx exclusion at the contacts (Supplementary Fig. 6b). Simulations confirmed that at the resolution limit of the microscope smaller contacts would have been reliably detected, emphasizing the uniformity and constrained size of the contacts (Supplementary Fig. 6c).Fig. 6Antigen discrimination requires size-constrained close-contact formation.a Calcium response curves for J8-GECI cells interacting with SLB2s presenting pMHCnull plus increasing densities of different-affinity pMHC. b Median time taken for cells to signal. From the same experiment as in (a). In a, b, each datapoint corresponds to a separate SLB; $$n = 14$$ SLBs with ≥148 cells analyzed per SLB. Data were fitted using a four-point dose-response curve, with the bottom of each curve in (a), and top of each curve in (b), constrained to responses on an SLB2 presenting pMHCnull. c Area (left y-axis) and diameter (right y-axis) of close contacts formed by J8-GECI cells interacting with an SLB2 presenting pMHCnull plus the indicated agonist pMHC at point of calcium release; $$n = 68$$ (9Vhi, from 18 cells), 148 (9Vlo, from 17 cells), 100 (3Ilo, from 16 cells), 9 (9Llo, from 5 cells), 42 (4Dlo, from 7 cells), and 58 (Null/gp100, from 11 cells) individual contacts. Data obtained for pMHCnull and pMHC9V-hi/lo SLBs are from the same experiment as in Fig. 3b–e. To measure the correlation between the affinity of low-density pMHC and the contact size, a linear least-squares regression was performed, and the p-value shown. The boxplots indicate the quartiles with a line at the median. Whiskers extend to points that lie within 1.5 IQRs of the lower and upper quartile. d TIRFM images of close-contact formation with/without calcium release for the indicated cell lines on SLB2s presenting pMHCnull. A white dashed circle is shown for the J8-GECI-CD2KO cell line to indicate the region of close-contact formation. e Fraction of cells that produce signals on an SLB2 presenting pMHCnull. Shown is the mean (± S.D.) of $$n = 3$$ SLBs with ≥309 cells analyzed per SLB. Means were compared using one-way ANOVA with Šídák correction. f Fraction of cells that signal on an SLB2 presenting pMHCnull plus ~10 molecules/μm2 of different-affinity pMHC. Data obtained for J8-GECI-CD2WThi cells are shown in dark blue with each point representing the mean ± S.D ($$n = 3$$ SLBs per agonist pMHC with 16-75 cells analyzed per SLB). The light blue line is taken from curves fitted in (a) for data obtained with the J8-GECI cells. The gray datapoint is taken from Fig. 5f for the J8-GECI-CD2KO cell line. Source data are provided in the *Source data* file.
To demonstrate the need for constrained contacts, as predicted by the dwell-time model of signaling3, we examined the effects of increasing contact size on pMHC discrimination. J8-GECI-CD2KO cells overexpressing CD2 by a factor of ~10 (J8-GECI-CD2WThi; Supplementary Fig. 6d) produced ~5–10-fold larger close contacts, on average, than J8-GECI-CD2WT-expressing cells (Fig. 6d, Supplementary Fig. 6e). Notably, J8-GECI-CD2KO, J8-GECI-CD2WT, and J8-GECI-CD2WThi cells on pMHCnull-presenting SLB2s produced calcium responses of $5\%$, $20\%$, and $65\%$, respectively (Fig. 6e), revealing a loss of discrimination as contact size increases. Receptor expression levels (Supplementary Fig. 6d), and levels of activation via direct engagement of the TCR (Supplementary Fig. 6f) were comparable for all cell lines, suggesting that the signaling differences were dependent on the extent of CD2 engagement. A TCR-deficient J8-GECI-CD2WThi cell line did not produce calcium signals on an SLB2 presenting pMHCnull (Fig. 6e) or via direct TCR engagement (Supplementary Fig. 6f), indicating that the signaling observed at high levels of CD2 expression was TCR-dependent.
To further confirm that ligand discrimination was lost by J8-GECI-CD2WThi cells, we examined the effect of varying pMHC affinity (7–3200 μM KD) at constant pMHC density, on the signaling responses. The larger contacts formed by J8-GECI-CD2WThi cells could be readily segmented using our automated analysis pipeline (Supplementary Fig. 6g). Using SLB2s presenting pMHCnull and ~10 molecules/μm2 of pMHC varying in affinity, we observed that contact size was unaffected by pMHC affinity or the presence of the TCR (Supplementary Fig. 6h), and that ~$70\%$ of J8-GECI-CD2WThi cells produced calcium responses irrespective of pMHC affinity (Fig. 6f, Supplementary Movie 24). This was in marked contrast to cells expressing native levels of CD2, which formed much smaller contacts and efficiently discriminated between different pMHC at this density (Fig. 6f). These results indicate that contact size underpins T-cell ligand discrimination.
## Discussion
Much of what we know about T-cell interactions with apposing surfaces has come from studies of T cells interacting with SLBs presenting both TCR and integrin (i.e., LFA-1) ligands. As invaluable as this approach has been, the interplay between small adhesion proteins and the target cell glycocalyx, and their contributions to antigen recognition and TCR signaling, could not be studied. More importantly, alongside this and until recently, the role of microvilli in contact formation and T-cell responsiveness has received limited attention.
We implemented a SLB-based surrogate of the APC surface, comprising the major glycoprotein elements of the glycocalyx formed by APCs, the ligands of small and large adhesion proteins, and both model self and agonistic pMHC. We discovered that, in this more physiological setting, T-cell responses were profoundly affected by the presence of a glycocalyx and the small adhesion molecules. Removing the model glycocalyx provoked strong signaling and a loss of T-cell specificity. Conversely, eliminating the adhesive proteins reduced and delayed signaling, confirming the barrier-like activity of the cell glycocalyx5. T cells overcame the model glycocalyx by ‘punching through’ the barrier, forming numerous contacts visualized as small black holes in the layer of fluorescently labeled CD45 and CD43 molecules on the SLB. We referred to these structures as ‘close contacts’. Contact formation was an active process that proceeded in four stages, i.e., sequential ‘searching’ and ‘scanning’ modes separated by the formation of close contacts, a TCR- and pMHC-dependent transition to a ‘spreading’ stage, and, finally, synapse formation. Each close contact was stabilized by the interaction of CD2 with CD58; removing CD58 reduced the sensitivity of recognition ~100-fold. Sensitivity was enhanced by CD2 via increased close-contact persistence, CD45 exclusion over a larger area, and increased Lck recruitment within close contacts. ICAM-1, on the other hand, which was excluded from close contacts, did not enhance antigen sensitivity or early close-contact formation but was required during the spreading stage following antigen recognition, presumably when LFA-1 ‘switches’ to a high-affinity state (reviewed in ref. 56). These results indicate that the primary adhesive protein functioning during antigen recognition is CD2 and explain how naïve T cells can respond to DCs in an antigen-dependent/ICAM independent manner57. CD2 is reported to be enriched at microvillar tips, which would potentiate CD2 engagement during initial contacts58. Increasing the glycocalyx density three-fold compromised the efficiency of the T-cell response, consistent with the effects, e.g., of the altered composition of the cancer cell glycocalyx5. We note that such effects could be further exaggerated for deeper as well as denser glycocalyces. A remarkable feature of close contacts formed under all conditions, including for bilayers with more dense glycocalyces, however, was their uniformity, each being confined to ~0.1 μm2 (~0.36 μm diameter) prior to signaling. Most importantly, a dramatic loss of TCR discrimination was observed when close-contact size was larger than ~0.1 μm2.
Previously, Cai et al. used SLBs presenting fluorescent quantum dots to study T-cell contact formation, a method they called synaptic contact mapping (SCM)2. Using SCM, they observed that 0.5 μm holes formed in the quantum dot layer. The surfaces of the T cells we studied were populated with numerous microvilli, and we assume that the contacts observed using SCM are analogous to the close contacts formed on our bilayers. As in our experiments, LFA-1 and ICAM-1 were excluded from the close contacts, but Cai et al. also observed that contact stabilization was TCR-dependent. In contrast, we observed that contact stabilization was TCR-independent, underscoring the important contribution of small adhesion proteins to antigen detection. Since the ligands of small adhesive proteins were not present in the bilayers used for SCM, TCR/pMHC interactions likely performed the important function of physically stabilizing contact formation in the experiments of Cai et al. Given that, in our experiments, TCR-deficient T cells formed close contacts normally but did not progress to spreading, the scanning-to-spreading transition must comprise a key early checkpoint, whereupon the cells will have already discriminated between different-affinity pMHC19. Remarkably, we found that a single close contact was sufficient for this transition. During the spreading stage, T cells were observed to form many additional contacts, i.e., three-fold more, with the SLB. This ‘intensified interrogation’ of the activating surface likely allows T cells to reach higher activation thresholds (e.g., NFAT translocation). Finally, whereas Cai et al. did not observe “profound” exclusion of CD45, 30-$40\%$ local exclusion of CD45 at close contacts was readily demonstrable in our experiments. We note, however, that it is also proposed that CD45 is “pre-excluded” from the tips of microvilli59.
Why do T cells use structures as exotic as microvilli to engage target cells? It could have been expected that the formation of intimate contacts over large areas would allow for more efficient and rapid antigen detection. One important factor is that the glycocalyx barrier needs to be overcome, which might be more easily achieved with small protrusions. An additional requirement, we propose, is the need for T cells to be discriminatory. We have shown that at physiological levels of CD2 expression and in the presence of a glycocalyx, T cells could discriminate readily between ligands varying in affinity from 7 μM to >2000 μM KD, using close contacts of ~0.1 μm2 (~0.36 μm diameter). However, increasing contact size by overexpressing CD2 led to a dramatic loss of antigen discrimination. But how can the TCR be triggered in the absence of strongly-binding ligands? Computational simulations based on the dwell-time model of TCR triggering have suggested that if large close contacts form, TCRs are unlikely to encounter phosphatases, favoring receptor phosphorylation and signaling3. In contrast, when contacts are small, i.e., ~0.1 μm2, the simulations showed that discriminatory signaling could be readily achieved. Indeed, the simulations predicted the relative potencies of pMHC ligands presented to CD4+ and CD8+ T cells with remarkable accuracy. Strongly supporting this treatment of receptor triggering, slowing the diffusion of the TCR suffices to initiate signaling54. Here, we directly confirmed that ligand discrimination by T cells is strictly reliant on the formation of CD2-stabilized, microvillus-sized contacts. Given these findings, it is conceivable that the immune system varies CD2 expression to tune responsiveness. Memory T cells have enhanced sensitivity to lower-affinity antigens which is perhaps explained by their higher expression of CD260,61. However, the large CD2-dependent contact that forms at mature synapses called the “corolla”, which is thought to amplify responses37, is unlikely also to contribute to discriminatory TCR signaling.
This study has explored the relationship between small and large adhesion molecules, a cell glycocalyx, close-contact formation, and self/non-self discrimination by T cells interacting with glass-supported bilayers facilitating imaging. Our findings need now to be confirmed for bona fide T cell/APC contact or for interactions of T cells with even better models of APC surfaces. We previously argued that the appearance of a phosphatase-containing glycocalyx on primitive lymphoid cells during their evolution may have allowed their utilization of a contact-sensitive, molecular segregation-based mechanism of receptor signaling49. Our data suggest that the use of microvilli could have been a necessary, complementary adaptation which ensured that individual contacts were small in area, allowing discriminatory signaling by these primitive cells.
## Cell culture
All cell culture was performed in HEPA-filtered cell culture cabinets. All media components were bought sterile or 0.22 μm filtered before use. Cells were grown at 37 °C in $5\%$ CO2. Cell density and viability were monitored with a 1:1 mix of cell culture medium and $0.4\%$ Trypan blue. Stained cells were analyzed with a Countess II automated cell counter (ThermoFisher). Cells were independently tested negative for mycoplasma (Human Immunology Unit, WIMM). Jurkat T cells and derived cell lines were cultured in RPMI-1640 medium supplemented with $10\%$ (v/v) fetal calf serum (FCS), $1\%$ (v/v) HEPES buffer, and $1\%$ (v/v) pen/strep/neo antibiotics (Sigma; complete RPMI). Cells were maintained between 0.1–1 × 106 ml−1. Human embryonic kidney 293T (HEK-293T) cells were cultured in Dulbecco’s modified *Eagle medium* (DMEM) supplemented with $10\%$ (v/v) FCS, $1\%$ (v/v) pen/strep/neo antibiotics (Sigma), and $1\%$ (v/v) glutamine (complete DMEM). The catalog numbers for commercial items can be found in Supplementary Table 1.
## Creation of cell lines
The J8 and TCRKO cell lines were created from Jurkats (clone E6-1) as previously described42,50. ZAP70-HaloTag was expressed in J8 cells via lentiviral transduction to produce the J8-ZAP70-Halo cell line. Expression of the jGCaMP7s43 in the J8 (referred to as J8-GECI) or TCRKO (referred to as TCRKO-GECI) cell lines was undertaken and validated as previously described50,54. To ablate CD2 in J8-GECI and TCRKO-GECI cells, CRISPR-Cas9 guides were designed and selected for high specificity with minimal off-targets using Benchling (hg38 reference genome). Oligonucleotides were then designed, annealed, and ligated into the LentiCRISPRv2 plasmid using the dual BsmBI sites, as previously described62,63. Cells were sorted on the CD2-negative population by FACS 7 days after lentiviral transduction to produce either J8-GECI-CD2KO or TCRKO-GECI-CD2KO cells. To produce the J8-GECI-CD2WT, J8-GECI-CD2WThi, and TCRKO-GECI-CD2WThi cell lines, J8-GECI-CD2KO or TCRKO-GECI-CD2KO cells were stably transduced with wild-type CD2 (CD2WT) cDNA or CD2 cDNA encoding a cytosolic domain deletion leaving only 3 aa of the intracellular domain (CD2ΔCYT), using lentiviruses. Cells were sorted for endogenous or high levels of CD2WT or CD2ΔCYT expression by FACS. Lck-HaloTag was inserted into CD2WT and CD2ΔCYT by lentiviral transduction to produce the J8-GECI-CD2WT-Lck-Halo and J8-GECI-CD2ΔCYT-Lck-Halo cell lines. For all cells used in this study, LFA-1 expression was increased by lentiviral expression of cDNA encoding CD11a and CD18 to match the expression of these proteins in human primary CD8+ T cells that were primed and rested.
## Lentivirus production and transduction
1 × 106 HEKs in complete DMEM were seeded onto 6-well plates 24 h prior to the addition of plasmids. After 24 h, 0.5 μg of pHR/LentiCRISPRv2-based plasmid was co-transfected with vectors containing the lentiviral packaging proteins (0.5 μg of p8.91 and 0.5 μg pMDG). GeneJuice (Merck) was used to transfect the plasmid mixture according to the manufacturer’s protocol for adherent cells. Forty-eight hours post-transfection, supernatant was collected, filtered (0.22 μm), and added to a 6-well plate containing 0.5–1 × 106 Jurkat cells or derivative cell lines in 2 ml complete RPMI. Forty-eight hours post infection 4 ml of complete RPMI was added. Cells were analyzed at least 72 h post transduction. After 72 h, CRISPR KO cells were further selected by treatment with 1 ng/μl puromycin in complete RPMI for 3 days and/or cells were sorted by FACS after 7 days.
## Human primary CD8+ T cells
T cells were obtained from blood leukocyte cones purchased from NHS Blood and Transplant, John Radcliffe Hospital, Oxford, UK. Blood cones were used under the ethical guidelines of NHS Blood and Transplant. CD8+ T cells were isolated by Ficoll-Paque density gradient centrifugation and the use of a CD8+ T Cell Isolation Kit (Miltenyi) according to the manufacturer’s protocol. Isolated CD8+ T cells were washed and resuspended in complete RPMI with IL-2 (50 U/ml, PeproTech) and CD3/CD28-coated Human T-Activator Dynabeads (ThermoFisher Scientific). Cells were resuspended in complete RPMI containing fresh IL-2 every 2 days. Dynabeads were removed after day 5, and cells left to further expand for 7 more days. On day 12, aliquots of cells were frozen for future use. Twenty-four hours before use, cells were thawed, washed, and resuspended in complete RPMI with IL-2 (50 U/ml). On the day of use, dead cells were removed using the Dead Cell Removal Kit (Miltenyi Biotec) according to the manufacturer’s protocol and the live cells placed back into complete RPMI and IL-2 (50 U/ml).
## Flow cytometry
Cells were stained using conjugated mouse anti-human antibodies specific for the indicated proteins. The following antibodies were used: Anti-CD2-PE (BioLegend, Cat# 300208, 1:100), Anti-CD3-PE (BioLegend, Cat# 300408, 1:100), Anti-CD4-PE (BioLegend, Cat# 300508, 1:100), Anti-CD8α-PE (BioLegend, Cat# 344706, 1:100), Anti-CD11a-PE (BioLegend, Cat# 350606, 1:100), Anti-CD45-PE (BioLegend, Cat# 304008, 1:100), Anti-HLA-A2-PE (BioLegend, Cat# 343306, 1:100), Anti-B2M-PE (BioLegend, Cat# 316306, 1:100), Anti-CD54-PE (BioLegend, Cat# 353105, 1:100), Anti-CD58-PE (BioLegend, Cat# 330905, 1:100), Anti-CD43-PE (BioLegend, Cat# 343203, 1:100), Anti-CD83-FITC (BioLegend, Cat# 305306, 1:100), Anti-CD1a-Pacific Blue (BioLegend, Cat# 300124, 1:100), Anti-CD14-PerCP/Cyanine 5.5 (BioLegend, Cat# 367110, 1:100), Anti-CD11c-AF647 (BioLegend, Cat# 301620, 1:100), Anti-HLA-DR-APC-Cy7 (BioLegend, Cat# 307618, 1:100), and IgG1 Isotype-PE (MOPC-21; BioLegend, Cat# 400114, dilution matched to the highest other PE-labeled antibody concentration in the staining experiment). 0.5 × 106 cells were counted, washed in PBS ($0.01\%$ NaN3; PBS azide), and then stained at 4 °C for 1 h using the indicated dilution of the antibody. Cells were washed ×2 in PBS azide, fixed in $2\%$ paraformaldehyde (PFA), and analyzed by flow cytometry. Cells were analyzed using the Attune NxT (Lifetechnologies) flow cytometer. Compensation was performed where required. Data were analyzed using FlowJo. All cell sorting was performed by the Weatherall Institute of Molecular Medicine FACS Facility.
## Protein production and labeling
Soluble forms of CD45, CD58, and ICAM-1 (CD54) were made as previously described64. Briefly, cDNA encoding the ECD of CD43 (residues 20-253 UniProtKB P16150) was cloned into the pHR plasmid downstream of the sequence encoding cRPTPσSP. The CD43 cDNA was modified to encode a H6-SRAWRHPQFGG-H6 ‘H6-spacer-H6’ tag at the C-terminus for purification and stable interaction with 1,2-dioleoyl-sn-glycero-3 (DGS)-NTA(Ni2+) containing SLBs65. Soluble protein expressed by lentiviral transduction of HEK 293T cells was purified using metal-chelate and size-exclusion chromatography with an AKTA Pure protein purification system. pMHC was produced as previously described66. Briefly, cDNA encoding the ECD of HLA-A2 (residues 25-304, UniProtKB P79603) and beta-2-microglobulin (β2M, residues 21-119, UniProtKB P61769) were ligated into pET28a (+; kanamycin resistant) vector for expression in Rosetta 2 (DE3)pLysS competent E. coli (Merck). The HLA-A2 cDNA was modified with a ‘H6-spacer-H6’ tag at the C-terminus to allow interaction with SLBs. HLA-A2 and β2M were purified from inclusion bodies and folded in the presence of either 9V, 3I, 9L, 4D, 5F, Tax WT, or gp100 peptide41,67. Monomeric pMHC was purified using the AKTA Pure protein purification system. UCHT-1 Fab was prepared from purified antibody using immobilized papain as directed by the manufacturer (ThermoFisher). Fab digestion and purity were confirmed by size exclusion chromatography (AKTA). A C-terminally histidine- and HaloTag-tagged form of the UCHT-1 Fab was expressed by lentiviral transduction in HEK 293T cells and purified using metal-chelate and size-exclusion chromatography (referred to as UCHT-1 Fab-HaloTag). OKT3 antibody was provided by the Medical Research Council Human Immunology Unit, Oxford. All proteins were snap-frozen in dry ice and stored at −80 °C prior to use. Proteins were labeled using either an Alexa Fluor (Alexa)-488, -555, or -647 Antibody Labeling Kit (ThermoFisher) according to the manufacturer’s instructions, or a similar custom protocol using the NHS dye derivatives. Briefly, $10\%$ (v/v) 1 M sodium bicarbonate solution was added to the protein and incubated with a 10-fold molar excess of the NHS dye (diluted in anhydrous dimethyl sulfoxide, DMSO) for 1 h, followed by purification in a SEC spin column (Bio-Rad). Proteins were labeled at ≥1 dye per molecule. Prior to use, the frozen protein was thawed at 4 °C, centrifuged (17,000 × g, 5 min, 4 °C) to remove any large aggregates formed during the thawing process, re-aliquoted to a fresh tube, and the concentration determined using a NanoDrop Spectrophotometer (Labtech). After thawing, the protein was labeled, divided into 10 μl aliquots, and snap-frozen or used and stored at 4 °C for 2 weeks before being discarded.
## Protein densities on moDC
Monocyte-derived dendritic cells (moDCs) were produced by isolating human monocytes from PBMCs using a Ficoll gradient and a CD14+ CD16- Magnetic Bead Isolation Kit (Miltenyi). Purified monocytes were grown in an uncoated 24-well plate at 1 × 106 ml−1 in RPMI 1640 supplemented with $10\%$ (v/v) FCS, $1\%$ (v/v) sodium pyruvate, $1\%$ (v/v) HEPES, $1\%$ (v/v) pen/strep antibiotics, $1\%$ (v/v) L-glutamine, $1\%$ NEAA (v/v), and $1\%$ β-mercaptoethanol at 37 °C with $5\%$ CO2. Monocytes grown in medium without cytokines were used as a negative control. To differentiate monocytes into immature moDCs recombinant human IL-4 (10 ng/ml) and GM-CSF (100 ng/ml; BioLegend) were added to the medium for 1 week, with fresh medium and cytokine mix added on day three of seven. On day 5, spent medium was replaced with fresh medium containing the cytokines and 200 ng/ml of LPS (Sigma-Aldrich) to create mature moDCs. These were analyzed by flow cytometry for mature moDC markers (CD14, CD1a, CD11c, CD83, HLA-DR, and Live/Dead stain68) in combination with either β2M, HLA-A2, CD43, CD45, CD58, or ICAM-1 PE-conjugated antibodies (BioLegend). BD Biosciences Quantibrite PE calibration beads were used to estimate total numbers of surface proteins per mature moDC, as per the manufacturer’s protocol. To convert total protein per moDC into a density, the surface area of moDCs was calculated from z-stack (3D) images taken using a ZEISS LSM 880 with Airyscan every 200 nm from the basal to apical plane in Airyscan mode. Briefly, moDCs were labeled with CellMask Red (ThermoFisher) according to the manufacturer’s protocols, fixed in $4\%$ PFA and $0.25\%$ glutaraldehyde at room temperature for 15 min, then placed on PLL-coated glass. Total surface area was calculated by thresholding and then quantifying the perimeter of the cell membrane (outer line) in each frame of a z-stack, multiplying the perimeter by the depth (i.e., 200 nm) to obtain the surface area per frame, and summing the ‘per frame surface area’ across the entire z-stack. This gave a median surface area of ~2000 μm2, consistent with other measurements69,70. Total protein amounts were divided by the median surface area to derive approximate protein densities.
## Glass-supported lipid bilayers
SLBs were prepared using vesicle fusion. A lipid mixture consisting of $98\%$ (mol%) POPC (Avanti Polar Lipids) and $2\%$ (mol%) DGS-NTA-Ni2+ (NiNTA; Avanti Polar Lipids) in chloroform were mixed in a cleaned glass vial and dried under a stream of nitrogen. $2\%$ NiNTA was chosen to minimize unwanted interactions of the lipid with biological material (e.g., non-tag histidine within proteins and negatively charged sugars on cells) whilst allowing physiological densities of multiple his-tagged proteins to be attached to the surface71. The dried lipid mix was resuspended in 0.22 μm filtered (0.02 μm filtered for TIRFM experiments) PBS at 1 mg/ml, vortexed, and sonicated (30 min on ice with a tip sonicator, or until transparent in an ultrasonicator bath) to produce small unilamellar vesicles (SUVs). Glass coverslips (25 mm, thickness no. 1.5; VWR) were cleaned for at least 2 h in 3:1 sulfuric acid/hydrogen peroxide at room temperature, rinsed in MQ water, and plasma cleaned for 1 min (oxygen plasma) or 20 min (argon plasma). CultureWell 50-well silicon covers (Grace Bio-Labs) were cut and placed on the washed coverslips (maximum 4/coverslip). SUVs were added to each well at a final concentration of 0.5 mg/ml (10 μl total volume) and left for 1 h at room temperature. Wells were washed at least five times by removing and adding PBS to each well. The amount of liquid in each well was adjusted to be level with the well edge before adding 5 μl of proteins mixed at the desired concentrations. This was done to ensure reproducible incubation concentrations. Protein mixes were incubated with the bilayer for an hour at room temperature and washed ten times in pre-warmed (37 °C) PBS or PBS plus 2 mM MgSO4 (37 °C) immediately before use. Point fluorescence correlation spectroscopy (pFCS) was used to relate protein concentration to density on the SLB.
## Point fluorescence correlation spectroscopy
The Zeiss 780 equipped with a ×40 water objective was used for pFCS measurements. Data were acquired in photon counting mode, using a $\frac{488}{594}$/633 MBS, and a pinhole set to 1 AU. Excitation was performed with a 633 nm He-Ne laser set at $1\%$ power for Alexa-647 labeled proteins. To calibrate the confocal volume size, 100 nM Alexa-647 dye in PBS was used. SLBs were allowed to acclimate to 37 °C before measurements were taken, and the z-axis was tuned to the highest intensity plane before imaging. Three 10 s measurements were typically taken for each SLB.
Data were fitted and processed using PyCorrFit software72. Autocorrelation values \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{{\rm{G}}}}}}\left({{{{{\rm{\tau }}}}}}\right)$$\end{document}Gτ were fitted to the following equation:1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{{\rm{G}}}}}}\left({{{{{\rm{\tau }}}}}}\right)={{{{{\rm{G}}}}}}\left(0\right)\cdot {{{{{\rm{GD}}}}}}\left({{{{{\rm{\tau }}}}}}\right)\cdot {{{{{\rm{GT}}}}}}\left({{{{{\rm{\tau }}}}}}\right)+{{{{{\rm{Off}}}}}}$$\end{document}Gτ=G0⋅GDτ⋅GTτ+Off The autocorrelation value at a given lag time \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{{\rm{\tau }}}}}}$$\end{document}τ is described by the correlation amplitude \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{{\rm{G}}}}}}\left(0\right)$$\end{document}G0, by diffusive processes \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{{\rm{GD}}}}}}\left({{{{{\rm{\tau }}}}}}\right)$$\end{document}GDτ, by the photophysics \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{{\rm{GT}}}}}}\left({{{{{\rm{\tau }}}}}}\right)$$\end{document}GTτ, and by the correlation offset \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{{\rm{Off}}}}}}$$\end{document}Off. The density of protein can be obtained because \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{{\rm{G}}}}}}\left(0\right)$$\end{document}G0 is inversely related to the average number of particles in the confocal volume \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$({{{{{\rm{G}}}}}}[0]\propto {{{{{{\rm{N}}}}}}}^{-1})$$\end{document}(G[0]∝N−1). To obtain absolute density values the size of the confocal volume must be known. This is obtained from the calibration of free dye in solution. To obtain details on the confocal volume for 2D surfaces the following equation was used:2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{{\rm{D}}}}}}={{{{{{\rm{d}}}}}}}^{2}/[8\cdot {{{{{\rm{ln}}}}}}[2]\cdot {{{{{\rm{\tau }}}}}}{{{{{\rm{D}}}}}}]$$\end{document}D=d2/[8⋅ln[2]⋅τD] The equation can be rearranged to identify \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{{\rm{d}}}}}}$$\end{document}d, which represents the beam diameter or observation spot size. This is obtained from an assumption of a circular cross-section at the center and highest intensity region of a laser beam (i.e., a gaussian laser beam; one with a transverse intensity profile that approximates a gaussian distribution). \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{{\rm{D}}}}}}$$\end{document}D is the diffusion coefficient that can be obtained from the literature. The transit time \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{{\rm{\tau }}}}}}{{{{{\rm{D}}}}}}$$\end{document}τD is the average time taken for a dye to pass through the confocal volume. This is obtained from fitting an autocorrelation plot and taking the decay time. Once the diameter is obtained, the 2D area can be calculated, and therefore the density of labeled and mobile proteins on an SLB. The diameter (d) when using Alexa-647 dye at 37 °C was ~290 nm.
## Altering SLB composition
We used high levels of pMHCnull to mimic the background of self pMHC presented on APCs but also to keep total protein density constant on the SLB. pMHCnull was exchanged for other proteins (e.g., CD58, CD43, CD45, and ICAM-1) during production of the protein mix to modify SLBs. To ensure protein densities on the SLB were physiological they were matched to the mature monocyte-derived dendritic cell surface. To obtain the proteins within the density ranges consistent with moDCs, different combinations of protein concentrations were tested and analyzed using pFCS, using a trial-and-error approach wherein one protein was labeled with Alexa-647 and the others were unstained. To obtain the other protein densities for a given set of protein concentrations, the Alexa-647 labeled protein was switched to an unstained one at the same concentration, and a previously unstained protein was switched for an Alexa-647 labeled protein. For example, to obtain protein densities for SLB1s, three separate SLBs were produced to test one concentration set, whereas SLB2s required six. The final concentrations used for SLBs (in a 5 μl volume) are shown in Table 1 below. Table 1Protein concentrations used for different SLB compositionsSLB1SLB1 + CD58SLB1 + glycocalyxSLB2SLB$\frac{2}{3}$x glycocalyxProteinConc. ( ng/μl)Conc. ( ng/μl)Conc. ( ng/μl)Conc. ( ng/μl)Conc. ( ng/μl)pMHC9V00000.5pMHCnull1214.811.05101.78ICAM-121.21.21.21.2CD5800.500.50.375CD45002.52.56.3CD43001.51.54.5
## Altering agonist pMHC density
For pMHC density measurements and prior to placing cells on SLBs, Alexa-647 labeled pMHC9V on the SLB was subject to 3 × 10 s pFCS measurements using a 40x water objective (NA1.2; with 633 nm laser at $1\%$ laser power). To alter the density of pMHC9V, ceteris paribus, it was exchanged directly for pMHCnull; e.g., 0.5 ng/μl pMHC9V and 9.5 ng/μl pMHCnull becomes 0.1 ng/μl pMHC9V and 9.9 ng/μl pMHCnull. As the lower limit of detection for pFCS in this setup was ~1 molecule/μm2, standard curves were produced to extrapolate densities below this. This was achieved by performing a four-point titration of density versus concentration. The three data points were measured above 1/μm2 and were used to extrapolate the 4th lower density by assuming a linear relationship between density and concentration that passes through 0,0. The regression line equation could then be used to calculate concentration and density of agonist pMHC. For SLB1, the equation was Y (molecules/μm2) = 132.6*X(concentration in ng/μl). For SLB1 + glycocalyx, the equation was Y (molecules/μm2) = 327*X(concentration in ng/μl). For SLB1 + CD58, the equation was Y (molecules/μm2) = 434.8 × X(concentration in ng/μl). For SLB2, the equation was Y (molecules/μm2) = 185.5*X(concentration in ng/μl).
## Calcium release assay
Jurkat-derived cell lines or human primary CD8+ T cells that were primed and rested were either labeled with Fluo-4 or expressed the genetically encoded calcium sensor (GECI) jGCaMP7s. For Fluo-4, ~0.5 × 106 cells were washed in PBS and placed in 1:1 mix of RPMI (no supplements). 25 μg/ml of Fluo-4 dye (final concentration; ThermoFisher) was added, and cells left for 15 min at 37 °C. Cells were washed in pre-warmed PBS plus 2 mM MgSO4 and resuspended in the same. For the GECI, ~0.5 × 106 cells were washed in pre-warmed PBS plus 2 mM MgSO4, resuspended in the same. In all cases cells were incubated for a further 5 min at 37 °C. After the 5-min incubation, cells were gently placed onto the target surface and imaged using a 10x magnification objective for a larger field of view to simultaneously analyze 102−103 cells. Cells were imaged every 1 s. Both Fluo-4 and GECI were excited using an argon 488 nm laser. Calcium imaging was performed on a Zeiss LSM 780 inverted confocal scanning microscope using a 10× objective. Cells were analyzed using a bespoke MATLAB code.
## Automated cell tracking and calcium analysis
To facilitate the analysis of calcium responses across hundreds to thousands of cells an automated MATLAB script was used (https://github.com/janehumphrey/calcium). The script tracks the movement and fluorescence intensity of the calcium dye (or any fluorescent label) for 100–1000 s of individual cells simultaneously. It first removes background, flat-field corrects, and gaussian smooths each video. Local maxima are then identified from the calcium signal, and the maxima from each frame are combined into contiguous tracks using a nearest-neighbor approach. The displacement and speed of each cell can be calculated. The time to adhesion is also calculated by a user-set threshold of speed, and when the cell speed drops and stays below this threshold for the remainder of its track a cell is classified as adhered. This was set at 0.2 μm/s based on the highly adherent interaction of Jurkats with unblocked glass. The mean intensity of the calcium signal is tracked over time for each cell. Time for $50\%$ of all cells to adhere was extrapolated from fits of cumulative distribution plots of fraction of cells adhered, versus time for each SLB. Curves were fitted to the cumulative distribution plots constrained only to a maximum value of $100\%$, i.e., that all cells would eventually have a speed that dropped below 0.2 μm/s. Of note, displacement/speed is typically due to the currents generated even when gently adding cells to the SLB, which we exploited to address the adhesiveness of different SLB compositions.
The start of a calcium release (i.e., a proxy for TCR triggering events) was identified from a positive rate of increase in calcium signal from an identified background, calculated individually for each cell i.e., the lowest $10\%$ of a cell’s fluorescence intensity typically derived from the first 30–60 s of a cell track. To control for any variation in baseline fluorescence levels, thus allowing a comparison of calcium responses within and between samples, each calcium intensity trace was normalized by setting the baseline to one. User-set thresholds for identifying triggering events were also set, e.g., minimum duration and fold change from baseline for each increase in calcium signal. These criteria were set stringently with a minimum duration of 10 s and a fold change from baseline at 3 to remove false-positive increases in fluorescent signal derived from noise or cells settling on top of/near each other. These thresholds were set based on comparing the results of automated versus manual analysis of cell activation from several videos. Cells that showed calcium release within 40 s of the start of their measurement track were excluded from the analysis. These cells (typically 5–$10\%$ of all cells) settled and produced a bright calcium signal that then faded over the course of the video. This gave the appearance of an initial calcium spike to the code and was erroneously detected as an activated cell. These cells may represent pre-activated cells derived from charge-based interaction with Eppendorf tubes or from pipetting. Time for $50\%$ of all cells to signal was extrapolated from fits of cumulative distribution plots of the fraction of cells signaling, versus time for each SLB. Curves were fitted to the cumulative distribution plots constrained to a minimum of 0 and a maximum value of ~$80\%$, based on the fraction of cells responding to OKT3 on glass on the day within a 10-min video; ~$80\%$ represented the upper limit of signaling given that most cells (>$75\%$ of all cells) would respond on OKT3-coated glass within 2 min. Overall, the code provides details on kinetics of movement, cell speed, adhesion, and cell activation as well as more in-depth details such as the strength of the calcium response, e.g., height, length, and integrated calcium increase. These metrics can be combined providing information about the time cells respond relative to settling, and the probability of a cell responding based on its time adhered.
## Calcium, IRM, and synapse imaging
Calcium release assays were performed as above. Immediately after, with the same SLB, and on the same microscope, the 10x objective was switched to a NA1.4 63x oil immersion objective to image cells in brightfield, the IRM contact area, and synapse formation. The confocal was set up to allow IRM images by choosing an appropriate emission filter that allows incident light reflected from the sample to be detected73. To visualize synapses, ICAM-1 was labeled with Alexa-555 prior to addition to the SLB. Cells were manually segmented and IRM contact area was quantified using Yen thresholding and the Analyze Particle feature in Fiji. Synapse formation was manually quantified using the fraction of all cells in a field of view that had formed an ICAM-1 ring.
## Scanning electron microscopy
13-mm diameter glass slides were incubated with poly L-lysine (PLL; 70–150 kDa MW; Sigma) according to the manufacturer’s protocols. 1 × 106 J8 cells were fixed in $4\%$ PFA $0.25\%$ glutaraldehyde in PBS overnight at 4 °C. Fixed cells were gently placed on coated slides and left to settle for 20 min at room temperature. Slides were gently rinsed in PBS and fixed using $1\%$ OsO4 incubated at room temperature for 30 min. The sample was washed in PBS and dehydrated using increasing concentrations of ethanol (50 to $100\%$ in $10\%$ increments) over an hour. Ethanol was removed and 0.5 ml hexamethyldisilazane added for 3 min. This was removed and cells dried in a fume cupboard. Glass slides were mounted on carbon adhesive tape, sputter coated with gold, and viewed using a JEOL-6390 Scanning Electron Microscope.
## Confocal imaging of fixed T cell microvilli
0.5–1 × 106 J8 cells were labeled with anti-CD62L (L-selectin) antibody labeled with Alexa-647 at 4 °C for 1 h. Cells were washed in PBS, then fixed in $4\%$ PFA and $0.25\%$ glutaraldehyde for 20 min at room temperature. During this time cells were labeled with 1x CellMask Green Plasma Membrane Stain (ThermoFisher). Cells were washed in PBS and immediately placed on PLL-coated glass surfaces (previously washed with ethanol) for 10 min prior to imaging. Imaging was performed on an LSM 880 scanning confocal with AiryScan within 30 min of plating. Cells were specifically imaged from the midplane to the apical surface to ensure microvilli in contact with the PLL (basal planes) were not included, in case the interaction altered protein organization even after fixation. Z-stacks were taken at 200 nm slices every 2 s. Cells were imaged using a NA1.4 63x oil immersion objective. Line scan averaging was set to a maximum of 8. Argon 488 nm, DPSS 561 nm, or He-Ne 633 nm laser power was set at appropriate levels to reduce bleaching and oversaturation. For measurements, the pinhole was set to 1 AU and a $\frac{488}{594}$/633 MBS was used. Images were processed using Fiji with the Z-projection (maximum intensity) feature. Fifteen frames at 200 nm spacing were used per cell analyzed. The microvillus diameters were quantified by manually choosing membrane protrusions with clear L-selectin enrichment, obtaining the cross-section line profile, and using the full width at half maximum of the increased signal as a measure of the diameter.
## Actin-drug modification of T-cell morphology (fixed-cell imaging)
0.5 × 106 J8-GECI cells were washed ×1 in PBS and resuspended in either 10 μM DMSO (control), 1 μM latrunculin B, and 10 μM cytochalasin D, or 100 nM jasplakinolide (Cayman Chemical) diluted in RPMI (no supplements) for 1 h at 37 °C with $5\%$ CO2. Cells were then fixed in PBS containing $4\%$ PFA and $0.25\%$ glutaraldehyde for 20 min at room temperature. During this period the cells were also labeled with CellMask Deep Red Plasma Membrane Stain (ThermoFisher). Cells were washed ×2 in PBS and immediately placed on PLL-coated glass surfaces (previously washed with ethanol) for 10 min prior to imaging at room temperature. Images were taken as described for imaging of fixed T cell microvilli.
## Actin-drug modification of T-cell morphology (live-cell imaging)
During drug treatment of the cells, SLBs were made containing either no glycocalyx elements (i.e., SLB1 + CD58; pMHCnull, pMHC9V, ICAM-1, and CD58) or with the glycocalyx (i.e., SLB2; pMHCnull, pMHC9V, ICAM-1, CD58, CD45, and CD43). The cells and SLBs were washed and resuspended in pre-warmed PBS plus 2 mM MgSO4 with the same concentration of actin-modifying drug as above, and incubated at 37 °C for 5 min prior to use. Calcium release was imaged using a 780 LSM scanning confocal for 10 min. To relate drug treatment to contact formation, prior to dropping cells on SLB2s, the cells were labeled with CellMask Deep Red. J8-GECI cells were labeled using 1x CellMask Deep Red for the final 10 min of actin-modifying drug treatment. CD45 and CD43 were labeled with Alexa-555 to observe close contacts. Cells were then washed and resuspended in pre-warmed PBS plus 2 mM MgSO4. SLBs and membrane-labeled cells were incubated at 37 °C for 5 min prior to being gently deposited onto the SLBs and immediately imaged by confocal microscopy. Calcium release, cell membrane, and close-contact formation were imaged every 10 s for 10 min. Confocal videos were analyzed using our custom contact analysis Python script given below, with user-input parameters modified to match the metadata from the confocal microscopes.
## TIRF microscopy
Three-color live cell imaging of SLB-cell contacts was performed on a bespoke TIRF microscope. The excitation path comprised laser lines of 488 nm (iBeam-SMART, Toptica), 561 nm (LaserBoxx, DPSS, Oxxius), and 641 nm (Obis, Coherent). Each beam was circularly polarized using quarter-wave plates, collimated, and expanded. Laser lines were combined using appropriate dichroic mirrors and expanded further to reduce the flat-field variation. The lasers were aligned off-axis at the edge of the objective lens (100x Plan Apo TIRF, NA 1.49 oil-immersion, Nikon Corporation) and focused onto the backfocal plane to achieve TIR illumination. The objective was mounted on an inverted optical microscope (Ti2, Eclipse, Nikon Corporation). The fluorescence emission was collected through the same objective and separated from the excitation light via a dichroic mirror (Di01-R$\frac{405}{488}$/$\frac{561}{635}$, Semrock). It was further passed through appropriate filters according to the excitation wavelength (FF01-$\frac{520}{44}$-25 + BLP01-488R, LP02-568RS-25 + FF01-$\frac{587}{35}$-25, FF01-$\frac{692}{40}$-25; Semrock), mounted in an automated filter wheel. The fluorescence was expanded (1.5x) and the image formed onto an electron-multiplying charge-coupled device (EMCCD, Evolve 512 Delta, Photometrics) with an electron multiplication gain of 250 ADU/photon operating in frame transfer mode. This resulted in an effective pixel size of 107 nm. To allow for live-cell imaging the microscope was enclosed with an incubator (DigitalPixel) and the temperature maintained at 37 °C. The focus was maintained with the Nikon Perfect Focus System. Image acquisition was automated using the open-source software Micro-Manager. Three-color images were acquired sequentially at 0.5–2 Hz with 100 ms exposure time. The laser powers were optimized to reduce bleaching while maximizing signal.
## TIRFM of calcium signaling, cell footprint, and close-contact formation
SLB2s presenting the background of pMHCnull and the indicated density of agonist pMHC were prepared as described above. To visualize contact structure, CD45 and CD43 were labeled with Alexa-555. All other proteins on the SLBs were unlabeled. SLBs were washed and prepared in pre-warmed PBS plus 2 mM MgSO4. To visualize the membrane of J8-GECI cells or primary CD8+ cells, the cells were labeled using 10 μM CellMask Deep Red for 10 min in RPMI (no supplements) at 37 °C. To visualize calcium release, primary cells were labeled with Fluo-4. J8-GECI or primary cells were then washed and resuspended in pre-warmed PBS plus 2 mM MgSO4. SLBs and cells were incubated at 37 °C for 5 min prior to being gently deposited onto the SLBs and immediately imaged by TIRFM. The SLB was imaged in the 5 min prior to depositing the cells in order to gain the background fluorescence for flat-field correction. A complete frame cycle was taken every 2 s for 10–20 min. Individual microvilli tips, observed as membrane puncta moving in and out of an evanescent field (as seen in Supplementary Fig. 2c), were analyzed using the Trackmate plugin in Fiji. The persistence time was defined as the time each membrane puncta/microvillus tip could be tracked before the fluorescent signal was lost. Only puncta visible prior to a cell settling on the surface were analyzed, i.e., prior to a close contact being formed.
## TIRFM of protein accumulation and exclusion at close contacts
SLB2s with Alexa-555 labeled CD45 and CD43 were prepared as described above. To image the TCR relative to close contacts, J8-GECI cells were incubated with 50 μg/ml of UCHT-1 Fab labeled with Alexa-488 for 10 min in RPMI (no supplements) at 37 °C. To image pMHC relative to close contacts (in a separate experiment), SLB2s were prepared using ~100 molecules/μm2 of pMHC9V-Alexa-647. To image ZAP70 (in a separate experiment), cells were labeled and prepared with 1x Janelia-Fluor 646 HaloTag Ligand in RPMI according to the manufacturer’s instructions. To image L-selectin (CD62L) relative to close contacts (in a separate experiment), cells were labeled with Alexa-647 tagged L-selectin antibody for 10 min in RPMI (no supplements) at 37 °C. After washing both the cells and SLBs in pre-warmed PBS plus 2 mM MgSO4, SLBs and membrane-labeled cells were incubated at 37 °C for 5 min prior to being gently deposited onto the SLBs and immediately imaged using TIRFM. The SLB was imaged in the 5 min prior to depositing the cells to obtain the background fluorescence for flat-field correction. A complete frame cycle was taken every 2 s for 10–20 min. For the enrichment analysis several FOV were taken afterward.
In analogous experiments, CD58 and ICAM-1 on the SLB2s were labeled with Alexa-647 to image their reorganization in relation to close contacts (in separate experiments). ICAM-1 was also labeled with Alexa-555 to be imaged alongside CD58 labeled with Alexa-647. Similarly, cells were labeled with ~15 μg/ml of anti-CD45 (Gap8.3 clone) tagged with silicon rhodamine (SiR) to label cell-expressed CD45. The use of the Gap8.3 Fab was impractical owing to its fast off-rate at 37 °C.
## Simulation of close-contact formation on supported lipid bilayers
Exclusion regions of varying extent of exclusion and size were simulated using Python and assuming a homogeneous intensity as ground truth. The ground truth SLB was an image with 1 nm pixel size and intensity 100 in which circular contacts with intensity values between 0 and 100 were added. The image formation was approximated by a Gaussian blur with sigma of the experimental point spread function of the bespoke TIRF microscope and pixels binned to 107 nm size. The experimental point spread function had been measured using TetraSpeck Microspheres and the sigma of the Gaussian approximation determined as 131 nm. Image brightness was adjusted to match experimental values and the EMCCD simulated using a noise model previously published74. The simulated images were segmented the same way as experimental TIRFM data to determine the smallest and least excluded contacts detectable.
## Analysis of close contacts
Quantitative image analysis of close contacts was performed with custom-written software in Python (https://www.python.org/) using the packages NumPy75, scikit-image76, scipy77, and pandas78, as well as matplotlib79 and seaborn80 for visualization. The software was divided into three parts: image segmentation, feature extraction and analysis, and advanced feature analysis for plotting. The full code is accessible at https://github.com/mkoerbel/contactanalysis_2D.
## Image segmentation
Three-color confocal or TIRFM time-lapse images (with dimensions (x, y, t, 3), acquired as described in “Actin-drug modification of T-cell morphology (live cell imaging)” or “TIRFM of calcium signaling, cell footprint and close-contact formation”) were used to analyze contact dynamics of SLB-interacting T cells. The 641-excitation channel showed the T cell membrane, the 561 channel a protein in the SLB indicating close contacts (glycocalyx, CD43 and CD45), and the 488 channel the intracellular calcium levels of the T cell. First, the cell membrane was used to identify a cell. Because the membrane dye over time also labels the SLB itself, a linearly increasing background was assumed and corrected for. A framewise Difference of Gaussian (DoG) filter was then applied to each frame of this channel to reduce noise and an inhomogeneous background. A binary mask of the cell membrane was created by combining a global, custom-set threshold with a threshold calculated by the Otsu method, which allowed a better segmentation of large, spread cells. Masks with an area below a threshold were removed. Assignment of cell labels to each mask was done with the Watershed algorithm including the time dimensions (i.e., in three dimensions). T cells that formed contacts with the SLB showed reduced lateral mobility, maintaining their position in (x) and (y). Masks of adhered cells thus overlapped in consecutive frames. Using a larger DoG filter on the last cell membrane frame, cell positions were defined by the detected local maxima, separated by at least the cell radius, and used as seeds for the Watershed segmentation. These seeds could be manually edited, and seed-regions added in case the automatic labeling could not distinguish between closely associated cells. Disconnected regions were assigned to the closest labeled region, not further than twice the cell radius measured as Euclidean distance in (x,y,t), and tracing backward in time.
For the analysis of close contacts based on glycocalyx exclusion, TIRFM images were first divided by the flat-field. The flat-field was obtained by summing and normalizing a separate image stack, taken before T cells were added to the SLB. A rolling-ball filter was applied in (t), followed by a Gaussian filter in (x,y) on the corrected image stack. The Laplacian of the filtered image was calculated (giving a Laplacian of Gaussian, LoG, filter overall). Two thresholds were applied to account for the variability of contact size across different conditions: to segment close contacts a hysteresis threshold was used. The upper and lower boundaries were defined as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{{\rm{h}}}}}}\cdot {{{{{\rm{std}}}}}}({{{{{{\rm{I}}}}}}\left({{{{{\rm{LoG}}}}}}\right)}_{{{{{{\rm{notCZ}}}}}}})$$\end{document}h⋅std(ILoGnotCZ), with ℎ being two user defined inputs and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{{{\rm{I}}}}}}\left({{{{{\rm{LoG}}}}}}\right)}_{{{{{{\rm{notCZ}}}}}}}$$\end{document}ILoGnotCZ all pixel intensities of the calculated LoG outside of contacts (segmented membrane areas). To segment late, unconstrained contacts an edge enhanced image (sum of the Gaussian filtered image and the LoG filtered image) was thresholded via \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{{\rm{mean}}}}}}({{{{{{\rm{I}}}}}}\left({{{{{\rm{Gauss}}}}}}\right)}_{{{{{{\rm{notCZ}}}}}}})-{{{{{{\rm{h}}}}}}}_{2}\cdot {{{{{\rm{std}}}}}}({{{{{{\rm{I}}}}}}\left({{{{{\rm{Gauss}}}}}}\right)}_{{{{{{\rm{notCZ}}}}}}})$$\end{document}mean(IGaussnotCZ)−h2⋅std(IGaussnotCZ) and morphologically eroded. Close contacts larger than a minimum size were labeled based on the detected cell labels.
## Feature extraction and analysis
To analyze the segmented images and extract contact features, the calcium response for each cell was first measured. The mean fluorescence intensity in a circle centered at the contact centroid was calculated for each timepoint. If a contact was missing at a given timepoint, it was linearly interpolated from the next neighboring centroids for that cell. The calcium trace was analyzed the same way as in the bulk calcium assay to obtain the timepoints of calcium triggering and adhesion. A quality control step was included to exclude cells that touched the edges of the FOV in the cell membrane channel either before they trigger or, if they do not trigger, within a given time period after the first close contact has been detected. Only cells passing this step were further analyzed. Several events that characterize the interaction of the SLB were defined at the cell level (see also the four interaction stages in the main text, and Table 2 for definitions of the interaction stages). Features for contacts and close contacts were based on area and intensity values and were calculated for each timepoint (see Table 3 for contact features and descriptors).Table 2Interaction eventsEventOutput stringDescriptionCell membrane detectedtime_CZ_first (s)Time when the first cell membrane signal was segmented. Initiation of “searching” stage. Cell adhesiontime_adhesion (s)Time when cell speed first falls below threshold. Close-contact formationtime_CCZ_first (s)Time when the first close contact was detected. Initiation of “scanning” stage. Calcium signalingtime_Ca (s)Time when the cell triggers. Initiation of “spreading” stage. Maximum cell membrane areatime_to_CZ_max (s)Time relative to maximum detected cell membrane area. Initiation of “synapsing” stage. Table 3Contact features and descriptorsFeatureOutput stringDescriptionAreaArea (μm2)*The area* of a contact based on the number of segmented pixels. Exclusionexclusion_10Only applicable to close contacts. The exclusion of glycocalyx from close contacts, defined as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1-{{{{{\rm{perc}}}}}}({{{{{{\rm{I}}}}}}}_{{{{{{\rm{inside}}}}}}})/{{{{{\rm{mean}}}}}}({{{{{{\rm{I}}}}}}}_{{{{{{\rm{outside}}}}}}})$$\end{document}1−perc(Iinside)/mean(Ioutside), with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{{\rm{perc}}}}}}({{{{{{\rm{I}}}}}}}_{{{{{{\rm{inside}}}}}}})$$\end{document}perc(Iinside) being the 10th percentile of intensity values inside the close contact, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{{\rm{mean}}}}}}({{{{{{\rm{I}}}}}}}_{{{{{{\rm{outside}}}}}}})$$\end{document}mean(Ioutside) being the mean of intensities on the 1 pixels wide outline of the segmented cell membrane. Intensity values were taken from the flat-field corrected images. Contact timecontact_time (s)Time the individual close contact has been detected for. In order to be “tracked” from frame to frame, the segmented areas need to overlap temporally. “ time_evol” classifies the end and start of the contact. StagestageFor a contact at a given timepoint, in which stage of the interaction the cell currently is.
## Derived feature analysis
The total cell membrane time trace was smoothed using a mean filter and the maximum detected. Using this, and in the previous step defined event, each timepoint was assigned one of the four interaction stages of the cell. A summary of all calcium traces and total cell membrane areas was produced.
## Sensitivity and resolution limit
The sensitivity to detect close contacts with the presented close-contact analysis was evaluated by simulations (Supplementary Fig. 6c). Simulations showed that contacts with sizes below the diffraction limit could be detected, especially if the exclusion was higher than $40\%$. The filtering applied (temporal filtering and minimal size threshold) ensured that the segmentation of a small number of pixels was not due to noise.
The resolution limit was defined by the optical system, in our case an objective with NA 1.49. Diffraction caused small contacts to appear larger than their actual size. This effect was convoluted with the exclusion at that contact. Simple thresholding of a less excluded contact would result in a smaller contact. This effect was minimized by choosing an edge-based filter (LoG). Segmentation was done on a pixel basis, thus the smallest detectable contact has the area of 1 pixel (for 107 nm pixel size that area was 0.011 μm2) which results in the reporting of close-contact areas below the diffraction limit. Any resulting bias would affect all data equally and not change the presented comparative differences or trends in our data.
## Cumulative density plots for different stages (searching to scanning, scanning to spreading)
Cumulative density function for the transition of each cell between stages was estimated using the Kaplan–Meier estimate for the survival function provided by the lifelines package in Python. The transition to the next stage was considered a “death” event. Median survival times were compared using the log-rank test.
## Detection of single-bound pMHC9V at close contacts
SLB2s with Alexa-555 labeled CD45 and CD43 and Alexa-647 labeled pMHC9V at a concentration of 1.25 pg/μl to achieve single-molecule densities, were prepared as described above. To identify engaged single pMHC9V, the exposure time to image pMHC was set to 200 ms to achieve motion-blur for unbound molecules. Initially 5 frames of the whole FOV were taken without cells for flat-field correction of the CD45/CD43 channel and to estimate the analysis threshold. For analysis, a rolling-ball average of 5 frames was calculated and each frame Gaussian blurred with a sigma of 1 pixel. The local maxima above a set threshold were counted to report the number of bound pMHC per timepoint. The threshold was set so that fewer than 2 pMHC were detected in the initial image stack. The timepoint of signaling was determined as the maximum in the calcium signal gradient and cells aligned with it. Close contacts were segmented using a rolling-ball average, LoG filter, and hysteresis threshold as described above. If the detected local maximum was within the segmented close-contact area, it counted towards the number of bound pMHC within that contact.
## Enrichment analysis of proteins on SLB2s
The enrichment of proteins relative to close contacts was calculated from TIRFM images with at least one channel devoted to the glycocalyx in the SLB. The glycocalyx channel was flat-field corrected and segmented with the same double-threshold approach as in ‘Analysis of close contacts.’ The thresholds were adjusted for each set of images from the same SLB. This defined regions “inside” the close contacts. Each region was morphologically diluted by 10 pixels. In this way, obtained “outside” regions were further refined by excluding parts that overlapped with another “inside” region. If the protein of interest was on the cell, the cell was segmented and “outside” regions that were not within the segmented cell region were excluded as well to avoid an apparent reduction in signal at the cell membrane edge. To calculate the enrichment of the protein of interest, which had been imaged in another color channel, the mean intensity of the protein signal from the “inside” regions, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{{{\rm{I}}}}}}}_{{{{{{\rm{inside}}}}}}}$$\end{document}Iinside, and “outside” regions, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{{{{{\rm{I}}}}}}}_{{{{{{\rm{outside}}}}}}}$$\end{document}Ioutside, for each close contact, was calculated. The enrichment was plotted for each close contact as mean(Iinside)/mean(Ioutside).
## Close-contact area as a fraction of total membrane area at signaling
Total membrane area was quantified for J8-GECI cells and primary CD8+ T cells in the same manner for moDCs, as described in the methods section titled ‘Protein densities on moDC’. The surface area was ~525 μm2 for J8-GECI and ~300 μm2 for primary T cells. Total close-contact area at signaling was divided by the total surface area of a cell in solution to provide the close-contact area as a fraction of total membrane area at signaling.
## Analysis of CD45 exclusion at close contacts
The TIRFM videos obtained were first analyzed as described in “Analysis of close contacts”, with the CD45 channel used as a cell membrane stain, adjusting segmentation accordingly, and not restricting the segmented close contacts to segmented cell membrane areas. CD45 exclusion was further analyzed by first obtaining “inside” and “outside” regions as defined in “*Enrichment analysis* of proteins on SLB2s” using the already segmented close contacts. The exclusion was calculated as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1-{{{{{\rm{perc}}}}}}({{{{{\rm{I}}}}}}_{{{{{{\rm{inside}}}}}}})/{{{{{\rm{mean}}}}}}({{{{{{\rm{I}}}}}}}_{{{{{{\rm{outside}}}}}}})$$\end{document}1−perc(Iinside)/mean(Ioutside), with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1-{{{{{\rm{perc}}}}}}({{{{{\rm{I}}}}}}_{{{{{{\rm{inside}}}}}}})$$\end{document}1−perc(Iinside) being the 10th percentile of intensity values inside the close contact.
## Tracking CD58 accumulation under the TCRKO-GECI cells
Individual sites of CD58 accumulation (a proxy for close contacts) were analyzed using the Trackmate plugin in Fiji.
## Image display
Imaging data were visualized using Fiji (https://imagej.net/Fiji81). All images within a figure panel were contrast matched. TIRFM images of the glycocalyx in the bilayer were flat-field corrected as described in ‘Analysis of close contacts.’ For proteins that accumulate under cells on the bilayer (i.e., CD58, ICAM-1, pMHC), and for image presentation purposes, background fluorescence was subtracted across the whole video. Background fluorescence was removed using a z-projection (average) of a 60 s video (taken at 0.5 Hz) of the bilayer taken prior to adding cells. All TIRFM images of the cell or bilayer proteins were z-projected (average) across 3 frames giving a temporal resolution of 6 s.
## Cartoon representations of proteins attached to the SLBs
Protein structure data were obtained from the Protein Data Bank (https://www.rcsb.org/)82: 2BNQ (1G4-TCR in complex with pMHC40), 1HNF (CD2)83, 5FMV (CD45)49, 1CCZ (CD58)84, 1Z7Z (ICAM-1)85, 5ES4 (LFA-1)86, 5T78 (MUC1 repeated for CD43)87. The structures were rendered using the online “Illustrate” tool (https://ccsb.scripps.edu/illustrate/)88 and assembled in Adobe Illustrator. The CD2/CD58 complex was based on the structure of the complex of their N-terminal domains (1QA9)89, and superpositions of the structure of the full-length ECD of CD2 (1HNF).
## Statistical analysis
Statistical significance was calculated for all quantitative data using GraphPad Prism or the Python programming language and statistical tool packages scipy and scikit-posthocs90. The number of cells per experiment, the number of experiments per measurement, the statistical significance of the measurement, and the statistical test used to determine the significance are indicated in each figure legend where quantification is reported. *In* general, significance was defined based on either Student’s t-test (two-sided), for comparing replicates, or Mann–Whitney U test, for comparing individual cells, computed on the mean or mean rank values from independent experimental replicates or one-way ANOVA with an appropriate multiple comparisons test. In most cases p values were reported in the figure. For comparing contact features across different conditions either the p values of the Kruskal–Wallis H test are reported or, if there was a significant difference in the group ($p \leq 0.05$), individual p values of the pairwise Mann–Whitney U test are reported if they were below 0.05.
## Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.
## Supplementary information
Supplementary Information Peer Review File Description of Additional Supplementary Files Supplementary Movie 1 Supplementary Movie 2 Supplementary Movie 3 Supplementary Movie 4 Supplementary Movie 5 Supplementary Movie 6 Supplementary Movie 7 Supplementary Movie 8 Supplementary Movie 9 Supplementary Movie 10 Supplementary Movie 11 Supplementary Movie 12 Supplementary Movie 13 Supplementary Movie 14 Supplementary Movie 15 Supplementary Movie 16 Supplementary Movie 17 Supplementary Movie 18 Supplementary Movie 19 Supplementary Movie 20 Supplementary Movie 21 Supplementary Movie 22 Supplementary Movie 23 Supplementary Movie 24 Reporting Summary The online version contains supplementary material available at 10.1038/s41467-023-36855-9.
## Source data
Source Data
## Peer review information
Nature Communications thanks the anonymous reviewer(s) for their contribution to the peer review of this work. Peer reviewer reports are available.
## References
1. Polliack A, de Harven E. **Surface features of normal and leukemic lymphocytes as seen by scanning electron microscopy**. *Clin. Immunol. Immunopath* (1975.0) **3** 412-430. DOI: 10.1016/0090-1229(75)90029-X
2. Cai E. **Visualizing dynamic microvillar search and stabilization during ligand detection by T cells**. *Science* (2017.0) **356** eaal3118. DOI: 10.1126/science.aal3118
3. Fernandes RA. **A cell topography-based mechanism for ligand discrimination by the T cell receptor**. *Proc. Natl Acad. Sci. USA* (2019.0) **116** 14002-14010. DOI: 10.1073/pnas.1817255116
4. Vink H, Duling BR. **Identification of distinct luminal domains for macromolecules, erythrocytes, and leukocytes within mammalian capillaries**. *Circ. Res* (1996.0) **79** 581-589. DOI: 10.1161/01.RES.79.3.581
5. Kuo JCH, Gandhi JG, Zia RN, Paszek MJ. **Physical biology of the cancer cell glycocalyx**. *Nat. Phys.* (2018.0) **14** 658-669. DOI: 10.1038/s41567-018-0186-9
6. Shurer CR. **Physical principles of membrane shape regulation by the glycocalyx**. *Cell* (2019.0) **177** 1757-1770.e21. DOI: 10.1016/j.cell.2019.04.017
7. Gandhi JG, Koch DL, Paszek MJ. **Equilibrium modeling of the mechanics and structure of the cancer glycocalyx**. *Biophys. J.* (2019.0) **116** 694-708. DOI: 10.1016/j.bpj.2018.12.023
8. Möckl L. **The emerging role of the mammalian glycocalyx in functional membrane organization and immune system regulation**. *Front Cell Dev. Biol.* (2020.0) **8** 253. DOI: 10.3389/fcell.2020.00253
9. 9.Williams, A. F. & Barclay, A. N. Handbook of experimental immunology. (Blackwell Scientific Publications, 1986).
10. Fanales-Belasio E, Zambruno G, Cavani A, Girolomoni G. **Antibodies against sialophorin (CD43) enhance the capacity of dendritic cells to cluster and activate T lymphocytes**. *J. Immunol.* (1997.0) **159** 2203-2211. DOI: 10.4049/jimmunol.159.5.2203
11. Brown WRA, Barclay AN, Sunderland CA, Williams AF. **Identification of a glycophorin-like molecule at the cell surface of rat thymocytes**. *Nature* (1981.0) **289** 456-460. DOI: 10.1038/289456a0
12. Hermiston ML. **Differential impact of the CD45 juxtamembrane wedge on central and peripheral T cell receptor responses**. *Proc. Natl Acad. Sci. USA* (2009.0) **106** 546-551. DOI: 10.1073/pnas.0811647106
13. Ziprin P, Alkhamesi NA, Ridgway PF, Peck DH, Darzi AW. **Tumour-expressed CD43 (sialophorin) mediates tumourmesothelial cell adhesion**. *Biol. Chem.* (2004.0) **385** 755-761. DOI: 10.1515/BC.2004.092
14. Fernandez-Rodriguez J. **The leukocyte antigen CD43 is expressed in different cell lines of nonhematopoietic origin**. *Tumour Biol.* (2002.0) **23** 193-201. DOI: 10.1159/000067252
15. Ardman B, Sikorski MA, Staunton DE. **CD43 interferes with T-lymphocyte adhesion**. *Proc. Natl Acad. Sci. USA* (1992.0) **89** 5001-5005. DOI: 10.1073/pnas.89.11.5001
16. Manjunath N, Johnson RS, Staunton DE, Pasqualini R, Ardman B. **Targeted disruption of CD43 gene enhances T lymphocyte adhesion**. *J. Immunol.* (1993.0) **151** 1528-1534. DOI: 10.4049/jimmunol.151.3.1528
17. Manjunath N, Correa M, Ardman M, Ardman B. **Negative regulation of T-cell adhesion and activation by CD43**. *Nature* (1995.0) **377** 535-538. DOI: 10.1038/377535a0
18. Newby J, Allard J. **First-passage time to clear the way for receptor-ligand binding in a crowded environment**. *Phys. Rev. Lett.* (2016.0) **116** 128101. DOI: 10.1103/PhysRevLett.116.128101
19. Brodovitch A. **T lymphocytes need less than 3 min to discriminate between peptide MHCs with similar TCR-binding parameters**. *Eur. J. Immunol.* (2015.0) **45** 1635-1642. DOI: 10.1002/eji.201445214
20. Mandl JN. **Quantification of lymph node transit times reveals differences in antigen surveillance strategies of naïve CD4+ and CD8+ T cells**. *Proc. Natl Acad. Sci. USA* (2012.0) **109** 18036-18041. DOI: 10.1073/pnas.1211717109
21. Dustin ML. **The immunological synapse**. *Cancer Immunol. Res* (2014.0) **2** 1023-1033. DOI: 10.1158/2326-6066.CIR-14-0161
22. Staunton DE, Dustin ML, Erickson HP, Springer TA. **The arrangement of the immunoglobulin-like domains of ICAM-1 and the binding sites for LFA-1 and rhinovirus**. *Cell* (1990.0) **61** 243-254. DOI: 10.1016/0092-8674(90)90805-O
23. Shimaoka M. **Reversibly locking a protein fold in an active conformation with a disulfide bond: Integrin αL I domains with high affinity and antagonist activity in vivo**. *Proc. Natl Acad. Sci. USA* (2001.0) **98** 6009-6014. DOI: 10.1073/pnas.101130498
24. Walling BL, Kim M. **LFA-1 in T cell migration and differentiation**. *Front Immunol.* (2018.0) **9** 952. DOI: 10.3389/fimmu.2018.00952
25. Bachmann MF, Barner M, Kopf M. **CD2 sets quantitative thresholds in T cell activation**. *J. Exp. Med* (1999.0) **190** 1383-1391. DOI: 10.1084/jem.190.10.1383
26. Weaver TA. **Alefacept promotes co-stimulation blockade based allograft survival in nonhuman primates**. *Nat. Med* (2009.0) **15** 746-749. DOI: 10.1038/nm.1993
27. Raychaudhuri S. **Genetic variants at CD28, PRDM1 and CD2/CD58 are associated with rheumatoid arthritis risk**. *Nat. Genet* (2009.0) **41** 1313-1318. DOI: 10.1038/ng.479
28. de Jager PL. **The role of the CD58 locus in multiple sclerosis**. *Proc. Natl Acad. Sci. USA* (2009.0) **106** 5264-5269. DOI: 10.1073/pnas.0813310106
29. Rigby MR. **Targeting of memory T cells with alefacept in new-onset type 1 diabetes (T1DAL study): 12 month results of a randomised, double-blind, placebo-controlled phase 2 trial**. *Lancet Diabetes Endocrinol.* (2013.0) **1** 284-294. DOI: 10.1016/S2213-8587(13)70111-6
30. McKinney EF, Lee JC, Jayne DRW, Lyons PA, Smith KGC. **T-cell exhaustion, co-stimulation and clinical outcome in autoimmunity and infection**. *Nature* (2015.0) **523** 612-616. DOI: 10.1038/nature14468
31. Patel SJ. **Identification of essential genes for cancer immunotherapy**. *Nature* (2017.0) **548** 537-542. DOI: 10.1038/nature23477
32. 32.Nishikori, M. et al. Diminished CD2 expression in T cells permits tumor immune escape. J. Clin. Cell Immunol. 7, 10.4172/2155-9899.1000406 (2016).
33. Abdul Razak FR, Diepstra A, Visser L, van den Berg A. **CD58 mutations are common in Hodgkin lymphoma cell lines and loss of CD58 expression in tumor cells occurs in Hodgkin lymphoma patients who relapse**. *Genes Immun.* (2016.0) **17** 363-366. DOI: 10.1038/gene.2016.30
34. Challa-Malladi M. **Combined genetic inactivation of β2-microglobulin and CD58 reveals frequent escape from immune recognition in diffuse large B cell lymphoma**. *Cancer Cell* (2011.0) **20** 728-740. DOI: 10.1016/j.ccr.2011.11.006
35. Binder C. **CD2 immunobiology**. *Front. Immunol.* (2020.0) **11** 1090. DOI: 10.3389/fimmu.2020.01090
36. Zurli V. **Phosphoproteomics of CD2 signaling reveals AMPK-dependent regulation of lytic granule polarization in cytotoxic T cells**. *Sci. Signal* (2020.0) **13** eaaz1965. DOI: 10.1126/scisignal.aaz1965
37. Demetriou P. **A dynamic CD2-rich compartment at the outer edge of the immunological synapse boosts and integrates signals**. *Nat. Immunol.* (2020.0) **21** 1232-1243. DOI: 10.1038/s41590-020-0770-x
38. Frangieh CJ. **Multimodal pooled Perturb-CITE-seq screens in patient models define mechanisms of cancer immune evasion**. *Nat. Genet* (2021.0) **53** 332-341. DOI: 10.1038/s41588-021-00779-1
39. Grakoui A. **The immunological synapse: a molecular machine controlling T cell activation**. *Science* (1999.0) **285** 221-227. DOI: 10.1126/science.285.5425.221
40. Chen JL. **Structural and kinetic basis for heightened immunogenicity of T cell vaccines**. *J. Exp. Med* (2005.0) **201** 1243-1255. DOI: 10.1084/jem.20042323
41. Aleksic M. **Dependence of T cell antigen recognition on T cell receptor-peptide MHC confinement time**. *Immunity* (2010.0) **32** 163-174. DOI: 10.1016/j.immuni.2009.11.013
42. Yang H. **HLA-E-restricted, Gag-specific CD8+T cells can suppress HIV-1 infection, offering vaccine opportunities**. *Sci. Immunol.* (2021.0) **6** eabg1703. DOI: 10.1126/sciimmunol.abg1703
43. Dana H. **High-performance calcium sensors for imaging activity in neuronal populations and microcompartments**. *Nat. Methods* (2019.0) **16** 649-657. DOI: 10.1038/s41592-019-0435-6
44. Yi J, Balagopalan L, Nguyen T, McIntire KM, Samelson LE. **TCR microclusters form spatially segregated domains and sequentially assemble in calcium-dependent kinetic steps**. *Nat. Commun.* (2019.0) **10** 277. DOI: 10.1038/s41467-018-08064-2
45. Fooksman DR. **Functional anatomy of T cell activation and synapse formation**. *Ann. Rev. Immunol.* (2010.0) **28** 79-105. DOI: 10.1146/annurev-immunol-030409-101308
46. Jung Y. **Three-dimensional localization of T-cell receptors in relation to microvilli using a combination of superresolution microscopies**. *Proc. Natl Acad. Sci. USA* (2016.0) **113** E5916-E5924. DOI: 10.1073/pnas.1605399113
47. Brodovitch A, Bongrand P, Pierres A. **T lymphocytes sense antigens within seconds and make a decision within one minute**. *J. Immunol.* (2013.0) **191** 2064-2071. DOI: 10.4049/jimmunol.1300523
48. Sage PT. **Antigen recognition is facilitated by invadosome-like protrusions formed by memory/effector T cells**. *J. Immunol.* (2012.0) **188** 3686-3699. DOI: 10.4049/jimmunol.1102594
49. Chang VT. **Initiation of T cell signaling by CD45 segregation at ‘close contacts’**. *Nat. Immunol.* (2016.0) **17** 574-582. DOI: 10.1038/ni.3392
50. Lippert AH. **Soft polydimethylsiloxane-supported lipid bilayers for studying T cell interactions**. *Biophys. J.* (2021.0) **120** 35-45. DOI: 10.1016/j.bpj.2020.11.021
51. van der Merwe PA, McNamee PN, Davies EA, Barclay AN, Davis SJ. **Topology of the CD2–CD48 cell-adhesion molecule complex: implications for antigen recognition by T cells**. *Curr. Biol.* (1995.0) **5** 74-84. DOI: 10.1016/S0960-9822(95)00019-4
52. Sušac L. **Structure of a fully assembled tumor-specific T cell receptor ligated by pMHC**. *Cell* (2022.0) **185** 3201-3213.e19. DOI: 10.1016/j.cell.2022.07.010
53. Dustin ML. **A novel adaptor protein orchestrates receptor patterning and cytoskeletal polarity in T-cell contacts**. *Cell* (1998.0) **94** 667-677. DOI: 10.1016/S0092-8674(00)81608-6
54. Chen KY. **Trapping or slowing the diffusion of T cell receptors at close contacts initiates T cell signaling**. *Proc. Natl Acad. Sci. USA* (2021.0) **118** e2024250118. DOI: 10.1073/pnas.2024250118
55. Kaizuka Y, Douglass AD, Vardhana S, Dustin ML, Vale RD. **The coreceptor CD2 uses plasma membrane microdomains to transduce signals in T cells**. *J. Cell Biol.* (2009.0) **185** 521-534. DOI: 10.1083/jcb.200809136
56. Gérard A, Cope AP, Kemper C, Alon R, Köchl R. **LFA-1 in T cell priming, differentiation, and effector functions**. *Trends Immunol.* (2021.0) **42** 706-722. DOI: 10.1016/j.it.2021.06.004
57. Feigelson SW. **ICAMs are not obligatory for functional immune synapses between naive CD4 T cells and lymph node DCs**. *Cell Rep.* (2018.0) **22** 849-859. DOI: 10.1016/j.celrep.2017.12.103
58. Ghosh S, di Bartolo V, Alon R, Alcover A, Haran G. **ERM-dependent assembly of T cell receptor signaling and co-stimulatory molecules on microvilli prior to activation**. *Cell Rep.* (2020.0) **30** 3434-3447. DOI: 10.1016/j.celrep.2020.02.069
59. Jung Y, Wen L, Altman A, Ley K. **CD45 pre-exclusion from the tips of T cell microvilli prior to antigen recognition**. *Nat. Commun.* (2021.0) **12** 1-16. DOI: 10.1038/s41467-021-23792-8
60. Sanders ME. **Human memory T lymphocytes express increased levels of three cell adhesion molecules (LFA-3, CD2, and LFA-1) and three other molecules (UCHL1, CDw29, and Pgp-1) and have enhanced IFN-gamma production**. *J. Immunol.* (1988.0) **140** 1401-1407. DOI: 10.4049/jimmunol.140.5.1401
61. Kimachi K, Sugie K, Grey HM. **Effector T cells have a lower ligand affinity threshold for activation than naive T cells**. *Int Immunol.* (2003.0) **15** 885-892. DOI: 10.1093/intimm/dxg087
62. Sanjana NE, Shalem O, Zhang F. **Improved vectors and genome-wide libraries for CRISPR screening**. *Nat. Methods* (2014.0) **11** 783-784. DOI: 10.1038/nmeth.3047
63. Shalem O. **Genome-scale CRISPR-Cas9 knockout screening in human cells**. *Science* (2014.0) **343** 84-87. DOI: 10.1126/science.1247005
64. Jenkins E. **Reconstitution of immune cell interactions in free-standing membranes**. *J. Cell Sci.* (2019.0) **132** jcs219709
65. Khan F, He M, Taussig MJ. **Double-hexahistidine tag with high-affinity binding for protein immobilization, purification, and detection on Ni-nitrilotriacetic acid surfaces**. *Anal. Chem.* (2006.0) **78** 3072-3079. DOI: 10.1021/ac060184l
66. Garboczi DN, Hung DT, Wiley DC. **HLA-A2-peptide complexes: refolding and crystallization of molecules expressed in Escherichia coli and complexed with single antigenic peptides**. *Proc. Natl Acad. Sci. USA* (1992.0) **89** 3429-3433. DOI: 10.1073/pnas.89.8.3429
67. Pettmann J. **The discriminatory power of the T cell receptor**. *Elife* (2021.0) **10** e67092. DOI: 10.7554/eLife.67092
68. Hashimoto SI. **Identification of genes specifically expressed in human activated and mature dendritic cells through serial analysis of gene expression**. *Blood* (2000.0) **96** 2206-2214. DOI: 10.1182/blood.V96.6.2206.h8002206_2206_2214
69. Dumortier H. **Antigen presentation by an immature myeloid dendritic cell line does not cause CTL deletion in vivo, but generates CD8+ central memory-like T cells that can be rescued for full effector function**. *J. Immunol.* (2005.0) **175** 855-863. DOI: 10.4049/jimmunol.175.2.855
70. Wiśniewski JR, Hein MY, Cox J, Mann M. **A ‘proteomic ruler’ for protein copy number and concentration estimation without spike-in standards**. *Mol. Cell Prot.* (2014.0) **13** 3497-3506. DOI: 10.1074/mcp.M113.037309
71. Nye JA, Groves JT. **Kinetic control of histidine-tagged protein surface density on supported lipid bilayers**. *Langmuir* (2008.0) **24** 4145-4149. DOI: 10.1021/la703788h
72. Müller P, Schwille P, Weidemann T. **PyCorrFit-generic data evaluation for fluorescence correlation spectroscopy**. *Bioinformatics* (2014.0) **30** 2532-2533. DOI: 10.1093/bioinformatics/btu328
73. Barr VA, Bunnell SC. **Interference reflection microscopy. Curr Prot Cell**. *Biol* (2009.0) **45** 23.1-4.23.19
74. Sage D. **Super-resolution fight club: assessment of 2D and 3D single-molecule localization microscopy software**. *Nat. Methods* (2019.0) **16** 387-395. DOI: 10.1038/s41592-019-0364-4
75. Harris CR. **Array programming with NumPy**. *Nature* (2020.0) **585** 357-362. DOI: 10.1038/s41586-020-2649-2
76. van der Walt S. **Scikit-image: image processing in Python**. *PeerJ* (2014.0) **2014** e453. DOI: 10.7717/peerj.453
77. Virtanen P. **SciPy 1.0: fundamental algorithms for scientific computing in Python**. *Nat. Methods* (2020.0) **17** 261-272. DOI: 10.1038/s41592-019-0686-2
78. 78.Reback, J. et al. pandas-dev/pandas: Pandas 1.0.3. 10.5281/ZENODO.3715232 (2020).
79. Hunter JD. **Matplotlib: a 2D graphics environment**. *Comput. Sci. Eng.* (2007.0) **9** 90-95. DOI: 10.1109/MCSE.2007.55
80. Waskom M. **seaborn: statistical data visualization**. *J. Open Source Softw.* (2021.0) **6** 3021. DOI: 10.21105/joss.03021
81. Schindelin J. **Fiji: an open-source platform for biological-image analysis**. *Nat. Methods* (2012.0) **9** 676-682. DOI: 10.1038/nmeth.2019
82. Berman HM. **The Protein Data Bank**. *Nucl. Acids Res.* (2000.0) **28** 235-242. DOI: 10.1093/nar/28.1.235
83. Bodian DL, Jones EY, Harlos K, Stuart DI, Davis SJ. **Crystal structure of the extracellular region of the human cell adhesion molecule CD2 at 2.5 A resolution**. *Structure* (1994.0) **2** 755-766. DOI: 10.1016/S0969-2126(94)00076-X
84. Ikemizu S. **Crystal structure of the CD2-binding domain of CD58 (lymphocyte function-associated antigen 3) at 1.8-Å resolution**. *Biochemistry* (1999.0) **96** 4289-4294
85. Xiao C. **The crystal structure of coxsackievirus A21 and its interaction with ICAM-1**. *Structure* (2005.0) **13** 1019-1033. DOI: 10.1016/j.str.2005.04.011
86. Springer TA, Sen M. **Leukocyte integrin αlβ2 headpiece structures: The αi domain, the pocket for the internal ligand, and concerted movements of its loops**. *Proc. Natl Acad. Sci. USA* (2016.0) **113** 2940-2945. DOI: 10.1073/pnas.1601379113
87. Movahedin M. **Glycosylation of MUC1 influences the binding of a therapeutic antibody by altering the conformational equilibrium of the antigen**. *Glycobiology* (2017.0) **27** 677-687. PMID: 28025250
88. Goodsell DS, Autin L, Olson AJ. **Illustrate: software for biomolecular illustration**. *Structure* (2019.0) **27** 1716-1720.e1. DOI: 10.1016/j.str.2019.08.011
89. Wang JH. **Structure of a heterophilic adhesion complex between the human CD2 and CD58 (LFA-3) counterreceptors**. *Cell* (1999.0) **97** 791-803. DOI: 10.1016/S0092-8674(00)80790-4
90. Terpilowski M. **scikit-posthocs: pairwise multiple comparison tests in Python**. *J. Open Source Softw.* (2019.0) **4** 1169. DOI: 10.21105/joss.01169
|
---
title: A polygenic and family risk score are both independently associated with risk
of type 2 diabetes in a population-based study
authors:
- Elena Duschek
- Lukas Forer
- Sebastian Schönherr
- Christian Gieger
- Annette Peters
- Florian Kronenberg
- Harald Grallert
- Claudia Lamina
journal: Scientific Reports
year: 2023
pmcid: PMC10036612
doi: 10.1038/s41598-023-31496-w
license: CC BY 4.0
---
# A polygenic and family risk score are both independently associated with risk of type 2 diabetes in a population-based study
## Abstract
The availability of polygenic scores for type 2 diabetes (T2D) raises the question, whether assessing family history might become redundant. However, family history not only involves shared genetics, but also shared environment. It was the aim of this study to assess the independent and combined effects of one family risk score (FamRS) and a polygenic score (PGS) on prevalent and incident T2D risk in a population-based study from Germany ($$n = 3071$$). The study was conducted in $\frac{2004}{2005}$ with up to 12 years of follow-up. The FamRS takes into account not only the number of diseased first grade relatives, but also age at onset of the relatives and age of participants. 256 prevalent and additional 163 incident T2D cases were recorded. Prevalence of T2D increased sharply for those within the top quantile of the PGS distribution resulting in an OR of 19.16 ($p \leq 2$ × 10–16) for the top $20\%$ compared to the remainder of the population, independent of age, sex, BMI, physical activity and FamRS. On the other hand, having a very strong family risk compared to average was still associated with an OR of 2.78 ($$p \leq 0.001$$), independent of the aforementioned factors and the PGS. The PGS and FamRS were only slightly correlated (r2Spearman = 0.018). The combined contribution of both factors varied with varying age-groups, though, with decreasing influence of the PGS with increasing age. To conclude, both, genetic information and family history are relevant for the prediction of T2D risk and might be used for identification of high risk groups to personalize prevention measures.
## Introduction
Type 2 diabetes (T2D) is a growing problem in modern societies. The proportion of affected individuals has been gradually increasing for the past few decades1. In addition to lifestyle factors such as obesity, disadvantageous eating behaviours, and a lack of physical activity, it has been shown that genetics plays a big role too in susceptibility to T2D2. Several studies investigated the heritability of T2D resulting in estimates between 20 and $80\%$ with a median of $40\%$3–5.
This rise in prevalence needs to be addressed and halted by improving effectiveness of prevention measures. It has been suggested that those measures should be more personalized and especially target high risk groups6. Knowledge on which individual is at high genetic risk to develop T2D in the future might be one of the key points.
A cluster of T2D cases in a family can be an indication for a high genetic risk for T2D. Having first degree relatives with T2D increased the risk of individuals to develop T2D themselves by three times compared to individuals without a positive family history7. The number of affected parents, but also the age at which the parents were diagnosed greatly influenced the estimation of a person’s diabetes risk8. Therefore, we used a family risk score incorporating not only the number of affected parents and siblings, but also weights for their disease onset and takes into account the participants’ age9.
A more puristic approach to a persons’ genetic risk lies in the measurement of genotypes and calculation of genetic risk scores. *With* genetic information becoming more easily available, the finding that genes seem to play a role in T2D has led to studies investigating polygenic scores (PGS). Khera et al.10, showed that such a genome-wide PGS can be used to identify more than $3\%$ of the population with a more than threefold risk for T2D than the remaining part of the general population10. Thus, such scores might help to target high-risk individuals and offer personalized prevention measures.
We have recently shown that both family history, represented as the aforementioned family risk score9, and genome-wide polygenic scores are independent predictors for the risk of stroke11 and myocardial infarction12. There are no such studies for T2D, yet.
Therefore, we aimed to investigate the independent and combined effect of a genome-wide polygenic score (PGS) and a family risk score (FamRS) on prevalent and incident T2D risk in the population-based KORA F3 study.
## Study population
The KORA studies are a series of population-based studies that were conducted in the city of Augsburg, Germany, and surrounding counties. KORA F3 was performed in $\frac{2004}{2005}$, which was the baseline visit for the present investigation. Participants were chosen randomly, stratified for 10 year age-groups and sex, resulting in 3184 participants. The analysis dataset, with available genotype data and variables of interest (family history and prevalent and incident diabetes) comprised of 3071 participants. The applied study methods included a standardized computer-assisted interview, a physical examination and a blood sampling at baseline13,14. The participants were followed-up for mortality and morbidities (including diabetes status) until 2016. The median follow-up time was 11 years with 25,363 person years.
The study was approved by the Ethics Committee of the Bavarian Medical Association (Bayerische Landesärztekammer), all research was performed in accordance with relevant guidelines/regulations and written informed consent was obtained from all participants.
## Definition of diabetes
All information given in the interview at baseline ($\frac{2004}{2005}$) on diabetes and types of diabetes were validated by patient-records and/or by the patients’ general practitioners, resulting in 256 participants with type 2 diabetes. Age at diagnosis was self-declared in a standardized interview. The incident diabetes cases ($$n = 163$$) were defined as those who did not have T2D at the time of the baseline KORA F3 study ($\frac{2004}{2005}$), but who developed it in the timespan between KORA F3 and the follow-up in 2016. For incident cases, diabetes status, type of diabetes and age at diagnosis was validated by medical charge review or contactin the respionsible physician. This information was not derived for the 256 prevalent cases and was additionally missing for 280 participants. Therefore, the analysis dataset for the incident analysis comprised of 2535 participants.
## Definition and calculation of the FamRS
In order to obtain the FamRS, participants were asked the following questions for both parents and all siblings in a standardized interview: “Does or did your … (e.g. father) have one of the following diseases?” with diabetes as one of the diseases in the list. No specification of diabetes subtypes was made. If the question for diabetes was answered with yes, it was followed-up by the question, if it was “before the age of 60”, “at the age of 60 or later” or “age unknown” to obtain an age-dependent weight. The FamRS score was then obtained by comparing this weighted observed number of affected family members with the number that would have been expected giving the number of relatives and age of the respective participant. More details and the algorithm can be found in9,15 and the Supplementary Material.
The calculated FamRS was then used both as a continuous variable as well as a categorical variable. For the categorical variable, the following categories were applied9: “average risk” for FamRS ≤ 0.5, “positive risk” for 0.5 < FamRS ≤ 1, “strong positive risk” for 1 < FamRS ≤ 2 and “very strong positive risk” for FamRS > 2. For most analyses, categories “positive risk” and “strong positive risk” were joined due to low numbers of participants in the “positive risk” group.
## Genotyping and calculation of the PGS
In the KORA-F3 study, genotypes were derived using the Illumina Omni 2.5 and Illumina Omni Express array. Genotypes were imputed using the HRC reference panel16 and the Michigan imputation Server17. With that method, about 40 million SNPs over the whole genome are available for each of the KORA-F3 study participants. The PGS for T2D that was used in this study was developed and validated by Khera et al.10 in more than 400,000 European samples (UK Biobank). The weights for the 6,917,436 Mio SNPs included in this PGS were derived from the Polygenic Score (PGS) Catalog18 as score number “PGS000014” and calculated using PGS-Calc (https://github.com/lukfor/pgs-calc) by summing up the product of all imputed genotype scores with their respective weights.
## Statistical analysis
Several logistic regression models were performed with both prevalent and incident T2D as outcome. FamRS and PGS were analyzed individually using three adjustment models each: (a) unadjusted model, (b) adjusted for sex, age, BMI and physical activity (PA) and c) additionally adjusting for PGS (in the FamRS model) or FamRS (in the PGS model). All covariates were assessed at the baseline visit. Since FamRS is highly right-skewed, it is not considered as a continuous variable, but was divided into three categories (average risk, positive or strong positive (FamRS 2) or very strong positive risk (FamRS 3)). PGS was considered as a continuous variable with the effect size given for one standard-deviation (sd) increase (1 sd of PGS = 0.123). Since a non-linear relationship of PGS with the prevalence of T2D was observed, the ORs are also given for several upper thresholds of the PGS ($20\%$, $10\%$, $5\%$ or $1\%$). Logistic regression models were also stratified for 10-year age-groups. Due to lower numbers of cases in these subgroups, these models are not adjusted for any other covariates except FamRS and PGS, which were included mutually. Further, FamRS was only divided into two categories in this analysis: average and above average (FamRS > 0.5). Since there were only 6 prevalent cases in the 35–44 y group, this age-group was omitted from this analysis. To account for the limited follow-up time for those, who died during follow-up, inverse probability of censoring weighting was applied for the incident analyses using package riskRegression in R. For each participant, the inverse of the probability of not being censored is calculated using a Cox-model. This is then used as weights in the logistic regression models.
To evaluate best thresholds for discrimination, ROC curves were used and AUC calculated (using function pROC in R). The predictive capacity of PGS and FamRS was determined by continuous Net Reclassification Index (NRI) and Integrated Discrimination Improvement (IDI) using the function improveProb in package Hmisc. Both adjustment models—adjusted for age, sex, BMI, physical activity plus an additional model including either PGS or FamRS – were used as baseline models to compare with. All analyses were performed using the programme R version 4.1.0. p-values < 0.05 were considered to be statistically significant.
## Population characteristics
Descriptive statistics can be found in Table 1. The KORA-F3 analysis dataset consists of 3071 participants, 1575 of which ($51.2\%$) are female, in the age range 35–84 years (mean age 57.4 years) and mean BMI of 27.66. $51.7\%$ were physically active (defined as regular activity for ≥ 1 h per week). At baseline, there were 256 participants with T2D ($8.3\%$), 10 with type 1 diabetes and 9 other or unclear types. From the 2535 participants without prevalent T2D included in the Follow-up analysis, 163 developed T2D after the baseline visit. 35 of those participants with incident T2D and 264 without incident T2D died within the Follow-up period. Both absolute and relative numbers of T2D increased with age at baseline (Supplementary Table S1). The average age at onset (i.e. age at diagnosis as given in the interview) for participants with T2D at baseline was 58 years. Average age at baseline for those participants was 67 years. For incident T2D cases, average age of diagnosis was 67.5 years. Table 1Characteristics of the KORA F3 study ($$n = 3$$,071).VariableAge (in years)57.4 ± 12.88 [46.0, 57.0, 67.0]Women1575 ($51.2\%$)Anthropometric measurementsWeight (in kg)77.93 ± 15.00 [66.90,76.80,87.60]BMI27.66 ± 4.61 [24.42, 27.13, 30.28]Height (in cm)167.7 ± 9.46 [160.5, 167.4, 174.6]Waist circumference (in cm)94.93 ± 13.15 [85.6, 95.10, 103.60]Hip circumference (in cm)106.9 ± 8.98 [100.9, 105.6, 111.8]Waist-Hip-Ratio0.8867 ± 0.085 [0.8240, 0.8890, 0.9470]LifestylePhysically active1589 ($51.7\%$)Categorizations of physical activity Regularly > 2 h per week688 ($22.4\%$) Regularly active for ~ 1 h per week901 ($29.3\%$) Irregularly active for ~ 1 h per week460 ($15\%$) Little to no physical activity1,022 ($33.3\%$)Smoking categories Regular smoker490 ($16.0\%$) Irregular smoker52 ($1.7\%$) Ex-smoker1078 ($35.1\%$) Never-smoker1283 ($41.8\%$)Family history for diabetesT2D in any parent777 ($25.5\%$)Categorization of Family Risk score FamRS Average2582 ($84.7\%$) Positive FamRS343 ($11.2\%$) Strong positive FamRS125 ($4.1\%$)Blood values Glucose (mg/dl)107.2 ± 32.27 [91.0, 100.0, 113.0] HbA1c-values (%)5.369 ± 0.54 [5.100, 5.300, 5.500] Total cholesterol (mg/dl)218.3 ± 39.94 [191.0, 216.0, 243.0] HDL cholesterol (mg/dl)58.77 ± 17.15 [46.00, 56.00, 69.00] LDL cholesterol (mg/dl)128 ± 32.63 [105, 127, 148] Triglycerides (mg/dl)165.2 ± 126.05 [88.0, 136.0, 201.0]DiabetesNumber of prevalent Type 2 cases, n (%)256 ($8.3\%$)Number of incident Type 2 cases, n (%)163 ($5.3\%$)Continuous variables are shown as mean ± SD and [$25\%$, $50\%$, $75\%$] percentile; Categorical variable in n (%).
## Distribution of the FamRS
Supplementary Fig. S1 shows the increasing proportion of T2D for increasing number of relatives affected by diabetes. The FamRS distribution markedly differs between participants with or without T2D at baseline with a high peak at 1 for non-cases and a shift to higher values for cases (Fig. 1). For the incident cases, there is an overrepresentation of peaks above zero compared to those participants not developing diabetes during follow-up (Supplementary Fig. S2). Supplementary Table S1 further shows an increase of prevalent and incident T2D over the FamRS risk categories. Supplementary Fig. S3 gives the distribution of the FamRS in T2D cases stratified in 10-year age-groups, showing a decreasing trend for increasing age. Therefore, FamRS seems to be highest for younger participants with T2D.Figure 1Density plot showing the FamRS distribution in participants without (lightblue, panel A) and with T2D (pink, panel B) at baseline; the lines depict the thresholds of the FamRS categories: positive family risk (blue line), strong positive family risk (purple), very strong positive family risk (red line).
## Distribution of the PGS
The overall distribution of the PGS in the KORA F3 study is given in Supplementary Fig. S4 together with corresponding distributions in 5 different populations from the 1000 Genomes reference population19. It closely matches the distribution of the Europeans, as expected. A clear right shift of the distribution can be observed in participants with T2D compared to those without T2D at baseline (Fig. 2), in consequence leading also to an increase of T2D cases for increasing percentile groups of the PGS (Supplementary Table S1). For incident cases, the shift is also notable but less pronounced (Supplementary Fig. S5). As was observed for the FamRS, the PGS is decreasing with age at baseline for T2D cases (Supplementary Fig. S6). Supplementary Fig. S7 depicts the prevalence of T2D in percentile groups of the PGS (in $5\%$ groups from 0–5 to 95–$100\%$). Up until the $75\%$-mark, there was no obvious difference in prevalence between the groups, with a steep increase above the $85\%$-percentile. Figure 2Density plot showing the PGS distribution in participants without (lightblue) and with T2D (pink) at baseline.
## Relationship between FamRS and PGS
The PGS was found to increase with increasing number of family members affected by T2D (Supplementary Fig. S8), which is only diluted in those with more than 6 affected family members. This group only consists of three participants, though.
The relative amount of individuals with positive to very strong positive familial risk increases with increasing PGS-percentiles (Fig. 3). The frequency of individuals with more than average FamRS is more than doubling from about $10\%$ in the lowest $20\%$ of the PGS distribution to $22\%$ in the highest $20\%$.Figure 3Relative amount of individuals in each FamRS category across the PGS percentile groups.
All in all, both measures are only slightly, although significantly, correlated (r2Spearman = 0.018, $$p \leq 5$$ × 10–14).
## Results of logistic regression models
Table 2 shows the results of the logistic regression models of continuous PGS on both prevalent and incident T2D risk. The effect of the PGS was quite stable and highly significant across all models, with ORs ranging from 5.95 to 6.21 for prevalent and 1.66 and 1.68 for incident cases (per 1 sd increase). Due to the nonlinear and steep increase of T2D prevalence especially in the upper $20\%$ of the PGS distribution (Supplementary Fig. S7), the logistic regression was repeated for several upper thresholds for PGS, with the respective remaining of the distribution as the reference category. The ORs increased from 19 to 47 for the upper $20\%$ to $5\%$ of the PGS distribution, even adjusted for age, sex, BMI, PA and FamRS (Table 3). For incident cases, a significant association can be observed for all presented top percentiles, with an OR of about 5 for the highest $5\%$ of the PGS distribution compared to the remainder. Table 2Results of Logistic regression models of the effect of PGS on risk of prevalent and incident diabetes. Outcome: Prevalent T2DOutcome: Incident T2DOR*CI ($95\%$)p-valueOR*CI ($95\%$)p-valueUnadjusted5.95[4.97–7.20] < 2 × 10–161.66[1.38–1.99]4.49 × 10–8Adjusted for Age + Sex + BMI + Physical activity6.35[5.18–7.89] < 2 × 10–161.68[1.39–2.04]1.08 × 10–7Adjusted for Age + Sex + BMI + Physical activity + FamRS6.21[5.06–7.74] < 2 × 10–161.67[1.37–2.03]2.63 × 10–7* OR are given for increase in 1 SD of PGS (0.123).Table 3Results of Logistic regression models of the effect of PGS cut-offs on risk of prevalent and incident diabetes. Outcome: Prevalent T2DOutcome: Incident T2DOR*CI ($95\%$)p-valueOR*CI ($95\%$)p-valueUpper $20\%$ of PGS19.16[13.65–27.27] < 2 × 10–162.16[1.45–3.18]1.21 × 10–4Upper $10\%$ of PGS32.09[22.21–47.05] < 2 × 10–162.95[1.68–4.99]9.35 × 10–5Upper $5\%$ of PGS47.03[29.69–76.26] < 2 × 10–164.94[2.00–11.35]2.67 × 10–4All models were adjusted for age, sex, BMI, physical activity and FamRS.*OR is given for those above this cut-off with the remainder of the population as the reference.
The respective results for FamRS are given in Table 4. Having a positive or strong positive family risk is significantly associated with T2D risk with an OR of 2.08, with more than double the risk (OR 4.73) for a very strong positive family risk. The ORs do not markedly change, if the model is adjusted for age, sex, BMI and PA. It is attenuated, though, but still significant, when the PGS is additionally included in the model. Table 4Results of Logistic regression models of the effect of FamRS on risk of prevalent and incident diabetesFamRS categoriesOutcome: Prevalent T2DOutcome: Incident T2DOR*CI ($95\%$)p-valueOR*CI ($95\%$)p-valueUnadjustedFamRS 22.08[1.45–2.92]4.05 × 10–51.40[0.85–2.19]0.165FamRS 34.73[3.04–7.20]1.77 × 10–143.46[1.85–6.09]5.89 × 10–5Adjusted for Age + Sex + BMI + Physical activityFamRS 22.47[1.67–3.61]4.30 × 10–61.60[0.95–2.58]0.062FamRS 34.33[2.65–6.95]2.17 × 10–93.74[1.94–6.82]3.49 × 10–5Adjusted for Age + Sex + BMI + Physical activity + PGSFamRS 21.67[1.01–2.71]0.0411.45[0.86–2.35]0.149FamRS 32.78[1.49–5.09]0.0013.47[1.79–6.35]1.07 × 10–4*ORs are given for FamRS categories “positive and strong positive family risk” (FamRS 2) and “very strong positive family risk” (FamRS 3) compared to average family risk.
The effect on incident diabetes was more robust with hardly a difference in ORs between the adjustment models. However, only very strong positive family risk remains significantly associated with an OR of 3.47 in the fully adjusted model. Simplifying the family history to “parents with T2D yes/no” also yields significant association with T2D risk (Supplementary Table S2). The ORs are in about the same range as the one for (strong) positive family risk (OR ranging from 2.14 to 2.68 in the prevalent models and 1.49–1.67 in the incident models), therefore missing the higher risk for those with very strong positive FamRS.
Since Supplementary Figs. S3 and S6 showed decreasing FamRS and PGS with increasing age, it was also evaluated, whether the effect of both scores changed with age. Therefore, logistic regression models were further stratified into 10y age-groups. The effect of PGS is slightly higher for those being 45–64 years at baseline than for older participants with no significant difference, though. The effect also doesn’t change, if additionally adjusted for FamRS (Supplementary Fig. S9, panel B). In contrast, the direction of interaction between age and FamRS on T2D changes, when PGS is taken into account (Supplementary Fig. S9, panel A). In line with the PGS, the OR for FamRS decreases slightly with increasing age in the unadjusted model. It increases, though, especially for the highest age-group, when PGS is adjusted for.
## Discrimination and reclassification measures
When the PGS is used as a classifier for prevalent diabetes, the area under the curve (AUC) amounts to 0.869, and the calculated best threshold is estimated to be 0.016 (~ $75\%$ percentile), (Supplementary Table S3. For incident cases, AUC is markedly lower, but still significantly different from 0.5.
For FamRS the AUC for prevalent cases is 0.617 with the best threshold at 0.109. This threshold reflects the boundary of the high peak at around 0 especially for participants without T2D (Fig. 1A). Therefore, most non-cases are captured (specificity = 0.841), but many cases missed (sensitivity = 0.384) (Supplementary Table SS3 The AUC for incident cases (0.538) does not significantly differ from 0.5.
To evaluate whether PGS or FamRS do improve reclassification compared to a model already including several explanatory variables, NRI and IDI were additionally calculated. However, this was restricted to prevalent cases, since AUCs indicated a low predictive power for incident cases.
Adding the PGS to the baseline model could correctly increase T2D probability for $80\%$ of the prevalent cases and decrease for $83\%$ of non-cases, respectively (Supplementary Table S4). This results in an overall NRI of 1.261. The corresponding IDI was 0.328 ($$p \leq 1.02$$ × 10–66). These measures were nearly identical, when FamRS was also included to the baseline model.
Adding the FamRS to the baseline model could merely ameliorate prediction for non-cases by correctly down-shifting the probability for $85\%$ of the non-cases (NRI = 0.361, $$p \leq 3.38$$ × 10–9), although overall NRI and IDI were still significant (IDI = 0.025, $$p \leq 0.0002$$). IDI was not significant any more, when FamRS was compared to a model already including the PGS (Supplementary Table S4).
The simultaneous addition of PGS and FamRS to the baseline model enhances the prediction for both prevalent cases (NRIevents = 0.579) and non-cases (NRInon-events = 0.665), which leads to an overall NRI of 1.244 and IDI of 0.336 ($$p \leq 1.47$$ × 10–66).
## Discussion
In this study we found a strong association for both, a genome-wide polygenic score and family history, represented by a weighted family risk score, with the risk of prevalent and incident T2D in a population-based study from Southern Germany. Both scores were only slightly correlated with each other and were found to be independent predictors for T2D.
The PGS was found to be clearly right-shifted for prevalent and to a lesser extent incident T2D cases. The ORs were almost unaffected by any of the adjustments. While the finding of this association itself is in line with previous studies10,20, the extent of the effect size markedly differs. Mars et al.21 evaluated PGSs in different populations resulting in a pooled OR of 1.66 per 1 sd increase in mixed European populations, considering incident and prevalent cases jointly, while we found an OR of around 6 for prevalent and 1.66 in incident cases. In our study, the association was found to be non-linear, with a sharp increase of risk for the upper 20–$25\%$ of the distribution, which corresponds to the best discriminating cutoff. Also for these highest risk groups, a marked difference to the literature could be observed: For the top $5\%$ vs. remainder, both Khera et al.10 and Mahajan et al.20,22 observed an OR of 2.75, while our study resulted in an adjusted OR of 47 for the same group. One explanation could be difference in prevalence (~ $2\%$ in10, ~ $8\%$ in KORA-F3) or definition/validation of Type 2 diabetes cases.
PGS was also found to be a valuable predictor for T2D with an AUC of 0.869 for prevalent and 0.613 for incident cases. It also significantly improves reclassification in addition to an already adjusted model and correctly shifts the probabilities for the majority of cases and non-cases to higher respectively lower values. Also here, reported AUCs for PGS were markedly lower (0.64–0.6620).
The FamRS was found to be significantly associated with both prevalent and incident T2D, supporting results of previous studies, which defined family history as any first-degree relative with T2D8 or by also taking into account the number of affected relatives23. The occurrence of T2D in either one of the parents was significantly associated with T2D risk with an OR of about 2.4 for prevalent and 1.6 for incident cases, even in the fully adjusted model. This is comparable to a Dutch study24, which found a HR of 2.2 for parental T2D, also after adjustment for lifestyle factors and obesity. When only looking at parental T2D, the risk discrimination between those with (strong) positive and very strong positive family risk would have been missed. A “very strong positive family risk” (FamRS > 2) occurs, if there are two or more T2D cases in a family with an age of onset before the age of 60. With such a measure, someone with very high familial risk might be identified, but a high fraction of individuals at risk might be missed, especially in small families. This is reflected in the prediction and reclassification measures. Although FamRS can significantly improve prediction, it is a better measure for correctly down-classifying non-cases than up-classifying cases.
However, family history scores taking into account family structure and age of onset, as the FamRS we used, have been shown to be more efficient than only looking at positive family history yes/no or even only disease status in parents9,11,12,25.
Unlike PGS, the effect size of FamRS is not stable across the different adjustment models for prevalent T2D. Although there is no difference from the unadjusted model to the model adjusting for age, sex, BMI and PA, adjustment for the PGS leads to an attenuation of the OR for FamRS risk categories, but it is still significantly associated. This means that a part of the information of FamRS is likely to be of genetic nature, while another part is independent from the PGS.
Cornelis et al.26 estimated by simulation and comparison with empirical data that about one third of the association of parental T2D with T2D risk is due to shared environment, while the remainder is due to shared genetics. This finding is supported by our data. Similar results have already been found for myocardial infarction12,27 and stroke11. For T2D, however, such data are sparse. Chatterjee et al.28 predicted—based on a theoretical model and observed SNP effect size distribution from GWAS—that family history, defined as presence of any affected first-grade relative, hardly improves polygenic risk prediction. In this theoretical framework, however, family history was assumed to represent only shared genetics. In UK-Biobank, however, including the family history of each of the relatives separately in a model improved prediction accuracy of polygenic risk scores29.
In this context, it should be noted that a recall bias cannot be excluded for the determination of the FamRS. It has been shown that both older individuals and those not affected from Diabetes were less accurate reporting the disease status of relatives than younger individuals and individuals with T2D30. This recall bias cannot play a role for incident cases, though, for which the FamRS was also shown to be significantly associated. In these incident models, the effect sizes for FamRS stay quite stable over different adjustment models. Especially for the very strong family risk group, there is also no decrease in effect size from the prevalent to the incident models. This is different for the PGS.
It has been shown that the heritability for T2D is twice as high in patients with age at onset 35–60 years (h2 = 0.69) compared to patients with onset up to 75 years (h2 = 0.31)31. Since the age of onset was about 10 years higher for incident cases in our study than for prevalent cases, the drop in effect sizes for the PGS can likely be explained by that. This assumption is further strengthened by the finding that effect sizes decrease for increasing 10y age-groups. However, confidence intervals overlap to a high degree, which advises against overinterpretations. However, previous studies32,33 have also found that polygenic scores play a greater role for those developing T2D at a younger age. For FamRS, however, there is only a drop in effect sizes for increasing age-groups, if it is unadjusted for the PGS. The drop cannot be observed any more, if it is adjusted for PGS, therefore getting rid of the “shared genetics” component. For higher age, the “shared environment” component, which also partly results in acquired habits and lifestyle, might thus play a bigger role in the FamRS-variable in older participants.
In conclusion, can and should PGS and FamRS be used to identify individuals with high T2D risk? Both scores have been shown to be quite independent from each other, and contribute to risk prediction in varying extent, depending primarily on age of onset. Even though a healthy lifestyle and especially having a normal body weight is beneficial for all in terms of T2D risk, independent of the genetic risk34, it has been shown that those being at high genetic risk defined by either high PGS35, but also high PGS and positive family history36 benefit most from adhering to a healthy lifestyle.
## Supplementary Information
Supplementary Information. The online version contains supplementary material available at 10.1038/s41598-023-31496-w.
## References
1. Thibault V. **Factors that could explain the increasing prevalence of type 2 diabetes among adults in a Canadian province: A critical review and analysis**. *Diabetol. Metab. Syndr.* (2016.0) **8** 1-10. DOI: 10.1186/S13098-016-0186-9
2. Cuschieri S. **The genetic side of type 2 diabetes—A review**. *Diabetes Metab. Syndr. Clin. Res. Rev.* (2019.0) **13** 2503-2506. DOI: 10.1016/j.dsx.2019.07.012
3. Poulsen P, Ohm Kyvik K, Vaag A, Beck-Nielsen H. **Heritability of type II (non-insulin-dependent) diabetes mellitus and abnormal glucose tolerance—A population-based twin study**. *Diabetologia* (1999.0) **42** 139-145. DOI: 10.1007/S001250051131
4. 4.J. B. Meigs, L. A. Cupples, & P. W. F. Wilson, Parental transmission of type 2 diabetes the Framingham offspring study, 2000. [Online]. http://diabetesjournals.org/diabetes/article-pdf/49/12/2201/365516/11118026.pdf.
5. Willemsen G. **The concordance and heritability of type 2 diabetes in 34,166 twin pairs from international twin registers: The discordant twin (DISCOTWIN) consortium**. *Twin Res. Hum. Genet.* (2015.0) **18** 762-771. DOI: 10.1017/THG.2015.83
6. Wareham NJ. **Personalised prevention of type 2 diabetes**. *Diabetologia* (2022.0) **65** 1796-1803. DOI: 10.1007/S00125-022-05774-7
7. Florez JC, Hirschhorn J, Altshuler D. **The inherited basis of diabetes mellitus: implications for the genetic analysis of complex traits**. *Annu. Rev. Genomics Hum. Genet.* (2003.0) **4** 257-291. DOI: 10.1146/ANNUREV.GENOM.4.070802.110436
8. Scott RA. **The link between family history and risk of type 2 diabetes is not explained by anthropometric, lifestyle or genetic risk factors: The EPIC-InterAct study**. *Diabetologia* (2013.0) **56** 60-69. DOI: 10.1007/s00125-012-2715-x
9. Williams RR. **Usefulness of cardiovascular family history data for population-based preventive medicine and medical research (the Health Family Tree Study and the NHLBI Family Heart Study)**. *Am. J. Cardiol.* (2001.0) **87** 129-135. DOI: 10.1016/s0002-9149(00)01303-5
10. Khera AV. **Genome-wide polygenic scores for common diseases identify individuals with risk equivalent to monogenic mutations**. *Nat. Genet.* (2018.0) **50** 1219-1224. DOI: 10.1038/s41588-018-0183-z
11. Hämmerle M. **A family and a genome-wide polygenic risk score are independently associated with stroke in a population-based study**. *Stroke* (2022.0) **53** 2331-2339. DOI: 10.1161/STROKEAHA.121.036551
12. Schnitzer F. **Association between a polygenic and family risk score on the prevalence and incidence of myocardial infarction in the KORA-F3 study**. *Atherosclerosis* (2022.0) **352** 10-17. DOI: 10.1016/j.atherosclerosis.2022.05.014
13. Löwel H, Döring A, Schneider A, Heier M, Thorand B, Meisinger C. **The MONICA Augsburg surveys—Basis for prospective cohort studies**. *Gesundheitswesen* (2005.0) **67** S13-18. DOI: 10.1055/s-2005-858234
14. Holle R, Happich M, Löwel H, Wichmann HE. **KORA—A research platform for population based health research**. *Gesundheitswesen* (2005.0) **67** S19-25. DOI: 10.1055/s-2005-858235
15. Lamina C. **Correlation between a positive family risk score and peripheral artery disease in one case-control and two population-based studies**. *Atherosclerosis* (2014.0) **237** 243-250. DOI: 10.1016/j.atherosclerosis.2014.08.032
16. McCarthy S. **A reference panel of 64,976 haplotypes for genotype imputation**. *Nat. Genet.* (2016.0) **48** 1279-1283. DOI: 10.1038/ng.3643
17. Das S. **Next-generation genotype imputation service and methods**. *Nat. Genet.* (2016.0) **48** 1284-1287. DOI: 10.1038/NG.3656
18. Lambert SA. **The Polygenic Score Catalog as an open database for reproducibility and systematic evaluation**. *Nat. Genet.* (2021.0) **53** 420-425. DOI: 10.1038/S41588-021-00783-5
19. Auton A. **A global reference for human genetic variation**. *Nature* (2015.0) **526** 68-74. DOI: 10.1038/nature15393
20. Udler MS, McCarthy MI, Florez JC, Mahajan A. **Genetic risk scores for diabetes diagnosis and precision medicine**. *Endocr. Rev.* (2019.0) **40** 1500-1520. DOI: 10.1210/ER.2019-00088
21. Mars N. **Genome-wide risk prediction of common diseases across ancestries in one million people**. *Cell Genomics* (2022.0) **2** 100118. DOI: 10.1016/J.XGEN.2022.100118
22. Mahajan A. **Fine-mapping type 2 diabetes loci to single-variant resolution using high-density imputation and islet-specific epigenome maps**. *Nat. Genet.* (2018.0) **50** 1505-1513. DOI: 10.1038/S41588-018-0241-6
23. Hariri S, Yoon PW, Qureshi N, Valdez R, Scheuner MT, Khoury MJ. **Family history of type 2 diabetes: A population-based screening tool for prevention?**. *Genet. Med.* (2006.0) **8** 102-108. DOI: 10.1097/01.gim.0000200949.52795.df
24. Abbasi A. **Maternal and paternal transmission of type 2 diabetes: Influence of diet, lifestyle and adiposity**. *J. Int. Med.* (2011.0) **270** 388-396. DOI: 10.1111/J.1365-2796.2011.02347.X
25. Feng R, McClure LA, Tiwari HK, Howard G. **A new estimate of family disease history providing improved prediction of disease risks**. *Stat. Med.* (2009.0) **28** 1269. DOI: 10.1002/SIM.3526
26. Cornelis MC, Zaitlen N, Hu FB, Kraft P, Price AL. **Genetic and environmental components of family history in type 2 diabetes**. *Hum. Genet.* (2015.0) **134** 259. DOI: 10.1007/S00439-014-1519-0
27. Tada H. **Risk prediction by genetic risk scores for coronary heart disease is independent of self-reported family history**. *Eur. Heart J.* (2016.0) **37** 561-567. DOI: 10.1093/eurheartj/ehv462
28. Chatterjee N, Wheeler B, Sampson J, Hartge P, Chanock SJ, Park JH. **Projecting the performance of risk prediction based on polygenic analyses of genome-wide association studies**. *Nat. Genet.* (2013.0) **45** 400-405. DOI: 10.1038/ng.2579
29. Hujoel MLA, Loh P-R, Neale BM, Price AL. **Incorporating family history of disease improves polygenic risk scores in diverse populations**. *Cell Genomics* (2022.0) **2** 100152. DOI: 10.1016/J.XGEN.2022.100152
30. Bensen JT. **Accuracy of proband reported family history: The NHLBI family heart study (FHS)**. *Genet. Epidemiol.* (1999.0) **17** 141-150. DOI: 10.1002/(SICI)1098-2272(1999)17:2
31. Almgren P. **Heritability and familiality of type 2 diabetes and related quantitative traits in the Botnia study**. *Diabetologia* (2011.0) **54** 2811-2819. DOI: 10.1007/S00125-011-2267-5
32. Kong X, Xing X, Zhang X, Hong J, Yang W. **Early-onset type 2 diabetes is associated with genetic variants of β-cell function in the Chinese Han population**. *Diabetes. Metab. Res. Rev.* (2020.0) **36** e3214. DOI: 10.1002/dmrr.3214
33. Piko P, Werissa NA, Fiatal S, Sandor J, Adany R. **Impact of genetic factors on the age of onset for type 2 diabetes mellitus in addition to the conventional risk factors**. *J. Pers. Med.* (2021.0) **11** 1-17. DOI: 10.3390/jpm11010006
34. Schnurr TM. **Obesity, unfavourable lifestyle and genetic risk of type 2 diabetes: A case-cohort study**. *Diabetologia* (2020.0) **63** 1324-1332. DOI: 10.1007/S00125-020-05140-5
35. Ye Y, Chen X, Han J, Jiang W, Natarajan P, Zhao H. **Interactions between enhanced polygenic risk scores and lifestyle for cardiovascular disease, diabetes, and lipid levels**. *Circ. Genomic Precis. Med.* (2021.0) **14** E003128. DOI: 10.1161/CIRCGEN.120.003128
36. Ding M. **Additive and multiplicative interactions between genetic risk score and family history and lifestyle in relation to risk of type 2 diabetes**. *Am. J. Epidemiol.* (2020.0) **189** 445-460. DOI: 10.1093/AJE/KWZ251
|
---
title: 'Associations of physical activity domains and muscle strength exercise with
non-alcoholic fatty liver disease: a nation-wide cohort study'
authors:
- Yewan Park
- Dong Hyun Sinn
- Kyunga Kim
- Geum-Youn Gwak
journal: Scientific Reports
year: 2023
pmcid: PMC10036618
doi: 10.1038/s41598-023-31686-6
license: CC BY 4.0
---
# Associations of physical activity domains and muscle strength exercise with non-alcoholic fatty liver disease: a nation-wide cohort study
## Abstract
It is unclear if various types and domains of exercise have an identical effect on non-alcoholic fatty liver disease (NAFLD). Thus, this study aimed to investigate associations of different physical activity domains and muscle strength exercise with NAFLD using a nation-wide cohort database. Adults aged 20–79 years who participated in the Korean National Health and Nutrition Examination Survey between 2014 and 2018 were analyzed. Hepatic steatosis index was used to identify NAFLD. Physical activity was assessed with the Global Physical Activity Questionnaire. Of 21,015 participants, 4942 ($23.5\%$) had NAFLD. Participants with ≥ 150 min/week of total physical activity had a lower risk of NAFLD than those with < 150 min/week (the fully adjusted OR: 0.86, $95\%$ CI 0.78–0.95). When the individual domain of physical activity was assessed, ≥ 150 min/week of recreation activity was associated with a reduced risk of NAFLD (OR: 0.77, $95\%$ CI 0.67–0.88), whereas ≥ 150 min/week of travel or work activity was not. The fully adjusted OR for NAFLD comparing participants with ≥ 2/week to those with < 2/week of muscle strength exercise was 0.83 ($95\%$ CI 0.73–0.94). Muscle strength exercise ≥ 2/week showed a lower risk of NAFLD for all levels of total and each specific domains of physical activity except for ≥ 150 min/week of work activity. An increased level of physical activity and muscle strength exercise was associated with a reduced risk of NAFLD, albeit the effect varied depending on domains of physical activity. Thus, physical activity should be differentiated by domains for the management of NAFLD. Muscle strength exercise could also be a good option for individuals who could not perform moderate-to-vigorous physical activity.
## Introduction
Non-alcoholic fatty liver disease (NAFLD) is characterized by the accumulation of fat in the liver without secondary causes1. It is closely related to obesity, diabetes, dyslipidemia, and metabolic syndrome2–4. NAFLD is one of the most prevalent liver diseases worldwide, with an estimated prevalence of $24\%$5.
Lifestyle modification such as weight reduction through a hypocaloric diet and exercise serves as the basis for the treatment of NAFLD in the absence of pharmacological agents1,6,7. The majority of studies suggesting a favorable benefit have focused on recreational physical activity, leaving the kind, intensity, and duration of physical activity necessary for optimal therapeutic outcomes in the management of NAFLD unclear7,8. The World Health Organization (WHO) 2020 guideline has stated that moderate-to-vigorous physical activity in any domain is beneficial. Until now, evidence to conclude whether health benefits of physical activity vary by type or domain is insufficient9. In addition, the WHO 2020 guideline recommends muscle strength exercise more than twice a week for all adults9. However, recent studies have reported contradictory effects of work-related physical activity on health, such as HOMA-IR10. Diabetes11, blood pressure12, coronary heart disease13,14, and cardiovascular mortality14–16. Likewise, growing evidence indicates that work-related physical activity is not protective against NAFLD17,18, prompting further research to investigate whether NAFLD is influenced differently by the domain of physical activity. Also, how muscle strength exercise interacts with physical activity to affect NAFLD remains unclear.
Therefore, the objective of this study was to investigate associations of different physical activity domains and muscle strength exercise with NAFLD using the Korean National Health and Nutrition Examination Survey (KNHANES).
## Study design, setting, and participants
The KNHANES is a nation-wide surveillance system to monitor the health and nutritional status of the general population of South Korea19. Each year, representative samples of approximately 10,000 people are selected. Health examination, health interview, and nutritional survey are then conducted.
We screened a total of 28,194 adult men or women aged 20–79 years who participated in the KNHANES from January 2014 to December 2018. Among them, we excluded 4446 participants who met the following exclusion criteria to include participants without chronic viral hepatitis, liver cirrhosis, heavy alcohol use, or malignancy: [1] chronic hepatitis B ($$n = 893$$, determined by the presence of hepatitis B surface antigen); [2] chronic hepatitis C ($$n = 73$$, determined by the presence of hepatitis C virus RNA test or history of chronic hepatitis C); [3] liver cirrhosis ($$n = 42$$, determined by a history of liver cirrhosis); 4) heavy alcohol intake ($$n = 2096$$, 30 g or more for a day for men and 20 g or more for a day for women)6; [5] history of malignancy ($$n = 1200$$); and [6] pregnant women ($$n = 142$$). Of these participants, we further excluded 2723 participants missing key variables for assessing NAFLD [$$n = 1714$$: missing values for alanine aminotransferase ($$n = 1126$$), heights ($$n = 44$$), body weights ($$n = 2$$), and alcohol intake ($$n = 542$$)] or missing key information on physical activity ($$n = 1019$$). Finally, a total of 21,025 participants were analyzed (Fig. 1). The survey was conducted after receiving written informed consent from all study participants. The study protocol was reviewed and approved by the Institutional Review Board of the Korea Disease Control and Prevention Agency (No: 2013-12EXP-03-5C, 2018-01-03-P-A) and the Samsung Medical Center (No: 2021-01-013). The study was performed in accordance with the Declaration of Helsinki. Figure 1Flowchart showing the selection of study participants.
## Study outcomes, variables, and measurements
The diagnosis of NAFLD was made using hepatic steatosis index (HSI)20. HSI consists of aspartate aminotransferase, alanine aminotransferase, sex, body mass index (BMI), and diabetes mellitus. Participants with HSI of 36 or higher were considered to have NAFLD. The health interview including physical activity was conducted by trained surveyors consisting of nurses and epidemiologists. To gather comprehensive physical activity information, the Global Physical Activity Questionnaire (GPAQ) and frequency of muscle strength exercise were collected. The level of physical activity was interviewed using the Korean version of GPAQ21. The GPAQ was originally developed by WHO to monitor physical activity in numerous countries. It is grouped into three domains of physical activity: recreation, travel, and work activities. The recreation domain includes sports, fitness, and leisure activities. The travel domain includes transport to and from places. The work domain has paid or unpaid work, study/training, household chores, harvesting food/crops, fishing or hunting for food, and seeking employment. The GPAQ provides information on the frequency (times in a week) and duration (minutes at a day) of each domain of physical activity. In recreation and work domains, the intensity of physical activity was also provided (moderate or vigorous). Results were analyzed as suggested by the WHO: [1] the duration of vigorous physical activity was doubled and added to the duration of moderate physical activity, [2] three domains of physical activity were summed to calculate the duration of total physical activity. Since the WHO guideline states that all adults should do at least 150 min/week of moderate-intensity aerobic physical activity, total and each domain of physical activity were divided into < 150 min/week and ≥ 150 min/week. To investigate additional benefits of ≥ 300 min/week of physical activity, we performed further categorization: 0 min/week, 1–149 min/week, 150–299 min/week, and ≥ 300 min/week. The frequency of muscle strength exercise was determined by the number of muscle strength exercise in a week. Since the WHO 2020 guideline recommends that all adults should perform muscle strength exercise at least twice a week9, muscle strength exercise was categorized as < 2/week and ≥ 2/week.
Variables collected were age, sex, height, body weight, waist circumference, triglyceride, high density cholesterol, systolic blood pressure, diastolic blood pressure, fasting serum glucose, hepatitis B surface antigen, anti-HCV antibody, history of chronic hepatitis C, history of malignancy, history of liver cirrhosis, use of antihypertensive medications, antidiabetic medications, antidyslipidemic medications, alcohol use behavior, smoking status, pregnancy, household income information, and education level. BMI was calculated as weight in kilograms/height in square meters (kg/m2). Household income information was classified into quartiles: low, low-intermediate, intermediate-high, and high. Education levels were stratified into four categories: elementary school or lower, middle school, high school, college or higher. Alcohol intake was categorized into < 10 g/day and ≥ 10 g/day. Metabolic syndrome was defined for participants having three or more of the followings: [1] elevated waist circumference: ≥ 90 cm for men and ≥ 85 cm for women, [2] elevated triglycerides: ≥ 150 mg/dL or use of antidyslipidemic medications, [3] low high-density HDL-C: < 40 mg/dL for men,< 50 mg/dL for women, [4] elevated blood pressure: ≥ $\frac{130}{85}$ mmHg or use of blood pressure lowering agents, [5] elevated fasting glucose: ≥ 100 mg/dL or on treatment for elevated glucose.
## Statistical analysis
Descriptive statistics for continuous variables are presented as median and interquartile range (IQR). Categorical variables are presented as numbers and proportions (%). Comparison of variables between groups was performed using Student’s t-test or Wilcox rank-sum test for continuous variables and Chi-square test for categorical variables. Generalized logistic regression was performed to determine whether the prevalence of NAFLD was different depending on physical activity after adjusting for potential confounding or mediating factors. When adjusting for age and sex, we used age per year as a continuous variable. In the fully adjusted model, we further adjusted for BMI (continuous), metabolic syndrome (yes vs. no), income levels (low, low-intermediate, intermediate-high, high), education levels (elementary or lower, middle school, high school, college or higher), smoking (current, ex-smoker, and never smoker), alcohol intake (< 10 g/day vs. ≥ 10 g/day), total physical activity (< 150 min/week vs. ≥ 150 min/week), and muscle strength exercise (< 2/week vs. ≥ 2/week). When specific domains of physical activity were assessed, other specific domains were adjusted. For recreation, travel (< 150 min/week vs. ≥ 150 min/week) and work activity (< 150 min/week vs. ≥ 150 min/week) were adjusted. For travel, recreation (< 150 min/week vs. ≥ 150 min/week) and work activity (< 150 min/week vs. ≥ 150 min/week) were adjusted. For work, recreation (< 150 min/week vs. ≥ 150 min/week) and travel activity (< 150 min/week vs. ≥ 150 min/week) were adjusted.
Subgroup analysis was performed to evaluate the relationship between physical activity or muscle strength exercise and NAFLD within each subgroup. Subgroups were predefined as follows: by age (< 65 years vs. ≥ 65 years), sex (male vs. female), BMI (< 25 kg/m2 vs. ≥ 25 kg/m2), metabolic syndrome (yes vs. no), muscle strength exercise (< 2/week vs. ≥ 2/week), and total physical activity (< 150 min/week vs. ≥ 150 min/week). All variables with a p value < 0.05 were considered statistically significant. All statistical analyses were performed using R version 3.6.3 (The R Foundation for Statistical Computing, Vienna, Austria).
## Results
Baseline characteristics of study participants are summarized in Table 1. Among 21,015 participants, 4942 ($23.5\%$) had NAFLD. Participants with NAFLD were more likely to be older, male, current/ex-smokers, metabolically unhealthy, and have lower income, lower education, higher BMI than those without NAFLD. Participants with NAFLD also consisted of more participants who did not perform WHO recommended level (≥ 150 min/week) of physical activity and more participants who did not perform WHO recommended level (≥ 2/week) of muscle strength exercise (Table 2). When specific domain of physical activity was assessed, NAFLD participants were more likely to be inactive in recreation and travel domain activities, but not in work domain activity. Table 1Baseline characteristics of study participants by non-alcoholic fatty liver disease status ($$n = 21$$,015).Overall ($$n = 21$$,015)NAFLD (+) ($$n = 4942$$)NAFLD (−) ($$n = 16$$,073)p valueAge (year)51 (38–63)52 (39–63)50 (37–63)< 0.001Male8659 (41.2)2354 (47.6)6305 (39.2)< 0.001Income† Low4971 (23.7)1283 (26.0)3688 (22.9)< 0.001 Low-intermediate4963 (23.6)1164 (23.6)3799 (23.6) Intermediate-high5343 (25.4)1294 (26.2)4049 (25.2) High5686 (27.1)1194 (24.2)4492 (27.9)Education† Elementary or lower4043 (19.2)1151 (23.3)2892 (18.0)< 0.001 Middle school2159 (10.3)551 (11.1)1608 (10.0) High school6800 (32.4)1590 (32.2)5210 (32.4) College or higher7987 (38.0)1640 (33.2)6347 (39.5)Occupation†13,068 (62.2)3056 (61.8)10,012 (62.3)0.3Smoking Current smoker3468 (16.5)973 (19.7)2495 (15.5)< 0.001 Ex-smoker4174 (19.9)1025 (20.7)3149 (19.6) Never smoker13,373 (63.6)2944 (59.6)10,429 (64.9)Alcohol consumption < 10 g/day13,025 (62.0)3137 (63.5)9888 (61.5)0.014 ≥ 10 g/day7990 (38.0)1805 (36.5)6185 (38.5)Body mass index (kg/m2)23.6 (21.4–25.9)27.6 (26.0–29.6)22.6 (20.8–24.4)< 0.001Metabolic syndrome*†6359 (30.3)3150 (63.7)3209 (20.0)< 0.001 Elevated waist circumference5928 (28.2)3618 (73.2)2310 (14.4)< 0.001 Elevated triglycerides7571 (36.0)2889 (58.4)4682 (29.1)< 0.001 Reduced HDL-C7178 (34.2)2402 (48.6)4776 (29.7)< 0.001 Elevated blood pressure8131 (38.7)2795 (56.6)5336 (33.2)< 0.001 Elevated fasting glucose7309 (34.8)2825 (57.2)4484 (27.9)< 0.001Values are expressed as number (%) or median (quartile).NAFLD non-alcoholic fatty liver disease, HDL-C high-density lipoprotein.*Metabolic syndrome was defined when any three of five risk factors were present: elevated waist circumference: ≥ 90 cm for men, ≥ 85 cm for women; elevated triglycerides: ≥ 150 mg/dL or use of antidyslipidemic medications; reduced HDL-C: < 40 mg/dL for men, < 50 mg/dL for women; elevated blood pressure: ≥ $\frac{130}{85}$ mmHg or use of antihypertensive medications; elevated fasting glucose: ≥ 100 mg/dL or use of antidiabetic medications.†These variables had missing value. Number of participants with missing value were as follow: income ($$n = 52$$), education ($$n = 26$$), occupation ($$n = 12$$), metabolic syndrome ($$n = 64$$).Table 2Risk of non-alcoholic fatty liver disease according to the level of physical activity and muscle strength exercise. No. of subjectsNAFLD (%)p valueAge and sex adjusted OR ($95\%$ CI)Fully adjusted OR ($95\%$CI)Total physical activity (min/week)< 0.001 < 15011,19524.8RefRef ≥ 150982022.10.86 (0.81–0.92)0.86 (0.78–0.95)Domains of physical activityRecreation (min/week)< 0.001 < 15017,28924.1RefRef ≥ 150372620.70.79 (0.73–0.87)0.77 (0.67–0.88)Travel (min/week)0.002 < 15014,60824.1RefRef ≥ 150640722.20.91 (0.85–0.98)0.90 (0.81–1.01)Work (min/week)0.3 < 15019,67823.4RefRef ≥ 150133724.81.09 (0.95–1.25)0.90 (0.73–1.10)Muscle strength exercise (/week)< 0.001 < 216,73224.6RefRef ≥ 2428319.40.67 (0.62–0.73)0.83 (0.73–0.94)Fully adjusted model was adjusted for age (continuous), sex, body mass index (continuous), elevated waist circumference (yes vs. no), elevated triglycerides (yes vs. no), reduced high-density lipoprotein cholesterol (yes vs. no), diabetes mellitus (yes vs. no), hypertension (yes vs. no), income levels (Q1, Q2, Q3, and Q4), education (elementary or lower, middle school, high school, college or higher), smoking (current, ex-smoker, and never smoker), alcohol consumption (< 10 g/day vs. ≥ 10 g/day), total physical activity (< 150 min/week vs. ≥ 150 min/week), and muscle strength exercise (< 2/week vs. ≥ 2/week). For specific domains of physical activity, other domains were adjusted as follows: Recreation: travel and work activity (< 150 min/week vs. ≥ 150 min/week), Travel: recreation and work activity (< 150 min/week vs. ≥ 150 min/week), Work: recreation and travel activity (< 150 min/week vs. ≥ 150 min/week).NAFLD non-alcoholic fatty liver disease, OR odds ratio, CI confidence interval.
NAFLD prevalence was the highest among participants with 0 min/week of total physical activity. It was the lowest among those with ≥ 300 min/week of total physical activity (Fig. 2A). For the specific physical activity domain, NAFLD prevalence was the highest among participants with 0 min/week of recreation activity. It showed a dose-dependent decrease with an increase in recreation activity time. NAFLD prevalence was the highest among participants with 0 min/week of travel activity, showing no dose-dependent decrease with an increase in travel activity time. NAFLD prevalence was not different by work activity time. NAFLD was more prevalent in participants with 0 times of muscle strength exercise/week than in those with 1 time or ≥ 2 times of muscle strength exercise/week (Fig. 2B).Figure 2Prevalence of non-alcoholic fatty liver disease according to the level of (A) physical activity (total, recreation, travel, work) and (B) muscle strength exercise.
The fully adjusted odds ratio (OR) for NAFLD was 0.86 [$95\%$ confidence interval (CI): 0.78–0.95] when participants with ≥ 150 min/week of total physical activity were compared to those with < 150 min/week (Table 2). When the domain of physical activity was assessed separately, ≥ 150 min/week of recreation activity was negatively associated with the presence of NAFLD (OR: 0.77, $95\%$ CI 0.67–0.88), whereas ≥ 150 min/week of travel activity and ≥ 150 min/week of work activity were not. The fully adjusted OR for NAFLD comparing participants with ≥ 2/week to those with < 2/week of muscle strength exercise was 0.83 ($95\%$ CI 0.73–0.94). When the level of physical activity was subdivided further, using 0 min/week activity as a reference group, ≥ 300 min/week of total physical activity, ≥ 300 min/week of recreation activity, ≥ 300 min/week of travel activity, 150–299 min/week of work activity, and ≥ 2/week of muscle strength exercise were negatively associated with the presence of NAFLD (Supplementary Table S1). The inverse association between the level of physical activity and NAFLD was dose-dependent for recreation and travel activities.
NAFLD prevalence was higher for participants with < 2/week of muscle strength exercise than those with ≥ 2/week for all levels of total and each specific domains of physical activity except for ≥ 150 min/week of work activity (Fig. 3).Figure 3Prevalence of non-alcoholic fatty liver disease according to the level of muscle strength exercise (< 2/week and ≥ 2/week) in each physical activity group.
In subgroup analysis, the association between physical activity or muscle strength exercise and NAFLD had no interaction in all predefined subgroups (Fig. 4).Figure 4Odds ratio for the risk of NAFLD by the level of physical activity (< 150 min/week vs. ≥ 150 min/week) or muscle strength exercise (< 2/week vs. ≥ 2/week) in predefined subgroups. Models were adjusted for age (continuous), sex, body mass index (continuous), elevated waist circumference (yes vs. no), elevated triglycerides (yes vs. no), reduced high-density lipoprotein cholesterol (yes vs. no), diabetes mellitus (yes vs. no), hypertension (yes vs. no), income levels (Q1, Q2, Q3, and Q4), education (elementary or lower, middle school, high school, college or higher), smoking (current, ex-smoker, and never smoker), alcohol consumption (< 10 g/day vs. 10 g/day), total physical activity (< 150 min/week vs. ≥ 150 min/week), and muscle strength exercise (< 2/week vs. ≥ 2/week). For specific domains of physical activity, other domains were adjusted as follows: [1] For recreation activity, travel and work activities (< 150 vs. ≥ 150 min/week) were adjusted; [2] For travel activity, recreation and work activities (< 150 vs. ≥ 150 min/week) were adjusted; and [3] for work activity, recreation and travel activities (< 150 vs. ≥ 150 min/week) were adjusted.
## Discussion
In this nation-wide cross-sectional study, we found that WHO-recommended levels of total physical activity (≥ 150 min/week) and muscle strength exercise (≥ 2/week) were associated with a lower risk of NAFLD. When individual domains of physical activity were assessed, recreation activity (≥ 150 min/week), but not travel or work activity, was found to have a significant association with NAFLD. When the level of physical activity was subdivided further, ≥ 300 min/week of total physical activity, recreation activity, or travel activity, and 150–299 min/week of work activity were negatively associated with NAFLD. Muscle strength exercise ≥ 2/week showed a lower risk of NAFLD for all levels of total and each specific domains of physical activity except for ≥ 150 min/week of work activity. Subgroup analysis showed significant associations of NAFLD with total physical activity (≥ 150 min/week), recreation activity (≥ 150 min/week), and muscle strength exercise (≥ 2/week) in all predefined subgroups. These findings indicate that moderate-to-vigorous physical activity can reduce the risk of NAFLD. However, the impact varies by physical activity domains. Muscle strength exercise can also reduce the risk of NAFLD in most cases but except for ≥ 150 min/week of work activity.
In the present study, we demonstrated that the risk of NAFLD varied depending on the domains of physical activity. So far, most studies revealing the relationship between physical activity and NAFLD have focused on exercise intensity and recreation domain physical activity22–25. Only a few studies have examined effects of domains of physical activity on NAFLD17,18. In a cross-sectional study using NHANES, which assessed physical activity using GPAQ, work domain activity did not appear to be protective against NAFLD17. A population-based cohort study with 42,661 participants also showed that moderate to vigorous physical activity in the work domain had no discernible benefit on NAFLD18. In the present study, we observed that ≥ 150 min/week of recreation domain activity, ≥ 300 min/week of travel domain activity, and 150–299 min/week of work domain activity were associated with a reduced risk of NAFLD. Thus, in contrast to recreation or travel domain activity, vigorous level (≥ 300 min/week) of work domain activity might not be protective against NAFLD. Although the exact biological mechanism for these findings is unclear, the ‘physical activity paradox14,26, might partly explain it. At vigorous level, work physical activity is either excessively strenuous or excessively prolonged to be cardiorespiratory beneficial, resulting in persistently elevated blood pressure and heart rate. Furthermore, repeated actions without appropriate recuperation time increased the level of inflammation14,27. In addition, it has been shown that physical activity triggers beta-oxidation to promote damaged mitochondrial clearance28, but the anti-oxidative impact does not increase during intense occupational physical activity. 29 These findings indicate that physical activity should be differentiated by domains for health promotion effect. Especially, work activity should not be simply regarded as a substitute for recreation activity or as a measure of health-enhancing daily-life physical activity.
We also demonstrated that participants with ≥ 2/week of muscle strength exercise had a $17\%$ lower risk of having NAFLD than those with < 2/week. Additionally, muscle strength exercise ≥ 2/week was associated with a reduced risk of NAFLD for all levels of total and each specific domains of physical activity except for ≥ 150 min/week of work activity. A previous randomized controlled trial comparing effects of resistance and aerobic training on NAFLD showed that both types of exercise were equally beneficial30. Muscle strength exercise might alter muscle properties by increasing glycolysis and decreasing insulin resistance, hence lowering hepatic steatosis31. According to a systematic review, muscle strength exercise consumes less energy than aerobic exercise to decrease steatosis32. Thus, muscle strength exercise could be a good option for individuals who lack motivation for aerobic exercise or who have limited cardiovascular fitness to do moderate-to-vigorous physical activity.
When interpreting our findings, certain restrictions must be acknowledged. Since it was a cross-sectional study, we could not infer the causal relationship between physical activity or muscle strength exercise and NAFLD. Furthermore, we used HSI to define NAFLD rather than biopsy or abdominal imaging, which might have resulted in a classification bias. Because data on physical activity were acquired based on remembrance, there might be a recall bias, which could lead to misclassification. However, since the health interview was conducted with the assistance of a trained surveyor, we were able to reduce nonresponse bias and measurement bias more effectively than other survey methods33. In addition, since oxidative stress or circulating endotoxin were not included in the KNHANES dataset, the pathophysiological mechanisms underlying the risk of NAFLD in relation to specific physical activity could not be investigated. Despite the aforementioned limitations, our study has a number of advantages. We utilized nation-wide representative data, which might have reduced selection bias. Also, we used a validated GPAQ collected by trained personnel to gather information on physical activity. Finally, with a large sample size, we could perform subgroup analyses with multiple risk factors adjustment.
In summary, increased physical activity and muscle strength exercise were associated with a reduced risk of NAFLD. However, the impact varied by the domain of physical activity. Moderate to vigorous recreation activity and vigorous travel activity were associated with reduced risk of NAFLD. However, work activity was associated with reduced risk of NAFLD only at a moderate level. Muscle strength exercise was associated with a reduced risk of NAFLD in most cases. These findings suggest that physical activity should be differentiated by domains for the management of NAFLD and that vigorous level of work activity might need to be avoided as a measure of health-enhancing daily-life physical activity. Also, muscle strength exercise could be a good option for individuals who could not perform moderate-to-vigorous physical activity.
## Supplementary Information
Supplementary Table S1. The online version contains supplementary material available at 10.1038/s41598-023-31686-6.
## References
1. **KASL clinical practice guidelines: Management of nonalcoholic fatty liver disease**. *Clin. Mol. Hepatol.* (2013) **19** 325-348. DOI: 10.3350/cmh.2013.19.4.325
2. Sinn DH. **Lean non-alcoholic fatty liver disease and development of diabetes: A cohort study**. *Eur. J. Endocrinol.* (2019) **181** 185-192. DOI: 10.1530/EJE-19-0143
3. Yki-Jarvinen H. **Non-alcoholic fatty liver disease as a cause and a consequence of metabolic syndrome**. *Lancet Diabetes Endocrinol.* (2014) **2** 901-910. DOI: 10.1016/S2213-8587(14)70032-4
4. Targher G, Day CP, Bonora E. **Risk of cardiovascular disease in patients with nonalcoholic fatty liver disease**. *N. Engl. J. Med.* (2010) **363** 1341-1350. DOI: 10.1056/NEJMra0912063
5. Younossi Z. **Global burden of NAFLD and NASH: Trends, predictions, risk factors and prevention**. *Nat. Rev. Gastroenterol. Hepatol.* (2018) **15** 11-20. DOI: 10.1038/nrgastro.2017.109
6. **EASL-EASD-EASO clinical practice guidelines for the management of non-alcoholic fatty liver disease**. *Obes. Facts* (2016) **9** 65-90. DOI: 10.1159/000443344
7. Chalasani N. **The diagnosis and management of nonalcoholic fatty liver disease: Practice guidance from the American association for the study of liver diseases**. *Hepatology* (2018) **67** 328-357. DOI: 10.1002/hep.29367
8. Hallsworth K, Adams LA. **Lifestyle modification in NAFLD/NASH: Facts and figures**. *JHEP Rep.* (2019) **1** 468-479. DOI: 10.1016/j.jhepr.2019.10.008
9. Bull FC. **World Health Organization 2020 guidelines on physical activity and sedentary behaviour**. *Br. J. Sports Med.* (2020) **54** 1451-1462. DOI: 10.1136/bjsports-2020-102955
10. Tsenkova VK. **Leisure-time, occupational, household physical activity and insulin resistance (HOMAIR) in the Midlife in the United States (MIDUS) national study of adults**. *Prev. Med. Rep.* (2017) **5** 224-227. DOI: 10.1016/j.pmedr.2016.12.025
11. Lee EB. **Association between domain-specific physical activity and diabetes in Korean adults**. *Sci. Rep.* (2021) **11** 13066. DOI: 10.1038/s41598-021-92560-x
12. Byambasukh O, Snieder H, Corpeleijn E. **Relation between leisure time, commuting, and occupational physical activity with blood pressure in 125 402 adults: The lifelines cohort**. *J. Am. Heart Assoc.* (2020) **9** e014313. DOI: 10.1161/JAHA.119.014313
13. Hu G. **Occupational, commuting and leisure-time physical activity in relation to coronary heart disease among middle-aged Finnish men and women**. *Atherosclerosis* (2007) **194** 490-497. DOI: 10.1016/j.atherosclerosis.2006.08.051
14. Holtermann A, Krause N, van der Beek AJ, Straker L. **The physical activity paradox: Six reasons why occupational physical activity (OPA) does not confer the cardiovascular health benefits that leisure time physical activity does**. *Br. J. Sports Med.* (2018) **52** 149-150. DOI: 10.1136/bjsports-2017-097965
15. Holtermann A, Schnohr P, Nordestgaard BG, Marott JL. **The physical activity paradox in cardiovascular disease and all-cause mortality: The contemporary Copenhagen general population study with 104 046 adults**. *Eur. Heart J.* (2021) **42** 1499-1511. DOI: 10.1093/eurheartj/ehab087
16. Holtermann A. **Occupational and leisure time physical activity: Risk of all-cause mortality and myocardial infarction in the Copenhagen City Heart Study. A prospective cohort study**. *BMJ Open* (2012) **2** e000556. DOI: 10.1136/bmjopen-2011-000556
17. Kim D, Vazquez-Montesino LM, Li AA, Cholankeril G, Ahmed A. **Inadequate physical activity and sedentary behavior are independent predictors of nonalcoholic fatty liver disease**. *Hepatology* (2020) **72** 1556-1568. DOI: 10.1002/hep.31158
18. Byambasukh O, Zelle D, Corpeleijn E. **Physical activity, fatty liver, and glucose metabolism over the life course: The lifelines cohort**. *Am. J. Gastroenterol.* (2019) **114** 907-915. DOI: 10.14309/ajg.0000000000000168
19. Kweon S. **Data resource profile: The Korea National Health and Nutrition Examination Survey (KNHANES)**. *Int. J. Epidemiol.* (2014) **43** 69-77. DOI: 10.1093/ije/dyt228
20. Lee JH. **Hepatic steatosis index: A simple screening tool reflecting nonalcoholic fatty liver disease**. *Dig. Liver Dis.* (2010) **42** 503-508. DOI: 10.1016/j.dld.2009.08.002
21. Lee J. **Development of the Korean Global Physical Activity Questionnaire: Reliability and validity study**. *Glob. Health Promot.* (2020) **27** 44-55. DOI: 10.1177/1757975919854301
22. Kwak MS. **Role of physical activity in nonalcoholic fatty liver disease in terms of visceral obesity and insulin resistance**. *Liver Int.* (2015) **35** 944-952. DOI: 10.1111/liv.12552
23. Zelber-Sagi S. **Role of leisure-time physical activity in nonalcoholic fatty liver disease: A population-based study**. *Hepatology* (2008) **48** 1791-1798. DOI: 10.1002/hep.22525
24. Ueno T. **Therapeutic effects of restricted diet and exercise in obese patients with fatty liver**. *J. Hepatol.* (1997) **27** 103-107. DOI: 10.1016/s0168-8278(97)80287-5
25. Keating SE. **Effect of aerobic exercise training dose on liver fat and visceral adiposity**. *J. Hepatol.* (2015) **63** 174-182. DOI: 10.1016/j.jhep.2015.02.022
26. Holtermann A, Hansen JV, Burr H, Sogaard K, Sjogaard G. **The health paradox of occupational and leisure-time physical activity**. *Br. J. Sports Med.* (2012) **46** 291-295. DOI: 10.1136/bjsm.2010.079582
27. Lee J. **Occupational physical activity, not leisure-time physical activity, is associated with increased high-sensitivity C reactive protein levels**. *Occup. Environ. Med.* (2021) **78** 86-91. DOI: 10.1136/oemed-2020-106753
28. Chun SK, Lee S, Yang M-J, Leeuwenburgh C, Kim J-S. **Exercise-induced autophagy in fatty liver disease**. *Exerc. Sport Sci. Rev.* (2017) **45** 181-186. DOI: 10.1249/jes.0000000000000116
29. Ma J, Liu Z, Ling W. **Physical activity, diet and cardiovascular disease risks in Chinese women**. *Public Health Nutr.* (2007) **6** 139-146. DOI: 10.1079/phn2002393
30. Bacchi E. **Both resistance training and aerobic training reduce hepatic fat content in type 2 diabetic subjects with nonalcoholic fatty liver disease (the RAED2 Randomized Trial)**. *Hepatology* (2013) **58** 1287-1295. DOI: 10.1002/hep.26393
31. Hallsworth K. **Resistance exercise reduces liver fat and its mediators in non-alcoholic fatty liver disease independent of weight loss**. *Gut* (2011) **60** 1278-1283. DOI: 10.1136/gut.2011.242073
32. Hashida R. **Aerobic vs. resistance exercise in non-alcoholic fatty liver disease: A systematic review**. *J. Hepatol.* (2017) **66** 142-152. DOI: 10.1016/j.jhep.2016.08.023
33. Christensen AI, Ekholm O, Glumer C, Juel K. **Effect of survey mode on response patterns: Comparison of face-to-face and self-administered modes in health surveys**. *Eur. J. Public Health* (2014) **24** 327-332. DOI: 10.1093/eurpub/ckt067
|
---
title: Integrative transcriptomic and metabolomic analysis reveals alterations in
energy metabolism and mitochondrial functionality in broiler chickens with wooden
breast
authors:
- Ziqing Wang
- Erin Brannick
- Behnam Abasht
journal: Scientific Reports
year: 2023
pmcid: PMC10036619
doi: 10.1038/s41598-023-31429-7
license: CC BY 4.0
---
# Integrative transcriptomic and metabolomic analysis reveals alterations in energy metabolism and mitochondrial functionality in broiler chickens with wooden breast
## Abstract
This integrative study of transcriptomics and metabolomics aimed to improve our understanding of Wooden Breast myopathy (WB). Breast muscle samples from 8 WB affected and 8 unaffected male broiler chickens of 47 days of age were harvested for metabolite profiling. Among these 16 samples, 5 affected and 6 unaffected also underwent gene expression profiling. The Joint Pathway Analysis was applied on 119 metabolites and 3444 genes exhibiting differential abundance or expression between WB affected and unaffected chickens. Mitochondrial dysfunctions in WB was suggested by higher levels of monoacylglycerols and down-regulated genes involved in lipid production, fatty acid beta oxidation, and oxidative phosphorylation. Lower levels of carnosine and anserine, along with down-regulated carnosine synthase 1 suggested decreased carnosine synthesis and hence impaired antioxidant capacity in WB. Additionally, Weighted Gene Co-expression Network Analysis results indicated that abundance of inosine monophosphate, significantly lower in WB muscle, was correlated with mRNA expression levels of numerous genes related to focal adhesion, extracellular matrix and intercellular signaling, implying its function in connecting and possibly regulating multiple key biological pathways. Overall, this study showed not only the consistency between transcript and metabolite profiles, but also the potential in gaining further insights from analyzing multi-omics data.
## Introduction
Concurrent with the rapid growth and high muscle yield, modern broilers are prone to muscle diseases that can negatively affect the well-being of the birds as well as meat quality, such as Wooden Breast myopathy (WB)1,2. WB is characterized by pale focally or diffused stiffened areas from cranial to caudal regions on the pectoralis major (p. major) muscle with or without petechial hemorrhage2–4. WB is mostly subclinical and asymptomatic, but it can greatly deteriorate muscle health and meat quality, causing a significant negative economic impact to the poultry industry2,5.
Although the etiology of this disease is still under investigation, previous research has established association between certain factors and WB. Two main conditions are localized hypoxia and oxidative stress, as suggested by RNA-sequencing and metabolomic studies6–8. Oxidative stress is initiated by excess of reactive oxygen species (ROS), which induce long-term cellular damage and impair muscle contractility6. The high growth rate of commercial broilers is achieved by allocating more energy towards anabolic growth rather than other key metabolic processes, which could eventually have a negative effect on energy-demanding activities such as muscle contraction9. Specifically, altered energy homeostasis and partitioning in broilers may exhibit as dysregulated metabolism of energy substrates including lipid and glucose7, resulting in mitochondrial damage and ROS buildup. Various studies have observed susceptibility to WB in fast-growing chickens7,8,10, suggesting the importance of energy metabolism to the disease development.
Incorporation of different levels of biological variation has great potential in improving our understanding of mechanisms behind complex traits and diseases. Banerjee et al. applied integrated metabolomic and RNA-seq analysis to identify novel gene-metabolite pairs related to feed efficiency in pigs11. A similar approach has also been applied to identify biomarkers for type II diabetes, as well as gaining insights into pathophysiology of metabolic diseases12.
This study aimed to advance current understanding of biological mechanisms behind the WB pathology with the long-term goal of mitigating the disease. Previously, WB has been studied using one omics-technique at a time7–9,13–15. However, combining both transcriptomics and metabolomics allows us to link gene expression to its metabolic products, showing a bigger picture by studying multiple layers of information. Our integrated metabolomic and RNA-seq analyses revealed that WB is associated with alterations in amino acid and energy metabolism, as well as mitochondrial functionality. These changes in metabolism combined with impaired antioxidant capacity have possibly contributed to adverse myofibrillar changes and oxidative stress in WB. Additionally, our weighted gene-co-expression network analysis (WGCNA) identified the focal adhesion pathway as a strong driving factor for WB status distinction. To the best of our knowledge, there has been no published work on WB combining these two omics data.
## Chicken experiment and tissue collection
This research used publicly available data from our laboratory6,7, and no new animal experiment was conducted for the current study. However, for the benefit of readers, we provided a description of the animal experiment. Chickens in this study were all males and sampled real time during necropsy of about 300 birds from a commercial broiler line (referred to as Line 2 in the previous study from our laboratory7) with high breast muscle yield at Heritage Breeders (Princess Anne, MD), where birds were fed ad libitum and housed under optimal industry growing standards. At 47 days of age, the p. major muscle of live chickens was clinically examined by manual palpation. Specifically, chickens exhibiting severe or moderate p. major muscle firmness were classified as affected with WB, whereas those without any palpable signs of firmness were classified as unaffected. Birds were euthanized by cervical dislocation, and the p. major muscle from each bird was visually observed for macroscopic gross lesions such as areas of hemorrhage, firm and discolored muscle tissue at necropsy to confirm the WB affected or unaffected classification. After gross examination, roughly 1–2 g of muscle tissue were harvested from the caudal aspect of the right p. major muscle from 8 WB affected and 8 unaffected birds with comparable breast muscle weight, immediately frozen in liquid nitrogen, and stored at − 80 °C until further processing. For tissue processing, the frozen breast muscle samples were pulverized within plastic bags by hammering. Two sub-samples from the same pulverized tissue were taken, one for RNA-seq and the other for metabolomic analysis. Tissues were maintained in the frozen state during handling and processing. All methods were carried out in accordance with relevant guidelines and regulations. The animal protocol (#44 12-15-13R) for this experiment was approved by the University of Delaware Agricultural Animal Care and Use Committee.
## RNA-seq samples
Out of 16 total samples, 5 affected and 6 unaffected ones were processed for RNA sequencing as described previously by Mutryn et al.6. Obtained RNA-seq data are available at the Sequence Read *Archive via* accession number NCBI-SRA: SRP224368. Downstream analysis of RNA-seq data for this study was performed in the Biomix High Performance Computing Cluster16 at the Delaware Biotechnology Institute, University of Delaware. Raw sequence reads underwent quality check using FastQC v0.11.917 and were mapped to the current chicken reference genome Gallus_gallus-6a (Ensembl, database version 99) using Hisat2 v2.2.018, followed by HTSeq v0.11.219 to categorize the mapped reads using default parameters.
## Metabolomics analysis
Since variance in metabolomics data was expected to be larger than that in RNA-seq data, all 16 breast muscle samples, 8 affected and 8 unaffected, were used for metabolomics profiling by Metabolon Inc. (Durham, NC), as described in prior publication from our laboratory7. Briefly, Metabolon utilized liquid chromatography mass spectrometry and a large, well-annotated spectral library to identify the metabolites20. Metabolites with missing value across more than $80\%$ of samples were excluded for future analysis since missing values in metabolite data are often not random (Figure S1) but fell below the detection limit threshold21. Subsequently, the data were log2 transformed, standardized, and missing values were replaced by the minimum value observed for each metabolite. These data processing steps were performed in R v3.5.222. Numeric representations of the spectral entries for the metabolites identified in our sample set are provided as a supplementary file in Abasht et al.7.
## Sample consistency
Sample consistency was assessed by the Principal Component Analysis (PCA) without centering or scaling the variables using R stats package (v3.5.2)22. PCA analysis placed the sample C51 which had been clinically and grossly classified as “unaffected” adjacent to the affected samples (Figure S2), suggesting a pattern consistent with WB samples based on both metabolomic and transcriptomic profiles even though this sample exhibited no gross lesions or palpable stiffness. This is consistent with the previous studies where this sample was denoted as a subclinical WB case6,7. For the purpose of improving statistical power and accuracy, status of sample C51 was moved from the unaffected to the affected group for further analysis.
## Statistical analysis
HTSeq count data were filtered for low count genes based on the number of chickens in each group, followed by normalization and differential expression analysis using edgeR v3.24.323. The aforementioned analysis and one-way analysis of variance (ANOVA) of metabolites were performed in R v3.5.222. Differentially expressed genes (DEGs) and differentially abundant metabolites were then submitted to MetaboAnalyst v5.024 for analysis using the Joint Pathway Analysis module, where integrated metabolic pathways consisting of both metabolites and metabolic genes were used as a database. Both tight and loose integration methods were performed in order to be comprehensive. Specifically, genes and metabolites were pooled together in a single query for the tight integration method, whereas for the loose integration method, separate analyses were performed for each list and individual p values were weighted at the pathway level before combining to account for different sizes of input data24.
To exploit novel relationships between genes and metabolites pertinent to WB, weighted gene co-expression network analysis (WGCNA)25 was also performed. This time neither transcriptomic nor metabolomic data were filtered for differential metabolite abundance or differential gene expression. For data processing, transcriptomic data underwent filtering low count genes and by interquartile range, while metabolomic data only underwent missing value filtering and imputation by minimum value. Then, the two datasets were merged at the sample level followed by variance stabilization through log2(x + 1) transformation and standardization. WGCNA was performed in R v3.5.222 using package WGCNA v1.70-326. A soft threshold of 9 was selected based on the recommendation for small sample size from developers27. To consider both positive and negative correlations between features, unsigned networks were constructed with a minimum module size of 20 and module eigengene dissimilarity threshold of 0.15. To visualize the relationships among modules and WB, an eigengene network was constructed in R v3.5.222 using package WGCNA v1.70-326. Candidate modules were selected based on the module size, significant correlation (FDR < 0.05) with the WB status and inclusion of both genes and metabolites. Hub features in a module were identified as those with the highest module membership (MM) and gene significance (GS) values. The study was carried out in compliance with the ARRIVE guidelines.
## Joint pathway analysis
In total, 3444 genes were determined to be differentially expressed between WB-affected and unaffected birds with a fold-change (FC) greater than 1.3 and false discovery rate (FDR) threshold smaller than 0.05. Specifically, 1875 were up-regulated and 1569 were down-regulated in WB birds. Through one-way ANOVA, 119 metabolites were found differentially abundant between WB affected and unaffected p. major muscles (FDR < 0.05), among which 90 had higher concentration in WB affected samples. A less stringent FDR threshold value (0.1) was applied for joint pathway analysis in order to include more informative pathways. As a result, 7 out of 9 pathways identified by the loose integration method were overlapped with those selected by the tight integration method, suggesting a rather balanced power between transcriptomic and metabolic profiles in identification of significant features (Fig. 1). Notably, many of the identified pathways are related to amino acid, lipid and energy metabolism. The DEGs pertinent to the identified pathways are reported in Table 1.Figure 1Biological pathways identified by Joint Pathway Analysis of differentially expressed genes and differentially abundant metabolites between WB-affected and unaffected chickens. FDR false discovery rate. Table 1Subset of differentially expressed genes in energy metabolism and peroxisome biosynthesis. Gene NamePathwayLog2FCCarnosine synthase 1 (CARNS1)Histidine metabolism↓2.8Aconitase 1 (ACO1)TCA cycle↑0.5Isocitrate dehydrogenase (nad(+)) 3 catalytic subunit (IDH3A, IDH3B)TCA cycle↓0.8, ↓0.7Oxoglutarate dehydrogenase (OGDH)TCA cycle↓0.8Succinate-CoA ligase GDP/ADP-forming subunit alpha (SUCLG1)TCA cycle↓0.8Succinate dehydrogenase complex flavoprotein subunit (SDHA, SDHC)TCA cycle↓0.8, ↓0.5Malate dehydrogenase 2 (MDH2)TCA cycle↓0.6Lactate dehydrogenase (LDHA, LDHB, LDHD)Pyruvate metabolism↓1.2, ↑1.9, ↓1d-Aspartate oxidase (DDO)Ala, Asp, Glu metabolism*↓1.4Glutamic-oxaloacetic transaminase (GOT1,GOT2)Ala, Asp, Glu metabolism↓0.7, ↓0.8Asparagine synthetase (ASNS)Ala, Asp, Glu metabolism↑2.1Aspartoacylase (ASPA)Ala, Asp, Glu metabolism↓1.4N-Acetyltransferase 8 like (NAT8L)Ala, Asp, Glu metabolism↑1.3Ribosomal modification protein rimk like family member (RIMKLB)Ala, Asp, Glu metabolism↑0.7Glutamate-ammonia ligase (GLUL)Ala, Asp, Glu metabolism↓1.5Glutaminase 2 (GLS2)Ala, Asp, Glu metabolism↓1.9CD36 molecule (CD36)Fatty acid metabolism↓0.7Malonyl-CoA decarboxylase (MLYCD)Fatty acid metabolism↓0.7Carnitine palmitoyltransferase (CPT1A,CPT2)Fatty acid metabolism↓0.8, ↑0.6Acyl-CoA synthetase long chain family member 4 (ACSL4)Fatty acid metabolism↓0.5Solute carrier family 16 member 1 (SLC16A1)Fatty acid metabolism↓0.73-Oxoacid CoA-transferase 1 (OXCT1)Fatty acid metabolism↓0.9Uncoupling protein 3 (UCP3)Fatty acid metabolism↓1.8Lipin-1 (LPIN1)Glycerolipid metabolism↓0.5Adipose triglyceride lipase (PNPLA2)Glycerolipid metabolism↓0.6Adiponutrin (PNPLA3)Glycerolipid metabolism↓0.8Diacylglycerol o-acyltransferase 2 (DGAT2)Glycerolipid metabolism↓1.9Acyl-CoA wax alcohol acyltransferase 1 (AWAT1)Glycerolipid metabolism↓1.7Hypoxia inducible factor 1 subunit alpha (HIF1A)Hypoxia↑0.4RXR alpha (RXRA)Hypoxia↓1.4Forkhead box o1 (FOXO1)Hypoxia↓1.3Glycerone-phosphate O-acyltransferase (GNPAT)Ether lipid synthesis↓0.6Acyl-CoA reductase (FAR1)Ether lipid synthesis↓0.6Acyl-CoA oxidase 2 (ACOX2)Fatty acid metabolism↓0.5Phytanoyl-CoA 2-Hydroxylase (PHYH)Fatty acid metabolism↓0.8Peroxisome biogenesis factor 1 (PEX1)Peroxisome biosynthesis↓0.5Peroxisome biogenesis factor 5 (PEX5)Peroxisome biosynthesis↓0.5Peroxisome biogenesis factor 7 (PEX7)Peroxisome biosynthesis↓0.7Peroxisome biogenesis factor 10 (PEX10)Peroxisome biosynthesis↓0.5*Alanine, aspartate and glutamate metabolism.
## Weighted gene co-expression network analysis
Given the scale-free topology index being lower than 0.8 and the high connectivity (Figure S4) between features, the results indicated the existence of an overall strong profile driving the divergence between WB affected and unaffected broilers. In total, WGCNA yielded 19 modules (Fig. 2a), but only the lightyellow module was selected for further analysis based on its positive and significant correlation of 0.9 with WB (Fig. 2b, Fig. S3). Overall, 2842 genes and 172 metabolites were included in this module. Figure 2Weighted gene co-expression network analysis of combined RNA-seq and metabolomics data. a. Hierarchical cluster tree of co-expression modules; b. Eigengene network showing relationships among modules and wooden breast myopathy (WB); c. Scatterplots of gene significance (GS) for WB versus module membership (MM) in the lightyellow module. This figure was generated in R v3.5.2 (https://www.R-project.org/) using package WGCNA v1.70-3 (http://horvath.genetics.ucla.edu/html/CoexpressionNetwork/Rpackages/WGCNA/).
Module membership (MM) is used to quantify the distance between a feature and module eigenvalue based on their correlation, while gene significance (GS) is defined as the correlation between a feature and external traits. Features with both highest MM and GS values were considered as biologically interesting candidates for studying WB27. MM and GS in lightyellow module had a strong positive correlation (Fig. 2c), indicating that those highly connected features in this module were also highly correlated with the WB disease. Therefore, intramodular hub features were selected using a more stringent criteria, with both GS and MM > 0.8, as well as falling in the top $10\%$ ranked connectivity. In total, 301 hub features were identified, among which 295 were genes and 6 were metabolites. Top 10 hub features with the highest connectivity were further distinguished (Table 2). Notably, the sign of MM and GS is consistent with the direction of differential abundance of a feature (i.e., if a feature has a positive sign, it is also significantly more abundant in affected chickens compared with unaffected chickens). Interestingly, these top hub features had a high overlap of 2593 connected genes. Functional analysis on these overlapped genes revealed many signaling pathways including focal adhesion, phagosome, regulation of actin cytoskeleton, ECM-receptor interaction, FoxO signaling pathway, apoptosis and so on. Table 2Module membership and gene significance of the top 10 hub features in lightyellow module of weighted gene co-expression network analysis in wooden breast myopathy (WB) muscle samples. FeatureMMGSBrain Abundant Membrane Attached Signal Protein 1 (BASP1)0.990.92Zyxin (ZYX)0.990.87Actinin Alpha 1 (ACTN1)0.990.89Frizzled Class Receptor 1 (FZD1)0.990.91Complement C1r (C1R)0.990.89Inosine 5′-monophosphate (IMP)− 0.97− 0.89Coenzyme Q9 (COQ9)− 0.96− 0.87Carnosine Synthase 1 (CARNS1)− 0.96− 0.89Glutamic-Oxaloacetic Transaminase 2 (GOT2)− 0.95− 0.87Phosphofructokinase, Muscle (PFKM)− 0.95− 0.90MM module membership, GS gene significance.
## Discussion
In accordance with previous research on WB6,7, closer examination of the pathways obtained above revealed a rather broad range of biological perturbations in the WB p. major muscle, from mitochondrial functionality, energy metabolism to hypoxia and oxidative stress. The following sections discuss these metabolic pathways in play in more detail.
## Histidine metabolism
Histidine is an essential amino acid in chickens as a limiting factor for carnosine and anserine synthesis28. As histidine related compounds, carnosine and anserine are important for their roles in scavenging reactive oxygen species (ROS) and buffering intracellular pH28. In accordance with the previous study using the same dataset7, histidine and related metabolites 1-methyl-histidine and 3-methyl-histidine were significantly higher in WB affected breast muscle, whereas carnosine and anserine were found significantly lower (Fig. 3; Table S5). Our integrative analysis identified a potential mechanism for this decrease, attributed to the down-regulation of carnosine synthase 1 (CARNS1) in WB affected birds. Since CARNS1 synthesizes carnosine and anserine from histidine, 1-methyl-histidine and beta-alanine29,30, decreased carnosine levels could result in impaired antioxidant capacity and hence the widely reported oxidative stress in WB6,7,31. Khumpeerawat and colleagues reported moderate correlation between CARNS1 levels and carnosine content in chickens, with both having an inverse relationship with bird age29. In addition, 3-methyl-histidine was considered a marker for myofibrillar breakdown and found to have high prediction accuracy for WB32. Dietary supplementation of beta-alanine and histidine led to increased carnosine levels synthesized in both slow and fast-growing chickens30,33, which in turn may potentially increase antioxidant capacity30,33,34. However, such increase was not observed in anserine29,33, suggesting that anserine synthesis in chicken pectoral muscle relies rather heavily on the methylation of carnosine from S-adenosylmethionine by carnosine-N-methyltransferase (CARNMT1) instead of 1-methyl-histidine35. While methionine content was significantly higher in WB affected muscle, the enzyme in charge of the production of S-adenosylmethionine, methionine adenosyltransferase (MAT2A), was down-regulated by a FC of 1.7. This could explain why increased levels of carnosine failed to boost anserine levels in chickens. That said, it is worthwhile for future studies to confirm the scarcity of S-adenosylmethionine and whether the reinstated level of carnosine and anserine could ameliorate WB prevalence and severity in broilers. Figure 3Reduced levels of carnosine and anserine in wooden breast affected chickens, likely due to down-regulation of carnosine synthase 1 (CARNS1). Different colors indicate higher (blue), lower (green) and not statistically different (black) abundance of a metabolite or gene transcripts in wooden breast affected p. major muscle. This figure was created with BioRender.com (http://biorender.com).
## Remodeling of energy metabolism
As shown in Fig. 1, our Joint Pathway Analysis identified several pathways involved in energy metabolism including glycolysis or gluconeogenesis; alanine, aspartate and glutamate metabolism; glycerolipid metabolism; pyruvate metabolism and histidine metabolism. WB tissues manifested accumulation of monoacylglycerols 2-linoleoylglycerol, 1-palmitoylglycerol and glycerol-3-phosphate (Table S5), which is consistent with decreased expression of genes involved in TAG synthesis [diacylglycerol o-acyltransferase 2 (DGAT2), acyl-CoA wax alcohol acyltransferase 1 (AWAT1)] and degradation [adipose triglyceride lipase (PNPLA2), adiponutrin (PNPLA3)] in affected tissues (Fig. 4). Besides the potential reduction in TAG metabolism, the affected muscles also exhibited down-regulation of key enzymes participating in mitochondrial FA oxidation (Fig. 4), including carnitine palmitoyltransferase I (CPT1A), malonyl-CoA decarboxylase (MLYCD), uncoupling protein 3 (UCP3), acyl-CoA synthetase long chain family member 4 (ACSL4) and FA translocase (CD36). Downregulation of these genes agreed with the significantly lower levels of their substrates, carnitines and acyl-carnitines, as well as higher levels of palmitate (Table S5). It is noteworthy to mention that broilers of 2–3 weeks of age with early stage WB exhibited upregulation of CD36 and some other lipid genes13–15. Similarly, at 7 weeks of age, expression of a key enzyme in lipid metabolism, lipoprotein lipase (LPL), was higher in moderately affected chickens when compared with chickens severely affected with WB36. These findings suggested dysregulated and even shifting role of lipid metabolism in different stages of WB.Figure 4Proposed schematic representation of remodeled energy metabolism in WB p. major muscles. Colors indicate higher (blue), lower (green) and not statistically different (black) abundance of a metabolite or gene transcripts in wooden breast affected p. major muscles in broiler chickens. * Not detected in metabolome data. CD36 CD36 molecule, MLYCD Malonyl-CoA decarboxylase, CPT1A, CPT2 Carnitine palmitoyltransferase, ACSL4 Acyl-CoA synthetase long chain family member 4, SLC16A1 Solute carrier family 16 member 1, OXCT1 3-oxoacid CoA-transferase 1, UCP3 Uncoupling protein 3, LPIN1 Lipin-1, PNPLA2 Adipose triglyceride lipase, PNPLA3 Adiponutrin, DGAT2 Diacylglycerol o-acyltransferase 2, AWAT1 Acyl-CoA wax alcohol acyltransferase 1, HIF1A Hypoxia inducible factor 1 subunit alpha, RXRA RXR alpha, FOXO1 Forkhead box o1, GNPAT Glycerone-phosphate O-acyltransferase, FAR1 Acyl-CoA reductase, ACOX2 Acyl-CoA oxidase 2, PHYH Phytanoyl-CoA 2-Hydroxylase. This figure was created with BioRender.com (http://biorender.com).
Additional evidence indicating abnormal energy metabolism in WB is the build-up of 3-hydroxybutyrate in affected tissues (Table S5; Fig. 4). This accumulation could be caused by the down-regulation of solute carrier family 16 member 1 (SLC16A1; FC of 1.7), denoting less transporters available for the entry of 3-hydroxybutyrate into mitochondria37, and hence restricting its use as a fuel for muscle. In addition to this ketone body, the use of acetoacetate for energy production may also be restricted in WB, as indicated by the down-regulation of an essential enzyme in ketolysis, 3-oxoacid CoA-transferase 1 (OXCT1; Fig. 4), which catalyzes the production of acetyl-CoA for the TCA cycle from acetoacetate37.
Differential expression of numerous transcription factors (Fig. 4) relevant to energy metabolism were noted in WB affected muscle, hypoxia inducible factor 1 subunit alpha (HIF1A), RXR alpha (RXRA) and forkhead box o1 (FOXO1). Under hypoxia, diminished FOXO1 and RXRA level suppress expression of their target genes such as UCP3, CD36 and ACSL4 (Fig. 4), resulting in declined FA uptake and oxidation38,39.
Subsequently, oxidative phosphorylation may have also been hindered, as shown by related DEGs (Table 3) and a significantly lower level of phosphate (Table S5), which is a cytosolic regulator of this process through the cellular energy state feedback40,41. The down-regulation of ATP synthase mitochondrial F1 complex assembly factor 1 (ATPAF1) is associated with a lower respiratory capacity and degenerated mitochondria because of impaired ATP synthase assembly4,42. Overall, these metabolic and transcriptomic variations implied a negative energy balance due to impaired mitochondrial capacity in WB affected p. major muscles. Table 3Differentially expressed genes involved in oxidative phosphorylation. ComplexGene symbolUp-regulatedDown-regulatedI–ND4L, ND4, ND3, ND1, NDUFS8, NDUFS7, NDUFS6, NDUFS1, NDUFA7, NDUFA11, NDUFA12, NDUFB10II–SDHC, SDHAIII–UQCRFS1, CYTB, UQCRC2, UQCRQIVCOX6A1, COX7A2–VATP6V1C2, ATP6V1A, ATP6V0D2ATPAF1, ATP5L, ATP6, ATP6V0A1 In response to a negative energy balance, normally, “starved” myocytes may increase the rate of amino acid catabolism for the purpose of refilling the pools of intermediates in the TCA cycle, via a process termed anaplerosis3,43. However, this may not be occurring in WB affected broiler chickens (Fig. 5). Although amino acids serving as gluconeogenic and ketogenic precursors43 were significantly higher in WB tissue, down-regulation of D-aspartate oxidase (DDO) and glutamic–oxaloacetic transaminase (GOT1, GOT2) suggested an obstructed path to replenish the TCA cycle. Additionally, GOT1 was identified as a candidate gene for WB in a genome-wide association study44, further corroborating an attenuated energy metabolism from genetic basis. Proline dehydrogenase (PRODH) is the rate limiting enzyme for proline degradation and catalyzes the electron transfer from proline to flavine adenine dinucleotide (FAD) before passing on to cytochrome c in electron transport chain to generate ATP45. Accordingly, the upregulation of PRODH by a FC of 3.3, higher levels of FAD and lower levels of proline (Table S5) implied ATP shortage, even nutrient and inflammatory stress in the WB p. major muscle45. One potential explanation for this malfunctioned anaplerosis is the modern broilers’ genetic predisposition for muscular hypertrophy, driving most of its available amino acids towards protein synthesis. Figure 5Schematic representation of altered TCA cycle and increase in amino acids involved in anaplerosis in breast muscles affected with wooden breast myopathy. Colors indicate higher (blue), lower (green) and not statistically different (black) abundance of a metabolite or gene transcripts in wooden breast affected p. major muscles in broiler chickens. * Not detected in metabolome data. ACO1 Aconitase 1, nad(+)Isocitrate dehydrogenase, IDH3A, IDH3B 3 catalytic subunit, OGDH Oxoglutarate dehydrogenase, SUCLG1 Succinate-CoA ligase GDP/ADP-forming subunit alpha, SDHA, SDHC Succinate dehydrogenase complex flavoprotein subunit, MDH2 Malate dehydrogenase 2, LDHA, LDHB, LDHD Lactate dehydrogenase, DDO D-aspartate oxidase, GOT1, GOT2 Glutamic-oxaloacetic transaminase. ASNS Asparagine synthetase, ASPA Aspartoacylase, NAT8L N-acetyltransferase 8 like, RIMKLB Ribosomal modification protein rimk like family member, GLUL Glutamate-ammonia ligase, GLS2 Glutaminase 2. This figure was created with BioRender.com (http://biorender.com).
## Peroxisomal function
Peroxisomes are unique organelles indispensable to various vital metabolic pathways including FA alpha- and beta-oxidation, ROS metabolism and etherphospholipid synthesis46,47. More precisely, other subcellular organelles rely on the functional interplay with peroxisomes to fulfill their role in metabolism. For instance, very-long-chain FAs need to undergo stepwise shortening within peroxisomes before they can be shuttled into mitochondria for full oxidation46. In addition to compromised mitochondrial functions, our study also suggests impaired peroxisomal function in WB muscle.
Glycerone-phosphate O-acyltransferase (GNPAT) and acyl-CoA reductase (FAR1) are responsible for the synthesis of dihydroxyacetone phosphate and long-chain alcohols, respectively, before the formation of ether bond between them48. Therefore, the down-regulation of these two enzymes (Table 1) could hamper etherphospholipid synthesis by reducing the substrate level. In humans, the deficiency of GNPAT and/or FAR1 is linked to plasmalogen, or cell membrane glycerophospholipids, deprivation due to insufficient etherphospholipid synthesis46. Furthermore, significantly higher levels of glycerophosphorylcholine (GPC) and glycerophosphoethanolamine (GPE) in WB affected muscles (Table S5) could be indicating lower plasmalogen biosynthesis as well as turnover49. Given that plasmalogens are essential components of cell and subcellular membranes49, its reduction could result in more frequent membrane damage, and hence loss of cellular integrity. Considering that membrane biosynthesis increases as muscles grow50, a potential reduction in plasmalogens and the ensuing membrane instability could be even more problematic in the face of the increased hypertrophy in fast-growing broiler chickens.
Additionally, it is likely that peroxisome biogenesis was diminished in WB chickens, as shown by the down-regulation of peroxisome biogenesis factors (PEX1, PEX5, PEX7, PEX10) by an average FC of 1.5 (Table 1). Specifically, cytoplasmic receptors PEX5 and PEX7 recognize and bind to proteins containing peroxisome targeting signal (PTS)51, which are then imported into peroxisomes by a complex consisting of PEX2, PEX10 and PEX1252. These PTS matrix protein receptors exit peroxisomes to be recycled back to the cytosol with the help from PEX1, PEX6 and integral membrane protein PEX2652. An aberrant mitochondrial ultrastructure and respiratory chain activity were observed in PEX5 knockout mice, resulting from defective peroxisome biogenesis53.
Considering the supportive role of peroxisomes in metabolism, we hypothesize that albeit in a chronic manner, a potential decrease in their biogenesis in WB could eventually be detrimental, due to an increase in ROS levels and mitochondrial defects51,53. A study in Drosophila suggested a linkage between selective peroxisome loss and muscle function and physiology by disrupting energy metabolism51. Consequently, it is worthwhile to obtain a deeper understanding of the particular role played by peroxisomes in WB disease progression.
## Cellular signaling
Functional analysis of the interconnected network surrounding the hub features implied differences in cell–cell communication between WB affected and unaffected chickens, which are in line with results from a recent study on 7-week-old WB affected broilers54. Focal adhesions are the linking bridges between cells and extracellular matrix (ECM), where integrin, proteoglycan and actin cytoskeleton mediate mechanical and biochemical signaling55. In skeletal muscles, the cell-ECM crosstalk is not only vital for muscle contraction but can also modulate myogenesis and muscle repair55. It is likely that myofibrillar changes in the WB muscle happened concurrently with alterations in focal adhesion, as shown by DEGs in ECM-receptor interaction, cytokine-cytokine receptor interaction and Wnt signaling pathways (Tables S2–S4). Specifically, actinin alpha 1 (ACTN1) is responsible for crosslinking integrin to actin filaments in the cytoskeleton, and zyxin (ZYX) is recruited to newly formed focal complexes to stabilize membrane protrusion56,57. In the meantime, calpain (CAPN2) cleaves these proteins within focal adhesions to mediate their overall turnover and cell migration57. The ECM relevant DEGs such as integrin subunit alpha and beta and collagens (Table S2) also suggested aberrant ECM composition and interactions. Additionally, ZYX and ACTN1 are associated with cytoskeletal remodeling in vascular smooth muscle cells and vessel stiffness58,59. The upregulated ACTN1 and collagen type I alpha 1 chain (COL1A1) observed in WB chickens were concordant with changes in vascular smooth muscle cells during vascular regression59, which could possibly contribute to WB’s signature lesion phlebitis4.
One of the top hub features in WGCNA is inosine 5′-monophosphate (IMP), which can mediate diverse processes including inflammation as an extracellular signaling molecule60. Although the role of IMP in cellular signaling remains unclear, its importance was suggested by the high connectivity between IMP and other features in the network. Compared with unaffected samples, WB affected muscles showed a significant difference in the levels of IMP, adenosine 5′-monophosphate (AMP), guanosine 5′-monophosphate (GMP) and purine bases except for inosine (Table S5). The significantly lower level of IMP and AMP, as well as the down-regulation of AMP deaminase 1 (AMPD1) in WB tissues may even impose a negative effect in purine biosynthesis, as it is likely that the replenishment of IMP from AMP was impeded60. Moreover, catabolic product hypoxanthine from purine metabolism was linked to muscle fatigue and depletion of energy substrates61. However, unlike in mice61, the accumulation of hypoxanthine in affected broilers was not accompanied with enhanced glycolysis, mitochondrial biogenesis or oxidative phosphorylation, as shown by down-regulated DEGs (Table 3, Table S1), perhaps because of a genetic predisposition to diverting resources toward protein metabolism.
## Conclusion
The integration of transcriptomics and metabolomics enabled us to study pathways relevant to the WB progression by possibly linking genes and their metabolites within biological pathways. In particular, we showed a potential mechanism for impaired antioxidant capacity in WB, which involves the down-regulation of CARNS1, resulting in the depletion of its antioxidant products carnosine and anserine. Furthermore, this study hypothesized that the buildup of lipid intermediates presumably due to down-regulated lipid genes leads to lipid toxicity62 and mitochondrial dysfunction in WB. Our results also suggest an increased use of some energy precursors for protein synthesis in WB, so it may be of interest to study the mechanism behind the switch between energy metabolism and anabolic muscle growth in WB. *Correlated* genes related to focal adhesion indicated that intercellular signaling may play a role in regulating, even altering, different biological pathways during WB development.
## Supplementary Information
Supplementary Information. The online version contains supplementary material available at 10.1038/s41598-023-31429-7.
## References
1. Caplen G. **Kinematic analysis quantifies gait abnormalities associated with lameness in broiler chickens and identifies evolutionary gait differences**. *PLoS One* (2012.0) **7** e40800. DOI: 10.1371/journal.pone.0040800
2. Kuttappan VA, Hargis BM, Owens CM. **White striping and woody breast myopathies in the modern poultry industry: A review**. *Poult. Sci.* (2016.0) **95** 2724-2733. DOI: 10.3382/ps/pew216
3. Sihvo H-K, Immonen K, Puolanne E. **Myodegeneration with fibrosis and regeneration in the pectoralis major muscle of broilers**. *Vet. Pathol.* (2013.0) **51** 619-623. DOI: 10.1177/0300985813497488
4. Papah MB, Brannick EM, Schmidt CJ, Abasht B. **Evidence and role of phlebitis and lipid infiltration in the onset and pathogenesis of Wooden Breast Disease in modern broiler chickens**. *Avian Pathol.* (2017.0) **46** 623-643. DOI: 10.1080/03079457.2017.1339346
5. Hubert SM, Athrey G. **Energy metabolism and sources of oxidative stress in wooden breast—a review**. *F1000Res* (2020.0) **9** 319. DOI: 10.12688/f1000research.23296.1
6. Mutryn MF, Brannick EM, Fu W, Lee WR, Abasht B. **Characterization of a novel chicken muscle disorder through differential gene expression and pathway analysis using RNA-sequencing**. *BMC Genom.* (2015.0) **16** 399. DOI: 10.1186/s12864-015-1623-0
7. Abasht B, Mutryn MF, Michalek RD, Lee WR. **Oxidative stress and metabolic perturbations in Wooden Breast disorder in chickens**. *PLoS One* (2016.0) **11** e0153750. DOI: 10.1371/journal.pone.0153750
8. Marie Hubert S, Williams TJ, Athrey G. **Insights into the molecular basis of wooden breast based on comparative analysis of fast- and slow-growth broilers**. *biorxiv* (2018.0). DOI: 10.1101/356683
9. Tallentire CW, Leinonen I, Kyriazakis I. **Breeding for efficiency in the broiler chicken: A review**. *Agron. Sustain. Dev.* (2016.0) **36** 25. DOI: 10.1007/s13593-016-0398-2
10. Kong B-W. **RNA sequencing for global gene expression associated with muscle growth in a single male modern broiler line compared to a foundational Barred Plymouth Rock chicken line**. *BMC Genom.* (2017.0) **18** 82. DOI: 10.1186/s12864-016-3471-y
11. Banerjee P, Carmelo VAO, Kadarmideen HN. **Integrative analysis of metabolomic and transcriptomic profiles uncovers biological pathways of feed efficiency in pigs**. *Metabolites* (2020.0) **10** 275. DOI: 10.3390/metabo10070275
12. Connor SC, Hansen MK, Corner A, Smith RF, Ryan TE. **Integration of metabolomics and transcriptomics data to aid biomarker discovery in type 2 diabetes**. *Mol. Biosyst.* (2010.0) **6** 909-921. DOI: 10.1039/b914182k
13. Papah MB, Abasht B. **Dysregulation of lipid metabolism and appearance of slow myofiber-specific isoforms accompany the development of Wooden Breast myopathy in modern broiler chickens**. *Sci. Rep.* (2019.0) **9** 17170. DOI: 10.1038/s41598-019-53728-8
14. Papah MB, Brannick EM, Schmidt CJ, Abasht B. **Gene expression profiling of the early pathogenesis of wooden breast disease in commercial broiler chickens using RNA-sequencing**. *PLoS One* (2018.0) **13** e0207346. DOI: 10.1371/journal.pone.0207346
15. Lake JA, Papah MB, Abasht B. **Increased expression of lipid metabolism genes in early stages of wooden breast links myopathy of broilers to metabolic syndrome in humans**. *Genes (Basel)* (2019.0) **10** 746. DOI: 10.3390/genes10100746
16. 16.“Support from the University of Delaware CBCB Bioinformatics Core Facility and use of the BIOMIX compute cluster was made possible through funding from Delaware INBRE (NIH NIGMS P20 GM103446), the State of Delaware, and the Delaware Biotechnology Institut.”
17. 17.Andrews, S. FastQC: A Quality Control Tool for High Throughput Sequence Data (2010).
18. Kim D, Langmead B, Salzberg SL. **HISAT: A fast spliced aligner with low memory requirements**. *Nat. Methods* (2015.0) **12** 357-360. DOI: 10.1038/nmeth.3317
19. Anders S, Pyl PT, Huber W. **HTSeq—a Python framework to work with high-throughput sequencing data**. *Bioinformatics* (2015.0) **31** 166-169. DOI: 10.1093/bioinformatics/btu638
20. Evans AM, DeHaven CD, Barrett T, Mitchell M, Milgram E. **Integrated, nontargeted ultrahigh performance liquid chromatography/electrospray ionization tandem mass spectrometry platform for the identification and relative quantification of the small-molecule complement of biological systems**. *Anal. Chem.* (2009.0) **81** 6656-6667. DOI: 10.1021/ac901536h
21. Xia J, Psychogios N, Young N, Wishart DS. **MetaboAnalyst: A web server for metabolomic data analysis and interpretation**. *Nucleic Acids Res.* (2009.0) **37** W652-W660. DOI: 10.1093/nar/gkp356
22. Ripley BD. **The R project in statistical computing**. *MSOR Connect.* (2001.0) **1** 23-25. DOI: 10.11120/msor.2001.01010023
23. Robinson MD, Oshlack A. **A scaling normalization method for differential expression analysis of RNA-seq data**. *Genome Biol.* (2010.0) **11** R25. DOI: 10.1186/gb-2010-11-3-r25
24. Pang Z. **MetaboAnalyst 5.0: Narrowing the gap between raw spectra and functional insights**. *Nucleic Acids Res.* (2021.0) **49** W388-W396. DOI: 10.1093/nar/gkab382
25. Zhang B, Horvath S. **A general framework for weighted gene co-expression network analysis**. *Stat. Appl. Genet. Mol. Biol.* (2005.0) **4** Article17. DOI: 10.2202/1544-6115.1128
26. Langfelder P, Horvath S. **WGCNA: An R package for weighted correlation network analysis**. *BMC Bioinform.* (2008.0) **9** 559. DOI: 10.1186/1471-2105-9-559
27. Wang W. **Weighted gene co-expression network analysis of expression data of monozygotic twins identifies specific modules and hub genes related to BMI**. *BMC Genomics* (2017.0) **18** 872. DOI: 10.1186/s12864-017-4257-6
28. Moro J, Tome D, Schmidely P, Demersay TC, Azzout-Marniche D. **Histidine: A systematic review on metabolism and physiological effects in human and different animal species.**. *Nutrients* (2020.0) **12** 1414. DOI: 10.3390/nu12051414
29. Khumpeerawat P, Duangjinda M, Phasuk Y. **Carnosine content and its association with carnosine-related gene expression in breast meat of Thai native and black-bone chicken**. *Animals (Basel)* (2021.0) **11** 1987. DOI: 10.3390/ani11071987
30. Suwanvichanee C. **Effects of β-alanine and L-histidine supplementation on carnosine contents in and quality and secondary structure of proteins in slow-growing Korat chicken meat**. *Poult. Sci* (2022.0) **101** 101776. DOI: 10.1016/j.psj.2022.101776
31. 31.Zhang, X. et al. Proteomic characterization of normal and woody breast meat from broilers of five genetic strains. Meat Muscle Biol. 4, (2020).
32. Lake JA. **Identification of circulating metabolites associated with wooden breast and white striping**. *PLoS One* (2022.0) **17** e0274208. DOI: 10.1371/journal.pone.0274208
33. Kopec W. **Antioxidative characteristics of chicken breast meat and blood after diet supplementation with carnosine, L-histidine, and β-alanine**. *Antioxidants (Basel)* (2020.0) **9** 1093. DOI: 10.3390/antiox9111093
34. Hu X. **Effect of carnosine on growth performance, carcass characteristics, meat quality and oxidative stability in broiler chickens**. *J. Poult. Sci* (2009.0) **46** 296-302. DOI: 10.2141/jpsa.46.296
35. McManus IR. **Enzymatic synthesis of anserine in skeletal muscle by N-methylation of carnosine**. *J. Biol. Chem* (1962.0) **237** 1207-1211. DOI: 10.1016/S0021-9258(18)60310-2
36. Abasht B, Papah MB, Qiu J. **Evidence of vascular endothelial dysfunction in Wooden Breast disorder in chickens: Insights through gene expression analysis, ultra-structural evaluation and supervised machine learning methods**. *PLoS One* (2021.0) **16** e0243983. DOI: 10.1371/journal.pone.0243983
37. Evans M, Cogan KE, Egan B. **Metabolism of ketone bodies during exercise and training: physiological basis for exogenous supplementation: Ketone bodies and exercise**. *J. Physiol* (2017.0) **595** 2857-2871. DOI: 10.1113/JP273185
38. Szanto A. **Retinoid X receptors: X-ploring their (patho)physiological functions**. *Cell Death Differ.* (2004.0) **11** S126-S143. DOI: 10.1038/sj.cdd.4401533
39. Bastie CC. **FoxO1 stimulates fatty acid uptake and oxidation in muscle cells through CD36-dependent and -independent mechanisms**. *J. Biol. Chem.* (2005.0) **280** 14222-14229. DOI: 10.1074/jbc.M413625200
40. Bose S, French S, Evans FJ, Joubert F, Balaban RS. **Metabolic network control of oxidative phosphorylation: Multiple rolevascs of inorganic phosphate**. *J. Biol. Chem* (2003.0) **278** 39155-39165. DOI: 10.1074/jbc.M306409200
41. Wilson DF. **Oxidative phosphorylation: regulation and role in cellular and tissue metabolism: Oxidative phosphorylation: role in cell and tissue metabolism**. *J. Physiol* (2017.0) **595** 7023-7038. DOI: 10.1113/JP273839
42. Zhou Z. **ATPAF1 deficiency impairs ATP synthase assembly and mitochondrial respiration**. *Mitochondrion* (2021.0) **60** 129-141. DOI: 10.1016/j.mito.2021.08.005
43. Owen OE, Kalhan SC, Hanson RW. **The key role of anaplerosis and cataplerosis for citric acid cycle function**. *J. Biol. Chem* (2002.0) **277** 30409-30412. DOI: 10.1074/jbc.R200006200
44. Lake JA, Dekkers JCM, Abasht B. **Genetic basis and identification of candidate genes for wooden breast and white striping in commercial broiler chickens**. *Sci. Rep* (2021.0) **11** 6785. DOI: 10.1038/s41598-021-86176-4
45. Phang JM, Pandhare J, Liu Y. **The metabolism of proline as microenvironmental stress substrate**. *J. Nutr* (2008.0) **138** S2008-S2015. DOI: 10.1093/jn/138.10.2008S
46. Wanders RJA, Waterham HR, Ferdinandusse S. **Metabolic interplay between peroxisomes and other subcellular organelles including mitochondria and the endoplasmic reticulum**. *Front. Cell Dev. Biol* (2015.0) **3** 83. PMID: 26858947
47. Liu J, Lu W, Shi B, Klein S, Su X. **Peroxisomal regulation of redox homeostasis and adipocyte metabolism**. *Redox Biol* (2019.0) **24** 101167. DOI: 10.1016/j.redox.2019.101167
48. Cheng JB, Russell DW. **Mammalian wax biosynthesis. I. Identification of two fatty acyl-Coenzyme A reductases with different substrate specificities and tissue distributions.**. *J. Biol. Chem* (2004.0) **279** 37789-37794. DOI: 10.1074/jbc.M406225200
49. Braverman NE, Moser AB. **Functions of plasmalogen lipids in health and disease**. *Biochim. Biophys. Acta* (2012.0) **1822** 1422-1452
50. Carta G, Murru E, Banni S, Manca C. **Palmitic acid: Physiological role, metabolism and nutritional implications.**. *Front. Physiol.* (2017.0) **8** 902. DOI: 10.3389/fphys.2017.00902
51. Faust JE. **Peroxisomes are required for lipid metabolism and muscle function in Drosophila melanogaster**. *PLoS One* (2014.0) **9** e100213. DOI: 10.1371/journal.pone.0100213
52. Steinberg SJ. **Peroxisome biogenesis disorders**. *Biochim. Biophys. Acta* (2006.0) **1763** 1733-1748. DOI: 10.1016/j.bbamcr.2006.09.010
53. Martens K. **Peroxisome deficient aP2-Pex5 knockout mice display impaired white adipocyte and muscle function concomitant with reduced adrenergic tone**. *Mol. Genet. Metab* (2012.0) **107** 735-747. DOI: 10.1016/j.ymgme.2012.10.015
54. Malila Y. **Insights into transcriptome profiles associated with wooden breast myopathy in broilers slaughtered at the age of 6 or 7 weeks**. *Front. Physiol* (2021.0) **12** 691194. DOI: 10.3389/fphys.2021.691194
55. Csapo R, Gumpenberger M, Wessner B. **Skeletal muscle extracellular matrix - what do we know about its composition, regulation, and physiological roles? A narrative review**. *Front. Physiol* (2020.0) **11** 253. DOI: 10.3389/fphys.2020.00253
56. Hsu CP, Moghadaszadeh B, Hartwig JH, Beggs AH. **Sarcomeric and nonmuscle α-actinin isoforms exhibit differential dynamics at skeletal muscle Z-lines**. *Cytoskeleton (Hoboken)* (2018.0) **75** 213-228. DOI: 10.1002/cm.21442
57. Wozniak MA, Modzelewska K, Kwong L, Keely PJ. **Focal adhesion regulation of cell behavior**. *Biochim. Biophys. Acta* (2004.0) **1692** 103-119. DOI: 10.1016/j.bbamcr.2004.04.007
58. Suphamungmee W, Lehman W, Morgan KG. **Functional remodeling of the contractile smooth muscle cell cortex, a provocative concept, supported by direct visualization of cortical remodeling**. *Biology (Basel)* (2022.0) **11** 662. PMID: 35625390
59. 59.Rodriguez, D. et al. Vascular Aging in the Invertebrate Chordate, Botryllus schlosseri. Front. Mol. Biosci.8, 626827 (2021).
60. 60.Lovászi, M. et al. Inosine monophosphate and inosine differentially regulate endotoxemia and bacterial sepsis. FASEB J.35, e21935 (2021).
61. 61.Yin, C. et al. Hypoxanthine induces muscular ATP depletion and fatigue via UCP2. Front. Physiol.12, 647743 (2021).
62. 62.Listenberger, L. L. et al. Triglyceride accumulation protects against fatty acid-induced lipotoxicity. Proc. Natl. Acad. Sci. U. S. A.100, 3077–3082 (2003).
|
---
title: Molecular simulations of SSTR2 dynamics and interaction with ligands
authors:
- Silvia Gervasoni
- Camilla Guccione
- Viviana Fanti
- Andrea Bosin
- Giancarlo Cappellini
- Bruno Golosio
- Paolo Ruggerone
- Giuliano Malloci
journal: Scientific Reports
year: 2023
pmcid: PMC10036620
doi: 10.1038/s41598-023-31823-1
license: CC BY 4.0
---
# Molecular simulations of SSTR2 dynamics and interaction with ligands
## Abstract
The cyclic peptide hormone somatostatin regulates physiological processes involved in growth and metabolism, through its binding to G-protein coupled somatostatin receptors. The isoform 2 (SSTR2) is of particular relevance for the therapy of neuroendocrine tumours for which different analogues to somatostatin are currently in clinical use. We present an extensive and systematic computational study on the dynamics of SSTR2 in three different states: active agonist-bound, inactive antagonist-bound and apo inactive. We exploited the recent burst of SSTR2 experimental structures to perform μs-long multi-copy molecular dynamics simulations to sample conformational changes of the receptor and rationalize its binding to different ligands (the agonists somatostatin and octreotide, and the antagonist CYN154806). Our findings suggest that the apo form is more flexible compared to the holo ones, and confirm that the extracellular loop 2 closes upon the agonist octreotide but not upon the antagonist CYN154806. Based on interaction fingerprint analyses and free energy calculations, we found that all peptides similarly interact with residues buried into the binding pocket. Conversely, specific patterns of interactions are found with residues located in the external portion of the pocket, at the basis of the extracellular loops, particularly distinguishing the agonists from the antagonist. This study will help in the design of new somatostatin-based compounds for theranostics of neuroendocrine tumours.
## Introduction
Somatostatin is a cyclic disulphide bond-containing hormone expressed in two splicing variants of 14 (SST14) and 28 amino acids. The former is the predominant form in the brain, while the latter is found primary in the gut1,2. Somatostatin plays a crucial role in regulating the release of different hormones such as insulin, growth hormone and secretin, through its inhibitory activity3,4. SST is an agonist of somatostatin receptors (SSTRs), which belong to class A G-protein coupled receptors (GPCRs). Most GPCRs present seven transmembrane alpha helices (TM1-7), three extracellular (ECL1-3) and three intracellular (ICL1-3) loops (Fig. 1, left panel. The Ballesteros-Weinstein numbering scheme for class A GPCRs is adopted throughout the paper5). SSTRs are coupled with inhibitory G-protein (i.e., G\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$_i$$\end{document}i or G\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$_0$$\end{document}0)6, and can be divided into five families, from 1 to 5, among which the isoform 2 is the most expressed in human neuroendocrine tumours (NETs)7,8. Such an over-expression makes SSTR2 an important target for both anti-tumour therapy and diagnostic (i.e., theranostic)9. As a result, the development of drugs targeting SSTR2 is of great interest and a few molecules are in clinical use. Generally, these compounds mimic the structure of the endogenous agonist SST1410, and can be either peptides (e.g., octreotide11, lanreotide, pasireotide12, Fig. 1, right panel) or non-peptides (e.g., L-054,522 and L-054,26413,14). SST14 has a short half-life (< 3 min) and lacks selectivity towards the five SSTR isoforms. Differently, octreotide and lanreotide show higher affinity towards SSTR2 and a longer half-life (2 and 1 h, respectively)10,15,16. The non-peptide/peptidomimetic analogues of SST14 have shown increased half-life and potency towards specific SSTR isoforms as well17. Furthermore, they can function as agonist or antagonist18,19. Noteworthy, various studies have shown that antagonists have favourable pharmacokinetic profiles and better tumour visualization compared to agonists, thanks to their ability to bind multiple conformational states of SSTR2, despite poor internalization rates19,20.Figure 1Left panel: Schematic representation of GPCR structure and main domains. Transmembrane helices (green): TM1 = A44\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{1.33}$$\end{document}1.33-R70\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{1.59}$$\end{document}1.59, TM2 = I77\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{2.38}$$\end{document}2.38-A104\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{2.65}$$\end{document}2.65, TM3 = K112\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{3.22}$$\end{document}3.22-V145\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{3.55}$$\end{document}3.55, TM4 = I150\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{4.33}$$\end{document}4.33-A181\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{4.64}$$\end{document}4.64, TM5 = G202\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{5.32}$$\end{document}5.32-S237\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{5.67}$$\end{document}5.67, TM6 = R245\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{6.24}$$\end{document}6.24-V280\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{6.59}$$\end{document}6.59, TM7 = P288\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{7.29}$$\end{document}7.29-L315\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{7.56}$$\end{document}7.56. Extracellular loops (red): ECL1 = L105-G111, ECL2 = G182-S201, ECL3 = S281-T287. Intracellular loops (yellow): ICL1 = Y71-T76, ICL2 = H146-P149, ICL3 = V238-S244. Right panel: 2D sketches of SSTR2 agonists: (A) SST14 and (B) octreotide (T-ol = threoninol), and (C) the antagonist CYN154806 (Ac = acetyl group, PPN = Phe-NO\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$_2$$\end{document}2). Conserved residues are highlighted.
Given the great interest in development of SST14 analogues, several studies addressing the structural basis of ligand binding to SSTR2 have been recently published. First, computational works that exploited homology modelling21–23. Then, some experimental structures started to appear. Robertson and co-workers released the first cryo-EM structures of SSTR2 bound with SST14 and octreotide24. New structures soon followed, also in complex with other ligands14,15,25,26 and one in the apo inactive form27 (Table S1). An important feature of the SSTR2 structure (common to other GPCRs) is the high flexibility of ECL228, that was found completely open when in complex with SST14, folded down with octreotide, and in a middle position in the apo form24. Furthermore, all transmembrane helices engage rearrangements when moving from inactive to active states of the receptor. In particular, TM6 shows a hallmark outward movement29, TM5 is a common switch that moves closer to TM7, stabilizing TM630, and TM3 is a hub for stabilization31.
The main pharmacophore features characterizing the SSTR2 peptide ligands are represented by [1] a \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\beta$$\end{document}β-sheet shape, at which tip are located [2] an aromatic group (often a tryptophan residue), and [3] a basic positively charged moiety (such as a lysine residue). This portion is inserted into the cavity of the transmembrane pocket, while the other part of the molecule faces the external loops of the receptor14. The increasing availability of experimental structures of SSTR2 has shed light on the structural features of the receptor and on the flexible elements that play a crucial role in the interaction with ligands. Molecular dynamics (MD) simulations performed by Robertson and co-workers24 for apo-SSTR2 revealed that ECL2 does not spontaneously fold over the orthosteric site, while it closes upon the binding pocket when SSTR2 was in complex with octreotide. However, an in-depth characterization of the dynamics of SSTR2 in the inactive antagonist-bound form in comparison with the apo-inactive and active agonist-bound ones is needed to achieve a more complete picture.
In this work, we performed MD simulations of SSTR2 in both apo and holo forms and explored the binding modes of the complexed ligands. We focused on the endogenous ligand SST14 and its peptide analogues octreotide (OCT) and CYN154806 (CYN), behaving as agonist and antagonist, respectively. We thoroughly quantified the opening and closing movements of ECL2. Similarly, following previous studies on GPCR dynamics, we determined structural metrics able to discriminate between active (i.e., structures solved in complex with the G-protein and an agonist ligand) and inactive (i.e., structures solved without the G-protein both in the apo and in the antagonist-bound forms) conformations of SSTR2. In the active and inactive conformations, the receptor is able to trigger or not the intracellular signal, respectively. Our results confirm that, as expected, the apo form of SSTR2 is characterised by a higher flexibility compared to the holo forms. We found that OCT induces the closure of ECL2, confirming previous findings, while the loop does not close upon CYN, despite their similar steric hindrance. Combination of binding free energies calculations and interaction fingerprints evaluations reveals peculiar behaviors of the three ligands in terms of intra- and inter-molecular interactions.
## Results and Discussion
To explore the dynamics of SSTR2 in different conformations and to elucidate agonist and antagonist binding at the molecular level, we performed multi-copy MD simulations, with an overall simulation time of 5 μs per system. In detail, we simulated the receptor in the active and inactive states. Conformational clusters extracted from MD simulations provided the basis for further characterizations both in energetic terms and patterns of detailed interactions.
## The binding of ligand stabilizes SSTR2 loops
Given the well-known challenge of GPCR experimental resolution32–35, the systems selected for this work were obtained using stabilizing agents: SSTR2-SST14 and SSTR2-OCT were in complex with the G-protein, SSTR2 apo with a nanobody, and SSTR2-CYN was expressed as a chimera. For this reason, we first assessed the overall stability of the systems during the MD simulations. For computational time reasons, we performed the MD simulations without including the G-protein. Since the allosteric effect of the G-protein is expected to have an impact on major conformational rearrangements, it is safe to assume that the lack of the G-protein in our plain μs-long MD simulations did not affect much the system structural features under investigation. Indeed, the average RMSD values of the backbone C\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha$$\end{document}α atoms (Fig. S1) were all below 2.5 Å, suggesting that the removal of the stabilizing agents did not alter the main conformation of the systems. Consistently, the RMSF of the backbone C\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha$$\end{document}α atoms (Fig. S2) reached the highest values for the intra and extracellular loops, while the transmembrane helices were highly stable in all cases. As expected, ECL2, which is the longest loop and interacts with the ligands, and ICL3, which is responsible for the recognition of the G-protein, were the most flexible domains, generally reaching or exceeding an RMSF value of 4 Å. However, we found some differences in loops movements through the four systems: the apo structure showed the largest loops flexibility (especially ECL2 and ECL3) compared to the other systems, while the SSTR2-SST14 complex is the one with the less flexible loops. These results agree with experimental findings for a generic GPCR, according to which its apo structure is highly unstable and prone to easily explore different conformational states, whereas the binding with a ligand stabilizes an active conformation of the receptor32,36. In particular, our data suggest that the endogenous ligand SST14 is more stabilizing than the synthetic agonist OCT and antagonist CYN (mean values of the backbone C\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha$$\end{document}α atoms RMSF integral for SST14 = 279 Å, OCT = 294 Å, CYN = 305 Å, apo = 307 Å).
## ECL2 closes upon octreotide but not CYN154806
The ECL2 is known to be highly flexible and its closure upon the binding pocket is believed to facilitate the interaction with ligands37,38. The RMSF values discussed above highlighted the involvement of this loop in the main differences between the dynamics of the four systems. Therefore, we analyzed in detail the opening and closing movements of ECL2 during the MD simulations. To this aim, we identified two geometric variables describing ECL2 movements. We computed [1] the distance \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\delta$$\end{document}δ between the loop tip (center of mass of Q187, W188, G189 C\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha$$\end{document}α atoms) and the center of mass of the seven TM helices C\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha$$\end{document}α atoms, and [2] the angle \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\beta$$\end{document}β between the W188 C\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha$$\end{document}α atom, the base of ECL2 (center of mass of A181 and I195 C\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha$$\end{document}α atoms), and the base of ECL3 (center of mass of S281 and P288 C\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha$$\end{document}α atoms) (Fig. S3). These two values were computed for all MD frames of each replica (saved every 1 ps), and combined in the scatter plot shown in Fig. 2. After a careful visual inspection of the MD trajectories, we identified threshold values of distance \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\delta$$\end{document}δ and angle \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\beta$$\end{document}β allowing for a distinction between open and close configuration of the ECL2 loop (i.e., \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$24< \delta < 33$$\end{document}24<δ<33 Å and 29 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$< \beta < 48^\circ$$\end{document}<β<48∘ for the close conformation). According to these values, as shown in Fig. 2, the experimental structures in complex with OCT (7XAU and 7T11), lanreotide (7XAV), and L-054,522 (7XN9) have a closed loop, while all the others feature an open loop. These results agree with previous findings24.Figure 2ECL2 opening and closing movements. Each point of the plot refers to a frame of the MD trajectories. The results of SST14, OCT, CYN and apo structures are coloured in yellow, green, magenta and blue, respectively. Threshold values of distance and angle are indicated. The red box includes the frames in which the ECL2 is in the closed conformation, the orange box indicates the border between open and closed conformations. The distance values range from 24 to 44 Å, the angle values from 29\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^\circ$$\end{document}∘ to 135\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^\circ$$\end{document}∘. The areas enclose all the frames (dots) corresponding to the four systems. The triangles and squares represent the experimental structures of SSTR2 (list of all experimental structures in Table S1).
Each point in Fig. 2 refers to one MD frame, which is associated with single values of distance and angle. All the frames included in the red/orange boxes are those in which ECL2 is in the closed conformation according to the above criteria. It is interesting to compare the corresponding percentages of occurrence in the four cases: $28.2\%$ for SSTR2-OCT, $1.5\%$ for SSTR2-CYN, $0.1\%$ for the apo form and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sim$$\end{document}∼$0\%$ for SST14.
The distribution of points in the plot mirrors the RMSF trend: SSTR2-SST14 and SSTR2-CYN show the less flexible loop, followed by SSTR2-OCT in which \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sim$$\end{document}∼$28\%$ of the frames present a closed ECL2. In the case of SST14, ECL2 cannot close upon the binding pocket due to the steric hindrance of the ligand (mean Connolly surface area39 collected during the MD trajectories for SST14, OCT and CYN: 1273 ± 26 Å\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^2$$\end{document}2, 887 ± 30 Å\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^2$$\end{document}2, and 895 ± 28 Å\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^2$$\end{document}2, respectively). Whereas, the marked difference between the loop motions in the complexes with OCT and CYN can be found in the different interaction patterns between the two peptides (see below). The points corresponding to the apo form are the most scattered, reinforcing the higher flexibility compared to the holo structures, and suggesting that ligands play a key role in the stabilization of ECL2.
## Key features distinguishing active from inactive conformations of SSTR2
GPCRs are able to trigger the intracellular signal through the presence of the G-protein (or arrestins), and thanks to both microswitches of aminoacids29 and major conformational changes. These receptors are indeed able to assume multi-conformational states, that can be stabilized by the presence of a ligand28.
The transition from inactive to active form of class A GPCRs involves the movement of TM5, TM6 and TM7: an outward displacement of TM5 and TM6, an inward movement of TM7 at the intracellular side, and an additional inwards shift of TM5 and TM7 at the extracellular side29–31. In a recent work, Lu and co-workers identified two geometric variables able to distinguish between active and inactive forms of the class A GPCR angiotensin II40, suggesting that these parameters could be likewise measured for other GPCRs of the same class. On this ground, we computed these parameters on our MD trajectories, by considering [1] the distance between the C\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha$$\end{document}α atoms of C225\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{5.55}$$\end{document}5.55 and S305\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{7.46}$$\end{document}7.46, and [2] the angle between the C\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\alpha$$\end{document}α atoms of T255\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{6.34}$$\end{document}6.34, C268\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{6.47}$$\end{document}6.47 and I80\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{2.41}$$\end{document}2.41 (Fig. S4). The first quantity accounts for the conformational changes of TM5 and TM7 (active \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$< \sim$$\end{document}<∼19 Å), while the second one reflects the outward movement of TM6 (active > \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sim$$\end{document}∼45\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^\circ$$\end{document}∘). Figure 3 reports a scatter plot in which each point represents a frame of the MD trajectories associated with the distance and angle values. The active conformations of SSTR2 in complex with SST14 and OCT, and the inactive apo conformation are well clustered into two distinct regions of the plot, spanning from 16 to 22 Å and from 48\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^\circ$$\end{document}∘ to 77\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^\circ$$\end{document}∘ (active), and from 22 to 28 Å and from 30\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^\circ$$\end{document}∘ to 47\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^\circ$$\end{document}∘ (inactive). On the contrary, the inactive conformation of the receptor in complex with the antagonist CYN is located in an intermediate region, ranging from 19 to 25 Å and from 34\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^\circ$$\end{document}∘ to 48\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^\circ$$\end{document}∘, partly overlapping with the apo region and, at the same time, extending towards the active one.
Interestingly, the experimental structures corresponding to the active complexes are located in the center of the active region, while the experimental conformation of the apo form (i.e., PDB 7UL5) is located at the edge of the inactive area. These results further confirm the high instability of SSTR2 in the apo state, that during the 5 μs of MD simulations explored multiple conformations, moving away from that reported in the cryoEM structure. The only exception is represented by the PDB structure 7XN9 (i.e., SSTR2 in complex with a non-peptide agonist L-054,522), which is located in the antagonist/inactive region. It is noteworthy that, this is the only SSTR2-agonist complex in which the receptor is not coupled with the G-protein, which can explain its peculiar behaviour. Figure 3Scatter plot separating active from inactive conformations in SSTR2. The yellow, green, magenta and blue dots refer to each frame of the SSTR2-SST14, SSTR2-OCT, SSTR2-CYN and SSTR2 apo trajectories, respectively. The areas around dots represent the frame extension. The triangles and squares represent the experimental structures of SSTR2 (list of all experimental structures in Table S1).
## Intramolecular interactions of peptides influence the binding with SSTR2
The ligands share the same cyclic \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\beta$$\end{document}β-sheet structure, presenting the characteristic residues Trp (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^D$$\end{document}DTrp for OCT and CYN) and Lys that interact at the bottom of the binding pocket with I177\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{4.60}$$\end{document}4.60, F208\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{5.38}$$\end{document}5.38 and D122\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{3.32}$$\end{document}3.32, Q126\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{3.36}$$\end{document}3.36, Y302\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{7.43}$$\end{document}7.43, respectively. The terminal portions of all peptides are in contact with the extracellular loops of the receptor, a structural feature that is believed to confer the isoform selectivity14, together with some elements of the TM bundle41. Despite the generally high flexibility of peptides, all ligands firmly interact with SSTR2, as reported by the average RMSD values below 3 Å (Fig. S5). Although these values indicate a great stability of the binding modes, the RMSF analysis revealed that some portions of the ligands are still highly flexible (Fig. S6). In SST14, besides a high RMSF at the N-terminal, F7 fluctuated significantly during the simulations (Fig. S6A). Similarly, in OCT the N-terminal threoninol and the F3 reached the highest RMSF values (Fig. S6B). CYN shows a high stability, with the exception of one replica, in which it reaches an average RMSD of 4.1 ± 0.4 Å (Fig. S5C). This high value is due to the loss of the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\pi$$\end{document}π-\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\pi$$\end{document}π intramolecular interaction between Y3 and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^D$$\end{document}DY8, which are the residues that fluctuated the most (Figs. S6C, S7). As expected, in all ligands the conserved (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^D$$\end{document}D)Trp and Lys residues were highly stable in all trajectories.
The ligands were also stabilized by intramolecular interactions between the side chains atoms. In particular, F11 of SST14 interacts with F6 and N5, and in turn N5 interacts with T12, while \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^D$$\end{document}DF1 of OCT interacts with F3 and C7 (Fig. S8A,B). The intramolecular interactions of CYN involved Y3 and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^D$$\end{document}DY8, which is favoured by the presence of a cis amide bond that forces the conformation of Y3 towards \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^D$$\end{document}DY8, engaging a persistent aromatic interaction. Further CYN intramolecular interactions can be found between \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^D$$\end{document}DW4 with the N-terminal (i.e., Ac-PPN1-\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^D$$\end{document}DC2), T6 with \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^D$$\end{document}DC2 and Y3, and PPN1 with C7 (Fig. S8C).
The five trajectories of each SSTR2 holo system were first concatenated and then subjected to a cluster analysis (see “Methods”, Table S2). Figure 4 shows the representatives of the most populated cluster for each system, to highlight the different binding modes. In agreement with the low RMSD/F values, these conformations are comparable to those of the corresponding experimental structures (i.e., RMSD of heavy atoms with respect to the experimental structure of SSTR2-SST14: 2.5 Å, SSTR2-OCT: 1.7 Å, SSTR2-CYN: 2.3 Å, Table S2).Figure 4Representatives of the most populated cluster of (A) SSTR2-SST14 ($64.5\%$) in yellow, (B) SSTR2-OCT ($42.9\%$) in green, (C) SSTR2-CYN ($79.1\%$) in magenta. On the left: 3D representation, ECL2 is coloured in orange, the interacting residues are represented in sticks. On the right: 2D interaction diagram, generated by Discovery Studio42.
Interestingly, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sim$$\end{document}∼$79\%$ of the frames of SSTR2-CYN trajectories belong to the most populated cluster, which is the most similar to the experimental binding mode. While the second cluster represents the conformation lacking the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\pi$$\end{document}π-\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\pi$$\end{document}π interaction between Y3 and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^D$$\end{document}DY8 (Fig. S7, RMSD 4.6 Å).
A binding free energy analysis on the conformational clusters extracted from the trajectories was performed using the MM-GBSA method (see “Methods”). Table S2 lists the average values, weighted on the cluster population. The endogenous ligand SST14 is the compound that binds SSTR2 with the higher affinity (− 90.1 ± 10.7 kcal/mol), while OCT and CYN show a comparable binding affinity (− 70.0 ± 12.4 kcal/mol and − 73.4 ± 10.2 kcal/mol, respectively). These results are consistent with experimental data reporting an IC\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$_{50}$$\end{document}50 value of 0.2 nM for SST14, and 0.6 nM for OCT43.
## SSTR2 agonists and antagonist show different interaction patterns
We divided the binding pocket into three different regions, according to the spatial distribution of protein residues: bottom, middle and top. The complete list of residues and their spatial distribution is reported in Table S3. Then, we computed for the three peptides the partial contribution to the binding free energy of residues located in each region (Fig. 5).Figure 5Partial contribution of SSTR2 residues to the MM-GBSA binding free energy (kcal/mol) for SST14, OCT and CYN. Values are reported in kcal/mol according to the pocket regions: (A) top, (B) middle, (C) bottom and ordered according to their spatial localization. Only residues in which at least one contribution reached an energy value of − 1.0 kcal/mol are reported. The cells are colored from the highest to the lowest values of the binding free energy (i.e., from yellow to red).
All peptides strongly interact with F294\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{7.35}$$\end{document}7.35, V298\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{7.39}$$\end{document}7.39 (middle, TM7) and P286 (top, ECL3). SST14 and OCT share a similar interaction pattern, both interacting with F92\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{2.53}$$\end{document}2.53, S279\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{6.58}$$\end{document}6.58 and H107. However, some differences through ligands can be found. SST14 interacts with V121\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{3.31}$$\end{document}3.31, Y205\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{5.35}$$\end{document}5.35, S281\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{6.60}$$\end{document}6.60 and S201 with an energy of 1.3 kcal/mol (or more) lower than OCT and CYN. Similarly, interaction with V118\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{3.28}$$\end{document}3.28, M282 and E200 is stronger for OCT, and with L99\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{2.60}$$\end{document}2.60, A100\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{2.61}$$\end{document}2.61 and V280\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{6.59}$$\end{document}6.59 for CYN.
To further investigate the key SSTR2/peptides contacts we performed an interaction fingerprint analysis (Fig. 6). Generally, most interactions are hydrophobic, involving also aromatic side chains. SST14 interacts mainly (> $30\%$) with residues belonging to ECL3 (I284, S285, P286) in the top region, and with residues belonging to TM5 (Y205\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{5.35}$$\end{document}5.35, F208\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{5.38}$$\end{document}5.38), TM6 (F272\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{6.51}$$\end{document}6.51, F275\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{6.54}$$\end{document}6.54, N276\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{6.55}$$\end{document}6.55, S279\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{6.58}$$\end{document}6.58), and TM7 (K291\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{7.32}$$\end{document}7.32, F294\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{7.35}$$\end{document}7.35, V298\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{7.39}$$\end{document}7.39) in the middle region. Similarly, OCT interacts mainly (>$50\%$) with domains in the top and middle regions: ECL3 (I284, P286), TM5 (Y205\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{5.35}$$\end{document}5.35, F208\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{5.38}$$\end{document}5.38), TM7 (F294\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{7.35}$$\end{document}7.35, V298\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{7.39}$$\end{document}7.39), TM6 (F272\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{6.51}$$\end{document}6.51, N276\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{6.55}$$\end{document}6.55, V280\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{6.59}$$\end{document}6.59) common to SST14, and I195 in the ECL2. Conversely, fewer overlaps are observed for CYN: TM5 (Y205\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{5.35}$$\end{document}5.35, F208\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{5.38}$$\end{document}5.38), TM6 (F272\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{6.51}$$\end{document}6.51) and TM7 (K291\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{7.32}$$\end{document}7.32, F294\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{7.35}$$\end{document}7.35). In this latter case the major interactions (> $50\%$) involve residues peculiar to this peptide: ECL1 (V106), ECL2 (S192, T194, I195), TM2 (Q102\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{2.63}$$\end{document}2.63, V103\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{2.64}$$\end{document}2.64).Figure 6Interaction fingerprints for (A) SSTR2-SST14, (B) SSTR2-OCT, (C) SSTR2-CYN. The numbers indicate the persistence of interaction (%) between the residue of SSTR2 (vertical axis) and that of the ligand (horizontal axis). Darker colors correspond to higher persistences. Interactions are coloured according to their type: yellow to orange for hydrophobic, blue for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\pi$$\end{document}π-\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\pi$$\end{document}π stacking, green for hydrogen bonds, violet for salt bridges, and magenta for cation-\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\pi$$\end{document}π. Only interactions equal or greater than $10\%$ are reported. The residues of SSTR2 are divided according to the pocket regions (i.e., top, middle, bottom) and ordered according to their spatial localization.
Furthermore, the Lys residue in each ligand (K9 for SST14 and K5 for both OCT and CYN) stably interacts with D122\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{3.32}$$\end{document}3.32, Q126\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{3.36}$$\end{document}3.36 and Y302\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{7.43}$$\end{document}7.43, as already reported in literature based on structural data25. Differently, we found that the Trp residue (W8 for SST14 and DW4 for both OCT and CYN) maintains the interaction with I177\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{4.60}$$\end{document}4.60, looses that with F208\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{5.38}$$\end{document}5.38, and gains an additional one, Q126\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{3.36}$$\end{document}3.36 (for SST14 and OCT) or D122\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{3.32}$$\end{document}3.32 (for CYN).
Interestingly, this last difference between SST14/OCT and CYN is not the only one: almost all residues interacting with SST14 are also common to OCT (33 out of 36, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sim$$\end{document}∼$92\%$). Whereas, CYN shared with SST14 and OCT the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sim$$\end{document}∼$81\%$ and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sim$$\end{document}∼$71\%$ of the interacting residues, respectively. Furthermore, by looking at Fig. 5A,B, a similar interaction pattern characterizing SST14 and OCT can be clearly identified throughout the top, middle and bottom regions. Differently, only the bottom region looks comparable in CYN (Fig. 5C).
Noteworthy, the patterns shown in Fig. 5 can contribute explaining the different dynamical behaviour of ECL2 in SSTR2/OCT and SSTR2/CYN systems. By comparing the interaction fingerprints we found the main differences in residues belonging to ECL2, TM1 and TM2, which interact the most with CYN, and in ECL3 and TM6, which on the contrary mostly interact with OCT. This means that OCT is mainly located in the pocket region opposite to that of the bottom of ECL2, resulting in much more freedom of movement for the loop. Differently, CYN resides most of the simulation time near the bottom of ECL2, thus hindering the closure of the loop (Fig. S9).
## Conclusions
Somatostatin receptors, especially the isoform SSTR2, represent a prominent target for NET theranostics. Lately, the burst of experimental structures of SSTR2 has given the chance to study the conformational features of this receptor and, as a consequence, further explore the key determinants of interaction with the endogenous ligand SST14 and its synthetic analogues. In this work we performed multi-copy μs-long MD simulations of SSTR2 in three different forms: active agonist-bound, inactive antagonist-bound, and apo inactive. Our results show that the apo state is characterized by a higher flexibility compared to the holo states, in particular for the EC and IC loops. Thanks to this higher flexibility, through our MD simulations we were able to extensively sample conformations of the apo form that differ to a certain extent from the experimental structure (see Figs. 2, 3). Furthermore, by monitoring the opening and closing movements of ECL2, we found that this loop is able to close upon OCT, but it mainly remains open in the apo form, with the endogenous ligand SST14 and, surprisingly, with the antagonist ligand CYN. In particular, ECL2 closes very rarely upon CYN, although its number of residues and surface area are like those of OCT. This suggests that steric hindrance alone is not the only feature driving ECL2 closure, but the specific residues involved in the ligand/SSTR2 interaction are crucial for orienting the ligands in a position that allows more space for loop movements. The binding free energy analysis combined with the interaction fingerprints reveal that all peptides similarly interact with the buried residues of the binding pocket (e.g., D122\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{3.32}$$\end{document}3.32, Q126\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{3.36}$$\end{document}3.36, F294\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{7.35}$$\end{document}7.35). However, differences were found in the interaction patterns with residues located in the external portion of the pocket (i.e., at the ECLs). These findings confirm that the well-defined pharmacophore region of the pocket is essential for ligand binding, while the interactions with the external residues appear to discriminate between the agonists and the antagonist. Further molecular-level studies for a larger set of compounds are needed to assess the above findings. Furthermore, the two agonists SST14 and OCT shared a similar intra-molecular pattern of interaction, which is different for the antagonist CYN thus reflecting a markedly distinct binding to the receptor.
To the best of our knowledge, this is the first computational study systematically exploring the dynamics of SSTR2 in different conformational states, exploiting the recently released experimental structures. Our findings contribute to drug design efforts towards the discovery of new somatostatin-based compounds for theranostic of neuroendocrine tumours.
## Methods
The starting 3D structures of SSTR2 in complex with SST14, OCT, CYN and in the apo form were retrieved from the PDB IDs 7T10, 7T1124, 7XNA14 and 7UL527. To include missing atoms each structure underwent structure refinement using Modeller10.244. 7XNA was resolved as a chimeric structure comprising the stabilizing endo-1,4-beta-xylanase from Niallia circulans at the ICL3. We removed this portion and modelled the missing six residues (i.e., from S238 to G243). To reduce computational costs we did not include the G-protein in the structures. The ionization state of the residue side chains, the tautomeric states of histidine residues and the Asn/Gln flipping were checked by the H++ server45. The CHARMM-GUI server46 was used to embed the protein into a double layer of phosphatidyl choline (POPC, $70\%$) and cholesterol ($30\%$)47. The system was inserted in an OPC water box48 and neutralized by adding K\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^+$$\end{document}+ and Cl\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^-$$\end{document}- ions, reaching a 0.15 M concentration.
The AmberTools20 software49 was used to assign the force field ff19SB to the protein and to SST1450, lipid17 to POPC and cholesterol51, and the hydrogen mass repartition scheme52. OCT and CYN contain non-standard residues (i.e., \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^D$$\end{document}DF1, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^D$$\end{document}DW4, threoninol for OCT and Phe-NO\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$_2$$\end{document}2, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^D$$\end{document}DC2, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^D$$\end{document}DW4, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^D$$\end{document}DY8 for CYN), therefore we generated the parameters adopting the following procedure: [1] Gaussian16 (Revision A.03)53 was employed to compute the electrostatic potential of non-standard residues (B3LYP/6-31G** level of theory), [2] we fitted the atomic partial charges using Antechamber54 and the RESP method55, [3] the final topological files were created using the ff19SB force field and employing the prepgen and parmchk2 programs49. Since the nitro group contained in Phe-NO\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$_2$$\end{document}2 of CYN is not parametrized in ff19SB, we added the corresponding parameters manually, according to the reference values reported in GAFF256. The PDB structure of CYN reports a cis amide bond between Y3 and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^D$$\end{document}DW4. To simulate this conformation, during the MD simulations we applied dihedral restraints to this bond by imposing the cis form (equilibrium dihedral= 0.0 ± 10.0\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^\circ$$\end{document}∘, force constant= 50.0 kcal/mol/Å\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^2$$\end{document}2). The parm7 and rst7 files of the OCT and CYN are available in the Supporting materials.
Each system underwent an energy minimization combining the steepest-descent and the conjugated gradient algorithms and applying positional restraints on the protein and membrane atoms. Two steps of NVT and four of NPT equilibration followed the minimization, in which the positional restraints were incrementally reduced. We used the Langevin thermostat (1 ps\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$^{-1}$$\end{document}-1 as collision frequency) and the Berendsen barostat (1 Atm), a cutoff of 9 Å, the time step was incremented from 1 to 2 fs with the SHAKE algorithm57, the Particle Mesh Ewald method for long-range electrostatics58. The production run was carried on for 1 μs, using the NPT ensemble and 4 fs as a time step. Five replicas were generated for each system, resulting in a 5 μs overall simulation time. The MD simulations were conducted using the PMEMD module of Amber2049.
CPPTRAJ59 was used to perform the cluster analysis of MD trajectories. In detail, a hierarchical algorithm60 was used to group the frames into four conformational clusters, according to the ligand RMSD.
Binding free energy calculations using the MM-GBSA method implemented in Amber2061 were applied to each cluster (see Supporting Information for additional details on the method). Interaction fingerprints were computed using the ProLIF Python library62 on all the frames of the MD trajectories. The numbers of interactions were combined for all replicas and converted into persistence of interactions (%).
## Supplementary Information
Supplementary Information. The online version contains supplementary material available at 10.1038/s41598-023-31823-1.
## References
1. Günther T. **International Union of Basic and Clinical Pharmacology. CV. Somatostatin receptors: Structure, function, ligands, and new nomenclature**. *Pharmacol. Rev.* (2018.0) **70** 763-835. DOI: 10.1124/pr.117.015388
2. Fani M, Mansi R, Nicolas GP, Wild D. **Radiolabeled somatostatin analogs—a continuously evolving class of radiopharmaceuticals**. *Cancers* (2022.0) **14** 1172. DOI: 10.3390/cancers14051172
3. Börzsei R. **Exploration of somatostatin binding mechanism to somatostatin receptor subtype 4**. *Int. J. Mol. Sci.* (2022.0) **23** 6878. DOI: 10.3390/ijms23136878
4. Shamsi BH, Chatoo M, Xu XK, Xu X, Chen XQ. **Versatile functions of somatostatin and somatostatin receptors in the gastrointestinal system**. *Front. Endocrinol.* (2021.0) **12** 652363. DOI: 10.3389/fendo.2021.652363
5. Isberg V. **Generic GPCR residue numbers—aligning topology maps minding the gaps**. *Trends Pharmacol. Sci.* (2015.0) **36** 22-31. DOI: 10.1016/j.tips.2014.11.001
6. Rocheville M. **Subtypes of the somatostatin receptor assemble as functional homo- and heterodimers**. *J. Biol. Chem.* (2000.0) **275** 7862-9. DOI: 10.1074/jbc.275.11.7862
7. Elf A-K. **Evaluation of SSTR2 expression in SI-NETs and relation to overall survival after PRRT**. *Cancers* (2021.0) **13** 2035. DOI: 10.3390/cancers13092035
8. Qian ZR. **Association between somatostatin receptor expression and clinical outcomes in neuroendocrine tumors**. *Pancreas* (2016.0) **45** 1386-1393. DOI: 10.1097/MPA.0000000000000700
9. Heidari P. **Somatostatin receptor type 2 as a radiotheranostic PET reporter gene for oncologic interventions**. *Theranostics* (2018.0) **8** 3380-3391. DOI: 10.7150/thno.24017
10. Gomes-Porras M, Cárdenas-Salas J, Álvarez Escolá C. **Somatostatin analogs in clinical practice: A review**. *Int. J. Mol. Sci.* (2020.0) **1682** 21. DOI: 10.3390/ijms21051682
11. Paragliola RM, Salvatori R. **Novel somatostatin receptor ligands therapies for acromegaly**. *Front. Endocrinol.* (2018.0) **9** 78. DOI: 10.3389/fendo.2018.00078
12. Vitale G. **Pasireotide in the treatment of neuroendocrine tumors: A review of the literature**. *Endocr. Relat. Cancer* (2018.0) **25** R351-R364. DOI: 10.1530/ERC-18-0010
13. Chen L-N. **Structures of the endogenous peptide- and selective non-peptide agonist-bound SSTR2 signaling complexes**. *Cell Res.* (2022.0) **32** 785-788. DOI: 10.1038/s41422-022-00669-z
14. Zhao W. **Structural insights into ligand recognition and selectivity of somatostatin receptors**. *Cell Res.* (2022.0) **32** 761-772. DOI: 10.1038/s41422-022-00679-x
15. Bo Q. **Structural insights into the activation of somatostatin receptor 2 by cyclic SST analogues**. *Cell Discov.* (2022.0) **8** 47. DOI: 10.1038/s41421-022-00405-2
16. Harris AG. **Somatostatin and somatostatin analogues: Pharmacokinetics and pharmacodynamic effects**. *Gut* (1994.0) **35** S1-4. DOI: 10.1136/gut.35.3_Suppl.S1
17. Yang L. **Synthesis and biological activities of potent peptidomimetics selective for somatostatin receptor subtype 2**. *Proc. Natl. Acad. Sci. USA* (1998.0) **95** 10836-41. DOI: 10.1073/pnas.95.18.10836
18. Karimian N. **Somatostatin receptor type 2 antagonism improves glucagon counterregulation in biobreeding diabetic rats**. *Diabetes* (2013.0) **62** 2968-77. DOI: 10.2337/db13-0164
19. Fani M, Nicolas GP, Wild D. **Somatostatin receptor antagonists for imaging and therapy**. *J. Nucl. Med.* (2017.0) **58** 61S-66S. DOI: 10.2967/jnumed.116.186783
20. Koustoulidou S. **Synthesis and evaluation of two long-acting SSTR2 antagonists for radionuclide therapy of neuroendocrine tumors**. *Pharmaceuticals* (2022.0) **15** 1155. DOI: 10.3390/ph15091155
21. Cai Z. **64Cu-labeled somatostatin analogues conjugated with cross-bridged phosphonate-based chelators via strain-promoted click chemistry for PET imaging: In silico through in vivo studies**. *J. Med. Chem.* (2014.0) **57** 6019-29. DOI: 10.1021/jm500416f
22. Nagarajan SK, Babu S, Sohn H, Devaraju P, Madhavan T. **Toward a better understanding of the interaction between somatostatin receptor 2 and its ligands: A structural characterization study using molecular dynamics and conceptual density functional theory**. *J. Biomol. Struct. Dyn.* (2019.0) **37** 3081-3102. DOI: 10.1080/07391102.2018.1508368
23. Nagarajan SK. **Understanding the influence of lipid bilayers and ligand molecules in determining the conformational dynamics of somatostatin receptor 2**. *Sci. Rep.* (2021.0) **11** 7677. DOI: 10.1038/s41598-021-87422-5
24. Robertson MJ, Meyerowitz JG, Panova O, Borrelli K, Skiniotis G. **Plasticity in ligand recognition at somatostatin receptors**. *Nat. Struct. Mol. Biol.* (2022.0) **29** 210-217. DOI: 10.1038/s41594-022-00727-5
25. Heo Y. **Cryo-EM structure of the human somatostatin receptor 2 complex with its agonist somatostatin delineates the ligand-binding specificity**. *Elife* (2022.0) **11** e76823. DOI: 10.7554/eLife.76823
26. Chen S, Teng X, Zheng S. **Molecular basis for the selective G protein signaling of somatostatin receptors**. *Nat. Chem. Biol.* (2022.0). DOI: 10.1038/s41589-022-01130-3
27. Robertson MJ. **Structure determination of inactive-state GPCRs with a universal nanobody**. *Nat. Struct. Mol. Biol.* (2022.0). DOI: 10.1038/s41594-022-00859-8
28. Latorraca NR, Venkatakrishnan AJ, Dror RO. **GPCR dynamics: Structures in motion**. *Chem. Rev.* (2017.0) **117** 139-155. DOI: 10.1021/acs.chemrev.6b00177
29. Zhou Q. **Common activation mechanism of class A GPCRs**. *Elife* (2019.0) **8** e50279. DOI: 10.7554/eLife.50279
30. Hauser AS. **GPCR activation mechanisms across classes and macro/microscales**. *Nat. Struct. Mol. Biol.* (2021.0) **28** 879-888. DOI: 10.1038/s41594-021-00674-7
31. Weis WI, Kobilka BK. **The molecular basis of G protein-coupled receptor activation**. *Annu. Rev. Biochem.* (2018.0) **87** 897-919. DOI: 10.1146/annurev-biochem-060614-033910
32. Heydenreich FM, Vuckovic Z, Matkovic M, Veprintsèv DB. **Stabilization of G protein-coupled receptors by point mutations**. *Front. Pharmacol.* (2015.0) **6** 82. DOI: 10.3389/fphar.2015.00082
33. Grisshammer R. **The quest for high-resolution G protein-coupled receptor-G protein structures**. *Proc. Natl. Acad. Sci. USA* (2020.0) **117** 6971-6973. DOI: 10.1073/pnas.2002665117
34. Grisshammer R. **New approaches towards the understanding of integral membrane proteins: A structural perspective on G protein-coupled receptors**. *Protein Sci.* (2017.0) **26** 1493-1504. DOI: 10.1002/pro.3200
35. Salom D, Padayatti PS, Palczewski K. **Crystallization of G protein-coupled receptors**. *Methods Cell Biol.* (2013.0) **117** 451-468. DOI: 10.1016/B978-0-12-408143-7.00024-4
36. Wacker D, Stevens RC, Roth BL. **How ligands illuminate GPCR molecular pharmacology**. *Cell* (2017.0) **170** 414-427. DOI: 10.1016/j.cell.2017.07.009
37. Nicoli A, Dunkel A, Giorgino T, de Graaf C, Pizio AD. **Classification model for the second extracellular loop of class A GPCRs**. *J. Chem. Inf. Model* (2022.0) **62** 511-522. DOI: 10.1021/acs.jcim.1c01056
38. Woolley MJ, Conner AC. **Understanding the common themes and diverse roles of the second extracellular loop (ECL2) of the GPCR super-family**. *Mol. Cell Endocrinol.* (2017.0) **449** 3-11. DOI: 10.1016/j.mce.2016.11.023
39. Connolly ML. **Analytical molecular surface calculation**. *J. Appl. Cryst.* (1983.0) **16** 548-558. DOI: 10.1107/S0021889883010985
40. Lu S. **Activation pathway of a G protein-coupled receptor uncovers conformational intermediates as targets for allosteric drug design**. *Nat. Commun.* (2021.0) **12** 4721. DOI: 10.1038/s41467-021-25020-9
41. Lee SM. **Structural insights into ligand recognition and selectivity for classes A, B, and C GPCRs**. *Eur. J. Pharmacol.* (2015.0) **763** 196-205. DOI: 10.1016/j.ejphar.2015.05.013
42. 42.Systémes, B. D. Discovery Studio, 12.1.0. (Dassault Systémes, 2021).
43. Anthony L, Freda PU. **From somatostatin to octreotide LAR: Evolution of a somatostatin analogue**. *Curr. Med. Res. Opin.* (2009.0) **25** 2989-99. DOI: 10.1185/03007990903328959
44. Eswar N. **Comparative protein structure modeling using Modeller**. *Curr. Protoc. Bioinformatics* (2006.0) **15** 5.6.1-5.6.30. DOI: 10.1002/0471250953.bi0506s15
45. Gordon JC. **H++: A server for estimating pKas and adding missing hydrogens to macromolecules**. *Nucleic Acids Res.* (2005.0) **33** W368-71. DOI: 10.1093/nar/gki464
46. Jo S, Kim T, Iyer VG, Im W. **CHARMM-GUI: A web-based graphical user interface for CHARMM**. *J. Comput. Chem.* (2008.0) **29** 1859-65. DOI: 10.1002/jcc.20945
47. Saeedimasine M, Montanino A, Kleiven S, Villa A. **Role of lipid composition on the structural and mechanical features of axonal membranes: A molecular simulation study**. *Sci. Rep.* (2019.0) **9** 8000. DOI: 10.1038/s41598-019-44318-9
48. Izadi S, Anandakrishnan R, Onufriev AV. **Building water models: A different approach**. *J. Phys. Chem. Lett.* (2014.0) **5** 3863-3871. DOI: 10.1021/jz501780a
49. 49.Case, D. et al. Amber (University of California, 2022).
50. Tian C. **ff19SB: Amino-acid-specific protein backbone parameters trained against quantum mechanics energy surfaces in solution**. *J. Chem. Theory Comput.* (2020.0) **16** 528-552. DOI: 10.1021/acs.jctc.9b00591
51. 51.Gould, I., Skjevik, A., Dickson, C., Madej, B. & Walker, R. Lipid17: A comprehensive AMBER force field for the simulation of zwitterionic and anionic lipids. in prep. (2018).
52. Hopkins CW, Grand SL, Walker RC, Roitberg AE. **Long-time-step molecular dynamics through hydrogen mass repartitioning**. *J. Chem. Theory Comput.* (2015.0) **11** 1864-1874. DOI: 10.1021/10.1021/ct5010406
53. 53.Frisch, M. J. et al. Gaussian 16, (Gaussian, Inc.).
54. Wang J, Wang W, Kollman PA, Case DA. **Automatic atom type and bond type perception in molecular mechanical calculations**. *J. Mol. Graph. Model.* (2006.0) **25** 247-60. DOI: 10.1016/j.jmgm.2005.12.005
55. Singh UC, Kollman PA. **An approach to computing electrostatic charges for molecules**. *J. Comput. Chem.* (1984.0) **5** 129-145. DOI: 10.1002/jcc.540050204
56. Wang J, Wolf RM, Caldwell JW, Kollman PA, Case DA. **Development and testing of a general amber force field**. *J. Comput. Chem.* (2004.0) **25** 1157-74. DOI: 10.1002/jcc.20035
57. Kräutler V, van Gunsteren WF, Hünenberger PH. **A fast SHAKE algorithm to solve distance constraint equations for small molecules in molecular dynamics simulations**. *J. Comput. Chem.* (2001.0) **22** 501-508. DOI: 10.1002/1096-987X(20010415)22:5<501::AID-JCC1021>3.0.CO;2-V
58. Darden T, York D, Pedersen L. **Particle mesh Ewald: An N**. *J. Chem. Phys.* (1993.0) **98** 10089. DOI: 10.1063/1.464397
59. Roe DR, Cheatham TE. **PTRAJ and CPPTRAJ: Software for processing and analysis of molecular dynamics trajectory data**. *J. Chem. Theory Comput.* (2013.0) **9** 3084-95. DOI: 10.1021/ct400341p
60. Shao J, Tanner SW, Thompson N, Cheatham TE. **Clustering molecular dynamics trajectories: 1. Characterizing the performance of different clustering algorithms**. *J. Chem. Theory Comput.* (2007.0) **3** 2312-34. DOI: 10.1021/ct700119m
61. Miller BR. **MMPBSA.py: An efficient program for end-state free energy calculations**. *J. Chem. Theory Comput.* (2012.0) **8** 3314-21. DOI: 10.1021/ct300418h
62. Bouysset C, Fiorucci S. **ProLIF: A library to encode molecular interactions as fingerprints**. *J. Cheminform.* (2021.0) **13** 72. DOI: 10.1186/s13321-021-00548-6
|
---
title: Comparison of hepatic responses to glucose perturbation between healthy and
obese mice based on the edge type of network structures
authors:
- Yuki Ito
- Shinsuke Uda
- Toshiya Kokaji
- Akiyoshi Hirayama
- Tomoyoshi Soga
- Yutaka Suzuki
- Shinya Kuroda
- Hiroyuki Kubota
journal: Scientific Reports
year: 2023
pmcid: PMC10036622
doi: 10.1038/s41598-023-31547-2
license: CC BY 4.0
---
# Comparison of hepatic responses to glucose perturbation between healthy and obese mice based on the edge type of network structures
## Abstract
Interactions between various molecular species in biological phenomena give rise to numerous networks. The investigation of these networks, including their statistical and biochemical interactions, supports a deeper understanding of biological phenomena. The clustering of nodes associated with molecular species and enrichment analysis is frequently applied to examine the biological significance of such network structures. However, these methods focus on delineating the function of a node. As such, in-depth investigations of the edges, which are the connections between the nodes, are rarely explored. In the current study, we aimed to investigate the functions of the edges rather than the nodes. To accomplish this, for each network, we categorized the edges and defined the edge type based on their biological annotations. Subsequently, we used the edge type to compare the network structures of the metabolome and transcriptome in the livers of healthy (wild-type) and obese (ob/ob) mice following oral glucose administration (OGTT). The findings demonstrate that the edge type can facilitate the characterization of the state of a network structure, thereby reducing the information available through datasets containing the OGTT response in the metabolome and transcriptome.
## Introduction
Biological phenomena include the interactions between various biochemical molecular species to form network structures. Therefore, an in-depth investigation of the network structures corresponding to statistical interactions, including biochemical interactions, will permit better understanding of biological phenomena1,2. In particular, when the biochemical reaction network is unknown, the network formed by statistical interactions that have been inferred from a dataset provides substantial information toward an understanding of these biological phenomena. Moreover, the clustering of nodes corresponding to molecular species and enrichment analysis are frequently used to determine the biological significance of network structures2–4. However, as these methods focus on the function of the nodes, the role of the edges, or the connections between nodes, has rarely been studied.
We consider that it is equally important to investigate the functions performed by closely connected nodes and the type of functional connections present within network structures. Moreover, additional information regarding the functions of the edges, such as the composition of the category of each edge as an indicator of the biological characteristics of the network, is helpful for examining the characteristics of the network structure.
To determine the nature of the functional connection, it is also necessary to perform functional annotation of the edges in the network. A pioneering study of “edge ontology,” to functionally annotate network edges, was reported by Lu et al. in5. Several classical pathway databases have been established over the past decade to document the published information on established pathways6,7. However, two problems arise when these classical pathways are documented. First, each database contains unique components of the same pathway. Second, as each pathway is visualized independently, the crosstalk between pathways remains unknown. Alternately, modern high-throughput techniques, such as large-scale yeast two-hybrid screens8–12 and large-scale databases, have allowed the construction of standard large-scale networks documenting protein–protein interactions; this development has ushered in a different perspective on pathway studies, including quantitative insights from the network structure analysis1,2.
Lu et al.5 attempted to embed an established classical pathway within a large-scale network; they embedded classical pathways in large-scale networks by superimposing classical pathways, including their edges, to preserve the informative edges associated with classical pathways. They attempted to elaborate on the process of network integration in this manner. However, as the same edge symbols had different meanings, the precise definition of the edge for the overlay was ambiguous. Thus, to conduct large-scale pathway mining, there is a critical need to develop accurate edge ontology that can represent the various types of relationships between pathway components. Lu et al. represented various edge types consistently and developed an edge ontology for their classification. Consequently, their proposed edge ontology enabled them to delineate four distinct types of pathways, as well as provide additional information demonstrating the significance of the edge ontology.
From the findings of Lu et al., we hypothesized that the network structure could be characterized by quantifying the distribution of the edge ontology, which could serve as a novel index for comparing network structures. For this purpose, we defined the edge ontology of the metabolites and genes to analyze the network structures. However, the use of edge ontology analysis has not been reported for multiple layers and metabolites. Additionally, the edge ontology analysis defined in this study differs slightly from conventional edge ontology. Consequently, to avoid confusion, it is hereafter referred to as “edge type (ET).” The network structure is an integral part of ET-based network analysis. Generally, it is difficult for an “omics” measurement dataset to construct an entire network structure from biological knowledge. Therefore, it was critical to infer the network structure from the omics dataset before the ET analysis on the network was performed. In contrast to other methods using ordinary differential equations13–15 or regression to infer the network structure16, we used mutual information17.
Blood glucose levels are strictly maintained and impaired glucose tolerance leads to various diseases18,19. The liver plays an essential role in maintaining blood glucose levels, and part of the control of the metabolic response depends on the regulation of hepatic gene expression19–21. A recent study examined hepatic glucose metabolism using multiomics measurements22,23. However, the experimental design was inadequate for the examination of statistical independence. We retrieved metabolome and transcriptome data, previously acquired by our group, from 11 or 12 healthy mice (WT) prior to and following an oral glucose tolerance test (OGTT), which is a standard test used to diagnose diabetes and in which glucose administration was performed to determine if the patient has impaired glucose tolerance24. In this study, to gain a better understanding of the comprehensive hepatic responses to glucose perturbation, we examined the network difference between prior to and following the OGTT using ET. Additionally, we used obese (ob/ob) mice, which are leptin-deficient mutants with abnormal glucose metabolism25,26. Our comparison of the network responses of the OGTT in ob/ob mice with those in healthy mice based on the ETs was intended to reveal the anomalous response to glucose perturbation in obese mice. We first inferred the network structure using mutual information in WT and ob/ob mice and determined the difference network between prior to and following the OGTT, which we considered as the OGTT response network. The OGTT response networks were compared between WT and ob/ob mice and characterized abnormal glucose metabolism by the ETs. In addition, the distribution of ET was obtained by counting the ET that appears in the network. Thus, we propose that the distribution of ET can serve as a novel index to characterize the state of the network for each condition in a reducible manner and we demonstrate the utility of this index through a comparison of the network responses to the OGTT in ob/ob mice with those in healthy mice. Because the distribution of ET is almost unaffected by the local structure of the network, a robust characterization and comparison of network states can be achieved.
## Overview of our network analysis based on edge type
In this study, we attempted to use the edge types of a network to determine the difference between conditions: prior to and following OGTT, WT and ob/ob, and the metabolome and transcriptome. We used eight datasets: WT prior to OGTT, WT following OGTT, ob/ob prior to OGTT, and ob/ob following OGTT, for the metabolome and transcriptome, respectively. First, we inferred the network structure using mutual information prior to and following the OGTT in WT metabolome, referred to as the OGTT [0] and OGTT [4] networks, respectively (Fig. 1A,B). The nodes are connected by the edges, given that the mutual information17 between nodes is not significantly 0 in terms of statistical hypothesis test ($p \leq 0.05$, permutation test). Subsequently, the OGTT response network of the WT metabolome was defined as the difference between the OTTT [0] and the OGTT [4] network (Fig. 1B). Similarly, we defined the OGTT response network of the ob/ob metabolome as the difference between the OGTT [0] and the OGTT [4] networks. Then, we examined the abnormal OGTT response in ob/ob mice by comparing the OGTT response network between WT and ob/ob. The OGTT response networks of WT and ob/ob were compared in the transcriptome as well. The abnormal glucose metabolism in ob/ob mice was precisely characterized by the ETs. In the ET analysis, nodes were categorized according to the information from the KEGG database, and each ET was determined by both the node types at each end (Fig. 1C). The edges of the OGTT response networks were categorized into ETs, and the edges’ response to OGTT for each ET was compared between WT and ob/ob, as well as between metabolome and transcriptome data (Fig. 1D–G). Moreover, we obtained the distribution of ET by sorting the edges in a network of one condition into each ET, and summarized the difference network using the distribution of ET as an indicator (Fig. 1H–K). Consequently, it was found that the state of the network structure is reflected in the distribution of ETs, and that the characteristics of the state can be reduced to the distribution of ETs. The algorithmic procedure is described in “Materials and methods”. Figure 1Schematic representation of the network inference, the network comparison, the definition of the edge type, and the comparison of ET distribution. ( A) Evaluation of the relationships between the molecular species (metabolite or gene) by statistical independence. The nodes are connected by the edges, given that the mutual information17 between nodes is not statistically 0 ($p \leq 0.05$, permutation test). ( B) Inference of the OGTT [0] network, the OGTT [4] network, and the OGTT response network and the comparison of the OGTT response network between WT and ob/ob. OGTT [0] and OGTT [4] represent for the conditions prior to and following the OGTT, respectively. ( C) Definition of ETs. ( D–G) Comparison of the OGTT response networks using the defined ETs. ( H–K) Comparison of each condition’s networks using the ET distribution. ( D, H) The OGTT response network and the network of OGTT [0] and OGTT [4]. Nodes with different colors are different node types. ( E) The number of the OGTT response edges. “ P → P,” “A → P,” “P → A,” and “A → A” stand for the edges that were are common to the OGTT [0] and OGTT [4] networks, the edges only existing in the OGTT [4] network, the edges only existing in the OGTT [0] network and the edges existing in neither OGTT [0] nor OGTT [4] networks, respectively. ( F) The ET distribution of each OGTT response edge. An example for P → P edges is shown. ( G) The comparison of OGTT response edge distribution between each OGTT response network. # 1 and #2 stand for the OGTT response network 1 and 2, respectively. ( I) The number of each condition’s edges. ( J) The ET distribution of each condition’s edges. An example for OGTT [0] is shown. ( K) The comparison of ET distribution between each condition.
## Distinct individual variations are present in WT and ob/ob mice
We had previously performed metabolomics and transcriptomics analyses of liver tissue from WT and ob/ob mice prior to ($$n = 11$$ for WT, $$n = 12$$ for ob/ob) and 4 h after ($$n = 12$$) an OGTT to determine the liver’s response to glucose perturbation24. We identified 262 metabolites and 22,463 genes from the metabolomics and transcriptomics analyses, respectively. OGTT [0] and OGTT [4] represent the conditions prior to and following the OGTT, respectively.
We performed principal component analysis to examine the outline of the data and found that WT formed a distinct group from ob/ob in the subspace of the first principal component, regardless of the OGTT, in the metabolome and transcriptome data (Fig. S1). This finding indicated that the metabolome and transcriptome data contained sufficient information to distinguish WT from ob/ob. In contrast, although the OGTT [0] and OGTT [4] in the WT and ob/ob groups could not be distinguished, minor variations in the metabolome were observed between individuals each of these groups. We observed minor variations in the OGTT [0] of WT and ob/ob groups and some large variations in the OGTT [4] of WT and ob/ob groups in the transcriptome. Thus, we found that the WT and ob/ob individuals were distinctly grouped, but individuals within these groups displayed variations in their metabolome and transcriptome.
## The OGTT response, based on the edges between WT and ob/ob, comprised considerable differences
In the metabolome, the mutual information of each pairwise metabolite was investigated under the following four conditions: WT OGTT [0], WT OGTT [4], ob/ob OGTT [0], and ob/ob OGTT [4]. The network structure was inferred depending on whether the mutual information was positive or not. Mutual information between two nodes equaling zero indicates that they are statistically independent. Given that mutual information is statistically positive ($p \leq 0.05$, permutation test), the two nodes with estimated mutual information are connected by an edge (see “Materials and methods”, Fig. 2). Similarly, we examined the mutual information for the transcriptome in the four conditions. Our investigation revealed that the metabolomic and transcriptomic data used for network inference contained 136 metabolites and 12,415 genes that were quantified in the four previously mentioned conditions. Figure 2The OGTT response network. The OGTT response network of the metabolome (left column) and transcriptome (right column) in WT (upper row) and ob/ob (lower row). All of the 136 metabolites and the 136 genes randomly extracted from the 12,415 genes used for analysis were visualized. The blue, yellow, and green edge represents the edge that existed only in the OGTT [0] network, only in the OGTT [4] network, and that are common to both the networks, respectively.
Subsequently, we examined the node degree distribution of the networks across the four distinct groups (Fig. S2). Our investigations revealed that the node degree distribution of OGTT [0] in the metabolome was not significantly different between WT and ob/ob (Table S1, Bonferroni corrected $p \leq 0.05$). In contrast, in OGTT [4], the networks for WT and ob/ob had significantly different distributions, with a wider tail and slightly narrower distribution, respectively. These findings suggested that the OGTT induced the network differences between WT and ob/ob and that the OGTT-induced changes in the structure of the metabolome network may be qualitatively distinct. In contrast, in the transcriptome, the node degree distribution in OGTT [0] was significantly different between WT and ob/ob, being bimodal in OGTT [0] in ob/ob, whereas that was unimodal in WT. In OGTT [4], the distribution of WT had a wider tail than the distribution of ob/ob, which was unimodal with a narrower tail; moreover, the difference between the two groups was significant. Thus, when compared with the metabolome, the transcriptomic analysis revealed a significant difference in the network structure between WT and ob/ob in OGTT [0]. In contrast, the effect of the OGTT on the structural change in the transcriptome network was significantly different between WT and ob/ob, similar to the metabolome network. Furthermore, only one node of zero degree was found in the metabolome of WT OGTT [4] and ob/ob OGTT [4].
In this study, we defined full-connection edges as the edges of a network of fully connected nodes (Fig. 3A). The number of full-connection edges was equal to the maximum number of edges of the network, which is NC2, where N is the number of nodes in the network. The metabolome and transcriptome had a total of 9,180 and 77,059,905 full-connection edges, respectively. We compared the number of estimated edges to the number of full-connection edges to determine the graph density in networks with varying node sizes. In the metabolome, the graph density was $13.28\%$ [1,219] for WT OGTT [0], $19.14\%$ [1,757] for WT OGTT [4], $13.74\%$ [1,261] for ob/ob OGTT [0], and $10.38\%$ [953] for ob/ob OGTT [4] (Fig. 3A, upper panel). In the transcriptome, the graph density was $14.11\%$ [10,874,216] for WT OGTT [0], $17.30\%$ [13,334,125] for WT OGTT [4], $23.79\%$ [18,330,266] for ob/ob OGTT [0], and $10.73\%$ [8,271,366] for ob/ob OGTT [4] (Fig. 3A, lower panel). Additionally, the OGTT increased the number of edges in WT metabolome and transcriptome, whereas the opposite effect was found in ob/ob. This finding was consistent with the observation that the tails of the WT and ob/ob node degree distributions became wider and narrower, respectively, in both the metabolome and transcriptome (Fig. S2). Collectively, these observations indicate that the edge alterations induced by the OGTT resulted in a change in the tail of the node degree distribution. Figure 3The edge amount quantification of each network. The graph density in each condition; (A) the graph density of the OGTT response network; (B) the comparison of the graph density of the OGTT response network between WT and ob/ob; (C) the graph density in the metabolome (upper panel) and transcriptome (lower panel). The middle panel in (A–C) are supplementary schematic diagrams that helps to interpret each graph. ( A) The graph density of the OGTT [0] and OGTT [4] networks in WT and ob/ob. ( B) The changes in the presence and absence of edges between the OGTT [0] and OGTT [4]. “ P → P,” “A → P,” “P → A,” and “A → A” stand for the edges, which are common to the OGTT [0] and OGTT [4] networks, the edges only existing in the OGTT [4] network, the edges only existing in the OGTT [0] network and the edges existing in neither OGTT [0] nor OGTT [4] networks, respectively. ( C) The comparison between WT and ob/ob in the graph density of each edge explained in figure (B).
Subsequently, we examined the differences between the WT and ob/ob in the presence and absence of edges between OGTT [0] and OGTT [4] (Figs. 1B, 3B). The OGTT response network was defined as the difference in network between OGTT [0] and OGTT [4]. “ P to P” denotes the edges that were shared by the OGTT [0] and OGTT [4] networks, “A to P” denotes the edges that were unique to the OGTT [4] network, “P to A” denotes the edges that were unique to the OGTT [0] network, and “A to A” denotes the edges existing in neither OGTT [0] nor OGTT [4] network. ( in the figure, “to” is indicated by an arrow). Additionally, P to A, A to P, and P to P edges are considered as the OGTT response edges. We found that P to A and A to P edges existed in the metabolome and transcriptome of both WT and ob/ob. Thus, we calculated the net changes in the total number of edges for each individual. In the metabolome, the graph density of P to P edges was higher for WT than for ob/ob, suggesting that the former was possibly more resistant to glucose perturbation (Fig. 3B). Furthermore, we examined the number of the OGTT response edges that were identical between the WT and ob/ob, based on the graph density (Fig. 3C). More than half of all edges in both the metabolome and transcriptome were A to A in both groups. This implied that more than half of all edges were absent and were unchanged by the OGTT. Thus, we found that the same changes in the OGTT response between WT and ob/ob occurred in a limited part of the metabolome and transcriptome networks. Furthermore, with the exception of the A to A edges, the OGTT responses varied substantially between the WT and ob/ob groups.
## Difference in mutual information for most P to P edges is nonsignificant
As described previously, we determined the network structure using mutual information by examining statistical independence. The value of mutual information corresponding to the strength of edge connections can aid in understanding the network structure. However, the small sample size of the dataset made the value of the mutual information unreliable. Thus, we performed a statistical test to determine whether the mutual information values of two nodes were different between the networks (see “Materials and methods”).
For the P to P edges of the metabolome, our investigation revealed that there were 16 edges in WT ($0.17\%$ full-connection edges and $3.63\%$ P to P edges) and 6 edges in ob/ob ($0.07\%$ full-connection edges and $2.41\%$ P to P edges), with a significant difference for the OGTT (Table 1). In the transcriptome, there were 105,076 edges in WT ($0.14\%$ full-connection edges and $4.20\%$ P to P edges) and 73,021 edges in ob/ob ($0.09\%$ full-connection edges and $2.85\%$ P to P edges), with a significant difference for the OGTT. The ratio of P to P edges with different mutual information values was low in both WT and ob/ob in the metabolome and in the transcriptome. This result indicates that the major difference between the OGTT [0] and OGTT [4] networks was the structure rather than the strength of the connections. However, owing to the small sample size, it is possible that the number of edges with significant differences between OGTT [0] and OGTT [4] was underestimated. Table 1The number of the edges with the difference in the value of mutual information. MetabolomeTranscriptomeWT16WT105,076ob/ob6ob/ob73,021The number of the edges, which showed significantly different value of mutual information ($p \leq 0.05$, permutation test) in the P to P edges (described in Fig. 3B) in the metabolome (left) and transcriptome (right), respectively.
## ET characterizes the state of the network structure
Each node can be annotated with the biological property of the molecular species. Using node annotation, we defined the ET by categorizing the combination of annotations of pairwise connected nodes. Given that the properties of a molecular species are related to its function or role, the edges categorized as the same ET have similar properties and functions. We obtained the distribution of ETs for each condition based on the inferred network structure (see “Materials and methods”, Fig. 1H–K). Consequently, the statistical properties of the ET distribution will help to explain the relationship between network structures and biological phenomena.
One of the benefits of introducing ETs is the robust extraction of the network structure characteristics for comparison. To examine the robust extraction, we defined the graph density in which the counted edges are limited to a specific ET as the “ET graph density.” Although the threshold of the p-value affects the inferred network structure, when the threshold of the p-value is perturbed around 0.05, the Spearman correlation of ET graph density changes less than the conventional network statistics that are frequently used and estimated from network structure without using ETs, such as characteristic path length and cluster coefficient (Fig. 4A). This finding demonstrates that the rank of ET graph density is more robust than the conventional network statistics. Although the ET graph density is robust to perturbation of the p-value threshold, it highlights the difference between the network structure states. The ET graph density changes dynamically in response to the condition, whereas the conventional network statistics, other than the degree centrality, change very little compared with the ET graph density (Fig. 4B). However, degree centrality is not robust to perturbation of the p-value threshold. Thus, we concluded that the ET graph density enables robust network structure comparisons. Figure 4Comparison of the ET graph density and the conventional network statistics. The relative change of Spearman correlation coefficient of the ET graph density and the conventional network statistics against the $p \leq 0.05$ network in each condition (A) and of different conditions against the WT OGTT [0] network (B) in the metabolome (left column) and transcriptome (right column). Each value is normalized such that the value of $p \leq 0.05$ becomes one in (A) and that of WT OGTT [0] become one in (B). ET, Charc. path len., Cluster coef., Dgree cent., Closeness cent., and Betweenness cent. stand for absolute value of Spearman correlation coefficient of ET graph density, characteristic path length, cluster coefficient, the average of degree centrality, the average of closeness centrality, and the average of betweenness centrality, respectively.
We examined the ET graph density of P to A, A to P, and P to P edges in the OGTT response network of WT and ob/ob (Fig. 5A, see also Fig. 1D–F). Based on the definitions in the KEGG database, metabolites and genes were annotated into 11 types, and we defined 66 ETs based on the combination of annotations for pairwise nodes (see “Materials and methods”). Additionally, the ET graph density of P to A, A to P, and P to P edges in each ET were examined in the metabolome and transcriptome (Fig. S3). Each ET is indicated by a number (No.), along with coupled annotations of the pairwise nodes that correspond to it. Coupled annotations for pairwise nodes are denoted by the initials of each node annotation separated by a hyphen (A: amino acid metabolism, C: carbohydrate metabolism, E: energy metabolism, L: lipid metabolism, and N: nucleotide metabolism). For example, the edge between amino acid metabolism and carbohydrate metabolism is represented as ET No. 2 (A–C).Figure 5The ratio of OGTT response edge and comparison of the distribution of OGTT response edge in each edge type. ( A) The ET graph density of each OGTT response edge (A to P, P to P, and P to A) in each ET in the metabolome (left column) and transcriptome (right column) of WT (upper row) and ob/ob (lower row), respectively. ( B) The Chi-squared histogram distance of distributions of the ET graph density for the OGTT response edges between WT and ob/ob. ( C) The Chi-squared histogram distance of distribution of the ET graph density for the OGTT response edges between the metabolome and transcriptome.
Subsequently, we focused on 15 of the total of 66 ETs that were combinations of the major metabolite groups. Almost all ETs in the WT metabolome followed a similar pattern to the overall edges (Fig. 3B), with A to P edges having a higher ET graph density than P to A edges (Fig. 5A). ET No. 1 (A–A), 13 (L–L), and 15 (N–N) were notable in that their ET graph density of P to P edges was comparatively greater than that of other ETs. It would be of biological interest if associations between metabolites with similar biological characteristics were robust prior to and following the OGTT.
Almost all ETs in the ob/ob metabolome followed a similar pattern to the overall edges, with P to A edges having a higher ET graph density than A to P edges. ET No. 1 (A–A) and 15 (N–N) were distinctive in that the ET graph density of P to P edges was comparatively greater than that of other ETs. A high ET graph density for these ETs was also found for the WT metabolome. This suggested that the ETs unaffected by glucose perturbation were typically comparable between WT and ob/ob.
Subsequently, we examined the subtraction ranking of P to A from A to P of the ET. The difference in the subtraction corresponds to the increase in the net number of edges of OGTT [4] compared with OGTT [0]. The smaller the difference, the smaller the increase in the number of edges in OGTT [4]; if the difference was negative, the number of edges was lower in OGTT [4]. The higher the difference was ranked in descending order, the more edges there were in OGTT [4]. The lower the ranking, the more edges there were in OGTT [0]. In the WT metabolome (Table S2), the top three ETs (in descending order) were ET No. 11 (E–L), 10 (E–E), and 2 (A–C) and the bottom three were ET No. 4 (A–L), 1 (A–A), and 12 (E–N). The difference in ET No. 4 (A–L), 1 (A–A), and 12 (E–N) was negative. Thus, the number of edges decreased following the OGTT. This result indicated an inverse relationship to the change in the overall edges following the OGTT. Given that edges reflect node interaction, the presence of ET No. 2 (A–C) in the top three indicates the possibility that the conversion between amino acid metabolism and carbohydrate metabolism was highly active in WT mice by the OGTT. Similar explanations existed for ET No. 11 (E–L) and 10 (E–E) in the top three ETs. If the conversions between energy metabolisms, as well as between energy metabolism and lipid metabolism were highly activated by the OGTT, this would be surprising given their associations with the OGTT. Similarly, glucose stimulation may have suppressed each conversion between amino acid metabolisms, between amino acid metabolism and lipid metabolism, and also between energy metabolism and nucleotide metabolism in OGTT [4] of WT mice.
In the ranking of the ob/ob metabolome (Table S3), the top three ETs (in descending order) were ET No. 8 (C–L), 2 (A–C), and 4 (A–L) and the bottom three were ET No. 10 (E–E), 15 (N–N), and 12 (E–N). The difference in ET No. 8 (C–L), 2 (A–C), and 4 (A–L) was positive. Thus, the number of edges increased following the OGTT. This result indicated an inverse relationship to the change in the overall edges by the OGTT. In both WT and ob/ob, ET No. 12 (E–N) appeared in the bottom three in and ET No. 2 (A–C) appeared in the top three. Given that an increase in the total number of edges in OGTT [4] indicates active metabolite conversion, the increase in the edge of ET No. 8 (C–L), 4 (A–L) in OGTT [4] suggested active conversion between carbohydrate metabolism and lipid metabolism, as well as between amino acid metabolism and lipid metabolism. However, it should be noted that the metabolites used in this study that are classified as lipid metabolism are short-chain fatty acids, rather than the long-chain fatty acids that accumulate as triglycerides. In contrast, the net decrease in the edges of ET No. 10 (E–E), 15 (N–N), and 12 (E–N) by the OGTT suggested the possibility that the conversion related to energy metabolism and nucleotide metabolism was inactive in OGTT [4] for the ob/ob metabolome.
In the WT transcriptome, some ETs followed a similar pattern to the overall edges, with A to P edges having a slightly higher ET graph density than P to A edges; others followed a different pattern. All ETs in the ob/ob transcriptome followed a similar pattern to the overall edges, with P to A edges having a higher ET graph density than A to P edges. ET No. 10 (E-E) was distinctive in that the ET graph density of P to P edges was comparably greater than that of other ETs in both the WT and ob/ob transcriptomes.
We examined the subtraction ranking of P to A from A to P of the ET in the transcriptome. In the WT transcriptome (Table S4), the top three ETs (in descending order) were ET No. 15 (N–N), 14 (L–N), and 9 (C–N) and the bottom 3 were ET No. 10 (E–E), 11 (E–L), and 3 (A–E). The top three, ETs No. 15 (N–N), 14 (L–N), and 9 (C–N), indicated the possibility that the transcriptome caused the changes in the amount of metabolic enzymes cooperating with each other between nucleotide metabolisms, and between lipid metabolism and nucleotide metabolism, and also between carbohydrate metabolism and nucleotide metabolism in conjunction between these metabolic pathways following the OGTT in WT mice. This result suggested that the gene expression related to nucleotide metabolism and glucose metabolism was possibly regulated by the OGTT as they are connected through the pentose phosphate pathway. In contrast, the bottom three, ETs No. 10 (E–E), 11 (E–L), and 3 (A–E), indicate a decrease in the cooperation of metabolic enzymes involved in energy metabolism. This result suggested that the coordination of gene expression regulation related to energy metabolism in OGTT [0] might have been disrupted by the transcriptional re-wiring mediated by the OGTT.
In the ob/ob transcriptome (Table S5), the top three ETs (in descending order) were ET No. 1 (A–A), 4 (A–L), and 13 (L–L) and the bottom three were ET No. 10 (E–E), 12 (E–N), and 7 (C–E). This result suggested that the cooperation of metabolic enzymes shifted away from energy metabolism to amino acid and lipid metabolism. The similarity between WT and ob/ob in the transcriptome was that ET No. 10 (E–E) appeared in the bottom three in descending order.
Certain ETs displayed a response to the OGTT that was consistent with the overall edges (Fig. 3A) in terms of increasing or decreasing the number of edges in the OGTT [4], whereas others displayed the opposite response (Table S6). By introducing ETs, each edge of the overall edges is grouped to a particular ET. Thus, we observed similar responses to the OGTT for WT and ob/ob mice in both the metabolome and transcriptome, but these could not be observed when overall edges were examined. ET No. 1 (A–A), 2 (A–C), 8 (C–L), and 12 (E–N) in the metabolome and ET No. 3 (A–E), 7 (C–E), 10 (E–E), 11 (E–L), and 12 (E–N) in the transcriptome showed similar responses to the OGTT between WT and ob/ob. This suggested that the OGTT response of ob/ob mice was not fully abnormal and the normal response, which was observed in WT mice, still occurred partially in ob/ob mice. Additionally, ET No. 1 (A–A), 3 (A–E), 4 (A–L), 7 (C–E), 10 (E–E), and 11 (E–L) in WT and ET No. 2 (A–C), 4 (A–L), and 8 (C–L) in ob/ob showed different response to the OGTT between the metabolome and transcriptome. Thus, one advantage of using ETs is that we can investigate similar properties shared by WT and ob/ob or different profile between the metabolome and transcriptome, which is difficult when individual edges are examined.
## Difference between WT and ob/ob reflects the distribution of OGTT response edges
We examined the divergence between distributions of P to A, A to P, P to P, and A to A edges for the OGTT response for each ET (Fig. 5B; see also Fig. 1D–G). The divergence between the distributions was measured by the Chi-squared histogram distance. This distance indicates the dissimilarity between two histograms, which is defined in the range between 0 and 1. When the Chi-squared histogram distance between two histograms is 0, the two histograms are completely equivalent. We compared the distribution of WT and ob/ob using the Chi-squared histogram distance.
In the metabolome, the top three ETs with the largest Chi-squared histogram distance between WT and ob/ob were ET No. 10 (E–E), 13 (L–L), and 11 (E–L), all of which were involved in energy and lipid metabolism (Table S7). This finding indicated that the distributions of edges’ responses to the OGTT in these ETs were considerably different between WT and ob/ob. Indeed, it has already been reported that lipid synthesis is abnormally enhanced in ob/ob mice27,28. Additionally, it has been reported that ob/ob mice have abnormal energy metabolism, including decreased energy expenditure25,29, decreased mitochondrial respiration30, and switching of the energy substrate from glucose to fatty acid31.
The top three ETs with the smallest Chi-squared histogram distance between WT and ob/ob were ET No. 8 (C–L), 3 (A–E), and 1 (A–A). Two of these were involved in amino acid metabolism, suggesting that the response of edges involved in amino acid metabolism to the OGTT was similar in WT and ob/ob. The small Chi-squared histogram distance found for ET No. 8 (C–L) is paradoxical because de novo lipid synthesis is increased and carbohydrate to lipid conversion is also active in ob/ob mice27,28,32, even though the lipid metabolites in this study were not long-chain fatty acids. This counterintuitive result obtained by our methodology is exciting, as it may have yielded new insight.
In the transcriptome, the top three ETs with the largest Chi-squared histogram distance between WT and ob/ob were ET No. 15 (N–N), 12 (E–N), and 9 (C–N) (Table S7). The Chi-squared histogram distance found for ET No. 10 (E–E) was slightly smaller than ET No. 9 (C–N). The top three ETs with the smallest Chi-squared histogram distance between WT and ob/ob were ET No. 1 (A–A), 4 (A–L), and 13 (L–L). ETs involved in energy metabolism were found in the top three largest Chi-squared histogram distances, whereas ETs involved in amino acid metabolism were found in the top three smallest Chi-squared histogram distances in both the metabolome and transcriptome. The pattern of such relationships to annotation would be instructive in revealing the points of convergence and divergence between WT and ob/ob. Indeed, abnormal energy metabolism has been reported in ob/ob mice, including altered abundance of oxidative phosphorylation proteins30.
The Chi-squared histogram distance found for ET No. 10 (E-E) was large in both the metabolome and transcriptome, indicating that the response of edges to the OGTT related to energy metabolism was different in the metabolome and transcriptome between WT and ob/ob. Thus, there is a possibility that the substantial difference in transcriptional regulation in response to the OGTT between WT and ob/ob mice in energy metabolism may be reflected in the metabolome.
ET No. 1 (A–A) appeared in the top three ETs with the smallest Chi-squared histogram distance in both the metabolome and transcriptome, suggesting that the response of edges related to amino acid metabolism to the OGTT was slightly different between WT and ob/ob in both the metabolome and transcriptome. The transcriptional regulation and effect of the OGTT on the metabolome related to amino acid metabolism should differ slightly between WT and ob/ob mice owing to an indirect impact of the OGTT on amino acid metabolism.
The scatter plot of the Chi-squared histogram distance between WT and ob/ob for the metabolome and transcriptome is shown in Fig. S4. We found that the Chi-squared histogram distance of ET No. 10 (E–E) was large, and that of ET No. 1 (A–A) was small in both the metabolome and transcriptome. Additionally, we found that the Chi-squared histogram distances associated with amino acid metabolism were small and so were located near the origin of the scatter plot. In contrast, the Chi-squared histogram distances associated with energy metabolism were large and far from the origin. Finally, we found that the Chi-squared histogram distance associated with nucleotide metabolism in the transcriptome was typically greater than that of the metabolome and that the distance associated with the lipid metabolism in the metabolome was typically greater than that of the transcriptome.
## Regulation of the relationship for OGTT response between the metabolome and the transcriptome described by the distribution of ET
We examined the Chi-squared distance between the distributions of the OGTT response edges of the metabolome and the transcriptome for each ET. The changes in the graph density were similar between the metabolomes and transcriptomes in WT and ob/ob (Fig. 3A), whereas the Chi-squared distances varied among ETs in WT and ob/ob (Fig. 5C). The top three ETs with the smallest distances in WT and ob/ob were ET No. 9 (C–N), 4 (A–L), and 12 (E–N), and ET No. 1 (A–A), 14 (L–N), and 6 (C–C), respectively (Table S8). The small distances suggest the existence of regulatory relationships between the metabolome and the transcriptome for the ETs in WT and ob/ob mice. Additionally, we found that the Chi-squared distance between ob/ob was greater than that between WT for each ET (Fig. S5). This observation implied that the regulatory relationship between the metabolome and transcriptome was disrupted in ob/ob mice in comparison to WT mice.
## Reduction of characteristics of OGTT response edges by multidimensional scaling
Multidimensional scaling (MDS) is a technique for visualizing the level of similarities of pairwise samples in a dataset that creates a map displaying the relative positions. We visualized the similarity between ETs, which are represented by the Chi-squared distance in the distributions of the OGTT response edges, such as P to A, A to P, and P to P, in a two-dimensional plot using MDS (Fig. 6). ET No. 1 (A–A) was located further from the other ETs in all four MDS plots, suggesting that the edges of ET No. 1 (A–A) indicate a distinct role in the OGTT response. Additionally, ET No. 10 (E–E) is located further from the other ETs, except for the WT metabolome, suggesting that the edges of ET No. 10 (E–E) indicate a distinct role in the OGTT response apart from in the WT metabolome. Additionally, we found that the MDS map of WT and ob/ob was more similar for the transcriptome than for the metabolome, which is consistent with the fact that the variance of the Chi-squared distance for the metabolome is greater than that for the transcriptome (Fig. 5B). Thus, an MDS map can be helpful to extract the characteristics of ET and reduce the information. Figure 6The MDS plot for the ETs. ETs are arranged according to Chi-squared histogram distance of the distributions of the OGTT response edges. The metabolome of WT (left upper panel) and ob/ob (left lower panel), the transcriptome of WT (right upper panel) and ob/ob (right lower panel).
## The distribution of ETs contains reduced information on the state of the network structure
The two-sample Cramér–von Mises test33 is a statistical hypothesis test in which the null hypothesis is that the two samples are drawn from the same distribution. It was used to determine whether the two ET distributions were distinct. We assumed that the ETs followed an identical distribution independently and used the two-sample Cramér–von Mises test to determine the difference between the distributions of ETs for WT OGTT [0], WT OGTT [4], ob/ob OGTT [0], and ob/ob OGTT [4] in the metabolome and transcriptome (Table 2, Bonferroni corrected $p \leq 0.05$; see also Fig. 1H–K). All pairwise distributions were significantly different from each other, except for the pairing of WT OGTT [0] and ob/ob OGTT [4] in the metabolome data. These findings suggest that the state of the network structure is reflected in the distribution of ETs, and that the characteristics of the state can be reduced to the distribution of ET. Additionally, the possibility of a difference between WT OGTT [0] and ob/ob OGTT [0] was greater in the metabolome and transcriptome than that for the other pairs (Fig. 7). Likewise, the possibility of a difference between WT OGTT [4] and ob/ob OGTT [4] was lower than that for the majority of other metabolomics and transcriptomic variables. This finding suggested that the glucose perturbation decreased the possibility of a difference between WT and ob/ob. In other words, this result might suggest that glucose perturbation (by the OGTT) and the mechanism by which the liver maintains glucose homeostasis reduced the difference in the hepatic metabolome and transcriptome between WT and ob/ob mice compared with steady-state conditions, which were WT OGTT [0] and ob/ob OGTT [0], in terms of the biological categorization of inferred molecular relationship based on mutual information. Although the possibility of a difference between WT OGTT [0] and WT OGTT [4] was similar to that between WT OGTT [0] and ob/ob OGTT [0] in the metabolome, the possibility of difference between WT OGTT [0] and WT OGTT [4] was markedly reduced compared with the possibility of a difference between WT OGTT [0] and ob/ob OGTT [0] in the transcriptome. In contrast, although the possibility of difference between ob/ob OGTT [0] and ob/ob OGTT [4] was similar to that between WT OGTT [0] and ob/ob OGTT [0] in the transcriptome, the possibility of difference between ob/ob OGTT [0] and ob/ob OGTT [4] in the metabolome was markedly reduced when compared to the possibility of difference between WT OGTT [0] and ob/ob OGTT [0] in the metabolome. In other words, following glucose perturbation (by the OGTT), the metabolome of WT and the transcriptome of ob/ob could be as dissimilar as WT and ob/ob under steady-state conditions (WT OGTT [0] and ob/ob OGTT [0]), respectively. Thus, the metabolome of ob/ob and the transcriptome of WT were relatively similar compared with the steady-state, following glucose perturbation. These observations collectively suggested that WT and ob/ob mice were affected by glucose perturbation with changes in the metabolome and transcriptome, respectively. Conversely, the transcriptome of WT mice and the metabolome of ob/ob mice are suggested to be less affected by the OGTT. Consequently, the positional relationship between “WT OGTT [0] and WT OGTT [4]” and “ob/ob OGTT [0] and ob/ob OGTT [4]” was symmetrical with respect to a line with a positive slope. The positional relationship between “WT OGTT [0] and ob/ob OGTT [4]” and “WT OGTT [4] and ob/ob OGTT [0]” was symmetric with respect to a line with a negative slope. Thus, the distribution of ETs contains reduced information on the state of the network structure. Table 2The distance of A2 statistics and p-values. MetabolomeTranscriptomeA2A2WT [0]WT [4]ob/ob [0]ob/ob [4]WT [0]WT [4]ob/ob [0]ob/ob [4]WT [0]25.74565427.97986673.33705359WT [0]37.9323031134.66383211.002001WT [4]7.210916267.77674533WT [4]1181.80553130.839894ob/ob [0]11.1609826ob/ob [0]363.202317ob/ob [4]ob/ob [4]p-valuep-valueWT [0]WT [4]ob/ob [0]ob/ob [4]WT [0]WT [4]ob/ob [0]ob/ob [4]WT [0]7.5057E-108.3855E-110.10291111WT [0]4.8327E-15 < 1.0E-16 < 1.0E-15WT [4]0.000131570.0003385WT [4] < 1.0E-16 < 1.0E-15ob/ob [0]0.00106993ob/ob [0] < 1.0E-15ob/ob [4]ob/ob [4]The left and right column represent the metabolome and transcriptome, respectively. The upper and lower table are the A2 statistics and Bonferroni corrected p-values. In both the tables, WT [0], WT [4], ob/ob [0], and ob/ob [4] stands for WT OGTT [0], WT OGTT [4], ob/ob OGTT [0], and ob/ob OGTT [4], respectively. Figure 7The scatter plot of − log10 (p-value) between the metabolome and transcriptome. The scatter plot of the − log10 (Bonferroni corrected p-value) between the metabolome and transcriptome. WT [0], WT [4], ob [0], and ob [4] represent WT OGTT [0], WT OGTT [4], ob/ob OGTT [0], and ob/ob OGTT [4], respectively.
## Discussion
In this study, we inferred the OGTT response networks using statistical independence for WT and ob/ob mice in the metabolome and transcriptome to determine the difference in response to the OGTT between WT and ob/ob mice. Our results showed that glucose perturbation by the OGTT increased the number of metabolome and transcriptome network edges in WT, but the number was decreased in ob/ob (Fig. 3A). Furthermore, we observed that the majority of the edge responses to the OGTT for the metabolome and transcriptome were different between WT and ob/ob, except for the A to A edges (Fig. 3C). Subsequently, we analyzed the ETs and found the following two points: [1] There was a similar OGTT response in terms of ET between WT and ob/ob in the metabolome and transcriptome data (Table S6); and [2] the OGTT responses were different in terms of ET between the metabolome and transcriptome, even though the responses of the overall edges were similar. These findings could not be obtained by analyses without introducing ETs such as the analysis of node degree distribution and overall edges.
Owing to the complexity and unreliability of the results obtained from inferring the network structure from a dataset with a small sample, we chose to use the results given using ET. Additionally, comparing the different network structures is quite challenging. However, our study is novel in that it uses information on ET to reduce the amount of information about the state of the network structure and to compare between states of network structure. We used a simple “top three” selection to determine the remarkable ETs in the evaluation of the edges’ response to OGTT (Fig. 5A, Tables S2–S5) or the comparison of the edges’ response between WT and ob/ob or between the metabolome and the transcriptome (Fig. 5B–C, Tables S7, S8). However, there is a possibility that we can more reasonably select remarkable ETs. In future work, we will aim to develop a procedure to select remarkable ETs.
Kokaji et al.22 previously examined the OGTT response in WT and ob/ob mice, but their analysis was based on the average of each molecule, which differs from our definition of “OGTT response network.” However, our results corroborate their findings, as we found that the OGTT response network were considerably different between WT and ob/ob. Additionally, both studies found that a small number of edges appeared in both WT and ob/ob, whereas a large number of edges were unique to either WT or ob/ob.
The nodes were annotated with information limited to the metabolic pathways in the KEGG database, which enabled us to compare the metabolome and transcriptome using the ET categorization. Annotation of the metabolome is also possible using the Human Metabolome Database. A single metabolite is annotated into multiple metabolic pathways in the KEGG database, whereas the Human Metabolome Database annotates metabolites without redundancy. Therefore, the Human Metabolite Database would have been a better choice for this study if we were not making comparisons between the metabolome and transcriptome. Except for metabolic pathways, the genes in the KEGG database are lacking for the transcriptome and do not cover all datasets. TRANSFAC and RegNetwork provide information about transcription factors, but their annotation coverage is insufficient for genes. Gene ontology (GO) is the most complete source of gene annotation. However, the use of GO requires reformulation of the hierarchical ontology for the ET. Consequently, we chose to use the KEGG database annotation, allowing comparison between the transcriptome and metabolome. Cellular systems consist of several layers, including the metabolome and transcriptome, as well as the epigenome and proteome. We emphasize that our method of analyzing networks using common ETs has great potential for network analysis across multiple omics layers.
In biological research, a dataset is usually preprocessed by averaging across individuals for each molecular species. The averaging preprocess is reasonable and simplifies analysis, given our interest in the average behavior of phenomena. However, the averaging preprocess carries the disadvantage of losing information about variations between individuals, which is necessary to establish relationships between molecules. Therefore, we inferred the network structure using individual information based on statistical independence between molecular species. Linear correlations, such as the Pearson correlation, are frequently used to examine relationships between molecular species; however, they are incapable of detecting nonlinear relationships. In contrast, mutual information17 enables the detection of statistical independence by evaluating nonlinear relationships, including linear relationships. Conditional independence can yield a more precise network structure because of the exclusion of the effects of other molecular species, which indirectly affects the relationships between targeted pairwise molecular species. However, detecting conditional independence is difficult in datasets with a high number of dimensions and a small sample size. Additionally, the computational burden scales exponentially with the number of dimensions in the datasets. Therefore, we examined statistical independence to infer network structure in this study and, in the future, plan to continue to adopt conditional independence to infer network structure.
Owing to the high cost of biological experiments, the dataset had a small sample size, necessitating the use of the permutation test to determine statistical independence. Thus, we inferred the network structure and omitted the intensity of connections because of the low reliability of the mutual information value.
The small sample size of the dataset reduces the reliability of the inferred whole network structure, indicating that an edge-by-edge comparison of whole network structures is not robust. Additionally, it is critical to define a quantitative comparison metric for network structures, which is not easy. Information about the entire network structure, however, can be reduced to the distribution of ETs, which is insensitive to changes in edges. Additionally, we demonstrated that this reduced information enables the comparison of two states of network structure, corresponding to the WT and ob/ob OGTT responses, or the metabolome and transcriptome, respectively. Collectively, these findings suggest that the reduced information obtained from the distribution of ET can be used as a “fingerprint,” indicating characteristics of the state of network structure. The fingerprint is analogous to the bag-of-words model used in the field of text analysis. To summarize, the bag-of-words model reduces the complexity of a text by counting the occurrences of words in documents. In this method, even though documents are characterized using word counts and the information about the text structure is reduced, the model is useful for document classification and other purposes. However, additional research is required to show the advantage of the fingerprint by applying it to a variety of biological phenomena.
## Mouse studies
Previously published mouse liver samples were used for this study24. Briefly, ten-week-old male C57BL/6 wild-type and ob/ob mice were purchased from Japan SLC Inc. After overnight fasting (16 h), mice were administered 2 g/kg body weight of glucose orally. After 4 h from the experiments, mice were sacrificed by cervical dislocation and the liver (whole or left lateral lobe) was dissected and immediately frozen in liquid nitrogen. Similarly, the livers from overnight fasted (16 h) mice were dissected and frozen as a control. The frozen liver was pulverized with dry ice to a fine powder with a blender and separated into tubes for omic analysis (metabolomics, and transcriptomics). We used twelve WT and twelve ob/ob mice following oral glucose administration, and eleven WT and twelve ob/ob mice in a control group. The metabolomics and transcriptomic data of all the samples were reported in previous study24 and the metabolites and genes used for our analysis were expanded to include other metabolic pathways in addition to central carbon metabolism. All methods were carried out in accordance with relevant guidelines and all the mouse experiments were approved by the animal ethics committee of The University of Tokyo. This study is reported in accordance with ARRIVE guidelines.
## Principal component analysis
We performed the principal component analysis for centralized datasets of the metabolome and transcriptome, which were reformulated as the fold changes for the average of each molecular species.
## Procedure of network inference and ET identification
Suppose that \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${X}_{i}$$\end{document}Xi are a random variable, which correspond to the amount of molecular species i, such as gene expression and metabolite concentration, where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$$i = 1$,\cdots,m$$\end{document}$i = 1$,⋯,m and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$m$$\end{document}m is the number of molecular species. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${V}_{i}$$\end{document}Vi represents a vertex of network, corresponding to molecular species i. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${E}_{ij}=\{\mathrm{0,1}\}$$\end{document}Eij={0,1} represents the edge between \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${V}_{i}$$\end{document}Vi and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${V}_{j}$$\end{document}Vj. If \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${E}_{ij}=1$$\end{document}Eij=1, the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${V}_{i}$$\end{document}Vi and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${V}_{j}$$\end{document}Vj are connected each other, otherwise they are not connected. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${p}_{0}$$\end{document}p0 represents the threshold on p-value to determine edge existence. The procedure of part of network inference is as follows.
Step 1: evaluate the p-value by statistical hypothesis test, whose null hypothesis is \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I\left({X}_{i};{X}_{j}\right)=0$$\end{document}IXi;Xj=0, for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$i>j, i,$j = 1$,\cdots m$$\end{document}i>j,i,$j = 1$,⋯m. The p-value of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${E}_{ij}$$\end{document}*Eij is* denoted by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${p}_{ij}$$\end{document}pij.
Step 2: if \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${p}_{ij}<{p}_{0}$$\end{document}pij<p0, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${E}_{ij}=1$$\end{document}Eij=1. Otherwise \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${E}_{ij}=0$$\end{document}Eij=0.
After the network structure is determined by network inference, the ET is identified. Suppose that the set of node type is denoted by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${N}_{type}=\{{n}_{1},\cdots,{n}_{s}\}$$\end{document}Ntype={n1,⋯,ns}, where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${n}_{k},$k = 1$,\cdots,s$$\end{document}nk,$k = 1$,⋯,s and s represent a node type and the number of node types, respectively. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${N}_{type}$$\end{document}*Ntype is* determined based on database and one can set it arbitrarily in accordance with the aim. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${V}_{type,i}$$\end{document}Vtype,i represents the node type of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${V}_{i}$$\end{document}Vi. If the node type of molecular spices i is \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${n}_{k}$$\end{document}nk, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${V}_{type,i}={n}_{k}$$\end{document}Vtype,i=nk. Next, suppose that the set of ET is denoted by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${E}_{type}=\{{e}_{1},\cdots,{e}_{t}\}$$\end{document}Etype={e1,⋯,et}, where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${e}_{k},$k = 1$,\cdots,t$$\end{document}ek,$k = 1$,⋯,t and t represent an ET and the number of ETs, respectively. The ET \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${e}_{k}$$\end{document}ek t does not exceed the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\frac{s\left(s-1\right)}{2}+s$$\end{document}ss-12+s. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${E}_{type,ij}$$\end{document}Etype,ij represents the edge type of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${E}_{ij}$$\end{document}Eij. If the edge type of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${E}_{ij}$$\end{document}*Eij is* \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${e}_{k}$$\end{document}ek, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${E}_{type,ij}={e}_{k}$$\end{document}Etype,ij=ek. The procedure of part of ET identification is as follows.
Step 3: identify the node type \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${V}_{type,i}$$\end{document}Vtype,i from \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${N}_{type}$$\end{document}Ntype for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$$i = 1$,\cdots,m$$\end{document}$i = 1$,⋯,m, which is determined by annotation of molecular species i based on database.
Step 4: identify the ET \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${E}_{type,ij}$$\end{document}Etype,ij from \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${E}_{type}$$\end{document}Etype for all pairs of i and j satisfying \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${E}_{ij}=1$$\end{document}Eij=1, according to the pair of node type \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${V}_{type,i}$$\end{document}Vtype,i and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${V}_{type,j}$$\end{document}Vtype,j.
Finally, we obtain the distribution of ET by counting the appearance of each ET \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${e}_{k}$$\end{document}ek, for \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${E}_{type,ij}$$\end{document}Etype,ij of all pairs of i and j satisfying \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${E}_{ij}=1$$\end{document}Eij=1. Note that there is a possibility that multiple node types assign to one node, for instance, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${V}_{type,i}=\{{e}_{k},{e}_{l}\}$$\end{document}Vtype,i={ek,el}. In the case of multiple node types, the ETs corresponding to all pairwise combination of edge node between \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${V}_{type,i}$$\end{document}Vtype,i and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${V}_{type,j}$$\end{document}Vtype,j assign to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${E}_{type,ij}$$\end{document}Etype,ij. And one counts the appearance of each ET in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${E}_{type,ij}$$\end{document}Etype,ij, which is equally weighted by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$1/{\#E}_{type,ij}$$\end{document}1/#Etype,ij, where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\#E}_{type,ij}$$\end{document}#Etype,ij represents the number of ETs contained in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${E}_{type,ij}$$\end{document}Etype,ij.
## OGTT response network definition
The OGTT response network is defined as the difference of network between OGTT [0] and OGTT [4]. The edges in the OGTT response network are denoted by four types, according to their presence or absence in the network of OGTT [0] and OGTT [4]: “P to P” denotes the edges that are shared by the OGTT [0] and OGTT [4] networks, “A to P” denotes the edges that are unique to the OGTT [4] network, “P to A” denotes the edges that are unique to the OGTT [0] network, and “A to A” denotes the edges existing in neither OGTT [0] nor OGTT [4] network. Additionally, P to A, A to P, and P to P edges are referred to as the OGTT response edges. The ET in the OGTT response network is identified in the same way of step3 and step4 in the procedure above.
## Calculation of the mutual information
The mutual information of two random variables X and Y is defined as\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I\left(X,Y\right)=\iint p(x,y)\mathrm{log}\frac{p(x,y)}{p(x)p(y)}dxdy$$\end{document}IX,Y=∬p(x,y)logp(x,y)p(x)p(y)dxdywhere p(x,y) is the joint probability distribution of X and Y, and p(x) and p(y) are marginal probability distribution of X and Y, respectively.
The estimation of joint probability distribution of X and Y from the dataset is required in order to calculate the mutual information. We estimated the joint probability distribution using B-spline function17. The spline order k was set to $k = 3$ and the number of bin M was determined in each combination of X and Y34. The metabolome and transcriptome data were standardized and subjected to calculation of the mutual information.
## Statistical hypothesis test on statistical independence
The permutation test is performed under the null hypothesis that the mutual information \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I\left(X;Y\right)$$\end{document}IX;Y is 035. If and only if \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I\left(X;Y\right)=0$$\end{document}IX;$Y = 0$, X and Y are statistically independent. The sample of null distribution is generated by permutation of original data X. We obtained the empirical null distribution by the random sampling, where the sample size was 500 in each permutation test. The p-value is obtained by the ratio of the number of samples, which are larger than the value of mutual information \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$I\left(X;Y\right)$$\end{document}IX;Y, to the sample size of empirical null distribution.
## Statistical hypothesis test on the difference of values of mutual information
To examine whether the values of the mutual information are different between prior to and following the OGTT, we performed the permutation test under the null hypothesis and alternative hypothesis described in following formula\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left\{\begin{array}{c}{H}_{0}:{I}_{OGTT [0]}\left(X;Y\right)={I}_{OGTT [4]}(X;Y)\\ {H}_{1}:{I}_{OGTT [0]}\left(X;Y\right)\ne {I}_{OGTT [4]}(X;Y)\end{array}\right.$$\end{document}H0:IOGTT[0]X;Y=IOGTT[4](X;Y)H1:IOGTT[0]X;Y≠IOGTT[4](X;Y) The sample size was 500 to generate null distribution empirically in each permutation test.
## Network visualization
Cytoscape36 was used for network visualization. We used the method that is as non-redundant as possible for node categorization in visualization. The metabolites were categorized using the information of the Human Metabolome Database37–40 in the metabolome. In the transcriptome, the genes whose OS and BS fields include both *Mus musculus* in the factor.dat file in TRANSFAC41 were categorized as transcription factor. *The* genes which were registered in RegNetwork42 as Type: regulator, Organism: mouse, Evidence: Experimental, Confidence: High were also categorized as transcription factor. *The* genes which were registered to the pathway included in “1. Metabolism” in KEGG PATHWAY database43–45 were categorized as metabolic enzyme. ENSMUSG00000017715 (Pgs1) and ENSMUSG00000020593 (Lpin1) categorized to both transcription factors and metabolic enzymes, but were classified as transcription factors. *The* genes not categorized to neither transcription factor and metabolic enzyme were categorized as Others.
## Edge type categorization
For node categorization, we used following 11 calcifications in “1. Metabolism” of KEGG PATHWAY database: 1.1 Carbohydrate metabolism, 1.2 Energy metabolism, 1.3 Lipid metabolism, 1.4 Nucleotide metabolism, 1.5 Amino acid metabolism, 1.6 Metabolism of other amino acids, 1.7 *Glycan biosynthesis* and metabolism, 1.8 Metabolism of cofactors and vitamins, 1.9 Metabolism of terpenoids and polyketides, 1.10 Biosynthesis of other secondary metabolites, 1.11 Xenobiotics biodegradation and metabolism. The metabolites or genes which belong to the pathway included in each classification were categorized into these 11 types. Next, 66 edge types connecting between 11 node types were defined.
## Calculation of graph density
The graph density D and the ET graph density \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{ET}$$\end{document}DET were calculated by flowing formulas.\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$D=\frac{2E }{N\left(N-1\right)}$$\end{document}$D = 2$ENN-1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{ET}=\frac{2{E}_{ET} }{{N}_{ET}\left({N}_{ET}-1\right)}$$\end{document}DET=2EETNETNET-1where E is the number of estimated edges, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${E}_{ET}$$\end{document}EET is the number of edges that estimated and categorized into each ET, N is the number of the node in the whole graph and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${N}_{ET}$$\end{document}NET is the number of the node in the subgraph categorized into each ET.
## Chi-squared histogram distance
The Chi-squared histogram distance between the histograms of P and Q is defined by\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\chi }^{2}\left(P,Q\right)=\frac{1}{2}\sum_{i}\frac{{\left({P}_{i}-{Q}_{i}\right)}^{2}}{\left({P}_{i}+{Q}_{i}\right)}$$\end{document}χ2P,$Q = 12$∑iPi-Qi2Pi+Qiwhere the subscript i indicates the i-th bin of the histogram of P or Q, Pi and Qi indicate the frequency of each i-th bin.
## MDS plot
Torgerson scaling method46 determines the coordinate of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${x}_{i}$$\end{document}xi so as to minimize\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\sum_{i,j}{\left({z}_{ij}-\sum_{m}{x}_{im}{x}_{jm}\right)}^{2}$$\end{document}∑i,jzij-∑mximxjm2 where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${z}_{ij}=\frac{1}{2}\left({d}_{io}^{2}+{d}_{jo}^{2}-{d}_{ij}^{2}\right)$$\end{document}zij=12dio2+djo2-dij2, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${d}_{ij}$$\end{document}dij is the distance between the coordinates of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${x}_{i}$$\end{document}xi and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${x}_{j}$$\end{document}xj, and o indicates the coordinate of the origin. In this study, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${d}_{ij}^{2}$$\end{document}dij2 corresponds to Chi squared histogram distance between distributions of ET.
## Implementation
All analysis was implemented by MATLAB 2016b (The Mathworks Inc.).
## Supplementary Information
Supplementary Information. The online version contains supplementary material available at 10.1038/s41598-023-31547-2.
## References
1. Barabási AL, Oltvai ZN. **Network biology: Understanding the cell’s functional organization**. *Nat. Rev. Genet.* (2004) **5** 101-113. DOI: 10.1038/nrg1272
2. Aittokallio T, Schwikowski B. **Graph-based methods for analysing networks in cell biology**. *Brief. Bioinform.* (2006) **7** 243-255. DOI: 10.1093/bib/bbl022
3. Fukushima A, Kusano M, Redestig H, Arita M, Saito K. **Metabolomic correlation-network modules in Arabidopsis based on a graph-clustering approach**. *BMC Syst. Biol.* (2011) **5** 1. DOI: 10.1186/1752-0509-5-1
4. Takehisa H, Sato Y, Antonio B, Nagamura Y. **Coexpression network analysis of macronutrient deficiency response genes in rice**. *Rice* (2015) **8** 24. DOI: 10.1186/s12284-015-0059-0
5. Lu LJ. **Comparing classical pathways and modern networks: towards the development of an edge ontology**. *Trends Biochem. Sci.* (2007) **32** 320-331. DOI: 10.1016/j.tibs.2007.06.003
6. Cary MP, Bader GD, Sander C. **Pathway information for systems biology**. *FEBS Lett.* (2005) **579** 1815-1820. DOI: 10.1016/j.febslet.2005.02.005
7. Schaefer CF. **Pathway databases**. *Ann. N. Y. Acad. Sci.* (2004) **1020** 77-91. DOI: 10.1196/annals.1310.009
8. Uetz P. **A comprehensive analysis of protein-protein interactions in**. *Nature* (2000) **403** 623-627. DOI: 10.1038/35001009
9. Ito T. **A comprehensive two-hybrid analysis to explore the yeast protein interactome**. *Proc. Natl. Acad. Sci. USA.* (2001) **98** 4569-4574. DOI: 10.1073/pnas.061034498
10. Gavin AC. **Functional organization of the yeast proteome by systematic analysis of protein complexes**. *Nature* (2002) **415** 141-147. DOI: 10.1038/415141a
11. Ho Y. **Systematic identification of protein complexes in Saccharomyces cerevisiae by mass spectrometry**. *Nature* (2002) **415** 180-183. DOI: 10.1038/415180a
12. Gavin AC. **Proteome survey reveals modularity of the yeast cell machinery**. *Nature* (2006) **440** 631-636. DOI: 10.1038/nature04532
13. Gardner TS, Di Bernardo D, Lorenz D, Collins JJ. **Inferring genetic networks and identifying compound mode of action via expression profiling**. *Science* (2003) **301** 102-105. DOI: 10.1126/science.1081900
14. Bansal M, Gatta GD, di Bernardo D. **Inference of gene regulatory networks and compound mode of action from time course gene expression profiles**. *Bioinformatics* (2006) **22** 815-822. DOI: 10.1093/bioinformatics/btl003
15. Kimura S, Sato M, Okada-Hatakeyama M. **An effective method for the inference of reduced S-system models of genetic networks**. *IPSJ Trans. Bioinforma.* (2014) **7** 30-38. DOI: 10.2197/ipsjtbio.7.30
16. Guo S, Jiang Q, Chen L, Guo D. **Gene regulatory network inference using PLS-based methods**. *BMC Bioinform.* (2016) **17** 545. DOI: 10.1186/s12859-016-1398-6
17. Daub CO, Steuer R, Selbig J, Kloska S. **Estimating mutual information using B-spline functions: An improved similarity measure for analysing gene expression data**. *BMC Bioinform.* (2004) **5** 118. DOI: 10.1186/1471-2105-5-118
18. Gerich JE. **Physiology of glucose homeostasis**. *Diabetes Obes. Metab.* (2000) **2** 345-350. DOI: 10.1046/j.1463-1326.2000.00085.x
19. Desvergne B, Michalik L, Wahli W. **Transcriptional regulation of metabolism**. *Physiol. Rev.* (2006) **86** 465-514. DOI: 10.1152/physrev.00025.2005
20. Han HS, Kang G, Kim JS, Choi BH, Koo SH. **Regulation of glucose metabolism from a liver-centric perspective**. *Exp. Mol. Med.* (2016) **48** e218. DOI: 10.1038/emm.2015.122
21. Chao HW, Chao SW, Lin H, Ku HC, Cheng CF. **Homeostasis of glucose and lipid in non-alcoholic fatty liver disease**. *Int. J. Mol. Sci.* (2019) **20** 298. DOI: 10.3390/ijms20020298
22. Kokaji T. **Transomics analysis reveals allosteric and gene regulation axes for altered hepatic glucose-responsive metabolism in obesity**. *Sci. Signal.* (2020) **13** 1236. DOI: 10.1126/scisignal.aaz1236
23. Matsuzaki F. **An extensive and dynamic trans-omic network illustrating prominent regulatory mechanisms in response to insulin in the liver**. *Cell Rep.* (2021) **36** 109569. DOI: 10.1016/j.celrep.2021.109569
24. Uematsu S. **Multi-omics-based label-free metabolic flux inference reveals obesity-associated dysregulatory mechanisms in liver glucose metabolism**. *Science* (2022) **25** 103787
25. Coleman DL. **Obese and diabetes: Two mutant genes causing diabetes-obesity syndromes in mice**. *Diabetologia* (1978) **14** 141-148. DOI: 10.1007/BF00429772
26. Lindström P. **The physiology of obese-hyperglycemic mice [ob/ob mice]**. *Sci. World J.* (2007) **7** 666-685. DOI: 10.1100/tsw.2007.117
27. Wolf G. **Insulin resistance associated with leptin deficiency in mice: A possible model for noninsulin-dependent diabetes mellitus**. *Nutr. Rev.* (2009) **59** 177-179. DOI: 10.1111/j.1753-4887.2001.tb07009.x
28. Anstee QM, Goldin RD. **Mouse models in non-alcoholic fatty liver disease and steatohepatitis research**. *Int. J. Exp. Pathol.* (2006) **87** 1-16. DOI: 10.1111/j.0959-9673.2006.00465.x
29. Kennedy AJ, Ellacott KLJ, King VL, Hasty AH. **Mouse models of the metabolic syndrome**. *DMM Dis. Models Mech.* (2010) **3** 156-166. DOI: 10.1242/dmm.003467
30. Holmström MH, Tom RZ, Björnholm M, Garcia-Roves PM, Zierath JR. **Effect of leptin treatment on mitochondrial function in obese leptin-deficient ob/ob mice**. *Metabolism* (2013) **62** 1258-1267. DOI: 10.1016/j.metabol.2013.04.001
31. Liang CP, Tall AR. **Transcriptional profiling reveals global defects in energy metabolism, lipoprotein, and bile acid synthesis and transport with reversal by leptin treatment in ob/ob mouse liver**. *J. Biol. Chem.* (2001) **276** 49066-49076. DOI: 10.1074/jbc.M107250200
32. Iizuka K, Miller B, Uyeda K. **Deficiency of carbohydrate-activated transcription factor ChREBP prevents obesity and improves plasma glucose control in leptin-deficient (ob/ob) mice**. *Am. J. Physiol. Endocrinol. Metab.* (2006) **291** 358-364. DOI: 10.1152/ajpendo.00027.2006
33. Arnold TB, Emerson JW. *Nonparametric Goodness-of-Fit Tests for Discrete Null Distributions* (2023)
34. Shimazaki H, Shinomoto S. *A Method for Selecting the Bin Size of a Time Histogram* (2023)
35. Good P. *Permutation Tests: A Practical Guide to Resampling Methods for Testing Hypotheses* (2013)
36. Shannon P. **Cytoscape: A software environment for integrated models of biomolecular interaction networks**. *Genome Res.* (2003) **13** 2498-2504. DOI: 10.1101/gr.1239303
37. Wishart DS. **HMDB: The human metabolome database**. *Nucleic Acids Res.* (2007) **35** D521-D526. DOI: 10.1093/nar/gkl923
38. Wishart DS. **HMDB: A knowledgebase for the human metabolome**. *Nucleic Acids Res.* (2009) **37** 603-610. DOI: 10.1093/nar/gkn810
39. Wishart DS. **HMDB 3.0-The human metabolome database in 2013**. *Nucleic Acids Res.* (2013) **41** 807
40. Wishart DS. **HMDB 4.0: The human metabolome database for 2018**. *Nucleic Acids Res.* (2018) **46** D608-D617. DOI: 10.1093/nar/gkx1089
41. Matys V. **TRANSFAC and its module TRANSCompel: Transcriptional gene regulation in eukaryotes**. *Nucleic Acids Res.* (2006) **34** D108-D110. DOI: 10.1093/nar/gkj143
42. Liu Z-P, Wu C, Miao H, Wu H. **RegNetwork: An integrated database of transcriptional and post-transcriptional regulatory networks in human and mouse**. *Database* (2015) **2015** 095. DOI: 10.1093/database/bav095
43. Kanehisa M, Goto S. **KEGG: Kyoto encyclopedia of genes and genomes**. *Nucleic Acids Res.* (2000) **28** 27-30. DOI: 10.1093/nar/28.1.27
44. Kanehisa M. **Toward understanding the origin and evolution of cellular organisms**. *Protein Sci.* (2019) **28** 1947-1951. DOI: 10.1002/pro.3715
45. Kanehisa M, Furumichi M, Sato Y, Ishiguro-Watanabe M, Tanabe M. **KEGG: Integrating viruses and cellular organisms**. *Nucleic Acids Res.* (2021) **49** D545-D551. DOI: 10.1093/nar/gkaa970
46. Torgerson WS. **Multidimensional scaling: I. Theory and method**. *Psychometrika* (1952) **17** 401-419. DOI: 10.1007/BF02288916
|
---
title: Correlation between meteorological factors and vitamin D status under different
season
authors:
- Xichao Wang
- Ke Lu
- Junjie Shen
- Shihan Xu
- Qi Wang
- Yaqin Gong
- Yunyu Xia
- Xiaochun Wang
- Lin Chen
- Shanjun Yan
- Zaixiang Tang
- Chong Li
journal: Scientific Reports
year: 2023
pmcid: PMC10036626
doi: 10.1038/s41598-023-31698-2
license: CC BY 4.0
---
# Correlation between meteorological factors and vitamin D status under different season
## Abstract
Pregnant women with low vitamin D levels tend to have poor clinical outcomes. Meteorological factors were associated with vitamin D. Here, we aimed to study the current status of 25-Hydroxy vitamin D (25(OH)D) concentrations in pregnant women in Kunshan city and investigate the meteorological factors associated with 25(OH)D levels under different seasons. The correlation between meteorological factors and 25(OH)D levels was estimated by cross-correlation analysis and multivariate logistic regression. A restrictive cubic spline method was used to estimate the non-linear relationship. From 2015 to 2020, a total of 22,090 pregnant women were enrolled in this study. Pregnant women with 25(OH)D concentrations below 50 nmol/l represent $65.85\%$ of the total study population. There is a positive correlation between temperature and 25(OH)D. And there is a protective effect of the higher temperature on vitamin D deficiency. However, in the subgroup analysis, we found that in autumn, high temperatures above 30 °C may lead to a decrease in 25(OH)D levels. This study shows that vitamin D deficiency in pregnant women may widespread in eastern China. There is a potential inverted U-shaped relationship between temperature and 25(OH)D levels, which has implications for understanding of vitamin D changes under different seasons.
## Introduction
Vitamin D is an essential fat-soluble vitamin. 25-Hydroxy vitamin D (25(OH)D) is an important form of vitamin D stored in the human body1. The standards for vitamin D levels in the population are still controversial. According to Endocrine Society clinical practice guideline, 25(OH)D concentrations below 30 nmol/l (12 ng/ml) are considered severe vitamin D deficiency. Concentrations below 50 nmol/l (20 ng/ml) are considered to be vitamin D deficiency2. Globally, vitamin D deficiency during pregnancy is common, including in China3–5. It is estimated that over $90\%$ of women of gestational age have vitamin D levels below 50 nmol/l4. Maternal vitamin D status, such as serum 25(OH)D, is essential for pregnancy and infant health outcomes6–8. Pregnant women may be at a higher risk of low 25(OH)D levels and related diseases, including preeclampsia9, gestational diabetes mellitus10, preterm birth11,12 and low birthweight13. Studies on vitamin D in pregnant women have important public health implications.
Few studies have investigated the prevalence of vitamin D deficiency in pregnant women in eastern China, and most of the available studies have examined the relationship between vitamin D levels and specific diseases. Examples include vitamin D levels and bone disease14, vitamin D levels and kidney-related ailments15,16. Meteorological factors are crucial to the impact of vitamin D levels. Researchers are increasingly focusing on the relationship between meteorological factors and vitamin D levels. Most previous studies on vitamin D consider the influence of a small number of meteorological factors17–20. For example, studies on the prediction of vitamin D deficiency generally ask subjects about their sun exposure habits or the approximate number of hours a week they are exposed to the sun using a questionnaire21. In observational studies, researchers generally use some proxy to measure the effect of meteorological factors, such as the number of hours of sunshine per day22,23. Whether sunshine hours is a factor affecting vitamin D levels remains a question. The study from southwest China showed a correlation between sunlight hours and vitamin D22. However, the survey from Hangzhou, China, gave the opposite result23. In recent years, more and more meteorological factors have received the attention of researchers. For example, researchers have looked at the effects of ozone on UV and vitamin D24. The relationship between vitamin D status and latitude, cloudiness, UV-B exposure and solar zenith angle was investigated25–28. Furthermore, the seasonality of vitamin D has been confirmed in most studies29–32. However, to our knowledge, no studies have analyzed the different effects of meteorological factors on vitamin D under different seasons.
Information on the epidemiology of vitamin D status in pregnant women is essential. Exploring the impact of meteorological factors on vitamin D status across seasons is also of great practical importance, particularly in guiding public health policy. However, this information and related studies are limited in the Chinese pregnant women population. The current study collected 25(OH)D data in 22,090 healthy pregnant women in Kunshan, China. The prevalence of vitamin D deficiency was assessed, the correlation between 25(OH)D levels and meteorological factors was explored. We also investigated the non-linear relationships of these factors with serum 25(OH)D in pregnant women under different seasons.
## Study population
The data for this study were obtained from Kunshan, China (northern latitude 31°), which has a typical subtropical monsoon climate. From June 2015 to December 2020, 30,523 women in mid-pregnancy (15–20 weeks of gestation) who underwent pregnancy monitoring and serum 25(OH)D concentration testing were collected. As shown in Fig. 1, we excluded pregnant women who are not local residents and excluded those with high-risk pregnancies. ( The effect of meteorological factors on vitamin D may require a longer time period. Exclusion of non-local residents would allow the study population to be in a similar meteorological environment as much as possible. Also, we did not collect additional information such as facial skin status of the study population. The inclusion of subjects who were all local residents was also intended to make the study population as consistent as possible.) Pregnant women with certain diseases that with consequences for the metabolism of 25(OH)D were also excluded. A total of 22,090 pregnant women were eventually included in this study. The study protocol, submitted for review by the ethical committee at the Affiliated Kunshan Hospital of Jiangsu University (approval No. 2020-03-046-K01), was approved, and it complied with the Declaration of Helsinki. Patient information was initially documented for hospital’s quality improvement purposes. The requirement for informed consent was waived because of the anonymous and observational design of this investigation, and the decision was approved by the Ethics Committee of Kunshan Hospital, Jiangsu University. Data analyzers were blinded to the identity of the patients. Figure 1Flow chart for this study.
## Assessment of 25(OH)D
All participants in this study were asked to fast before blood samples were taken. Serum 25(OH)D concentrations were measured immediately using an automated electrochemiluminescence immunoassay on a Roche Cobas 8000/e602 analyzer (Roche Diagnostics, Mannheim, Germany). All 25(OH)D assays in this paper were tested by the same machine and all measurements passed the National Health Council's Endocrine External Quality Assessment.
## Meteorological data
Meteorological data are from the Ecology and Environment Bureau of Kunshan City. Data was matched with patient information. The following meteorological information was collected: daily average temperature, °C; daily average atmospheric pressure, hPa; daily average relative humidity, %; daily average wind speed, m/s (These indicators refer to the arithmetic mean of the observations during one day (24 h).); daily total sunshine hours, hour (The daily total sunshine hours is defined as the sum of the periods of direct solar irradiance at or above 120 w/m2.); daily total precipitation, mm (The daily total precipitation represents the depth of precipitation accumulated on a horizontal surface during a day without evaporation, infiltration, or loss).
## Definitions
Based on previous study2, for the 25(OH)D levels, we took the following groups: 25(OH)D concentrations below 50 nmol/l are considered to be vitamin D deficiency. Concentrations below 75 nmol/l are considered to be vitamin D insufficient. Concentrations above 75 nmol/l are considered to be sufficient vitamin D levels. ( The above groupings are only for data analysis. In the data description section, readers can make their own judgments about vitamin D deficiency based on their preferred criteria.) Seasonal factors in the samples were defined and classified according to spring (March, April, and May), summer (June, July, and August), autumn (September, October, and November), and winter (December, January, and February). We determined age grouping criteria based on several previous studies and surveys33–36. Those younger than 25 years old were defined as young maternal, those between 25 and 35 years old were defined as age-appropriate maternal, and those older than 35 years old were defined as advanced maternal.
## Statistical analyses
The baseline characteristics were presented using means with the standard deviations or medians with the interquartile ranges for continuous variables and frequencies with percentages for categorical variables. The Kruskal–Wallis test was used to compare continuous variables, and the chi-square test was used to compare categorical variables. The correlation between meteorological factors and 25(OH)D concentrations was initially estimated by cross-correlation analysis, taking into account the strong correlation between meteorological factors. Univariate and multivariate logistic regression models were fitted separately to explore the relationship between 25(OH)D concentrations and meteorological factors.
To further confirm the consistency between observed meteorological factors and 25(OH)D concentrations, we conducted subgroup analyses according to different seasons (spring, summer, autumn and winter) and different age groups.
To detect any possible non-linear dependence in the regression model and allow flexibility in interpreting the relationship between continuous covariates and study outcomes, changes in meteorological factors were assessed by shape-constrained three-dimensional spline regression models. We fitted shape-constrained cubic spline regression models for the dichotomous dependent variable (presence of vitamin D deficiency) and the continuous dependent variable (25(OH)D levels) in separate waves. Considering our sample size, in order to balance the smoothness of the curve and avoid overfitting, we selected 5 knots. All statistical analyses were performed using R4.1.0 (https://www.r-project.org/). We considered a two-sided P value < 0.05 to be significant.
## Prevalence of serum 25(OH)D deficiency in pregnant women assessed in China
The median 25(OH)D levels in subjects were 40 nmol/L (IQR 30, 55). From 2015 to 2020, pregnant women with 25(OH)D concentrations below 50 nmol/L represent $65.85\%$ of the total study population. The annual rates were $81.30\%$, $71.17\%$, $61.28\%$, $65.60\%$, $62.96\%$ and $65.51\%$ respectively. Pregnant women with 25(OH)D concentrations below 75 nmol/L (but above 50 nmol/L) accounted for $25.93\%$ of the total study population. The annual rates were $15.56\%$, $23.71\%$, $31.02\%$, $28.85\%$, $29.54\%$ and $27.99\%$ respectively. In most cases, 25(OH)D levels in Kunshan pregnant women are low. The mean 25(OH)D levels reached 50 nmol/L (20 ng/ml) only for a few months. As shown in Table 1 and Fig. 2, 25(OH)D levels were significantly higher in summer and autumn than in spring and winter. Although pregnant women's 25(OH)D levels increased each year from 2015 to 2018, this trend became less pronounced from 2018 to 2020.Table 1Baseline characteristics of participants in the total population under different seasons. VariableSpring ($$n = 5588$$)Summer ($$n = 5905$$)Autumn ($$n = 5435$$)Winter ($$n = 5162$$)PAge(years)29 ± 429 ± 429 ± 429 ± 40.14*25(OH)D (nmol/L)35[28,48]48[35,63]48[35,63]23[30,43] < 0.001^ < 50 nmol/L4297 (76.9)3106 (52.6)2897 (53.3)4246 (82.3) < 0.001# 50 ~ 75 nmol/L1035 (18.5)2105 (35.6)1836 (33.8)752 (14.6) > 75 nmol/L256 (4.6)694 (11.8)702 (12.9)164 (3.2)Daily average temperature (°C)17 ± 528 ± 319 ± 57 ± 4 < 0.001*Daily total sunshine hours (h)5.1 ± 4.35.1 ± 4.34.2 ± 3.73.6 ± 3.6 < 0.001*Daily average pressure (hPa)1015 ± 61005 ± 31018 ± 61026 ± 5 < 0.001*Daily average relative humidity (%)70 ± 1679 ± 1278 ± 1274 ± 14 < 0.001*Daily total precipitation (mm)3 ± 77 ± 184 ± 112 ± 5 < 0.001*Daily average wind speed (m/s)2.2 ± 0.82.1 ± 0.91.6 ± 0.71.9 ± 0.8 < 0.001*25(OH)D, 25-hydroxyvitamin D.#P was calculated used the chi-squared test.^P was calculated used the Kruskal–Wallis test.*P was calculated used the one-way ANOVA test. $P \leq 0.05$ is considered statistically significance. P-values indicate significant differences for the given parameter between the seasons. Figure 2The changing trends of serum concentration of 25(OH)D by year and month.
## The relationship between meteorological factors and 25(OH)D levels
We used cross-correlation analysis to analyze the correlation between monthly median 25(OH)D values and monthly meteorological factors. Figure 3 shows the relationship between monthly meteorological data (including monthly average temperature, monthly average sunshine hours, and monthly total precipitation) and 25(OH)D levels. In the relationship between temperature and vitamin D, when the lag is equal to 1, the autocorrelation function (ACF) reached a maximum of 0.48, indicating a positive correlation between the temperature in the previous month and 25(OH)D. Similar cross-correlations were found for the mean monthly sunshine hours as well as the monthly total precipitation. No correlation was observed between other meteorological factors and 25(OH)D levels. Figure 3The correlation of meteorological factors with median serum 25(OH)D concentrations.
In addition, we performed univariate and multivariate logistic regression analyses. ( We recalculated the average temperature and average sunshine hours for the 30 days before the pregnant women were examined as the meteorological indicator for the previous month.) As shown in Table 2, we found that mean temperature in the previous month may be an influencing factor for vitamin D deficiency. ( OR = 0.92, $P \leq 0.001$), while no relationship was observed between monthly total precipitation, mean sunshine hours in the previous month and vitamin D deficiency. In the remaining parameters, vitamin D deficiency was significantly more probable in winter than in spring (OR = 1.47, $P \leq 0.001$). Age may be a risk factor for vitamin D deficiency (OR = 0.98, $P \leq 0.001$).Table 2Multivariate logistic regression analysis for associations with vitamin D deficiency. VariablesUnivariate odds ratio$95\%$ CIP valueMultivariate odds ratio$95\%$ CIP valueLowHighLowHighAge0.980.970.99 < 0.001*0.980.970.98 < 0.001*Season (vs spring) Summer0.370.340.41 < 0.001*0.990.871.110.83 Autumn0.290.260.31 < 0.001*0.960.831.100.54 Winter1.131.031.240.0091*1.471.321.62 < 0.001*Average temperature in previous month0.920.910.92 < 0.001*0.920.920.93 < 0.001*Average sunshine hours in previous month0.760.740.77 < 0.001*0.980.961.010.25Monthly total precipitation0.900.890.91 < 0.001*1.010.101.020.17
## Subgroup analysis of the association between meteorological factors and 25(OH)D levels
In Table 3, we could find that the relationship between mean temperature in the previous month and vitamin D deficiency persisted in different seasonal subgroups. ( OR < 1 in all season subgroups) The relationship between age and vitamin D deficiency was also consistently observed. ( OR < 1 in all season subgroups).Table 3Multivariate logistic regression analysis for associations with vitamin D deficiency in different season subgroups. Subgroup(season)VariablesUnivariate odds ratio$95\%$ CIP valueMultivariate odds ratio$95\%$ CIP valueLowHighLowHighSpring ($$n = 5588$$)Age0.980.970.990.006*0.980.971.000.01*Average temperature in previous month0.970.960.98 < 0.001*0.910.890.93 < 0.001*Average sunshine hours in previous month1.020.971.080.431.361.251.49 < 0.001*Monthly total precipitation1.061.011.110.03 *1.171.101.23 < 0.001*Summer ($$n = 5905$$)Age0.980.970.99 < 0.001*0.980.960.99 < 0.001*Average temperature in previous month0.910.900.92 < 0.001*0.910.900.92 < 0.001*Average sunshine hours in previous month0.910.880.95 < 0.001*0.980.941.030.40Monthly total precipitation1.011.001.030.101.031.011.040.008*Autumn ($$n = 5435$$)Age0.970.960.99 < 0.001*0.970.960.99 < 0.001*Average temperature in previous month0.860.840.88 < 0.001*0.840.810.86 < 0.001*Average sunshine hours in previous month0.850.820.87 < 0.001*1.071.011.130.02* ara>Monthly total precipitation0.960.940.97 < 0.001*1.010.981.030.61Winter ($$n = 5162$$)Age0.970.960.99 < 0.001*0.980.971.000.02*Average temperature in previous month0.930.920.94 < 0.001*0.950.930.96 < 0.001*Average sunshine hours in previous month0.690.640.73 < 0.001*0.850.780.92 < 0.001*Monthly total precipitation1.040.991.100.141.040.981.100.20 We divided the age into three subgroups (< 25 years old, 25–35 years old, and > 35 years old) to observe the association between meteorological factors and vitamin D deficiency in pregnant women of different ages. We consistently observed the relationship between the mean temperature in the previous month and vitamin D deficiency in Table 4. ( OR < 1 in all age subgroups).Table 4Multivariate logistic regression analysis for associations with vitamin D deficiency in different age subgroups. Subgroup (age)VariablesUnivariate odds ratio$95\%$ CIP valueMultivariate odds ratio$95\%$ CIP valueLowHighLowHigh < 25 ($$n = 2319$$)Season (vs spring)Summer0.280.190.42 < 0.001*0.980.531.820.95Autumn0.200.130.30 < 0.001*0.960.481.930.92Winter1.050.631.750.861.580.902.780.11Average temperature in previous month0.890.880.91 < 0.001*0.900.860.93 < 0.001*Average sunshine hours in previous month0.770.700.86 < 0.001*1.060.931.200.38Monthly total precipitation0.890.850.93 < 0.001*1.020.961.080.5525–35($$n = 16$$,531)Season (vs spring)Summer0.390.350.43 < 0.001*0.990.851.150.85Autumn0.300.270.33 < 0.001*0.960.811.140.63Winter1.161.031.300.02 *1.481.271.63 < 0.001*Average temperature in previous month0.920.920.92 < 0.001*0.930.920.94 < 0.001*Average sunshine hours in previous month0.770.750.79 < 0.001*1.000.961.030.83Monthly total precipitation0.900.890.91 < 0.001*1.010.991.020.36 > 35[2340]Season (vs spring)Summer0.360.310.42 < 0.001*1.000.811.240.98Autumn0.270.230.31 < 0.001*0.970.761.230.78Winter1.110.951.300.201.531.281.83 < 0.001*Average temperature in previous month0.910.910.92 < 0.001*0.920.910.94 < 0.001*Average sunshine hours in previous month0.730.700.76 < 0.001*0.950.911.000.04*Monthly total precipitation0.900.890.92 < 0.001*1.010.991.030.32
## Non-linear relationship between mean temperature in the previous month and 25(OH)D levels
In fitting the non-linear relationship, we first took the natural logarithm of 25(OH)D to ensure the normality of the data (a plot of the probability density distribution before and after log transformation can be seen in Figure S1). A multivariate-adjusted spline regression model established the non-linear relationship as Fig. 4A shows. ( In the model, we adjusted for age, season, and sunshine hours) Fig. 4B shows the non-linear relationship between average temperature in the previous month and log 25(OH)D level. Figure 4Dose–response relationship between mean temperature and serum 25(OH)D levels.
In subgroup analyses grouped by different ages (< 25 years old, 25–35 years old, > 35 years old), a non-linear relationship was observed between average temperature in the previous month and vitamin D deficiency. As shown in Fig. 5A,C and E, as the temperature increased to 20 °C, the OR value became less than 1. As shown in Fig. 5B,D and F, 25(OH)D levels increased with increasing temperature. Figure 5Dose–response relationship between mean temperature and serum 25(OH)D levels. ( Age subgroup).
In a subgroup analysis of the different seasons, a non-linear relationship was found between average temperature in the previous month and vitamin D deficiency. As shown in Fig. 6A,C,E G, with the increase of temperature, the OR value gradually becomes less than 1. As shown in Fig. 6B,D,F, and H, there was a non-linear relationship between average temperature in the previous month and log 25(OH)D levels in different seasons. *In* general, the level of 25(OH)D increased with increasing temperature. However, this upward trend became flat or unobservable after rising to a certain temperature in spring and summer. A downward trend was observed in the autumn. Figure 6Dose–response relationship between mean temperature and serum 25(OH)D levels. ( Seasonal subgroup).
## Relationship between mean temperature of the same month and 25(OH)D levels
We have additional research on the temperature of the same month in the supplement. We found that mean temperature of the same month is a protective factor for vitamin D deficiency. ( OR = 0.92(not adjusted), OR = 0.93(adjusted for age, season, sunshine hours and precipitation)) Neither the non-linear relationship between temperature of the same month and vitamin D deficiency nor the non-linear relationship between temperature and vitamin D levels was observed. In the subgroup analyses of summer and autumn, we also observed a potential inverted U-shaped relationship between temperature and vitamin D levels in the same month.
## Discussion
Demographic information on the vitamin D levels of pregnant women in eastern *China is* limited. From 2015 to 2020, $65.68\%$ of pregnant women in the *Kunshan area* had 25(OH)D levels above 50 nmol/l. Despite being located in a region of eastern China that receives adequate sunshine throughout the year, a considerable proportion of women in this region still have low vitamin D. As the economy and medical care have developed, the general health of pregnant women has improved, including their vitamin status. Overall 25(OH)D status in pregnant women increased from 2015 to 2018, but this upward trend has leveled off in recent years. One possible reason for the trend of slow growth in 25(OH)D status is the extreme hot weather in recent years. In extremely hot weather, we speculate that the outside habits of pregnant women may change, potentially affecting the amount of UV radiation they receive. This change in going outside habits may affect the vitamin D status.
By analyzing the relationship between meteorological factors and 25(OH)D levels, we found that temperature was an important influencing for vitamin D deficiency. The relationship between temperature and 25(OH)D levels persisted in different subgroups. Based on such findings, we further explored the non-linear relationship between mean temperature in the previous month and vitamin D deficiency (and log 25(OH)D). The change of 25(OH)D concentration with increasing temperature was small until the temperature reached 20℃. While between 20 °C and 25 °C, this variation is wide. It was probable that the pregnant women had more outdoor activities in this comparatively more comfortable temperature range. Then when the temperature exceeded 25 °C, pregnant women spent progressively less time outdoors, which may have contributed to the lower 25(OH)D change. The different definition of vitamin D deficiency had little effect on the non-linear relationship between temperature and vitamin D. However, different definitions of vitamin D can lead to huge variations in the OR of temperature. A low criterion will overestimate the influence of temperature (the OR will be particularly high) and a high criterion will underestimate the influence of temperature (the OR will be particularly close to 1).
Our study provides further population-based information on vitamin D levels in pregnant women in southeastern China. Despite being located in sunny eastern China, the vitamin D levels of pregnant women were generally inadequate. A previous study assessed the status of overall vitamin D levels in pregnant women in China based on data from the China Nutrition and Health Surveillance (CHNS), with median 25(OH)D concentrations of 15.48 (11.89–20.09) ng/mL in 2010–2012 and 2015 –2017 was 13.02 (10.17–17.01) ng/mL, Vitamin D adequacy was only $25.17\%$ in 2010–2011 and $12.57\%$ in 2015–201737. Although, we took different serum 25(OH)D thresholds to define vitamin D deficiency and insufficiency. We still observed chronic vitamin D deficiency and insufficiency in Chinese pregnant women. Studies from India38 and Bangladesh39 have also reported a very high prevalence of vitamin D deficiency and insufficiency in pregnant women. Vitamin D deficiency in pregnant women continues to be a severe public health problem in tropical and subtropical regions.
In addition, in a previous study, researchers found that maternal vitamin D deficiency was positively associated with Hui ethnicity ($$P \leq 0.02$$, relative to Han ethnicity), vitamin D supplementation ($$P \leq 0.02$$) and low ambient UV levels ($P \leq 0.001$). As most of the serum 25(OH)D samples were taken in autumn and winter, seasonal factors were not included in the logistic regression. Still, additional analyses were conducted for autumn and winter samples40. A study from Shenzhen, China, showed that season was also an associated factor for vitamin D deficiency in pregnant women. ( Winter vs autumn vs spring, OR = 3.69, $P \leq 0.001$)41. Other studies report the association between BMI42, gestational frequency43,44 and gestational age39,45–47 and maternal vitamin D deficiency. However, to our knowledge, few studies have focused on exploring the relationship between meteorological factors and vitamin D under different seasons.
The current study's novelty is that the relationship between average temperature in the previous month and 25(OH)D levels varies across the seasons. Specifically, the upward trend in 25(OH)D levels became less pronounced in the spring and summer after reaching certain temperatures. In particular, in autumn, when the average temperature reached 29 °C in the previous month, 25(OH)D levels started to decrease. In recent years, in southeastern China, maximum temperatures reached around 40 °C in summer and autumn. We speculate that excessively high temperatures may lead to fewer pregnant women going outside. They may therefore lack the necessary amount of sunshine and exercise, which may contribute to lower vitamin D levels48. In addition, external vitamin supplementation is an essential factor affecting the vitamin D levels in the human body. And temperature may affect the content of vitamin D supplements. Due to environmental factors such as temperature49, oxygen and light, vitamin D may be lost during food processing and storage50.
Vitamin D levels in pregnant women are critical to their clinical outcomes. Our study assessed the relationship between meteorological factors and 25(OH)D levels through a retrospective study of a large population. Our study took complete account of the seasonality of 25(OH)D levels in the people. However, some limitations of our analysis remain. Although our study included healthy women of the same gestational age, other information, such as BMI and whether sun protection, was not collected. Secondly, the clues obtained in this paper can only indicate the correlation between factors and vitamin D. The genuine causal relationship needs to be verified through further studies. Thirdly, the measurement of serum 25(OH)D concentrations is not the golden standard, which may have an influence on the analysis results.
In conclusion, our study found that vitamin D levels in pregnant women in southeast China remain inadequate. Temperature in the previous month is a factor associated with 25(OH)D levels, but it behaves differently under different seasons. In spring, summer and autumn, temperature and vitamin D levels in pregnant women showed a potential inverted U-shaped relationship: too high or too low temperatures led to lower 25(OH)D levels. Our findings may have important implications for public health strategies involving pregnant women in China.
## Supplementary Information
Supplementary Information. The online version contains supplementary material available at 10.1038/s41598-023-31698-2.
## References
1. Holick MF. **Resurrection of vitamin D deficiency and rickets**. *J. Clin. Invest.* (2006) **116** 2062-2072. DOI: 10.1172/jci29449
2. Holick MF. **Evaluation, treatment, and prevention of vitamin D deficiency: An endocrine society clinical practice guideline**. *J. Clin. Endocrinol. Metab.* (2011) **96** 1911-1930. DOI: 10.1210/jc.2011-0385
3. Holick MF. **Vitamin D deficiency**. *N. Engl. J. Med.* (2007) **357** 266-281. DOI: 10.1056/NEJMra070553
4. Woo J. **Very high rates of vitamin D insufficiency in women of child-bearing age living in Beijing and Hong Kong**. *Br. J. Nutr.* (2008) **99** 1330-1334. DOI: 10.1017/s0007114507844382
5. Hollis BW, Wagner CL. **Assessment of dietary vitamin D requirements during pregnancy and lactation**. *Am. J. Clin. Nutr.* (2004) **79** 717-726. DOI: 10.1093/ajcn/79.5.717
6. Palacios C, Kostiuk LK, Peña-Rosas JP. **Vitamin D supplementation for women during pregnancy**. *Cochrane Database Syst. Rev.* (2019) **7** Cd008873. DOI: 10.1002/14651858.CD008873.pub4
7. Rostami M. **Effectiveness of prenatal vitamin D deficiency screening and treatment program: A stratified randomized field trial**. *J. Clin. Endocrinol. Metab.* (2018) **103** 2936-2948. DOI: 10.1210/jc.2018-00109
8. Wagner CL, Hollis BW. **The implications of vitamin D status during pregnancy on mother and her developing child**. *Front. Endocrinol. (Lausanne)* (2018) **9** 500. DOI: 10.3389/fendo.2018.00500
9. Khaing W. **Calcium and vitamin D supplementation for prevention of preeclampsia: A systematic review and network meta-analysis**. *Nutrients* (2017). DOI: 10.3390/nu9101141
10. Rizzo G. **Vitamin D and gestational diabetes mellitus: Is there a link?**. *Antioxidants (Basel)* (2019). DOI: 10.3390/antiox8110511
11. Woo J, Giurgescu C, Wagner CL. **Evidence of an association between vitamin D deficiency and preterm birth and preeclampsia: A critical review**. *J. Midwifery Womens Health* (2019) **64** 613-629. DOI: 10.1111/jmwh.13014
12. McDonnell SL. **Maternal 25(OH)D concentrations ≥40 ng/mL associated with 60% lower preterm birth risk among general obstetrical patients at an urban medical center**. *PLoS One* (2017) **12** e0180483. DOI: 10.1371/journal.pone.0180483
13. Bi WG. **Association between vitamin D supplementation during pregnancy and offspring growth, morbidity, and mortality: A systematic review and meta-analysis**. *JAMA Pediatr.* (2018) **172** 635-645. DOI: 10.1001/jamapediatrics.2018.0302
14. Khazai N, Judd SE, Tangpricha V. **Calcium and vitamin D: Skeletal and extraskeletal health**. *Curr. Rheumatol. Rep.* (2008) **10** 110-117. DOI: 10.1007/s11926-008-0020-y
15. Cuppari L, Garcia Lopes MG, Kamimura MA. **Vitamin D biology: From the discovery to its significance in chronic kidney disease**. *J. Ren. Nutr.* (2011) **21** 113-116. DOI: 10.1053/j.jrn.2010.10.007
16. Jean G, Souberbielle JC, Chazot C. **Vitamin D in chronic kidney disease and dialysis patients**. *Nutrients* (2017). DOI: 10.3390/nu9040328
17. Cheng Q. **Factors associated to serum 25-hydroxyvitamin D levels among older adult populations in urban and suburban communities in Shanghai, China**. *BMC Geriatr* (2017) **17** 246. DOI: 10.1186/s12877-017-0632-z
18. Al-Musharaf S. **Vitamin D deficiency prevalence and predictors in early pregnancy among Arab women**. *Nutrients* (2018). DOI: 10.3390/nu10040489
19. Yan X. **Vitamin D status of residents in Taiyuan, China and Influencing Factors**. *Nutrients* (2017). DOI: 10.3390/nu9080898
20. Tønnesen R, Hovind PH, Jensen LT, Schwarz P. **Determinants of vitamin D status in young adults: Influence of lifestyle, sociodemographic and anthropometric factors**. *BMC Public Health* (2016) **16** 385. DOI: 10.1186/s12889-016-3042-9
21. Karamizadeh M, Seif M, Holick MF, Akbarzadeh M. **Developing a model for prediction of serum 25-Hydroxyvitamin D level: The use of linear regression and machine learning methods**. *J. Am. Nutr. Assoc.* (2022) **41** 191-200. DOI: 10.1080/07315724.2020.1869624
22. Ding F, Nie X, Li X, He Y, Li G. **Data mining: Biological and temporal factors associated with blood parathyroid hormone, vitamin D, and calcium concentrations in the Southwestern Chinese population**. *Clin. Biochem.* (2021) **90** 50-57. DOI: 10.1016/j.clinbiochem.2021.01.014
23. Shen M. **Seasonal variation and correlation analysis of vitamin D and parathyroid hormone in Hangzhou, Southeast China**. *J. Cell Mol. Med.* (2020) **24** 7370-7377. DOI: 10.1111/jcmm.15330
24. Corrêa MP, Yamamoto ALC, Moraes GR, Godin-Beekmann S, Mahé E. **Changes in the total ozone content over the period 2006 to 2100 and the effects on the erythemal and vitamin D effective UV doses for South America and Antarctica**. *Photochem. Photobiol. Sci.* (2019) **18** 2931-2941. DOI: 10.1039/c9pp00276f
25. Leal A, Corrêa MP, Holick MF, Melo EV, Lazaretti-Castro M. **Sun-induced production of vitamin D(3) throughout 1 year in tropical and subtropical regions: relationship with latitude, cloudiness, UV-B exposure and solar zenith angle**. *Photochem. Photobiol. Sci.* (2021) **20** 265-274. DOI: 10.1007/s43630-021-00015-z
26. Engelsen O. **The relationship between ultraviolet radiation exposure and vitamin D status**. *Nutrients* (2010) **2** 482-495. DOI: 10.3390/nu2050482
27. Chen W. **Relationship between cancer mortality/incidence and ambient ultraviolet B irradiance in China**. *Cancer Causes Control* (2010) **21** 1701-1709. DOI: 10.1007/s10552-010-9599-1
28. Chen W. **Relationship between cancer survival and ambient ultraviolet B irradiance in China**. *Cancer Causes Control* (2013) **24** 1323-1330. DOI: 10.1007/s10552-013-0210-4
29. Niino M. **Seasonal fluctuations in serum levels of vitamin D in Japanese patients with multiple sclerosis**. *J. Neuroimmunol.* (2021) **357** 577624. DOI: 10.1016/j.jneuroim.2021.577624
30. Niculescu DA. **Seasonal periodicity of serum parathyroid hormone and its relation with vitamin D in Romania**. *Arch. Osteoporos.* (2020) **15** 66. DOI: 10.1007/s11657-020-00744-1
31. Hyppönen E, Power C. **Hypovitaminosis D in British adults at age 45 y: Nationwide cohort study of dietary and lifestyle predictors**. *Am. J. Clin. Nutr.* (2007) **85** 860-868. DOI: 10.1093/ajcn/85.3.860
32. Kroll MH. **Temporal relationship between vitamin D status and parathyroid hormone in the United States**. *PLoS One* (2015) **10** e0118108. DOI: 10.1371/journal.pone.0118108
33. Sauer MV. **Reproduction at an advanced maternal age and maternal health**. *Fertil. Steril.* (2015) **103** 1136-1143. DOI: 10.1016/j.fertnstert.2015.03.004
34. Brehmer L, Alexanderson K, Schytt E. **Days of sick leave and inpatient care at the time of pregnancy and childbirth in relation to maternal age**. *Scand. J. Public Health* (2017) **45** 222-229. DOI: 10.1177/1403494817693456
35. Wu H, Zhao M, Liang Y, Liu F, Xi B. **Maternal age at birth and neonatal mortality: Associations from 67 low-income and middle-income countries**. *Paediatr. Perinat. Epidemiol.* (2021) **35** 318-327. DOI: 10.1111/ppe.12734
36. Kato T. **Association of maternal age with child health: A Japanese longitudinal study**. *PLoS One* (2017) **12** e0172544. DOI: 10.1371/journal.pone.0172544
37. Hu Y. **Vitamin D nutritional status of chinese pregnant women, comparing the Chinese national nutrition surveillance (CNHS) 2015–2017 with CNHS 2010–2012**. *Nutrients* (2021). DOI: 10.3390/nu13072237
38. Sahu M. **Vitamin D deficiency in rural girls and pregnant women despite abundant sunshine in northern India**. *Clin. Endocrinol. (Oxf)* (2009) **70** 680-684. DOI: 10.1111/j.1365-2265.2008.03360.x
39. Ahmed F, Khosravi-Boroujeni H, Khan MR, Roy AK, Raqib R. **Prevalence and Predictors of Vitamin D Deficiency and Insufficiency among Pregnant Rural Women in Bangladesh**. *Nutrients* (2021). DOI: 10.3390/nu13020449
40. Yun C. **Vitamin D deficiency prevalence and risk factors among pregnant Chinese women**. *Public Health Nutr.* (2017) **20** 1746-1754. DOI: 10.1017/s1368980015002980
41. Zhao L. **High prevalence of vitamin D deficiency in Shenzhen pregnant women**. *J. Matern. Fetal Neonatal. Med.* (2021). DOI: 10.1080/14767058.2021.1910667
42. Shen Y. **Vitamin D nutrient status during pregnancy and its influencing factors**. *Clin. Nutr.* (2020) **39** 1432-1439. DOI: 10.1016/j.clnu.2019.06.002
43. Johnson DD. **Vitamin D deficiency and insufficiency is common during pregnancy**. *Am. J. Perinatol.* (2011) **28** 7-12. DOI: 10.1055/s-0030-1262505
44. Parlak M. **Severe vitamin D deficiency among pregnant women and their newborns in Turkey**. *J. Matern. Fetal. Neonatal. Med.* (2015) **28** 548-551. DOI: 10.3109/14767058.2014.924103
45. Bowyer L. **Vitamin D, PTH and calcium levels in pregnant women and their neonates**. *Clin. Endocrinol. (Oxf)* (2009) **70** 372-377. DOI: 10.1111/j.1365-2265.2008.03316.x
46. Aji AS, Yusrawati Y, Malik SG, Lipoeto NI. **The association of maternal vitamin D status during pregnancy and neonatal anthropometric measurements: A longitudinal study in Minangkabau Pregnant Women, Indonesia**. *J. Nutr. Sci. Vitaminol. (Tokyo)* (2020) **66** S63-S70. DOI: 10.3177/jnsv.66.S63
47. Agudelo-Zapata Y. **Serum 25-hydroxyvitamin D levels throughout pregnancy: A longitudinal study in healthy and preeclamptic pregnant women**. *Endocr. Connect* (2018) **7** 698-707. DOI: 10.1530/ec-18-0055
48. Perreault M. **Structured diet and exercise guidance in pregnancy to improve health in women and their offspring: Study protocol for the Be Healthy in Pregnancy (BHIP) randomized controlled trial**. *Trials* (2018) **19** 691. DOI: 10.1186/s13063-018-3065-x
49. Temova Ž, Roškar R. **Shelf life after opening of prescription medicines and supplements with vitamin D(3) for paediatric use**. *Eur. J. Hosp. Pharm.* (2017) **24** 115-119. DOI: 10.1136/ejhpharm-2016-000895
50. Hanson AL, Metzger LE. **Evaluation of increased vitamin D fortification in high-temperature, short-time-processed 2% milk, UHT-processed 2% fat chocolate milk, and low-fat strawberry yogurt**. *J Dairy Sci* (2010) **93** 801-807. DOI: 10.3168/jds.2009-2694
|
---
title: ER stress induces upregulation of transcription factor Tbx20 and downstream
Bmp2 signaling to promote cardiomyocyte survival
authors:
- Shreya Das
- Arunima Mondal
- Chandrani Dey
- Santanu Chakraborty
- Rudranil Bhowmik
- Sanmoy Karmakar
- Arunima Sengupta
journal: The Journal of Biological Chemistry
year: 2023
pmcid: PMC10036653
doi: 10.1016/j.jbc.2023.103031
license: CC BY 4.0
---
# ER stress induces upregulation of transcription factor Tbx20 and downstream Bmp2 signaling to promote cardiomyocyte survival
## Body
In mammals, the developing heart is highly proliferative prior to birth, and it involves the interplay of multiple signaling pathways. However, after birth, the cardiomyocytes lose its plasticity, exit the cell cycle, its proliferative capacity dissipates, and the cells grow in size primarily by hypertrophy [1]. In the neonates, post 1 week after birth, the cardiomyocytes become binucleated, express adult contractile protein isoforms, and lose its ability to regenerate [2, 3, 4]. The notion that adult cardiomyocytes lose their capacity to proliferate because of cell cycle arrest was revoked by growing studies showing that resident adult myocardial cardiomyocyte re-enters cell cycle following myocardial injury by regulating key regulatory pathways [5].
T-box transcription factor 20 (Tbx20) is a member of the Tbx1 subfamily of T-box–containing genes and plays pivotal roles in development and maintenance of heart by driving cardiomyocyte proliferation [6]. Loss of function of Tbx20 leads to unlooped and severely hypoplastic heart with embryonic lethality [7, 8, 9]. Ablation of Tbx20 in adult cardiomyocytes leads to severe cardiomyopathy with arrhythmias and death [10]. Gain of function of Tbx20 leads to increased cardiomyocyte proliferation in fetal heart development [11].
Endoplasmic reticulum (ER) is an organelle that mediates production and folding of different secretory and membrane proteins [12]. Any sort of dysregulation in the machinery of the ER because of external factors or internal stimulus leads to accumulation of misfolded protein leading to generation of ER stress. ER stress activates the adaptive cellular response signaling cascade known as unfolded protein response (UPR), which consists of three pathways, activating transcription factor 6 (ATF6), inositol-requiring enzyme 1 alpha (IRE1α), and protein kinase RNA-activated-like ER kinase (PERK). The protective UPR is initially beneficial as it works for restoration of homeostasis; however, a severe ER stress leads to cell death via apoptosis. There is a very delicate balance between ER stress–induced prosurvival and proapoptosis [13]. Tbx20 overexpression was previously shown to induce proliferation of cardiomyocytes during oxidative stress and hypoxia [14]; however, its mechanistic role during ER stress–mediated cardiomyopathy is still elusive.
Our study for the first time identified the novel unknown function of Tbx20 that is able to directly enhance the protective responses of the UPR for restoration of ER homeostasis in the milieu of cardiac injury. Since ER stress have been implicated in the development of multiple cardiomyopathies; hence, we have used tunicamycin (Tun)-induced ER stress as our cardiomyopathy model system in vivo in order to look into a global phenomenon of ER stress–related cardiomyopathies. In the current study, we examine the function of Tbx20 and bone morphogenetic protein 2 (Bmp2) signaling during ER stress–induced cardiomyopathy. ER stress–mediated upregulation of Tbx20 resulted in increased cardiomyocyte proliferation via upregulating the Bmp2–pSmad$\frac{1}{5}$/8 signaling axis. Upregulation of Tbx20 also resulted in decreased cardiomyocyte apoptosis and fibrosis. However, increasing the intensity of ER stress or prolonging the ER stress resulted in decreased cardiomyocyte proliferation, increased apoptosis and fibrosis, thus disrupting cardiomyocyte homeostasis. In addition, we identify tbx20 as a direct target of Atf6 during ER stress–mediated cardiomyopathy. Our study reported an elevated level of Bmp2 protein both during the initial stages as well as prolonged ER stress response in adult murine hearts, thus making it a potential biomarker candidate for early detection of ER stress–mediated cardiomyopathies.
## Abstract
In the mammalian heart, fetal cardiomyocytes proliferate prior to birth; however, they exit the cell cycle shortly after birth. Recent studies show that adult cardiomyocytes re-enters the cell cycle postinjury to promote cardiac regeneration. The endoplasmic reticulum (ER) orchestrates the production and assembly of different types of proteins, and a disruption in this machinery leads to the generation of ER stress, which activates the unfolded protein response. There is a very fine balance between ER stress–mediated protective and proapoptotic responses. T-box transcription factor 20 (Tbx20) promotes embryonic and adult cardiomyocyte proliferation postinjury to restore cardiac homeostasis. However, the function and regulatory interactions of Tbx20 in ER stress–induced cardiomyopathy have not yet been reported. We show here that ER stress upregulates Tbx20, which activates downstream bone morphogenetic protein 2 (Bmp2)-pSmad$\frac{1}{5}$/8 signaling to induce cardiomyocyte proliferation and limit apoptosis. However, augmenting ER stress reverses this protective response. We also show that increased expression of tbx20 during ER stress is mediated by the activating transcription factor 6 arm of the unfolded protein response. Cardiomyocyte-specific loss of Tbx20 results in decreased cardiomyocyte proliferation and increased apoptosis. Administration of recombinant Bmp2 protein during ER stress upregulates Tbx20 leading to augmented proliferation, indicating a feed-forward loop mechanism. In in vivo ER stress, as well as in diabetic cardiomyopathy, the activity of Tbx20 is increased with concomitant increased cardiomyocyte proliferation and decreased apoptosis. These data support a critical role of Tbx20-Bmp2 signaling in promoting cardiomyocyte survival during ER stress–induced cardiomyopathies.
## Tbx20 activity and Bmp2 activity are increased upon induction of ER stress in H9c2 cells in vitro
The role of Tbx20 and Bmp2 signaling during ER stress–mediated cardiomyopathy has not been reported to date. To examine the expression profile of Tbx20 and Bmp2 upon induction of ER stress, we treated H9c2 cells with increasing concentration (2, 5, 10, 20, and 50 μg/ml) of ER stress–inducer Tun for 4, 8, 12, and 24 h. The 4 h time point did not show any significant change in the expression of Tbx20 and Bmp2 as compared with control (data not shown). Western blot analysis showed a gradual increase in the expression of Tbx20 and Bmp2 with increasing concentration of Tun with respect to control during 8 and 12 h (Fig. S1, A and B). Increased expression of cleaved form of Atf6 (Atf6-p50) at 24 h time point indicates the establishment of ER stress in our culture condition (Fig. 1, B and C). The expression of Atf6-p50 however decreased in the 50 μg/ml Tun-treated cells as compared with 20 μg/ml Tun-treated cells. Western blot analysis further showed a gradual increase in expression of Tbx20 up to a concentration of 20 μg/ml Tun during 24 h. However, increasing the concentration of Tun to 50 μg/ml resulted in significant decrease in the expression of Tbx20 (1.13 ± 0.07-fold) as compared with 20 μg/ml Tun-treated (4.26 ± 0.073-fold) cells. The Bmp2 signal transduction pathway was also increased gradually with increasing concentration of Tun. The expression of Bmp2 however decreased significantly in 50 μg/ml Tun-treated cells (0.66 ± 0.42-fold) as compared with 20 μg/ml Tun cell treatment (4.13 ± 0.05-fold; Fig. 1, B and C). Decrease in the expression of Bmp2 in the 50 μg/ml Tun-treated cells was accompanied by a decrease in its downstream signaling cascade molecule pSmad$\frac{1}{5}$/8 (1.06 ± 0.11-fold) as compared with 20 μg/ml Tun-treated cells (5.6 ± 0.39-fold; Fig. 1, B and C). Since the fold change increase in the expression of Tbx20 and Bmp2 was highest during 24 h as compared with 8 and 12 h, hence, this time point was used for further experiments. Figure 1T-box transcription factor 20 (Tbx20) and bone morphogenetic protein 2 (Bmp2) activity is increased upon endoplasmic reticulum (ER) stress induction in vitro. A, the expression of ER stress markers atf6 and grp78 is increased upon tunicamycin (Tun) treatment in H9c2 cells as determined by quantitative real-time PCR (qRT–PCR). The activity of Tbx20 and Bmp2 is concomitantly increased gradually upon ER stress induction (2 μg/ml Tun, 5 μg/ml Tun, 10 μg/ml Tun, and 20 μg/ml Tun). However, a 50 μg/ml Tun treatment resulted in decrease in the activity of Tbx20 and Bmp2 as determined by qRT–PCR. B, Western blot analysis showing a gradual increase in the expression of activating transcription factor 6 (Atf6)-p50 with increasing concentration of Tun. However, its expression decreased in 50 μg/ml Tun-treated cells. The expression of Tbx20, Bmp2, and its downstream cascade molecule pSmad$\frac{1}{5}$/8 also increased gradually up to a concentration of 20 μg/ml Tun. Increasing the concentration of Tun to 50 μg/ml resulted in significant decrease in the expression of Tbx20, Bmp2, and pSmad$\frac{1}{5}$/8. C, quantitative representation by ImageJ software of the proteins using three biological replicates from B. D, immunostaining analysis showing an increase in the expression of Tbx20, Bmp2, pSmad$\frac{1}{5}$/8, and Atf6 proteins upon increasing ER stress induction (a’, a’’, a’’’, and a’’’’), (b’, b’’, b’’’, and b’’’’), (c’, c’’, c’’’, and c’’’’), and (d’, d’’, d’’’, and d’’’’) as compared with control (a, b, c, and d), respectively. Increasing the intensity of ER stress by treatment with 50 μg/ml Tun however resulted in the decrease in expression of Tbx20 (a’’’’’), Bmp2 (b’’’’’), pSmad$\frac{1}{5}$/8 (c’’’’’), and Atf6 (d’’’’’). Insets in B show single-channel cropped images of indicated areas (white arrows). Scale bar of main images represent 50 μm. Scale bar of inset represents 20 μm. E–H, quantitative representation of D. Statistical significance was calculated by one-way ANOVA. Error bars represent SD from at least three biological replicates. ns, p: nonsignificant, ∗$p \leq 0.05$, ∗∗$p \leq 0.005$, ∗∗∗$p \leq 0.0005$, ##$p \leq 0.0001$; n ≥ 3 independent experiments.
Similar results were observed at mRNA level. The increased expression of ER stress markers atf6 and grp78 in the presence of 2, 5, 10, 20, and 50 μg/ml Tun, respectively, compared with control cell indicates ER stress induction (Fig. 1A). Interestingly, the expression of both tbx20 and bmp2 was also elevated gradually up to 20 μg/ml as compared with control. However, the expression of tbx20 and bmp2 was significantly decreased in 50 μg/ml Tun-treated cells as compared with 20 μg/ml Tun-treated cells (Fig. 1A).
To decipher whether other ER stress inducers have similar effect on the expression of Tbx20 and Bmp2, H9c2 cells were treated with DTT and thapsigargin (Tg). H9c2 cells were treated with DTT at a concentration of 1, 3, 5, and 10 mM for 24 h. The expression of Tbx20 was increased gradually up to 3 mM DTT-treated (2.5 ± 0.2-fold) cells as compared with control cells. However, it later decreased in 5 mM DTT- (0.82 ± 0.06-fold) and 10 mM DTT-treated (0.46 ± 0.04-fold) cells (Fig. S1, D and E). The expression of Bmp2 showed a similar pattern of increase up to 3 mM DTT treatment. The expression of Bmp2 in 5 mM DTT-treated cells is decreased; however, it was nonsignificant with respect to 3 mM. The expression of Bmp2 decreased significantly in 10 mM DTT-treated cells. H9c2 cardiomyocytes were treated with different concentrations (1.5, 3, 6, and 10 μM) of Tg. Tg treatment resulted in increased expression of both Tbx20 and Bmp2 up to 6 μM concentration. However, their expression later decreased in 10 μM Tg-treated cells (Fig. S1, F and G). Thus, all three ER stress inducers like Tun, DTT, and Tg result in increased expression of Tbx20 and Bmp2 up to a certain extent of ER stress. Increasing the ER stress further results in decrease in the expression of both Tbx20 and Bmp2.
Immunofluorescence technique was employed to determine the protein expression of the markers following ER stress induction. H9c2 cells treated with Tun showed significant increase in nuclear expression of Tbx20 up to 5.4 ± $0.18\%$, 9.5 ± $0.62\%$, 21.67 ± $2.3\%$, and 42.27 ± $2.6\%$ when treated with 2 μg/ml Tun (Fig. 1, Da’ and E), 5 μg/ml Tun (Fig. 1, Da’’ and E), 10 μg/ml Tun (Fig. 1, Da’’’ and E), and 20 μg/ml Tun (Fig. 1, Ba’’’’ and E), respectively, compared with control (1.06 ± $0.03\%$; Fig. 1, Da and E). However, its expression later decreased in 50 μg/ml Tun-treated cells (Fig. 1, Da’’’’’ and E). The immunoreactivity of Bmp2 also followed a similar trend of increase up to 20 μg/ml Tun (Fig. 1, Db’’’’ and F) as compared with control (Fig. 1, Db and F), which again decreased in 50 μg/ml Tun-treated (Fig. 1, Db’’’’’ and F) cells. Increase in Bmp2 signaling during ER stress was apparent by increased pSmad$\frac{1}{5}$/8-positive nuclei in 5 μg/ml Tun- (Fig. 1, Dc’’ and G), 10 μg/ml Tun- (Fig. 1, Dc’’’ and G), and 20 μg/ml Tun-treated (Fig. 1, Dc’’’’ and G) cells as compared with control cells (Fig. 1, Dc and G). Its expression later decreased in 50 μg/ml Tun-treated cells (Fig. 1, Dc’’’’’ and G). The nuclear localization of Atf6 also increased up to 3.2 ± $0.3\%$ in 2 μg/ml (Fig. 1, Dd’ and H), 10.27 ± $1.0\%$ in 5 μg/ml (Fig. 1, Dd’’ and H), 33.27 ± $4.9\%$ in 10 μg/ml (Fig. 1, Dd’’’ and H), and up to 54.73 ± $2.5\%$ in 20 μg/ml Tun-treated (Fig. 1, Dd’’’’ and H) cells in comparison to control (Fig. 1, Dd and H). However, its expression decreased to 8.86 ± $1.3\%$ in 50 μg/ml Tun-treated (Fig. 1, Dd’’’’’ and H) cells. These data further corroborate that Tbx20–Bmp2 signaling is elevated gradually with increasing ER stress, which is later decreased as the intensity of ER stress is increased in accordance with the cell death that is observed at higher concentration of Tun (Fig. S1C). The expression profile of Atf6 also correlated with the expression of Tbx20. These data thus suggest that the expression of Tbx20 and Bmp2 correlate with the viable status of the cells as well as the physiological status of ER stress.
## Atf6-mediated induction of Tbx20 promotes cardiomyocyte proliferation and limits cardiomyocyte apoptosis
Our study has shown an increase in the expression of Tbx20 during ER stress induced by Tun, DTT, and Tg; however, the mechanism behind this increase in the expression is still elusive. Since the increase in the expression of Tbx20 followed a similar pattern of increase to that of Atf6-p50 and with decrease in the expression of Atf6-p50 in 50 μg/ml Tun-treated cells, the expression of Tbx20 also decreased; this led us to speculate a possible role of Atf6 in the upregulation of Tbx20 during ER stress. Atf6, which is a basic leucine zipper family of transcription factor, was previously shown to bind to canonical UPRE TGACGTGG/A of various genes in order to transcribe them [15]. Atf6 was shown to impart its cardioprotective effect in the direction of prosurvival by ameliorating the extent of ER stress, and it also has role in compensatory myocyte growth. It was also shown to confer global protection of cardiomyocytes from ischemia/reperfusion injury by reprogramming cellular proteostasis [16]. Atf6 was also shown to play vital role in maintaining homeostasis of cardiomyocytes under both pathological and physiological states. In one study, chromatin immunoprecipitation (ChIP)-*Seq analysis* was performed to identify the putative targets of Atf6 [17]. This study identified Tbx20 as one of the possible targets of Atf6. Hence, tbx20 genomic sequences were examined for Atf6 binding consensus sequence. Bioinformatics analysis revealed the presence of canonical UPRE TGACGTG binding sequence for Atf6 in the promoter of rat tbx20 gene (Fig. 2A). To examine whether Atf6 controls the expression of tbx20 gene through direct binding, H9c2 cells were treated with 20 μg/ml Tun, and the DNA-binding ability of Atf6 to tbx20 promoter was performed using ChIP assay. The results revealed direct binding of Atf6 to the promoter region of tbx20 in ER stress–induced H9c2 cells (Fig. 2B). Real-time analysis revealed that in Tun-treated cardiomyocytes, Atf6 binds to the promoter of tbx20 with 19.85 ± 2.8-fold enrichment over immunoglobulin G (IgG) controls (Fig. 2C).Figure 2During endoplasmic reticulum (ER) stress, activating transcription factor 6 (Atf6)-mediated upregulation of T-box transcription factor 20 (Tbx20) promotes cardiomyocyte proliferation and limits cardiomyocyte apoptosis in cultured H9c2 cardiomyocytes. A, the rat tbx20 gene contains conserved canonical Atf6 DNA-binding sequence TGACGTG in the tbx20 promoter region. B, chromatin immunoprecipitation (ChIP) analysis showed direct binding of Atf6 in tbx20 promoter region. C, ChIP assay followed by quantitative RT–PCR (qRT–PCR) showed ∼19-fold enrichment of Atf6 binding to the tbx20 promoter region during ER stress induction. D, Western blot analysis revealed pretreatment of H9c2 cardiomyocytes with 300 μM AEBSF followed by ER stress induction resulted in significant decrease in the expression of Atf6-p50. Inhibition of nuclear translocation of Atf6 was accompanied by concomitant decrease in the expression of Tbx20. E, quantitative representation by ImageJ software of the proteins using three biological replicates from D. F, level of ROS generation upon induction of ER stress showed a gradual increase in the level of ROS production by DCFDA method. Decrease in the expression of Tbx20 in 50 μg/ml tunicamycin (Tun)-treated cells was accompanied by significant rise in ROS levels as compared with 20 μg/ml Tun-treated cells. G, immunofluorescence staining revealed that increase in expression of Tbx20 during ER stress is accompanied with increased expression of proliferative marker Ki67 (a’’, a’’’, and a’’’’) as compared with control (a). However, a Tun concentration of 50 μg/ml resulted in decrease in cardiomyocyte proliferation marked by decreased expression of Ki67 (a’’’’’). Decrease in expression of Tbx20 is accompanied by increased cardiomyocyte apoptosis. The expression of apoptosis inducer Chop and proapoptotic marker *Bax is* increased upon 50 μg/ml Tun treatment (b’’’’’ and c’’’’’) in comparison to lower concentrations of Tun (b’, b’’, b’’’, and b’’’’) and (c’, c’’, c’’’, and c’’’’). The expression of antiapoptotic marker BclXL is decreased during 50 μg/ml Tun treatment (d’’’’’) in comparison to lower concentrations of Tun (d’, d’’, d’’’, and d’’’’). Insets in G show single-channel cropped images of indicated areas (white arrows). Scale bar of main images represents 50 μm. Scale bar of inset represents 20 μm. H–K, quantitative representation of G. Statistical significance was calculated by one-way ANOVA. Error bars represent SD from $$n = 3$$ independent biological replicates. ns, p: nonsignificant, ∗$p \leq 0.05$, ∗∗$p \leq 0.005$, ∗∗∗$p \leq 0.0005$, ##$p \leq 0.0001.$ AEBSF, 4-(2-aminoethyl) benzenesulfonyl fluoride hydrochloride; DCFDA, 2′,7′-dichlorofluorescin diacetate; ROS, reactive oxygen species.
In order to corroborate the direct binding of Atf6 to tbx20 promoter, the H9c2 cardiomyocytes were treated with Atf6-specific inhibitor 4-(2-aminoethyl) benzenesulfonyl fluoride hydrochloride (AEBSF), which blocks the cleavage of Golgi-membrane-bound Atf6, thereby blocking its nuclear translocation followed by treatment with Tun (20 μg/ml). Western blot analysis showed decreased nuclear localization of Atf6-p50 by 0.01 ± 0.01-fold upon treatment with AEBSF followed by Tun as compared with Tun treatment alone (2.43 ± 0.02-fold; Fig. 2, D and E). Decrease in the nuclear translocation of Atf6 because of AEBSF treatment was accompanied by subsequent decrease in the expression of Tbx20 (0.21 ± 0.07-fold) during ER stress as compared with ER stress induction alone (1.97 ± 0.03-fold; Fig. 2, D and E). Therefore, our study showed that increase in the expression of Tbx20 during ER stress was mediated by Atf6 because of its direct DNA-binding ability to the promoter region of tbx20 gene.
Tun acts by inhibiting N-linked glycosylation, thus resulting in improper maturation of proteins [18]. Increasing ER stress results in increased disruption of the disulphide bonds resulting in increased reactive oxygen species (ROS) generation [19, 20]. We therefore looked into the ROS levels in our study. Increasing the ER stress resulted in a gradual increase in ROS generation. However, at a Tun concentration of 50 μg/ml when the expression of Tbx20 was reduced, the level of ROS increased drastically (Fig. 2F). Previous studies have shown that overexpression of Tbx20 results in decreased levels of ROS [21]. Thus, our study showed that ER stress–induced increase in the Tbx20 restricts ROS generation; however, decrease in the expression of Tbx20 results in drastic increase in ROS generation during ER stress.
Tbx20 was shown to promote fetal cardiomyocyte proliferation and inhibit cardiomyocyte apoptosis [22, 23]. To decipher the role of increased Tbx20 during ER stress–induced cardiomyopathy, we looked into the expression profile of proliferative (Ki67) and apoptotic markers (Chop and Bax) post ER stress induction in H9c2 cells. The Ki67-positive nuclei increased gradually up to 12.4 ± $2.4\%$ in 5 μg/ml Tun (Fig. 2, Ga’’ and H), 34.3 ± $5.5\%$ in 10 μg/ml Tun (Fig. 2, Ga’’’ and H), and 60.43 ± $4.04\%$ in 20 μg/ml Tun-treated (Fig. 2, Ga’’’’ and H) cells, respectively, as compared with control (Fig. 2, Ga and H). However, it later decreased to 3.5 ± $0.6\%$ in 50 μg/ml Tun-treated cells (Fig. 2, Ga’’’’’ and H). The increased expression for Ki67 correlated with that of Tbx20, with highest cardiomyocyte proliferation in 20 μg/ml Tun-treated cells where the expression of Tbx20 and Bmp2 was also the highest. Tg treatment also resulted in increased cardiomyocyte proliferation as marked by increased Ki67-positive cells at a concentration of 3 μM Tg-treated cells (Fig. S2, Aa’ and B) where the expression of Tbx20 and Bmp2 was also increased (Fig. S1, F and G) as compared with control (Fig. S2, Aa and B). However, it later decreased at a concentration of 10 μM Tg (Fig. S2, Aa’’ and B). Thus, Tbx20 works during ER stress by increasing cardiomyocyte proliferation.
Conversely, increased expression of Tbx20 limits cardiomyocyte apoptosis. Chop plays a pivotal role in ER stress–mediated apoptosis by downregulating the expression of antiapoptotic marker Bcl-XL and augmenting the expression of proapoptotic marker Bax [24]. The expression of Chop increased significantly up to 59.13 ± $4.8\%$ in 50 μg/ml Tun-treated (Fig. 2, Gb’’’’’ and I) cells as compared with 20 μg/ml Tun (15.17 ± $2.8\%$; Fig. 2, Gb’’’’ and I). Tg treatment also resulted in significant increase in the expression of Chop at a concentration of 10 μM Tg (Fig. S2, Ab’’ and B) where the expression of Tbx20 was decreased (Fig. S1, F and G) as compared with control (Fig. S2, Ab and B) and 3 μM Tg-treated (Fig. S2, Ab’ and B) cells. Increased expression of Chop resulted in increased immunoreactivity of Bax up to 33.23 ± $3.2\%$ in 50 μg/ml Tun-treated (Fig. 2, Gc’’’’’ and J) cells as compared with 20 μg/ml Tun (7.03 ± $2.9\%$; Fig. 2, Gc’’’’ and J), 10 μg/ml Tun (5.2 ± $0.8\%$; Fig. 2, Gc’’’ and J), and 5 μg/ml Tun (2.9 ± $0.37\%$; Fig. 2, Gc’’ and J), 2 μg/ml Tun (1.9 ± $0.85\%$; Fig. 2, Gc’ and J), and control (1.7 ± $0.62\%$; Fig. 2, Gc and J) cells. On the contrary, the expression of antiapoptotic marker Bcl-XL decreased to 18.67 ± $6.5\%$ in 50 μg/ml Tun-treated (Fig. 2, Gd’’’’’ and K) cells in comparison to 20 μg/ml Tun (55.83 ± $5.1\%$; Fig. 2, Gd’’’’ and K), 10 μg/ml Tun (60 ± $1.0\%$; Fig. 2, Gd’’’ and K), 5 μg/ml Tun (64.6 ± $4.5\%$; Fig. 2, Gd’’ and K), 2 μg/ml Tun (69.67 ± $3.1\%$; Fig. 2, Gd’ and K), and control (74 ± $5.0\%$; Fig. 2, Gd and K) cells. Thus, ER stress–induced expression of Tbx20 results in activation of Bmp2–pSmad$\frac{1}{5}$/8 signaling that increases cardiomyocyte proliferation. However, after a certain threshold, ER stress eventually leads to decrease in the expression of Tbx20 with concomitant increase in cardiomyocyte apoptosis and Tbx20 fails to impart its protective role.
## Tbx20 is necessary, and it acts upstream of Bmp2–pSmad1/5/8 signaling in protecting cardiomyocytes against Tun-induced ER stress
Our study has shown that Tbx20 promotes cardiomyocyte proliferation during ER stress by activating Bmp2–pSmad$\frac{1}{5}$/8 signaling axis. Previous studies have shown that Tbx20 acts upstream of Bmp2 during heart development [22, 25, 26]. To determine the molecular hierarchy between Tbx20–Bmp2–Smad$\frac{1}{5}$/8 signaling axis and to decipher the mode of action by which Tbx20 protects the cardiomyocyte against ER stress, Tun-treated H9c2 cells were pretreated with Tbx20-specific siRNA. Treatment of H9c2 cells with 100 nM Tbx20 siRNA resulted in $73\%$ reduction in the expression of endogenous level of Tbx20 (Fig. S2, H and I). Western blot analysis showed that pretreatment of H9c2 cells with Tbx20-specific siRNA followed by ER stress induction resulted in significant decrease in the expression of Tbx20 (0.18 ± 0.02-fold) as compared with Tun treatment alone (1.84 ± 0.15-fold) (Fig. 3, A and D). Knockdown of Tbx20 followed by ER stress induction resulted in concomitant decrease in the expression of Bmp2 (0.4 ± 0.03-fold) and its downstream signaling molecule pSmad$\frac{1}{5}$/8 (0.33 ± 0.03-fold) as compared with Tun treatment alone (Fig. 3, A and D). These data suggested that Tbx20 functions upstream of Bmp2 in imparting protection during ER stress. Figure 3T-box transcription factor 20 (Tbx20) acts upstream of Bmp2–pSmad$\frac{1}{5}$/8 signaling in protecting cultured H9c2 cardiomyocytes against endoplasmic reticulum (ER) stress. A, Western blot analysis showed a decrease in the expression of Tbx20 upon knockdown with Tbx20 siRNA followed by ER stress induction as compared with ER stress induction alone. Knockdown of Tbx20 followed by ER stress induction resulted in significant decrease in the expression of Bmp2, its downstream signal transducer pSmad$\frac{1}{5}$/8 and Atf6. B, knockdown of Tbx20 followed by ER stress induction resulted in increased cardiomyocyte apoptosis as marked by increased expression of Bax, Chop, p-JNK, and decreased expression of BclXL as compared with tunicamycin (Tun) treatment alone. C, knockdown of Tbx20 followed by ER stress induction also resulted in increased expression of fibrotic genes Collagen I (Col I), Collagen III (Col III), and Periostin. The expression of calcification marker (RUNX2) increased significantly from control group; however, its expression between knockdown group and ER stress induction alone group remained unchanged. D, quantitative representation by ImageJ software of the proteins using three biological replicates from A–C. E, immunofluorescence staining showed siRNA-mediated knockdown of Tbx20 followed by ER stress induction (20 μg/ml Tun) resulted in decreased expression of Bmp2 (a’’) as compared with 20 μg/ml Tun treatment alone (a’) and control cells (a). Knockdown of Tbx20 followed by Tun treatment is accompanied by decreased cardiomyocyte proliferation marked by decreased expression of Ki67 (b’’) and increased apoptosis marked by increased expression of Chop (c’’) as compared with Tun treatment alone (b’ and c’) and control cells (b and c), respectively. Knockdown of Tbx20 followed by Tun treatment resulted in decreased expression of Atf6 (d’’) as compared with Tun treatment alone (d’) and control cells (d). Insets in E show single-channel images of respective makers. Scale bar of main images and insets represents 50 μm. F, quantitative representation of E. G, measurement of reactive oxygen species (ROS) levels showed a significant increase upon knockdown of Tbx20 followed by ER stress induction. Statistical significance was calculated by one-way ANOVA. Error bars represent SD from three independent biological replicates ($$n = 3$$). ns, p: nonsignificant, ∗$p \leq 0.05$, ∗∗$p \leq 0.005$, ∗∗∗$p \leq 0.0005$, ##$p \leq 0.0001.$ Atf6, activating transcription factor 6; Bmp2, bone morphogenetic protein 2; p-JNK, phosphorylated form of c-Jun N-terminal kinase; Tbx20, T-box transcription factor 20.
Next, we looked in the mode of action of Tbx20 in imparting protection against ER stress–induced cardiomyopathy. Knockdown of Tbx20 followed by ER stress induction resulted in significant increase in the expression of proapoptotic marker Bax (2 ± 0.24-fold) as compared with Tun treatment alone (1.18 ± 0.09-fold; Fig. 3, B and D). However, the expression of antiapoptotic marker BclXL decreased significantly (0.4 ± 0.09-fold) upon knockdown of Tbx20 (Fig. 3, B and D). Knockdown of Tbx20 also resulted in significant increase in the expression of apoptosis inducer Chop (2.67 ± 0.19-fold) as compared with Tun treatment alone (1.4 ± 0.08-fold; Fig. 3, B and D). Phosphorylated form of c-Jun N-terminal kinase (p-JNK) was shown to induce the expression of proapoptotic genes during ER stress. JNK was previously shown to abrogate the antiapoptotic effect of Bcl2 during ER stress [27]. Tbx20 knockdown also resulted in significant increase in the expression of p-JNK (9.4 ± 0.56-fold) as compared with Tun treatment alone (2.4 ± 0.22-fold). Thus, our study showed that Tbx20 imparts its protection against ER stress by decreasing cardiomyocyte apoptosis.
Since previous studies have shown the involvement of ER stress during fibrosis and calcification [28, 29], therefore, we looked into the expression of fibrotic markers Collagen I, Collagen III, Periostin, and calcification marker RUNX2. The expression of all the three fibrotic markers increased significantly upon knockdown of Tbx20 followed by Tun treatment as compared with Tun treatment alone (Fig. 3, C and D). The expression of calcification marker RUNX2 however remained unchanged upon knockdown of Tbx20 (Fig. 3, C and D). Immunostaining study showed that knockdown of Tbx20 followed by 20 μg/ml Tun treatment resulted in decreased expression of Bmp2 (4.8 ± $1.0\%$; Fig. 3, Ea’’ and F) as compared with 20 μg/ml Tun treatment alone (61.27 ± $3.0\%$; Fig. 3, Ea’ and F). Next, we checked the proliferation profile of the cardiomyocytes following knockdown of Tbx20 and subsequent ER stress induction. siRNA-mediated knockdown of Tbx20 followed by 20 μg/ml Tun treatment resulted in decreased immunoreactivity of Ki67 (8.03 ± $1.0\%$; Fig. 3, Eb’’ and F) as compared with 20 μg/ml Tun treatment alone (54.33 ± $4.0\%$; Fig. 3, Eb’ and F).
Knockdown of Tbx20 followed by 20 μg/ml Tun administration caused increased cardiomyocyte apoptosis. The expression of Chop increased up to 72 ± $4.0\%$ (Fig. 3, Ec’’ and F) upon knockdown of Tbx20 followed by Tun (20 μg/ml) treatment as compared with 20 μg/ml Tun treatment alone (8.14 ± $3.1\%$; Fig. 3, Ec’ and F). Thus, decrease in the expression of Tbx20 followed by ER stress induction is accompanied by increased cardiomyocyte apoptosis. Similarly, knockdown of Tbx20 followed by 20 μg/ml Tun treatment also resulted in decreased expression of Atf6 (Fig. 3, Ed’’ and F) as compared with 20 μg/ml Tun (Fig. 3, Ed’ and F) treatment alone. Knockdown of Tbx20 followed by ER stress induction resulted in significant decrease in the expression of both total as well as cleaved form of Atf6 as evidenced by Western blot and immunostaining analysis. This led us to speculate a possible role of Tbx20 in maintaining the pool of Atf6 in cells during ER stress. ChIP assay followed by PCR analysis revealed that Tbx20 binds to atf6 promoter and induces its activity (Fig. S2, E and F). In 20 μg/ml Tun-treated cardiomyocytes, Tbx20 binds to the promoter of atf6 with 7.3 ± 1.2-fold enrichment over IgG controls (Fig. S2G). Hence, these data suggest that Tbx20 directly binds to and induces the expression of atf6 during ER stress–mediated cardiomyopathy, thus maintaining the pool of atf6 during stressed conditions.
Knockdown of Tbx20 followed by ER stress induction also resulted in significant increase in the level of ROS as compared with Tun treatment alone (Fig. 3G). Thus, this result strengthened the previous observation for the involvement of Tbx20 in restricting ROS generation during ER stress.
Together, these data suggest that Tbx20 is located upstream of Bmp2–pSmad$\frac{1}{5}$/8 signaling axis, and it is necessary in imparting protection against ER stress by increasing cardiomyocyte proliferation and decreasing cardiomyocyte apoptosis and fibrosis, which together results in restoration of cardiomyocyte homeostasis.
## Tbx20–Bmp2 signaling acts in a feed-forward loop mechanism in protecting cells against Tun-induced ER stress
To decipher the regulatory relationship between Tbx20 and Bmp2, H9c2 cardiomyocytes were treated with 50 μg/ml Tun followed by treatment with recombinant Bmp2 (RecBmp2) protein. Treatment of H9c2 cells with 200 ng/ml RecBmp2 protein resulted in $74.2\%$ increase in the expression of endogenous level of Bmp2 (Fig. S2, J and K). Western blot analysis showed a significant increase in the expression of Bmp2 (12.49 ± 1.4-fold) upon treatment of RecBmp2 protein following ER stress induction as compared with ER stress induction alone (2.1 ± 0.21-fold; Fig. 4, A and B). RecBmp2 treatment following ER stress induction also resulted in increased expression of Tbx20 (2.3 ± 0.23-fold) as compared with 50 μg/ml Tun treatment alone (1.2 ± 0.12-fold; Fig. 4, A and B). The expression of apoptotic marker Chop also decreased significantly upon administration of RecBmp2 protein as compared with ER stress induction alone (Fig. 4, A and B).Figure 4Tbx20–Bmp2 signaling acts in a feed-forward loop mechanism to protect cultured H9c2 cardiomyocytes against endoplasmic reticulum (ER) stress. A, Western blot analysis of H9c2 cells treated with 50 μg/ml tunicamycin (Tun) followed by administration of recombinant Bmp2 (RecBmp2) showed significant increase in the expression of Bmp2 even during increased ER stress. Administration of RecBmp2 protein following ER stress induction resulted in significant increase in the expression of Tbx20 as compared with 50 μg/ml Tun-treated group. Treatment of the ER-stressed cells with RecBmp2 also resulted in decrease in the expression of apoptotic marker Chop. B, quantitative representation by ImageJ software of the proteins using three biological replicates from A. C, immunofluorescence staining showed that RecBmp2 treatment following ER stress induction (50 μg/ml Tun) resulted in increase in the expression of Bmp2 (a’’) and Tbx20 (b’’) in comparison to 50 μg/ml Tun treatment alone (a’ and b’) and control cells (a and b), respectively. Increase in Bmp2 expression is accompanied by increased expression of Ki67 (c’’) and decreased expression of Chop (d’’) as compared with 50 μg/ml Tun treatment alone (c’ and d’) and control cells (c and d), respectively. D, quantitative representation of panels in C. E, Western blot analysis of H9c2 cells treated with Bmp2 inhibitor Noggin followed by ER stress induction (10 μg/ml Tun) caused significant decrease in the expression of Bmp2 as compared with ER stress induction group alone. Treatment with Noggin followed by ER stress induction however caused no significant change in the expression of Tbx20 from ER stress induction-alone group. F, quantitative representation by ImageJ software of the proteins using three biological replicates from E. G, immunostaining of H9c2 cells treated with Bmp2 inhibitor Noggin followed by Tun treatment (10 μg/ml Tun) resulted in decrease in the expression of Bmp2 (a’’) with reference to 10 μg/ml Tun treatment alone (a’) and control cells (a). Noggin treatment followed by ER stress induction caused no significant change in the expression of Tbx20 (b’’) as compared with ER stress induction alone (b’). Noggin administration followed by Tun treatment resulted in significant decrease in the expression of proliferative marker Ki67 (c’’) as compared with Tun treatment alone group (c’). The expression of apoptosis inducer Chop however remained unchanged between Noggin administered group (d’’) as compared with Tun treatment alone (d’). H, quantitative representation of panels in G. Scale bar represents 50 μm. Statistical significance was calculated by one-way ANOVA. Error bars represent SD from three independent biological replicates ($$n = 3$$); ns, p: nonsignificant, ∗$p \leq 0.05$, ∗∗$p \leq 0.005$, ∗∗∗$p \leq 0.0005$, ##$p \leq 0.0001.$ Bmp2, bone morphogenetic protein 2; Tbx20, T-box transcription factor 20.
Immunofluorescence study showed that the expression of Bmp2 was increased significantly (51.83 ± $3.5\%$; Fig. 4, Ca’’ and D) in RecBmp2 and Tun group as compared with Tun treatment alone (11.07 ± $1.8\%$; Fig. 4, Ca’ and D). The expression of Tbx20 was significantly augmented by approximately 29.80 ± $2.6\%$ (Fig. 4, Cb’’ and D) upon RecBmp2 and Tun treatment relative to Tun treatment alone (12.33 ± $1.6\%$; Fig. 4, Cb’ and D). Ki67 immunoreactivity, which is indicative of cardiac proliferation, also increased by approximately 31.57 ± $2.5\%$ (Fig. 4, Cc’’ and D) upon RecBmp2 and Tun treatment as compared with Tun treatment alone (2.8 ± $0.3\%$; Fig. 4, Cc’ and D). However, RecBmp2 and Tun treatment resulted in significant decrease in the expression of apoptosis inducer Chop by approximately 24.03 ± $2.1\%$ (Fig. 4, Cd’’ and D) as compared with only Tun treatment (61.37 ± $2.5\%$; Fig. 4, Cd’ and D).
To further corroborate our results, the H9c2 cells were pretreated with Noggin followed by induction of ER stress. Treatment of H9c2 cells with 200 ng/ml Noggin protein resulted in $67\%$ reduction in the expression of endogenous level of Bmp2 (Fig. S2, L and M). Pretreatment of 10 μg/ml Tun-treated H9c2 cardiomyocytes with Bmp2 inhibitor Noggin resulted in significant decrease in the expression of Bmp2 (1.36 ± 0.3-fold) as compared with ER stress induction alone (3.06 ± 0.25-fold; Fig. 4, E and F). However, Noggin treatment resulted in no significant change in the expression of Tbx20 (2.5 ± 0.13-fold) as compared with ER stress induction alone (2.62 ± 0.45-fold; Fig. 4, E and F).
Immunostaining analysis also corroborated with the Western blot results. Pretreatment of 10 μg/ml Tun-treated H9c2 cardiomyocytes with Bmp receptor inhibitor Noggin caused no significant change in the expression of Tbx20 (Fig. 4, Gb’’ and H) as compared with 10 μg/ml Tun treatment alone (Fig. 4, Gb’ and H). However, inhibition of Bmp2 resulted in significant attenuation in cardiomyocyte proliferation marked by reduced Ki67-positive nuclei by approximately 10.47 ± $1.8\%$ (Fig. 4, Gc’’ and H) relative to Tun treatment alone (29.97 ± $2.0\%$; Fig. 4, Gc’ and H). Inhibition of Bmp2 by Noggin followed by Tun treatment caused no significant change in the expression of Chop (15.7 ± $2.9\%$; Fig. 4, Gd’’ and H) as compared with Tun treatment alone (11.9 ± $1.1\%$; Fig. 4, Gd’ and H).
These results are consistent with our previous observation (Fig. 3) that Tbx20 is located upstream of Bmp2 in imparting protection against ER stress by increasing cardiomyocyte proliferation. These data also indicate the fact that exogenous administration of Bmp2 can impart protection during increased ER stress by increasing cardiomyocyte proliferation by a positive feed-forward mechanism.
## Prolonged Tun-induced ER stress in adult heart is accompanied by altered cardiac functions, increased cell size, and collagen deposition in vivo
In order to validate the direct protective effect imparted by Tbx20–Bmp2 signaling from ER stress–induced apoptosis in vivo, we administered Swiss Albino mice with 1 mg/kg body weight (BW) Tun intraperitoneal injections for 8 h and 2 days. The two time points were chosen to depict ER stress induction for short interval (8 h) and long interval (2 days) in order to corroborate the in vitro results.
Morphological abnormalities like hypertrophy (increased cardiomyocyte size) and fibrosis (increased collagen deposition) were detected upon prolonging ER stress in heart as compared with ER stress induction for short interval as well as control heart samples. Prolonged ER stress resulted in significant increase in heart weight (HW)/BW ratio (6.27 ± 0.36 mg/g) as compared with ER stress induction for short interval (4.92 ± 0.55 mg/g) and control group (4.42 ± 0.45 mg/g) (Fig. 5A).Figure 5Prolonged endoplasmic reticulum (ER) stress in adult murine heart results in altered cardiac function with increased cardiomyocyte size and collagen deposition in adult murine heart. A, the heart weight to body weight ratios indicative of cardiac hypertrophy is increased significantly during prolonged (2 days) ER stress as compared with ER stress induction for short duration (8 h) and control group. The change in heart weight to body weight ratio between 8 h ER stress and control group was however negligible ($$n = 9$$). Scale bar represents 20 μm. B, quantitative RT–PCR (qRT–PCR) analysis showed that prolonged (2 days) ER stress induction resulted in significant increase in the expression of cardiac function test markers bnp and β-mhc and significant decrease in the expression of serca2 as compared with 8 h ER stress induction group. C, prolonged (2 days) ER stress results in increased collagen deposition indicative of cardiac fibrosis as shown in Masson’s trichrome-stained adult heart sections as compared with 8 h ER stress group and control group ($$n = 6$$). D, in prolonged (2 days) ER stress–induced adult mice, cardiomyocyte cell size is increased marked by wheat-germ agglutinin (WGA) staining (green) in comparison to 8 h ER stress group and control group, respectively ($$n = 6$$). E, Western blot analysis of Tbx20 showed a significant increase in its expression in 8 h Tun-treated group as compared with control. The expression of Tbx20 later decreased significantly in the 2-day group. The expression of Bmp2 increased significantly during 2 days as compared with 8 h Tun-treated group. The expression of apoptotic marker Chop increased significantly in the 2-day group as compared with 8 h and control group. F, quantitative representation by ImageJ software of the proteins using three biological replicates from E. Scale bar represents 50 μm. Error bars represent SD from at least three independent biological replicates. Statistical significance was calculated by one-way ANOVA. ns, p: nonsignificant, ∗$p \leq 0.05$, ∗∗$p \leq 0.005$, ∗∗∗$p \leq 0.0005$, ##$p \leq 0.0001$; n ≥ 3 independent experiments. Tbx20, T-box transcription factor 20.
Cardiac function in mice following ER stress induction was checked by assessing the change in expression of brain natriuretic peptide (bnp), β myosin heavy chain (β-mhc), and sarcoendoplasmic reticulum calcium ATPase 2 (serca2). Previous studies reported bnp and β-mhc as a diagnostic biomarker for cardiac dysfunction [30, 31]. Cardiac serca2 was also previously reported as therapeutic targets for heart failure [32]. Our study showed an increase in the expression of bnp during 2 day Tun treatment (4.46 ± 0.47-fold) as compared with 8 h Tun treatment (1.08 ± 0.15-fold) and control mice (Fig. 5B). The expression of β-mhc was also increased during 2 day treatment as compared with 8 h and control mice. The expression of serca2, another therapeutic target for cardiomyopathy, was decreased significantly (0.7 ± 0.17-fold) during 2 day of ER stress induction in mice as compared with 8 h (1.15 ± 0.06-fold) and control mice (Fig. 5B).
Masson’s trichrome-stained adult heart sections revealed increased fibrotic regions marked by increased collagen deposition in prolonged ER stress (2 days)–induced heart tissue as compared with ER stress induced for short interval (8 h) heart tissue and control group (Fig. 5C). However, the presence of fibrotic regions in the 8 h ER stress–induced heart tissue was nonsignificant in comparison to control mice. Cardiomyocyte cell size was increased significantly in 2 day Tun-treated hearts (363.3 ± 35.74 μm2) as compared with 8 h ER stress–induced hearts (211.3 ± 36.89 μm2) and control group (165 ± 14.93 μm2) as indicated by wheat germ agglutinin staining (Fig. 5D). The increase in cell size was however nonsignificant between 8 h ER stress–induced heart and control group. These data show that prolonged ER stress (2 days) is accompanied by increased cardiomyocyte hypertrophy and fibrosis as compared with ER stress interval for shorter interval of time (8 h).
Western blot analysis revealed an increase in the expression of Tbx20 (2.86 ± 0.3-fold) during 8 h ER stress induction as compared with control group. However, a prolonged ER stress (2 days) resulted in significant decrease (1.2 ± 0.08-fold) in the expression of Tbx20 (Fig. 5, E and F). The expression of Bmp2 also increased during 8 h (2.46 ± 0.5-fold) ER stress induction as compared with control group. However, its expression increased significantly during 2 day ER stress induction group (5.3 ± 0.7-fold; Fig. 5, E and F). Decrease in the expression of Tbx20 during prolonged ER stress was accompanied by increased expression of apoptosis inducer Chop (2.9 ± 0.19-fold; Fig. 5, E and F) as compared with 8 h (1.32 ± 0.15-fold) and control group.
Together, these data suggest that ER stress induction for short interval (8 h), where the expression of Tbx20 is increased, resulted in no significant change in the expression of markers of cardiac function, hypertrophy, and fibrosis. However, prolonging the ER stress (2 days) resulted in significant decrease in the expression of Tbx20 with concomitant alteration in cardiac function of mice.
Next, we examined whether Tbx20 imparts its protective function during ER stress–mediated cardiomyopathy in rats also. Prolonged ER stress induction for 2 days in rats resulted in significant increase in HW to BW ratio (6.1 ± 0.1-fold) as compared with ER stress induction for 8 h (5.0 ± 0.05-fold) and control group (4.9 ± 0.16-fold; Fig. S3A). Western blot of Tbx20 showed an increase in its expression in 8 h Tun treatment group (2.7 ± 0.1-fold) as compared with control group. However, upon prolonging the ER stress to 2 days, the expression of Tbx20 was decreased significantly (0.6 ± 0.2-fold; Fig. S3, B and C). The expression of Bmp2 was also increased during 8 h (3.5 ± 0.34-fold) as compared with control. The expression of Bmp2 however increased drastically in 2 day Tun-treated group (8.6 ± 0.7-fold; Fig. S3, B and C). Decrease in the expression of Tbx20 in the 2-day group resulted in concomitant increase in the expression of apoptotic marker Chop (4.7 ± 0.5-fold) as compared with 8 h group (1.5 ± 0.09-fold; Fig. S3, B and C).
Cardiac function study of the rats revealed a significant increase in the expression of bnp and β-mhc in the 2-day group as compared with 8 h and control group. The change in the expression of bnp and β-mhc between 8 h and control group was nonsignificant (Fig. S3D). The expression of serca2 decreased in the 2 day group as compared with the 8 h and control group (Fig. S3D). ECG measurement of all the animals of each group was recorded. ECG recordings of the 2 day ER stress group resulted in significant increase in QT interval (0.093 ± 0.01 s) as compared with 8 h (0.077 ± 0.01 s) and control group (0.072 ± 0.01 s; Fig. S3, E and F). The RR interval of the 2 day group showed a significant decrease (0.19 ± 0.01 s) as compared with 8 h (0.24 ± 0.01 s) and control (0.25 ± 0.01 s) group (Fig. S3, E and G). The 2 day group also showed an elevation of the ST segment as compared with 8 h and control group (Fig. S3E).
Therefore, all these observations highlight the importance of Tbx20 in maintaining proper cardiac function. When the expression of Tbx20 is decreased during prolonged ER stress (2 days), the cardiac function is impaired resulting in progression of cardiomyopathy because of ER stress.
## ER stress–induced upregulation of Tbx20 activity is beneficial for cardiomyocyte viability and maintenance of cardiomyocyte homeostasis by regulating proliferation and apoptosis in adult murine heart
Next, the status of Tbx20 activity and its function in Tun-treated adult murine hearts in vivo was examined. Similar results as that of protein levels were observed at transcript level. The establishment of ER stress was accessed by checking the expression of ER stress markers grp78 and atf6 by quantitative RT–PCR (qRT–PCR). ER stress induction for short period (8 h) resulted in a significant increase in expression of grp78 (2.8 ± 1.2-fold) as compared with control. Its expression also increased during prolonged ER stress (4.2 ± 2.6-fold) as compared with control; however, the increase was nonsignificant in comparison to 8 h ER stress group (Fig. 6A). The expression of another ER stress marker atf6 was also increased during prolonged ER stress (6.0 ± 2.9-fold) as compared with 8 h ER stress treatment group (5.2 ± 1.9-fold); however, that change was nonsignificant. The expression of apoptosis inducer chop did not change significantly in the 8 h ER stress group (1.29 ± 0.6-fold) with respect to control group. However, its expression increased significantly in the 2 day ER stress group (7.7 ± 2.3-fold) (Fig. 6A). The expression of tbx20 increased significantly during 8 h ER stress (2.3 ± 1.1-fold) with reference to control group. On the contrary, a prolonged ER stress resulted in significant decrease in the expression of tbx20 (1.2 ± 0.4-fold) as compared with 8 h ER stress–induced (Fig. 6A) group. These results are consistent with the in vitro results. ER stress induction for a shorter interval leads to the upregulation of tbx20, which in turn accelerates the expression of protective ER gene atf6. Prolonging the ER stress eventually leads to decrease in the expression of tbx20. A nonsignificant increase in the expression of atf6 in the prolonged ER stress group might be attributed to the fact that the heart is composed of heterogeneous population of cells as compared with pure cardiomyocyte population of H9c2 cells; hence, the expression of atf6 might be regulated by factors other than tbx20 in murine heart. Another interesting observation is the drastic increase (11 ± 2.9-fold) in the expression of bmp2 in 2-day ER stress group with reference to 8 h ER stress group (3.96 ± 2.0-fold) as opposed to the in vitro data (Fig. 6A). The drastic increase of bmp2 in the prolonged ER stress group may also be due to the heterogeneity of the adult mice heart. Figure 6Endoplasmic reticulum (ER) stress–mediated upregulation of Tbx20–Bmp2 signaling results in increased proliferation and limits apoptosis in adult murine hearts. A, ER stress induction for short duration (8 h) resulted in increase in the expression of ER stress markers grp78 and atf6 compared with control group as determined by quantitative RT–PCR (qRT–PCR). However, the change in expression of atf6 and grp78 between 8 h and prolonged ER stress (2 days) group was found to be negligible. The expression of apoptosis inducer chop increased significantly in prolonged ER stress (2 days) group as compared with 8 h and control groups. The change in expression of chop between 8 h and control group was nonsignificant. The expression of tbx20 and bmp2 increased during ER stress induction for short duration (8 h) as compared with control; however, the expression of bmp2 increased significantly in the 2 day ER stress–induced group. B, ER stress induction for short duration (8 h) resulted in cardiomyocyte-specific increase in the expression of Tbx20 (a’) marked by Tbx20-positive nuclei (green) colabeled with cardiomyocyte-specific Mf20 (red) compared with control (a). However, a prolonged ER stress resulted in decrease in the expression of Tbx20 (a’’). The expression of proliferation marker Ki67 increased significantly as marked by increased Ki67-positive nuclei (b’; green) colabeled with cardiomyocyte-specific Mf20 (red) during 8 h of ER stress induction compared with control (b). Prolonged ER stress (2 days) resulted in decreased expression of Ki67 (b’’) compared with 8 h ER stress induction group. Cardiomyocyte-specific expression of Bmp2 was increased in 8 h ER stress (c’) as compared with control (c). However, its expression later decreased during 2 day ER stress (c’’) induction group. Bmp2 was also shown to colocalize with α-SMA with increased expression during prolonged ER stress (d’’) condition as compared with 8 h ER stress (d’) and control (d) groups. Increase in Bmp2 expression was accompanied by increased expression of Chop during prolonged ER stress (e’’). C–G, quantitative representation of panels in B. Scale bar of main images represents 50 μm. Scale bar of inset represents 20 μm. Statistical significance was calculated by one-way ANOVA. Error bars represent SD from at least three independent biological replicates (n ≥ 3); ns, p: nonsignificant, ∗$p \leq 0.05$, ∗∗$p \leq 0.005$, ∗∗∗$p \leq 0.0005$, ##$p \leq 0.0001.$ Bmp2, bone morphogenetic protein 2; Tbx20, T-box transcription factor 20; α-SMA, alpha-smooth muscle actin.
The requirement for Tbx20 in adult cardiomyocyte homeostasis post ER stress induction in vivo was determined by immunostaining. Colocalization of Tbx20 and Mf20+ (cardiomyocyte-specific marker) revealed a significant increase in the expression of Tbx20 in the cardiomyocytes of 8 h group (32 ± $6.6\%$; Fig. 6, Ba’ and C) with respect to control (5.02 ± $2.2\%$; Fig. 6, Ba and C). However, its expression decreased in the 2 day group (13.64 ± $3.2\%$; Fig. 6, Ba’’ and C) in comparison to 8 h group. The results correlated with the mRNA data. Increase in the expression of Tbx20 was accompanied with increased cardiomyocyte proliferation. Colocalization of Ki67 and Mf20+ showed an increase in cardiomyocyte proliferation marked by increased Ki67-positive nuclei in the 8 h group (30.06 ± $5.2\%$; Fig. 6, Bb’ and D) compared with control group (5.36 ± $1.8\%$; Fig. 6, Bb and D). However, it decreased in the 2 day group (10.24 ± $2.4\%$; Fig. 6, Bb’’ and D), which is suggestive of decreased proliferation because of decrease of Tbx20. Bmp2 was shown to colocalize along with Mf20+, and its expression was increased significantly in 8 h group (Fig. 6, Bc’ and E) with respect to control group (Fig. 6, Bc and E). However, its expression later decreased in 2 day group (Fig. 6, Bc’’ and E). In order to validate the drastic increase of bmp2 mRNA, colocalization of Bmp2 with α-SMA (myofibroblast-specific marker) was performed. Bmp2 was shown to colocalize with alpha-smooth muscle actin (α-SMA) in the 2 day group, and its expression was increased significantly (37.08 ± $2.9\%$; Fig. 6, Bd’’ and F) as compared with the 8 h group (3.7 ± $0.9\%$; Fig. 6, Bd’ and F). This observation supports the notion that drastic increase in Bmp2 expression is attributed to cell types other than cardiomyocytes in adult mice heart in vivo. Drastic increase in Bmp2 expression was accompanied by increased expression of Chop in 2 day group (Fig. 6, Be’’ and G). The increase in the expression of Bmp2 and subsequent increase in the expression of Chop during prolonged ER stress may be due to increased inflammatory response and is discussed in detail in the Discussion section.
Taken together, these observations further strengthen our hypothesis that there is a fine balance between ER stress–induced survival and death. ER stress induction for short interval leads to upregulation of Tbx20, which eventually caused increased cardiomyocyte proliferation because of increased expression of Bmp2 and limits cardiomyocyte apoptosis. However, a prolonged ER stress abrogates the expression of Tbx20 resulting in decreased proliferation and increased cardiomyocyte apoptosis, which leads to disruption of the homeostasis eventually leading to cardiomyocyte death.
## Hyperglycemia-induced ER stress upregulates activity of Tbx20 with concomitant increase in cardiomyocyte proliferation
Diabetic heart disease accounts for almost $80\%$ of deaths among the patients suffering from diabetes. The mechanisms that cause gradual cardiomyocyte apoptosis in chronic diabetes are multifactorial, but recent evidence suggest the involvement of the ER stress in the cardiac apoptosis in a streptozotocin-induced type 1 diabetic rat model and in hyperglycemia [33, 34]. ER stress has been implicated to induce fibrosis and cardiomyocyte death or apoptosis in diabetic cardiomyopathy. Therefore, to validate our results in a disease model, we have chosen diabetic cardiomyopathy as our model system.
The establishment of ER stress during diabetes was validated by the expression of ER stress markers. The mRNA level of atf6 was increased significantly (2.35 ± 0.8-fold) during diabetes as compared with control (Fig. 7A). The expression of tbx20 and bmp2 was also increased significantly during diabetes as compared with control (Fig. 7A). The protein expression of Atf6 was increased significantly during diabetes (25.12 ± $3.3\%$, Fig. 7, B and C) as compared with control (2.6 ± $0.8\%$). The expression of Tbx20 was also increased significantly during diabetes (23.28 ± $3.5\%$) as compared with control (3.68 ± $1.5\%$, Fig. 7, B and C). Bmp2 levels increased significantly in diabetes group (28.6 ± $2.6\%$) with reference to control group (4.16 ± $1.5\%$) (Fig. 7, B and C). Thus, diabetes augments ER stress with concomitant increase in the expression of Tbx20 and Bmp2. Diabetic cardiomyopathy was accompanied by altered cardiac function as marked by increased expression of bnp and β-mhc and decreased expression of serca2 as compared with control group (Fig. 7D). Western blot analysis showed a significant increase in the expression of Grp78 (3.35 ± 0.5-fold; Fig. 7, E and F) as compared with control group. The expression of Atf6 also increased significantly (4.9 ± 0.2-fold; Fig. 7, E and F). Diabetic cardiomyopathy also resulted in increased expression of Tbx20 (4.4 ± 0.5-fold) and Bmp2 (2.9 ± 0.5-fold) as compared with control group (Fig. 7, E and F).Figure 7Hyperglycemia-induced endoplasmic reticulum (ER) stress increases the activity of Tbx20–Bmp2 signaling axis with concomitant increase in proliferation and decrease in apoptosis. A, induction of diabetes in mice resulted in increased expression of atf6, tbx20, and bmp2 as determined by quantitative RT–PCR (qRT–PCR) analysis. B, immunohistochemical analysis revealed increased expression of Atf6, Tbx20, and Bmp2 upon diabetes induction in vivo. Scale bar of main images represents 50 μm. Scale bar of inset represents 20 μm. C, quantitative representation of panel in B. Statistical significance was calculated by Student’s t test (n ≥ 6). D, qRT–PCR analysis showed an increase in the expression of bnp and β-mhc and decrease in the expression of serca2 in the diabetes group as compared with control group. E, Western blot analysis showed an increase in the ER stress markers Grp78 and Atf6-p50 upon diabetes induction as compared with control. The expression of Tbx20 and Bmp2 is also increased during diabetes. F, quantitative representation by ImageJ software of the proteins using three biological replicates from E. G, qRT–PCR analysis showed an increase in the expression of atf6, grp78, tbx20, and bmp2 upon hyperglycemia induction in cultured H9c2 cells. H, Western blot analysis showed an increase in the expression of Tbx20 and Bmp2 during hyperglycemia induced for 2 days (25 mM 2d). Prolonging the hyperglycemic stress for 5 days (25 mM 5d) resulted in decrease in the expression of Tbx20 and Bmp2. I, quantitative representation by ImageJ software of the proteins using three biological replicates from H. J, immunofluorescence staining showed an increase in the expression of Tbx20 (a’) and Bmp2 (b’) upon hyperglycemia induction for 2 days in comparison to respective controls (a and b). However, prolonging the hyperglycemia (5 days) resulted in decrease in their expression (a’’ and b’’). Increase in Tbx20 resulted in concomitant increase in cardiomyocyte proliferation marked by increased Ki67 (c’) as compared with control cells (c). However, prolonging the stress resulted in decrease in its expression (c’’). Prolonged hyperglycemic stress resulted in increased cardiomyocyte apoptosis marked by increased Chop (d’’) expression compared with hyperglycemic stress for 2 days (d’) and control. Scale bar represents 50 μm. K, quantitative representation of panels in J. L, reactive oxygen species (ROS) levels were increased upon prolonging the hyperglycemic stress for 5 days (25 mM 5d) as compared with 2 days (25 mM 2d) and control. Statistical significance was calculated by one-way ANOVA for three independent biological experiments ($$n = 3$$). Error bars represent SD from three independent biological replicates. ns, p: nonsignificant, ∗$p \leq 0.05$, ∗∗$p \leq 0.005$, ∗∗∗$p \leq 0.0005$, ##$p \leq 0.0001.$ Atf6, activating transcription factor 6; Bmp2, bone morphogenetic protein 2; Tbx20, T-box transcription factor 20.
Similarly, the increased expression of ER stress genes atf6 and grp78 was observed in H9c2 cells with high concentration of glucose (25 mM) for 2 days. Increase in ER stress was accompanied by increase in the expression of tbx20 (2.0 ± 0.4-fold) and bmp2 (3.15 ± 0.7-fold) as compared with cells (Fig. 7G).
Western blot analysis revealed an increase in the expression of Tbx20 (4.3 ± 0.5-fold) and Bmp2 (4.4 ± 0.4-fold) when H9c2 cells were treated with higher glucose for 2 days. However, prolonging the ER stress for 5 days resulted in decreased expression of Tbx20 (1.4 ± 0.3-fold) and Bmp2 (2.7 ± 0.4-fold; Fig. 7, H and I).
The expression of Tbx20 increased up to 43.97 ± $4.5\%$ (Fig. 7, Ja’ and K) in the 2 day group as compared with control group (6.53 ± $1.2\%$; Fig. 7, Ja and K). However, prolonging the hyperglycemic stress for 5 days resulted in significant decrease in the immunoreactivity of Tbx20 (12.57 ± $2.4\%$; Fig. 7, Ja’’ and K) as compared with 2 days. The expression of Bmp2 also showed a similar increase in immunoreactivity in 2 day group (38.73 ± $6.6\%$; Fig. 7, Jb’ and K) as compared with control (10.5 ± $1.1\%$; Fig. 7, Jb and K) followed by decrease in 5 days (15.47 ± $1.4\%$; Fig. 7, Jb’’ and K). Increase in Tbx20 was accompanied by increased nuclear immunoreactivity of Ki67 in the 2 day group (59.37 ± $5.7\%$; Fig. 7, Jc’ and K) compared with control group (4.66 ± $0.6\%$; Fig. 7, Jc and K). However, with decreased expression of Tbx20, the expression of Ki67 also decreased in the 5 day group (6.92 ± $0.8\%$; Fig. 7, Jc’’ and K). Decrease in the activity of Tbx20 in 5 days was accompanied by increased apoptosis marked by augmented expression of Chop (57.97 ± $5.5\%$; Fig. 7, Jd’’ and K) as compared with the 2 day group (10.23 ± $1.6\%$; Fig. 7, Jd’ and K). Increase in the duration of hyperglycemic stress to 5 days in H9c2 cells resulted in increased ROS levels as compared with hyperglycemic stress induced for 2 days (Fig. 7L).
Thus, these observations further confirm that ER stress–mediated cardiomyopathy results in the upregulation of Tbx20 and Bmp2 with concomitant increase in cardiomyocyte proliferation. However, prolonging the stress eventually leads to decreased Tbx20 expression with concomitant decrease in proliferation and increase in apoptosis of cardiomyocytes.
## Discussion
The adult heart proliferates at a lower level strengthens the fact that cardiomyocyte can repair postinjury. Here, we show that ER stress–mediated activation of Tbx20 promotes cardiomyocyte proliferation and limits cardiomyocyte apoptosis by activating Bmp2–pSmad$\frac{1}{5}$/8 pathway and upregulating the expression of cardioprotective Atf6 arm of UPR. The balance between ER stress–mediated cardiomyocyte survival and ER stress–mediated cardiomyocyte apoptosis is a critical factor that directs the protective effect of Tbx20, and it must be taken into account while considering therapeutic approaches to ER stress–mediated cardiomyopathies. Tbx20 overexpressing cardiomyocyte or cardiomyocyte-specific induction of Bmp2 could provide protection even during prolonged ER stress and requires further studies. Thus, studies on induction of cardiogenic gene Tbx20 or upregulation of the Tbx20–Bmp2–pSmad$\frac{1}{5}$/8 pathway in adult cardiomyocytes can protect them from ER stress–mediated cardiomyopathy and promote their regeneration.
The adult mammalian heart possesses a little regenerative capacity, which is insufficient to compensate for the loss of cardiomyocyte because of pathophysiological conditions. Thus, triggering on the proliferative capacity of the pre-existing cardiomyocytes of adult heart represents promising strategy for restoration of cardiac homeostasis postinjury [35]. Tbx20 represses cell cycle inhibitory genes p21, meis1, and btg2, thereby promoting adult cardiomyocyte proliferation post myocardial infarction suggesting a critical mediator for cardiomyocyte proliferation postinjury [36]. Previous studies have shown that Bmp2 protects cardiomyocytes from ischemia/reperfusion injury via upregulating the Smad1 pathway, which in turn inhibits apoptosis [37]. The fact that Tbx20 and Bmp2 are essential factors for cardiomyocyte proliferation post ER stress–mediated injury was unknown so far. Our study has shown a regulatory mechanism whereby ER stress–mediated upregulation of Tbx20 leads to cardioprotection and restoration of cardiac homeostasis in H9c2 cardiomyocytes and adult mice heart by augmenting cardiomyocyte proliferation and limiting cardiomyocyte apoptosis via upregulating the Bmp2–pSmad$\frac{1}{5}$/8 pathway (Fig. 8). However, prolonging the extent of ER stress eventually results in reversal of this phenomenon with decreased Tbx20–Bmp2 expression, decreased cardiomyocyte proliferation, and increased cardiomyocyte apoptosis. Our study for the first time showed that Atf6 directly binds to the promoter of tbx20 gene during ER stress condition, thereby increasing its expression. Knockdown of Tbx20 followed by ER stress induction was shown to decrease Bmp2 signaling with concomitant decrease in cardiomyocyte proliferation and increase in cardiomyocyte apoptosis. Previous studies have shown that overexpression of Tbx20 results in reduction of fibrotic scars [36]. Knockout of Tbx20 on the other hand was shown to cause extensive fibrosis within a short period [10]. Bmp2 was also shown to decrease renal interstitial fibrosis and liver fibrosis [38, 39]. Our study showed the mode of action of Tbx20 during ER stress. First, ER stress–mediated increase in the expression of Tbx20 leads to the upregulation of Bmp2–pSmad$\frac{1}{5}$/8, which in turn increases cardiomyocyte proliferation. Second, Tbx20 was shown to decrease the expression of both apoptotic and fibrotic markers during ER stress, thereby restoring homeostasis. ER stress–mediated induction of Tbx20 was also shown to restrict the levels of ROS generation because of ER stress. The study also showed that Bmp2 acts downstream of Tbx20 in imparting protection against ER stress–induced cardiomyopathy. Figure 8Model for endoplasmic reticulum (ER) stress–mediated upregulation of T-box transcription factor 20 (Tbx20) resulting in increased cardiomyocyte survival. ER stress induction for short interval results in upregulation of activating transcription factor 6 (Atf6) pathway of unfolded protein response (UPR). Full-length Atf6 is cleaved (Atf6-p50), and it translocates to the nucleus where it increases the transcription of Tbx20. Increase in the level of Tbx20 results in upregulation of its downstream signaling cascade consisting of bone morphogenetic protein 2 (Bmp2)–pSmad$\frac{1}{5}$/8. Increased expression of Bmp2–pSmad$\frac{1}{5}$/8 further results in increased cardiomyocyte proliferation, thereby restoring homeostasis and increasing the survival of cardiomyocyte post ER stress induction. Tbx20 also helps in restoration of homeostasis by decreasing apoptosis, fibrosis, and reactive oxygen levels (ROS) levels during ER stress. Increased expression of Tbx20 during ER stress also helps to maintain the total pool of Atf6 during ER stress. A prolonged ER stress however results in decreased expression of cleaved Atf6 (Atf6-p50). This in turn results in decrease in the expression of Tbx20, and it downstreams Bmp2–pSmad$\frac{1}{5}$/8 signaling molecules resulting in decreased cardiomyocyte proliferation and increased cardiomyocyte apoptosis. Ectopic administration of recombinant Bmp2 protein results in upregulation of Bmp2, which in turn upregulates Tbx20 in a feed-forward mechanism, thereby restoring homeostasis by increasing cardiomyocyte proliferation.
Previous studies have shown the use of RecBmp2 protein for ectopic overexpression of Bmp2 [22, 38, 40]. Investigation of the relationship among this pathway by administration of RecBmp2 protein following ER stress induction (50 μg/ml Tun) revealed that Bmp2 can exert its synergistic effect on Tbx20 by increasing the expression of Tbx20 with concomitant increase in cardiomyocyte proliferation. However, Noggin treatment followed by ER stress induction (10 μg/ml Tun) caused no change in the expression of Tbx20 as compared with ER stress induction alone, thus proving the presence of a feed-forward loop mechanism in this pathway. We have also reported an increase in the activity of Tbx20 and Bmp2 during hyperglycemia in vitro and diabetic cardiomyopathy in vivo with concomitant increase in cardiomyocyte proliferation. Together, these data support a regulatory mechanism whereby Tbx20–Bmp2 signaling imparts its protective role in order to restore cardiomyocyte homeostasis during ER stress and during hyperglycemic condition.
A disruption in the protein assembly machinery results in the generation of ER stress with subsequent upregulation of the UPR. There is increasing evidence suggesting that the balance of protective UPR and ER stress–mediated apoptosis regulates the progression of cardiovascular diseases. Atf6 has mostly been considered as a protective molecule during ER stress. Hence, in our study, we have mainly focused on the Atf6 arm of UPR. Atf6 overexpression protected heart from ischemia/reperfusion injury by improving left ventricle–developed pressure, reducing expression of apoptotic markers, and ameliorating infarct size [41]. Constitutive expression of Atf6 was shown to improve cardiac function during ER stress–mediated cardiomyocyte apoptosis postinjury and during diabetes mellitus [42]. Atf6 exerts its cardioprotective effect by augmenting the expression of SERCA2a, antioxidants catalase (Cat), peroxiredoxin 5 (Prdx5), and VCP-interacting membrane protein (Vimp) [43, 44]. Here, we provide evidence that Atf6, a cardioprotective molecule of the UPR, is responsible for increase in the expression of Tbx20 during ER stress. Our study showed the direct DNA-binding ability of Atf6 to the promoter region of tbx20, thereby inducing its expression during ER stress. Our results were further corroborated by the use of Atf6 inhibitor AEBSF. We have shown by inhibiting the cleavage and nuclear translocation of active form of Atf6 (Atf6-p50) with AEBSF that it is indeed responsible for the increase in the expression of Tbx20 during ER stress.
Our study showed that the protective effect of Tbx20 during ER stress is mediated by Atf6 arm of the UPR signaling. However, mild ER stress also results in upregulation of the other two arms of UPR signaling pathway IRE1α and PERK. PERK pathway was shown to activate cytoprotective gene expression pathway, thus promoting stress-resistant state [45]. PERK knockout mice in response to transverse aortic constriction was shown to display altered cardiac function and enhanced cardiac apoptosis [46]. XBP1, which is a downstream molecule of IRE1α arm, was to increase vascular endothelial growth factor A–mediated angiogenesis in response to ER stress [47]. Despite having cardioprotective effects, these two pathways can induce the transcription of apoptotic molecules that results in detrimental consequences resulting in cardiomyopathy. PERK pathway results in the upregulation of Chop and p53-upregulated modulator of apoptosis (Puma), which induces cardiomyocyte apoptosis resulting in cardiomyopathy [48]. On the other hand, IRE1α results in the upregulation of apoptosis signal-regulating kinase 1 (ASK1), which is a critical molecule in eliciting cardiomyocyte death induced by ER stress [49]. Thus, these two pathways need to be studied further to decipher whether they could influence cell survival during ER stress.
Our study reported that knockdown of Tbx20 followed by Tun treatment resulted in decreased expression of both total and cleaved form of Atf6. ChIP analysis further revealed direct DNA binding of Tbx20 to the promoter of atf6 gene during ER stress, thereby showing that Tbx20 helps in maintaining the pool of total Atf6 during ER stress. Therefore, our study has showed the existence of a feedback mechanism between Tbx20 and Atf6 during ER stress. Tbx20 helps to maintain the pool of Atf6 during ER stress. On the other hand, during ER stress, Atf6 is cleaved, and it translocates to the nucleus where it drives the transcription of tbx20 gene.
Cardiomyocyte-specific deletion of Tbx20 results in embryonic lethality as it is required for the regulation of genes involved in fetal cardiomyocyte proliferation [26]. Loss of function of Tbx20 causes double outlet right ventricle and familial tetralogy of fallot [50, 51]. On the contrary, overexpression of Tbx20 promotes adult cardiomyocyte proliferation post myocardial infarction [36], thus highlighting its importance for proper heart development and function. Our study reported an increase in the expression of Tbx20 during 8 h of ER stress induction. However, prolonging the ER stress to 2 days resulted in significant downregulation in its expression in both mice and rat. ECG analysis revealed an increase in the QT interval in 2 day group as compared with control and 8 h ER stress induction group. On the other hand, RR interval was significantly decreased in the 2 day treatment group. ST segment elevation was also observed in the 2 day group. QT prolongation was shown to cause malignant arrythmia and sudden cardiac death [52]. On the other hand, shortening of the RR interval is indicative of increased heart rate [53]. ST segment elevation was shown to be associated with left ventricular hypertrophy and acute myocardial infarction [54]. All these parameters are established markers of cardiac function analysis. Since our study reported an alteration in these three parameters in the 2 day treatment group where the expression of Tbx20 is decreased in comparison to 8 h and control group, hence we can conclude that Tbx20 is indeed required for maintaining proper cardiac function during ER stress induction for short duration. Prolonging the ER stress eventually results in decreased expression of Tbx20 with concomitant alteration of cardiac function.
Our study reported a significant increase in the activity of Bmp2 during prolonged ER stress (2 days) in adult murine hearts. The increase in the expression of Bmp2 may be attributed to the fact that the adult murine heart is composed of multiple cell types as opposed to pure cardiomyocyte culture of H9c2 cells. The increase in Bmp2 expression may be due to the effect of ER stress on other cell types of adult heart. Bmp2 is expressed in multiple cell types of adult murine heart [55, 56]. Colocalization studies have shown that Bmp2 colocalizes with α-SMA (myofibroblast marker) in the ER stress (2 days) group with increased expression as compared with ER stress (8 h), thus strengthening our plausible explanation. Sustained ER stress was shown to recruit tumor necrosis factor (TNF) receptor–associated factor 2 (TRAF2) and ASK-1, which causes subsequent activation of JNK and NF-κβ and production of proinflammatory cytokines like interleukin-1, interleukin-6, and TNF-α [57]. Proinflammatory stimulus in turn regulates the activity of Bmp2. TNF-α and ROS generation increases the activity of Bmp2 [58]. Bmp2 in turn induces proinflammatory endothelial phenotype. Thus, sudden increase in Bmp2 expression during prolonged ER stress may be attributed to other cell types of murine heart and because of increased inflammatory response. Since in our study the level of Bmp2 is already increased, hence we can conclude that in vivo, upon prolonging the ER stress, Tbx20-independent Bmp2 function is overriding the Tbx20-dependent protective function of Bmp2. Therefore, prolonging or maintaining the expression of Tbx20 for longer duration during ER stress could result in restoration of normal cardiac functions even during prolonged ER stress–mediated cardiomyopathy.
The ER is an organelle responsible for folding of proinsulin, and ER stress is implicated with the pathogenesis of diabetes mellitus. Hyperglycemia, a causative factor of diabetes mellitus, disrupts ER homeostasis resulting in development of ER stress [34]. In order to replicate our observations in a disease model, we looked in the expression of Tbx20 during hyperglycemia in vitro and diabetic cardiomyopathy in vivo. Chronic hyperglycemia contributes to β-cell dysfunction by downregulating the expression of Atf6α/Ire1α, thus contributing to loss of homeostasis [59]. Our study showed an increase in the expression of ER stress markers Atf6 and Grp78 upon induction of diabetes. The expression of Tbx20 and Bmp2 also increased with concomitant increase in cardiomyocyte proliferation upon induction of hyperglycemia. Prolonged hyperglycemia caused augmentation of ER stress with concomitant decrease of Tbx20–Bmp2 signaling, decrease in cardiomyocyte proliferation, and increase in cardiomyocyte apoptosis. Thus, Tbx20–Bmp2 signaling imparts protection against hyperglycemia by augmenting cardiomyocyte proliferation.
It was shown previously that increased ER stress is associated with increased ROS generation because of augmented disruption of disulphide bonds [19, 20]. Our study reported a gradual increase in ROS generation with increase in the intensity of ER stress. However, when the expression of Tbx20–Bmp2 axis is decreased at a Tun concentration of 50 μg/ml, the ROS generation was increased drastically, and it almost correlated with positive control. Increased expression of Tbx20 was reported to result in decreased ROS generation in cardiomyocytes [21]. Our study showed that knockdown of Tbx20 followed by ER stress induction resulted in significant increase in ROS generation as compared with ER stress induction alone, thus highlighting the role of Tbx20 in limiting ROS generation. Prolonged hyperglycemic stress where the expression of Tbx20 was decreased resulted in significant increase in ROS generation as compared with hyperglycemic stress induced for short interval of time where the expression of Tbx20 was high. Therefore, our study showed that ROS generation increases gradually with increase in ER stress. However, when the expression of Tbx20 is decreased during ER stress or during hyperglycemia, the ROS levels were increased significantly and was almost equal to that of positive control, thus corroborating the role of Tbx20 in limiting ROS generation during ER stress.
In conclusion, it is inferred from our study that Tbx20–Bmp2 signaling acts during ER stress–mediated cardiomyopathy by increasing cardiomyocyte proliferation and limiting cardiomyocyte apoptosis. We predict that overexpression of Tbx20 or cardiomyocyte-specific expression of Bmp2 signaling could be exploited as a novel therapeutic approach to confer protection against prolonged ER stress and shift the balance toward prosurvival to restore cardiac homeostasis during ER stress–induced cardiomyopathy.
## Induction of ER stress in vivo
ER stress was induced in vivo by intraperitoneally injecting adult male Swiss Albino mice (8 weeks old) with Tun (catalog no.: T7765; Sigma–Aldrich) at a final concentration of 1 mg/kg BW diluted in sterile 150 mM dextrose [60], and the mice heart was harvested at 8 h and 2 days by euthanization using carbon dioxide, followed by cervical dislocation. The control mice were injected an equal volume of 150 mM dextrose containing $1\%$ dimethyl sulfoxide. In adult male Wister rats, ER stress was induced by intraperitoneal injection of 1 mg/kg BW Tun diluted in sterile 150 mM dextrose [61]. The rat hearts were harvested at 8 h and 2 days. All the animals were maintained as per Control and Supervision of Experiments on Laboratory Animals (CPCSEA) guidelines. All the animals were fed with normal chow diet and water ad libitum. All the experiments with animals were approved by the Institutional Animal Ethics Committee (IAEC), Jadavpur University (Ref no.: AEC/PHARM/$\frac{1701}{05}$/2017 dated November 12, 2020).
## Induction of diabetes in vivo
Male Swiss Albino mice (8 weeks old) was intraperitoneally injected with 150 mg/kg BW alloxan dissolved in $0.9\%$ saline to induce diabetes [62]. The control mice were treated with equal volume of $0.9\%$ saline. Mice with blood glucose levels >200 mg/dl were maintained for up to 2 weeks for experimental purpose [63, 64]. All the animals were maintained as per CPCSEA guidelines. All the animals were fed with normal chow diet and water ad libitum. All the experimental procedures were approved by the Institutional Ethical Committee, Presidency University (Registration PU/IAEC/SC/39), registered under “Committee for the purpose of CPCSEA, Ministry of Environment and forests, Govt. of India.
## Cell cultures and treatments
H9c2 cells were cultured in Dulbecco's modified Eagle's medium (Gibco) supplemented with $10\%$ fetal bovine serum (Himedia), 100 units/ml penicillin G sodium, and 100 μg/ml streptomycin sulfate (catalog no.: 15140122; Gibco) in the presence of $5\%$ CO2 at 37 °C. Upon reaching 60 to $70\%$ confluency, the cells were starved for 6 h and treated with different concentration of Tun (2, 5, 10, 20, and 50 μg/ml) [65, 66] in order to induce ER stress. The cells were harvested after 24 h and used for further analysis. H9c2 cells were treated with different concentrations of DTT (catalog no.: D9779; Sigma–Aldrich) (1, 3, 5, and 10 mM) and Tg (catalog no.: T9033; Sigma–Aldrich) (1.5, 2, 6, and 10 μM) for 24 h to induce ER stress [67, 68].
For induction of hyperglycemia in vitro, the H9c2 cells were starved overnight in serum-free glucose free media prior to treatment. The cells were supplemented with 25 mM glucose and 5 mM glucose serving as hyperglycemic and control conditions, respectively. The culture media were replenished with respective media every alternative day. The H9c2 cells were harvested at 2 and 5 days and used for further analysis.
## 3-[4,5-Dimethylthiazol-2-yl]-2,5 diphenyl tetrazolium bromide cell viability assay
3-[4,5-Dimethylthiazol-2-yl]-2,5 diphenyl tetrazolium bromide assay was performed to monitor cell viability upon treatment of Tun for different time points. Cultured H9c2 cells were seeded in 96-well plates and treated with different concentrations of Tun (2, 5, 10, 20, and 50 μg/ml) for 24 h to induce ER stress. About 5 mg/ml of 3-[4,5-dimethylthiazol-2-yl]-2,5 diphenyl tetrazolium bromide (catalog no.: TC191; Himedia) stock solution was diluted in a ratio of 1:10 in 1× PBS. About 40 μl of diluted stock solution was added to each well. The cells were then incubated for 3 h in $5\%$ CO2 at 37 °C. The solution was removed from each well, and 50 μl of extraction buffer ($80\%$ isopropanol, $20\%$ Triton X-100, and 12 (N) HCl) was added to each well. The absorbance was measured at 570 nm.
## RNA interference and cell treatments
For siRNA transfection, H9c2 cells grown at 60 to $70\%$ confluency were transfected with Tbx20 siRNA (assay ID: s164031; Ambion) at a final concentration of 100 nM (titrated for maximum downregulation) targeting the coding region of Tbx20 using Lipofectamine RNAiMAX (catalog no.: 13778-075; Invitrogen) reagent as per the manufacturer’s instructions after serum starvation overnight. The cells were maintained in transfection mix for 6 h in 37 °C in a $5\%$ CO2 incubator and then cells were maintained in complete growth media (Dulbecco's modified Eagle's medium + $10\%$ fetal bovine serum) for next 24 h. Control cells were transfected with scramble siRNA. About 24 h after siRNA transfection, the growth media were replenished with media containing 20 μg/ml Tun in order to induce ER stress and kept for another 24 h. The cells were harvested, and the cell lysate was used for Western blot analysis. Cultured cells treated with the aforementioned reagents were washed with 1× PBS and fixed with $4\%$ paraformaldehyde for immunostaining purpose.
H9c2 cells were treated with 300 μM of water-soluble serine protease inhibitor (AEBSF) (catalog no.: A8456; Sigma–Aldrich) to inhibit the cleavage of membrane-bound Atf6 for 6 h followed by treatment of 20 μg/ml Tun for 24 h [69]. The cells were harvested after 24 h, and the cell lysate was used for Western blot analysis.
Bmp2 recombinant protein (200 ng/ml, catalog no.: 355-BM; R&D Systems) and BMP inhibitor Noggin (200 ng/ml; catalog no.: 719-NG; R&D Systems) were used to overexpress and inhibit Bmp2, respectively. For overexpression of Bmp2, cultured H9c2 cardiomyocytes were first treated with 50 μg/ml Tun followed by treatment with Bmp2 recombinant protein for 24 h in parallel with vehicle control ($0.1\%$ bovine serum albumin [BSA] in 1× PBS). For inhibition of Bmp2, H9c2 cells were treated with Noggin for 24 h. This was followed by treatment with 10 μg/ml Tun to induce ER stress in parallel with vehicle control (1× PBS). The cells were harvested, and the cell lysate was used for Western blot analysis. Cultured cells treated with the aforementioned reagents were washed with 1× PBS and fixed with $4\%$ paraformaldehyde for immunostaining purpose.
## Western blot analysis
Western blot was performed as described previously [22]. Briefly, total protein was extracted from H9c2 cells from our different treatment conditions using protein lysis buffer (20 mM Tris–HCl, pH 7.5, 150 mM NaCl, 1 mM EDTA, 1 mM EGTA, $1\%$ glycerol, $1\%$ Nonidet P-40, 1 mM DTT, 100 mM NaF, 0.2 mM PMSF, and 1 mM Na3VO4) supplemented with protease inhibitor cocktail (catalog no.: GX-2811AR; Puregene) and phosphatase inhibitor cocktail (catalog no.: GX-1211AR; Puregene). A Bradford protein assay reagent (catalog no.: ML106; Himedia) was used to estimate the concentration of protein samples under different treatment conditions. About 60 to 100 μg of the protein extracts were fractionated using 7 to $12\%$ SDS-PAGE under reducing conditions. The gels were then transferred onto polyvinylidene difluoride membrane (catalog no.: 1620177; Bio-Rad), and the membranes was blocked with $5\%$ skimmed milk in Tris-buffered saline with Tween-20 for 1 h at room temperature. The membranes were then incubated with primary antibodies diluted in milk or BSA as per the manufacturer’s protocol at 4 °C overnight with constant shaking. The immunoblots were then incubated with horseradish peroxidase (HRP)–tagged secondary antibody and developed using Clarity Western ECL substrate (Luminol/enhancer solution and peroxide solution; catalog no.: 1610182; Bio-Rad).
## Immunostaining
The heart tissues sections were processed as described previously [22]. Briefly following deparaffinization and rehydration in graded ethanol ($100\%$, $95\%$, $75\%$, and $50\%$) and finally distilled water (two times), tissue sections were subjected to antigen retrieval in a microwave oven with citrate buffer (10 mM citric acid, $0.05\%$ Tween-20, pH 6.0). Following antigen retrieval, tissue sections were incubated with blocking buffer ($2\%$ BSA, $0.1\%$ Tween-20 in 1× PBS) for 1 h at room temperature. The sections were then incubated overnight at 4 °C with the respective primary antibodies as per experimental studies. Following incubation with the primary antibody, the sections were washed for three times with 1× PBS for 5 min each. The sections were then incubated with respective secondary antibody for 1 h at room temperature. Following antibody incubation, the sections were washed for three times with 1× PBS for 5 min each. The nuclei were counterstained with 4′,6-diamidino-2-phenylindole (catalog no.: D9542; Sigma–Aldrich) for 15 min at room temperature. The sections were washed for three times with 1× PBS for 5 min each and mounted in mounting media (20 mM Tris, pH 8.0, $0.5\%$ N-propyl gallate, $90\%$ glycerol). Images were taken by Leica DM2000 across different fields. For analysis of the tissue sections, at least three sections per mouse heart were used consisting a total of approximately 1000 to 1200 cardiomyocytes from experimental and littermate controls.
For in vitro studies, H9c2 cardiomyocytes cultured on coverslips were washed with PBS, followed by blocking ($2\%$ BSA and $0.1\%$ Tween-20 in 1× PBS) and subsequent antibody incubation. Images were taken by Leica DM2000 across different fields. In at least three independent experiments, a total of at least 200 cells were counted for each treatment (approximately 20 cells were counted per field, and a total of 10 number of fields per coverslip were examined).
## Antibodies
Ki67 (1 μg/ml, catalog no.: ab15580; Abcam), Tbx20 (5 μg/ml, catalog no.: PA5-40669; Thermo Fisher Scientific), Bmp2 (1:200 dilution, catalog no.: PA5-85956; Thermo Fisher Scientific), Atf6 (1:250 dilution, catalog no.: sc-166659; Santa Cruz Biotechnology), Chop (1:500 dilution, catalog no.: 2895S; Cell Signaling Technology), pSmad$\frac{1}{5}$/9 (1:500 dilution, catalog no.: 13820S; Cell Signaling Technology), Smad1 (1:1000 dilution, catalog no.: 9743S; Cell Signaling Technology), α-SMA (1 μg/ml, catalog no.: 14-9760-82; Thermo Fisher Scientific), Bax (1:200 dilution, catalog no.: 2772S; Cell Signaling Technology), BclXL (1:200 dilution, catalog no.: 2764S; Cell Signaling Technology), Mf20 (1:200 dilution; Developmental Studies Hybridoma Bank, University of Iowa), Grp78 (1:1000 dilution, catalog no.: 3177; Cell Signaling Technology), p-JNK (1:1000 dilution, catalog no.: 9251S; Cell Signaling Technology), JNK (1:1000 dilution, catalog no.: 9252S; Cell Signaling Technology), Col I (1:1000 dilution, catalog no.: PA5-95137; Thermo Fisher Scientific), Col III (1:1000 dilution, catalog no.: PA5-95595; Thermo Fisher Scientific), Periostin (1:1000 dilution, catalog no.: ab14041; Abcam), Runx2 (1:500 dilution, catalog no.: NBP2-67777; Novus Biologicals), Goat Anti-Rabbit IgG H&L (Alexa Fluor 488) (1:1000 dilution, catalog no.: ab150077; Abcam), Goat Antimouse IgG H&L (Texas Red) (1:1000 dilution, catalog no.: ab6787; Abcam), anti-rabbit IgG, HRP-linked Antibody (1:1000 dilution, catalog no.: 7074S; Cell Signaling Technology), antimouse IgG, HRP-linked Antibody (1:1000 dilution, catalog no.: 7076S; Cell Signaling Technology).
## ECG recording in anesthetic rats
Male Wister rats treated with Tun for different time intervals were anesthetized using ketamine (60 mg/kg BW) and xylazine (10 mg/kg BW) as previously mentioned [70]. ECG of the anesthetized rats was recorded for 10 min using standard lead II (metal ECG leads). The acquired ECG signals were analyzed by BIOPAC (Biosystems) MP36. The QT duration was measured from the onset of the QRS complex to the end of T wave. The time elapsed between two successive R waves of the QRS signal gave the measure of the RR interval.
## Determination of HW/BW ratio
HW normalized to BW was measured in milligram/gram units after euthanization using carbon dioxide, followed by cervical dislocation and harvested at 8 h and 2 days.
## Immunohistological analysis
Heart tissue from the three different groups was washed in PBS and fixed in $4\%$ paraformaldehyde and embedded as previously described [22, 71]. For histological staining, 5 μm tissue sections were deparaffinized, rehydrated, and used for subsequent analysis purpose.
## Cardiomyocyte size determination
The cross-sectional area of individual cardiomyocytes was determined by staining 5 μm tissue sections with FITC-conjugated wheat germ agglutinin (catalog no.: L4895; Sigma–Aldrich) for 1 h at room temperature. The nuclei were counterstained with nuclear stain 4′,6-diamidino-2-phenylindole (catalog no.: D9542; Sigma–Aldrich). The images were taken by Leica DM2000 across multiple fields. Cell size was quantified using ImageJ (National Institutes of Health) software.
## Fibrosis detection
Collagen deposition was determined by staining 5 μm tissue sections with Masson’s trichrome reagent. The collagen deposition was quantified using ImageJ (National Institutes of Health) software.
## Real-time qRT–PCR
Total RNA from cell and adult mouse and rat hearts was isolated using TRIzol Reagent (catalog no.: 15596026; Thermo Fisher Scientific). Following quantification using Qubit4 Fluorometer (Thermo Fisher Scientific), 1 μg RNA from each sample was used for complementary DNA (cDNA) preparation using iScript cDNA synthesis kit (catalog no.: 170889; Bio-Rad). About 1 μl of the synthesized cDNA was used for qRT–PCR using iTaq Universal SYBR Green Supermix (catalog no.: 1725121; Bio-Rad) in 7500 real-time PCR system (Applied Biosystems). The amplification was carried out using following thermal conditions initial holding at 95 °C for 10 min followed by 40 cycles of 95 °C for 15 s and 60 °C for 1 min and a dissociation stage of 95 °C for 15 s, 60 °C for 1 min, and then 95 °C for 30 s. Expression of β-actin mRNA was used as an endogenous control. The amount of RNA was quantified using the comparative CT method (ΔΔCt). The list of the primers used is mentioned in Table S1.
## ROS estimation
For quantification of the intracellular ROS generation because of various treatment conditions, the control and treated H9c2 cells were treated with 2′,7′-dichlorofluorescin diacetate (D6883; Sigma–Aldrich), which reacts with the intracellular ROS generated to give a green fluorescent compound dichlorofluorescein. Following different treatments, the H9c2 cells were washed with ice-cold Hanks balanced salt solutionand incubated with 100 μM dichlorofluorescein diacetate for 30 min at 37 °C. Following lysis of the cells with alkaline solution, the fluorescence intensity was measured at excitation of 485 nm and emission at 520 nm (Hitachi). For positive control, H9c2 cells were treated with $1\%$ H2O2 for 6 h.
## ChIP assay
DNA–protein complexes in cultured H9c2 cardiomyocytes treated with 20 μg/ml Tun were crosslinked for 10 min at room temperature by adding formaldehyde (Himedia) at a final concentration of $1\%$ to the culture media. 10× Glycine (Himedia) was added to each dish to quench unreacted formaldehyde. The fixed cells were lysed in SDS Lysis buffer and sonicated ten times for 30 s with a 1 min refractory period. The cell lysate was centrifuged 10,000g at 4 °C for 10 min to remove insoluble material. For immunoprecipitation, 10 μg of digested crosslinked chromatin was incubated with antibody against Atf6 (5 μg; catalog no.: sc-166659, Thermo Fisher Scientific) and incubated at 4 °C overnight. Immunoprecipitation with normal rabbit IgG was used as a negative control. Following incubation with respective antibodies, 60 μl of Protein AG Plus Agarose Beads (catalog no.: BB-PAG001PB, BioBharati) was added to each immunoprecipitate and incubated for 1 h at 4 °C. After centrifugation at 3000g for 1 min, the beads were washed with low salt immune complex wash buffer (one wash), high salt immune complex wash buffer (one wash), LiCl immune complex wash buffer (one wash), and TE buffer (one wash). The DNA–protein complexes were eluted in elution buffer. To free the DNA, the DNA–protein complexes were reversed crosslinked using 5 M NaCl and incubated at 65 °C for 5 h. The protein was removed by digestion with proteinase K at 65 °C for 2 h. The DNA was purified using phenol chloroform method. The immunoprecipitated and input DNA were subjected to real-time PCR using SYBR Green PCR reagent with the following primers: rtbx20 forward: 5′-GGAAGCAGTGACGTGAGAC′ and rtbx20 reverse: 5′-GCGACCTAAACTGTGCCT-3′ to amplify rat tbx20 promoter region. Fold enrichment relative to IgG (negative control) was calculated from three independent experiments ($$n = 3$$) using the comparative CT method (ΔΔCt) described previously [71].
In order to decipher the binding of Tbx20 to the promoter of atf6 gene, a similar procedure was used. For immunoprecipitation, 10 μg of digested crosslinked chromatin was incubated with antibody against Tbx20 (5 μg, catalog no.: PA5-40669; Thermo Fisher Scientific). The immunoprecipitated and input DNA were subjected to real-time PCR using SYBR Green PCR reagent with the following primers: ratf620 forward: 5′-TCCAGTCTAACGTGTGATGCA-3′ and ratf620 reverse: 5′-AAGAGTTAGGCTTCCCACCC-3′ to amplify rat atf6 promoter region. Fold enrichment relative to IgG (negative control) was calculated from three independent experiments ($$n = 3$$) using the comparative CT method (ΔΔCt) described previously [71].
## Statistical analysis
All the results were calculated as mean ± SD of at least three independent experiments. Statistical analysis between experimental groups was performed using Student's t test for two groups and one-way ANOVA for multiple groups using GraphPad Prism 9 Software (GraphPad Software, Inc). Results with $p \leq 0.05$ were considered significant.
## Data availability
All data are contained within the article.
## Supporting information
This article contains supporting information.
Supplemental Table S1, Figures S1–S3 Captions Supplemental Figure S1 Supplemental Figure S2 Supplemental Figure S3
## Conflict of interest
The authors declare that they have no conflicts of interest with the contents of this article.
## Author contributions
S. D. and A. S. conceptualization; S. D. methodology; S. D. validation; S. D. formal analysis; S. D., A. M., C. D., R. B., and S. K. investigation; S. D., S. C., and A. S. data curation; S. D. writing–original draft; S. D. and A. S. writing–review & editing; A. S. supervision; S. C. and A. S. funding acquisition.
## Funding and additional information
The work was supported by project grant sponsored by the $\frac{10.13039}{501100001409}$Department of Science and Technology-Science and Engineering Research Board to A. S. (DST-SERB, grant no.: CRG/$\frac{2020}{000348}$) and the $\frac{10.13039}{501100001409}$Department of Science and Technology-Science and Engineering Research Board to S. C. (DST-SERB; no.: EMR/$\frac{2017}{001382}$). S.D. is a recipient of a predoctoral fellowship from the $\frac{10.13039}{501100001501}$University Grant Commission, India.
## References
1. Li F., Wang X., Capasso J.M., Gerdes A.M.. **Rapid transition of cardiac myocytes from hyperplasia to hypertrophy during postnatal development**. *J. Mol. Cell. Cardiol.* (1996) **28** 1737-1746. PMID: 8877783
2. Porrello E.R., Mahmoud A.I., Simpson E., Hill J.A., Richardson J.A., Olson E.N.. **Transient regenerative potential of the neonatal mouse heart**. *Science* (2011) **331** 1078-1080. PMID: 21350179
3. Pasumarthi K.B., Field L.J.. **Cardiomyocyte cell cycle regulation**. *Circ. Res.* (2002) **90** 1044-1054. PMID: 12039793
4. Siedner S., Krüger M., Schroeter M., Metzler D., Roell W., Fleischmann B.K.. **Developmental changes in contractility and sarcomeric proteins from the early embryonic to the adult stage in the mouse heart**. *J. Physiol.* (2003) **548** 493-505. PMID: 12640016
5. Zhu F., Meng Q., Yu Y., Shao L., Shen Z.. **Adult cardiomyocyte proliferation: a new insight for myocardial infarction therapy**. *J. Cardiovasc. Transl. Res.* (2021) **14** 457-466. PMID: 32820393
6. Lu F., Langenbacher A., Chen J.N.. **Tbx20 drives cardiac progenitor formation and cardiomyocyte proliferation in zebrafish**. *Dev. Biol.* (2017) **421** 139-148. PMID: 27940156
7. Stennard F.A., Costa M.W., Lai D., Biben C., Furtado M.B., Solloway M.J.. **Murine T-box transcription factor Tbx20 acts as a repressor during heart development, and is essential for adult heart integrity, function and adaptation**. *Development* (2005) **132** 2451-2462. PMID: 15843414
8. Cai C.L., Zhou W., Yang L., Bu L., Qyang Y., Zhang X.. **T-box genes coordinate regional rates of proliferation and regional specification during cardiogenesis**. *Development* (2005) **132** 2475-2487. PMID: 15843407
9. Singh M.K., Christoffels V.M., Dias J.M., Trowe M.O., Petry M., Schuster-Gossler K.. **Tbx20 is essential for cardiac chamber differentiation and repression of Tbx2**. *Development* (2005) **132** 2697-2707. PMID: 15901664
10. Shen T., Aneas I., Sakabe N., Dirschinger R.J., Wang G., Smemo S.. **Tbx20 regulates a genetic program essential to adult mouse cardiomyocyte function**. *J. Clin. Invest.* (2011) **121** 4640-4654. PMID: 22080862
11. Chakraborty S., Yutzey K.E.. **Tbx20 regulation of cardiac cell proliferation and lineage specialization during embryonic and fetal development**. *Dev. Biol.* (2012) **363** 234-246. PMID: 22226977
12. Adams C.J., Kopp M.C., Larburu N., Nowak P.R., Ali M.. **Structure and molecular mechanism of ER stress signaling by the unfolded protein response signal activator IRE1**. *Front. Mol. Biosci.* (2019) **6** 11. PMID: 30931312
13. Das S., Mondal A., Samanta J., Chakraborty S., Sengupta A.. **Unfolded protein response during cardiovascular disorders: a tilt towards pro-survival and cellular homeostasis**. *Mol. Cell. Biochem.* (2021) **476** 4061-4080. PMID: 34259975
14. Sen P., Polley A., Sengupta A., Chakraborty S.. **Tbx20 promotes H9c2 cell survival against oxidative stress and hypoxia**. *Indian J. Exp. Biol.* (2019) **57** 643-655
15. Wang Y., Shen J., Arenzana N., Tirasophon W., Kaufman R.J., Prywes R.. **Activation of ATF6 and an ATF6 DNA binding site by the endoplasmic reticulum stress response**. *J. Biol. Chem.* (2000) **275** 27013-27020. PMID: 10856300
16. Blackwood E.A., Azizi K., Thuerauf D.J., Paxman R.J., Plate L., Kelly J.W.. **Pharmacologic ATF6 activation confers global protection in widespread disease models by reprograming cellular proteostasis**. *Nat. Commun.* (2019) **10** 187. PMID: 30643122
17. Ramdas Nair A., Lakhiani P., Zhang C., Macchi F., Sadler K.C.. **A permissive epigenetic landscape facilitates distinct transcriptional signatures of activating transcription factor 6 in the liver**. *Genomics* (2022) **114** 107-124. PMID: 34863900
18. Wu J., Chen S., Liu H., Zhang Z., Ni Z., Chen J.. **Tunicamycin specifically aggravates ER stress and overcomes chemoresistance in multidrug-resistant gastric cancer cells by inhibiting N-glycosylation**. *J. Exp. Clin. Cancer Res.* (2018) **37** 272. PMID: 30413206
19. Zeeshan H.M., Lee G.H., Kim H.R., Chae H.J.. **Endoplasmic reticulum stress and associated ROS**. *Int. J. Mol. Sci.* (2016) **17** 327. PMID: 26950115
20. Cao S.S., Kaufman R.J.. **Endoplasmic reticulum stress and oxidative stress in cell fate decision and human disease**. *Antioxid. Redox Signal.* (2014) **21** 396-413. PMID: 24702237
21. Shen T., Yang C., Ding L., Zhu Y., Ruan Y., Cheng H.. **Tbx20 functions as an important regulator of estrogen-mediated cardiomyocyte protection during oxidative stress**. *Int. J. Cardiol.* (2013) **168** 3704-3714. PMID: 23871353
22. Chakraborty S., Sengupta A., Yutzey K.E.. **Tbx20 promotes cardiomyocyte proliferation and persistence of fetal characteristics in adult mouse hearts**. *J. Mol. Cell Cardiol.* (2013) **62** 203-213. PMID: 23751911
23. Goetz S.C., Conlon F.L.. **Cardiac progenitors and the embryonic cell cycle**. *Cell Cycle* (2007) **6** 1974-1981. PMID: 17712230
24. Sano R., Reed J.C.. **ER stress-induced cell death mechanisms**. *Biochim. Biophys. Acta* (2013) **1833** 3460-3470. PMID: 23850759
25. Cai X., Nomura-Kitabayashi A., Cai W., Yan J., Christoffels V.M., Cai C.L.. **Myocardial Tbx20 regulates early atrioventricular canal formation and endocardial epithelial-mesenchymal transition via Bmp2**. *Dev. Biol.* (2011) **360** 381-390. PMID: 21983003
26. Boogerd C.J., Zhu X., Aneas I., Sakabe N., Zhang L., Sobreira D.R.. **Tbx20 is required in mid-gestation cardiomyocytes and plays a central role in atrial development**. *Circ. Res.* (2018) **123** 428-442. PMID: 29903739
27. Szegezdi E., Logue S.E., Gorman A.M., Samali A.. **Mediators of endoplasmic reticulum stress-induced apoptosis**. *EMBO Rep.* (2006) **7** 880-885. PMID: 16953201
28. Kropski J.A., Blackwell T.S.. **Endoplasmic reticulum stress in the pathogenesis of fibrotic disease**. *J. Clin. Invest.* (2018) **128** 64-73. PMID: 29293089
29. Furmanik M., van Gorp R., Whitehead M., Ahmad S., Bordoloi J., Kapustin A.. **Endoplasmic reticulum stress mediates vascular smooth muscle cell calcification via increased release of Grp78 (glucose-regulated protein, 78 kDa)-loaded extracellular vesicles**. *Arterioscler. Thromb. Vasc. Biol.* (2021) **41** 898-914. PMID: 33297752
30. Cao Z., Jia Y., Zhu B.. **BNP and NT-proBNP as diagnostic biomarkers for cardiac dysfunction in both clinical and forensic medicine**. *Int. J. Mol. Sci.* (2019) **20** 1820. PMID: 31013779
31. Waspe L.E., Ordahl C.P., Simpson P.C.. **The cardiac beta-myosin heavy chain isogene is induced selectively in alpha 1-adrenergic receptor-stimulated hypertrophy of cultured rat heart myocytes**. *J. Clin. Invest.* (1990) **85** 1206-1214. PMID: 2156896
32. Lipskaia L., Chemaly E.R., Hadri L., Lompre A.M., Hajjar R.J.. **Sarcoplasmic reticulum Ca(2+) ATPase as a therapeutic target for heart failure**. *Expert Opin. Biol. Ther.* (2010) **10** 29-41. PMID: 20078230
33. Li Z., Zhang T., Dai H., Liu G., Wang H., Sun Y.. **Involvement of endoplasmic reticulum stress in myocardial apoptosis of streptozocin-induced diabetic rats**. *J. Clin. Biochem. Nutr.* (2007) **41** 58-67. PMID: 18392099
34. Back S.H., Kaufman R.J.. **Endoplasmic reticulum stress and type 2 diabetes**. *Annu. Rev. Biochem.* (2012) **81** 767-793. PMID: 22443930
35. Senyo S.E., Steinhauser M.L., Pizzimenti C.L., Yang V.K., Cai L., Wang M.. **Mammalian heart renewal by pre-existing cardiomyocytes**. *Nature* (2013) **493** 433-436. PMID: 23222518
36. Xiang F.L., Guo M., Yutzey K.E.. **Overexpression of Tbx20 in adult cardiomyocytes promotes proliferation and improves cardiac function after myocardial infarction**. *Circulation* (2016) **133** 1081-1092. PMID: 26841808
37. Masaki M., Izumi M., Oshima Y., Nakaoka Y., Kuroda T., Kimura R.. **Smad1 protects cardiomyocytes from ischemia-reperfusion injury**. *Circulation* (2005) **111** 2752-2759. PMID: 15911698
38. Yang Y.L., Ju H.Z., Liu S.F., Lee T.C., Shih Y.W., Chuang L.Y.. **BMP-2 suppresses renal interstitial fibrosis by regulating epithelial-mesenchymal transition**. *J. Cell. Biochem.* (2011) **112** 2558-2565. PMID: 21590708
39. Chung Y.H., Huang Y.H., Chu T.H., Chen C.L., Lin P.R., Huang S.C.. **BMP-2 restoration aids in recovery from liver fibrosis by attenuating TGF-β1 signaling**. *Lab. Invest.* (2018) **98** 999-1013. PMID: 29789683
40. Zhou C., Chen D., Ren J., Huang D., Li R., Luo H.. **FGF8 and BMP2 mediated dynamic regulation of dental mesenchyme proliferation and differentiation via Lhx8/Suv39h1 complex**. *J. Cell. Mol. Med.* (2021) **25** 3051-3062. PMID: 33580754
41. Martindale J.J., Fernandez R., Thuerauf D., Whittaker R., Gude N., Sussman M.A.. **Endoplasmic reticulum stress gene induction and protection from ischemia/reperfusion injury in the hearts of transgenic mice with a tamoxifen-regulated form of ATF6**. *Circ. Res.* (2006) **98** 1186-1193. PMID: 16601230
42. Yan B., Liu S., Li X., Zhong Y., Tong F., Yang S.. **Preconditioning with endoplasmic reticulum stress alleviated heart ischemia/reperfusion injury via modulating IRE1/ATF6/RACK1/PERK and PGC-1α in diabetes mellitus**. *Biomed. Pharmacother.* (2019) **118** 109407. PMID: 31545290
43. Thuerauf D.J., Hoover H., Meller J., Hernandez J., Su L., Andrews C.. **Sarco/endoplasmic reticulum calcium ATPase-2 expression is regulated by ATF6 during the endoplasmic reticulum stress response: intracellular signaling of calcium stress in a cardiac myocyte model system**. *J. Biol. Chem.* (2001) **276** 48309-48317. PMID: 11595740
44. Jin J.K., Blackwood E.A., Azizi K., Thuerauf D.J., Fahem A.G., Hofmann C.. **ATF6 decreases myocardial ischemia/reperfusion damage and links ER stress and oxidative stress signaling pathways in the heart**. *Circ. Res.* (2017) **120** 862-875. PMID: 27932512
45. Lu P.D., Jousse C., Marciniak S.J., Zhang Y., Novoa I., Scheuner D.. **Cytoprotection by pre-emptive conditional phosphorylation of translation initiation factor 2**. *EMBO J.* (2004) **23** 169-179. PMID: 14713949
46. Liu X., Kwak D., Lu Z., Xu X., Fassett J., Wang H.. **Endoplasmic reticulum stress sensor protein kinase R-like endoplasmic reticulum kinase (PERK) protects against pressure overload-induced heart failure and lung remodeling**. *Hypertension* (2014) **64** 738-744. PMID: 24958502
47. Ghosh R., Lipson K.L., Sargent K.E., Mercurio A.M., Hunt J.S., Ron D.. **Transcriptional regulation of VEGF-A by the unfolded protein response pathway**. *PLoS One* (2010) **5** e9575. PMID: 20221394
48. Liu M., Dudley S.C.. **Role for the unfolded protein response in heart disease and cardiac arrhythmias**. *Int. J. Mol. Sci.* (2015) **17** 52. PMID: 26729106
49. Yamaguchi O., Higuchi Y., Hirotani S., Kashiwase K., Nakayama H., Hikoso S.. **Targeted deletion of apoptosis signal-regulating kinase 1 attenuates left ventricular remodeling**. *Proc. Natl. Acad. Sci. U. S. A.* (2003) **100** 15883-15888. PMID: 14665690
50. Huang R.T., Wang J., Xue S., Qiu X.B., Shi H.Y., Li R.G.. **TBX20 loss-of-function mutation responsible for familial tetralogy of Fallot or sporadic persistent truncus arteriosus**. *Int. J. Med. Sci.* (2017) **14** 323-332. PMID: 28553164
51. Pan Y., Geng R., Zhou N., Zheng G.F., Zhao H., Wang J.. **TBX20 loss-of-function mutation contributes to double outlet right ventricle**. *Int. J. Mol. Med.* (2015) **35** 1058-1066. PMID: 25625280
52. Nachimuthu S., Assar M.D., Schussler J.M.. **Drug-induced QT interval prolongation: mechanisms and clinical management**. *Ther. Adv. Drug Saf.* (2012) **3** 241-253. PMID: 25083239
53. Bahler R.C., Vrobel T.R., Martin P.. **The relation of heart rate and shortening fraction to echocardiographic indexes of left ventricular relaxation in normal subjects**. *J. Am. Coll. Cardiol.* (1983) **2** 926-933. PMID: 6630767
54. Coppola G., Carità P., Corrado E., Borrelli A., Rotolo A., Guglielmo M.. **ST segment elevations: always a marker of acute myocardial infarction?**. *Indian Heart J.* (2013) **65** 412-423. PMID: 23993002
55. Simões Sato A.Y., Bub G.L., Campos A.H.. **BMP-2 and -4 produced by vascular smooth muscle cells from atherosclerotic lesions induce monocyte chemotaxis through direct BMPRII activation**. *Atherosclerosis* (2014) **235** 45-55. PMID: 24814649
56. Pulkkinen H.H., Kiema M., Lappalainen J.P., Toropainen A., Beter M., Tirronen A.. **BMP6/TAZ-hippo signaling modulates angiogenesis and endothelial cell response to VEGF**. *Angiogenesis* (2021) **24** 129-144. PMID: 33021694
57. Chipurupalli S., Samavedam U., Robinson N.. **Crosstalk between ER stress, autophagy and inflammation**. *Front. Med. (Lausanne)* (2021) **8**
58. Csiszar A., Ahmad M., Smith K.E., Labinskyy N., Gao Q., Kaley G.. **Bone morphogenetic protein-2 induces proinflammatory endothelial phenotype**. *Am. J. Pathol.* (2006) **168** 629-638. PMID: 16436676
59. Walters S.N., Luzuriaga J., Chan J.Y., Grey S.T., Laybutt D.R.. **Influence of chronic hyperglycemia on the loss of the unfolded protein response in transplanted islets**. *J. Mol. Endocrinol.* (2013) **51** 225-232. PMID: 23833251
60. Crider A., Nelson T., Davis T., Fagan K., Vaibhav K., Luo M.. **Estrogen receptor β agonist attenuates endoplasmic reticulum stress-induced changes in social behavior and brain connectivity in mice**. *Mol. Neurobiol.* (2018) **55** 7606-7618. PMID: 29430617
61. Cybulsky A.V., Takano T., Papillon J., Khadir A., Liu J., Peng H.. **Complement C5b-9 membrane attack complex increases expression of endoplasmic reticulum stress proteins in glomerular epithelial cells**. *J. Biol. Chem.* (2002) **277** 41342-41351. PMID: 12191998
62. Du L., Liu C., Teng M., Meng Q., Lu J., Zhou Y.. **Anti-diabetic activities of Paecilomyces tenuipes N45 extract in alloxan-induced diabetic mice**. *Mol. Med. Rep.* (2016) **13** 1701-1708. PMID: 26718133
63. Mondal A., Das S., Samanta J., Chakraborty S., Sengupta A.. **YAP1 induces hyperglycemic stress-mediated cardiac hypertrophy and fibrosis in an AKT-FOXM1 dependent signaling pathway**. *Arch. Biochem. Biophys.* (2022) **722** 109198. PMID: 35358489
64. Li C., Zhang J., Xue M., Li X., Han F., Liu X.. **SGLT2 inhibition with empagliflozin attenuates myocardial oxidative stress and fibrosis in diabetic mice heart**. *Cardiovasc. Diabetol.* (2019) **18** 15. PMID: 30710997
65. Guha P., Kaptan E., Gade P., Kalvakolanu D.V., Ahmed H.. **Tunicamycin induced endoplasmic reticulum stress promotes apoptosis of prostate cancer cells by activating mTORC1**. *Oncotarget* (2017) **8** 68191-68207. PMID: 28978108
66. Abdullahi A., Stanojcic M., Parousis A., Patsouris D., Jeschke M.G.. **Modeling acute ER stress**. *Shock* (2017) **47** 506-513. PMID: 27755507
67. Xiang X.Y., Yang X.C., Su J., Kang J.S., Wu Y., Xue Y.N.. **Inhibition of autophagic flux by ROS promotes apoptosis during DTT-induced ER/oxidative stress in HeLa cells**. *Oncol. Rep.* (2016) **35** 3471-3479. PMID: 27035858
68. Li Z., Guo J., Bian Y., Zhang M.. **Intermedin protects thapsigargin-induced endoplasmic reticulum stress in cardiomyocytes by modulating protein kinase A and sarco/endoplasmic reticulum Ca2+-ATPase**. *Mol. Med. Rep.* (2021) **23** 107. PMID: 33300086
69. Okada T., Haze K., Nadanaka S., Yoshida H., Seidah N.G., Hirano Y.. **A serine protease inhibitor prevents endoplasmic reticulum stress-induced cleavage but not transport of the membrane-bound transcription factor ATF6**. *J. Biol. Chem.* (2003) **278** 31024-31032. PMID: 12782636
70. Hatzopoulos S., Petruccelli J., Laurell G., Finesso M., Martini A.. **Evaluation of anesthesia effects in a rat animal model using otoacoustic emission protocols**. *Hear. Res.* (2002) **170** 12-21. PMID: 12208537
71. Sengupta A., Molkentin J.D., Paik J.H., DePinho R.A., Yutzey K.E.. **FoxO transcription factors promote cardiomyocyte survival upon induction of oxidative stress**. *J. Biol. Chem.* (2011) **286** 7468-7478. PMID: 21159781
|
---
title: Alnus nitida and urea-doped Alnus nitida-based silver nanoparticles synthesis,
characterization, their effects on the biomass and elicitation of secondary metabolites
in wheat seeds under in vitro conditions
authors:
- Sajad Khan
- Raham Sher Khan
- Muhammad Zahoor
- Noor Ul Islam
- Tariq Khan
- Zar Muhammad
- Riaz Ullah
- Ahmed Bari
journal: Heliyon
year: 2023
pmcid: PMC10036665
doi: 10.1016/j.heliyon.2023.e14579
license: CC BY 4.0
---
# Alnus nitida and urea-doped Alnus nitida-based silver nanoparticles synthesis, characterization, their effects on the biomass and elicitation of secondary metabolites in wheat seeds under in vitro conditions
## Abstract
Nano-fertilizers are superior to conventional fertilizers, but their effectiveness has not yet been adequately explored in the field of agriculture. In this study, silver nanoparticles using leaves extract of an *Alnus nitida* plant were synthesized and further doped with urea to enhance the plant biomass and metabolic contents. The synthesized *Alnus nitida* silver nanoparticles (A.N-AgNPs) and urea-doped silver nanoparticles (U-AgNPs) were characterized using Scanning Electron Microscopy, Transmission Electron Microscopy, Powder X-ray Diffraction, and Energy Dispersive X-ray. The wheat seeds were grown in media under controlled conditions in the plant growth chamber. The effectiveness of nanoparticles was studied using different A.N-AgNPs and U-AgNPs concentrations (0.75 μg/ml, 1.5 μg/ml, 3 μg/ml, 6 μg/ml, and 15 μg/ml). They were compared with a control group that received no dose of nanoparticles. The plant biomass, yield parameters, and wheat quality were analyzed. The effect of silver nanoparticles and U-AgNPs were examined in developing wheat seeds and their potency in combating biotic stresses such as nematodes, herbivores, fungi, insects, weeds and bacteria; abiotic stresses such as salinity, ultraviolet radiation, heavy metals, temperature, drought, floods etc. In the seedlings, six possible phytochemicals at a spray dose of 6 μg/ml of U-AgNPs were identified such as dihydroxybenzoic acids, vanillic acid, apigenin glucosidase, p-coumaric acid, sinapic acid, and ferulic acid whereas in other treatments the number of phenolic compounds was lesser in number as well as in concentrations. Moreover, various parameters of the wheat plants, including their dry weight and fresh weight, were assessed and compared with control group. The findings of the study indicated that A.N-AgNPs and U-AgNPs act as metabolite elicitors that induced secondary metabolite production (total phenolic, flavonoid, and chlorophyll contents). In addition, U-AgNPs provided a nitrogen source and were considered a smart nitrogen fertilizer that enhanced the plant biomass, yields, and metabolite production.
## Introduction
The surging demand for food driven by a growing global population with elevated living standards poses considerable challenges for agriculture and has a profound impact on the environment. The world is experiencing an alarmingly high growth rate in its population. According to the latest statistics, the world's population has surpassed 7.884 billion peoples. The rapid growth rate become a serious concern for humanity to provide food to the massive population [1]. In the next few decades, global food demand will surge to $62\%$ and the hunger risk to $30\%$, which would be an alarming situation for the world [2]. The extensive application of chemical fertilizers has been a key contributor to improving crop productivity. Nevertheless, this practice has given rise to significant environmental pollution concerns [3]. Proper fertilizer management in the field is one of the most significant challenges, mainly focusing on the maximum nutritional efficacy of fertilizers to induce crop yield and ensure environmental safety [4]. Urea plays a vital role as a nitrogen fertilizer and serves as a critical component of chemical fertilizers that are indispensable for agricultural practices [5]. It is interesting to know that nitrogen fertilizer utilization had increased to $800\%$ since 1961 [6].
Elicitation is the most efficacious technique in vitro culture for inducing the production of secondary metabolites in various medicinal plants [7]. The nanoparticles act as abiotic elicitors to enhance the production of secondary metabolites by performing oxidative stress-relieving effects in the plant cells, improving antioxidant potential and metabolism [8,9]. The rapid expansion of the world's population and the recent global outbreak of COVID-19 are aggravating the problem of food scarcity, making it an increasingly pressing issue. These challenges are compelling researchers to develop more effective and sustainable techniques for crop production, which can cater to the growing food demands of humanity [10]. Thus, scientists have tried various approaches in which nanotechnology was proven highly effective with less toxicity. The environmental concerns associated with conventional fertilizers include water pollution, soil degradation, greenhouse gas emissions, and negative impacts on biodiversity. Developing a novel form of nanotechnologies-based nano-fertilizers is one of the promising approaches to significantly enhance wheat plant yield and quality, reducing the environmental issues (negative impacts on soil fertility, water, and air quality; causes greenhouse gas emissions, eutrophication, and health problem) associated with conventional fertilization. Smart fertilizers improve plant biomass, yields, and response against biotic and abiotic stresses [[11], [12], [13]]. In this way, fertilizers doped with nanoparticles act as nano fertilizers, as they not only supply the required amount of nitrogen to plants, but also protect them from microbial infections and environmental stresses [14,15].
The selection of *Alnus nitida* leaves extract for the biosynthesis of silver nanoparticles was based on various factors such as easily accessible, a rich source of plant secondary metabolites, such as flavonoids, alkaloids, tannins, and various biomolecules, such as proteins, enzymes, and sugars, reducing silver ions and synthesizing silver nanoparticles with a high degree of uniformity and stability [16]. The phytochemical compounds of *Alnus nitida* transfer their electrons to the metal ion resulting in the synthesis of nanoparticles, and quercetin has been identified as the compound responsible for the reduction process [17]. There are various physical, chemical, and biological techniques available for the synthesis of inorganic nanoparticles, including chemical reduction, sol-gel synthesis, hydrothermal synthesis, and laser ablation. However, green synthesis is a novel technique to produce nanoparticles, which employs natural, non-toxic, and sustainable materials like plant extracts, bacteria, or fungi. This approach provides significant benefits over conventional methods, which use hazardous chemicals and high-energy processes [18]. The central hypothesis driving green synthesis is that by utilizing natural, non-toxic, and sustainable materials, it is possible to develop a safer, more environmentally friendly, and cost-effective approach for producing nanoparticles that possess the desired properties for various applications [19]. It involves the identification and optimization of green synthesis methods to improve the efficiency, stability, and scalability of the process. Moreover, it allows for better control over the shape, size, and properties of the nanoparticles produced during the reaction. Therefore, biological synthesis holds significant promise as a superior alternative to traditional physical and chemical approaches to synthesized inorganic nanoparticles [20]. Silver nanoparticles are valuable in various fields, such as biomedicine, environmental remediation, and consumer products, due to their distinct physical and chemical properties. Their high surface area-to-volume ratio enables them to be highly reactive and interact with a broad range of molecules, making them a versatile material [21,22].
The present attempt made herein is novel as for the first time silver nanoparticles and urea-doped silver nanoparticles were synthesized using Alnus nitida-based leaves extract. The newly synthesized urea-doped silver nanoparticles (U-AgNPs) and silver nanoparticles (A.N-AgNPs) were subjected to wheat seeds under in vitro conditions. Moreover, we conducted a thorough examination of the biochemical and physiological effects of A.N-AgNPs and U-AgNPs in wheat seeds. The utilization of this innovative approach is a significant step forward in the field of silver nanoparticle applications. By utilizing plant extract and doping the nanoparticles with urea, the efficacy of the nanoparticles is significantly enhanced. The results of the study have the potential to revolutionize the field of nanoparticle synthesis and its applications in agriculture.
## Preparation of leaf in powdered form
To prepare the powder from Alnus nitida, fresh leaves were collected and washed meticulously with distilled water to eliminate any impurities. Afterward, they were dried in a well-ventilated area or a low-temperature oven (below 60 °C) to prevent the loss of bioactive compounds. Once dry, the leaves were finely ground into a powder using a mortar and pestle. The resulting powder was then transferred into a clean, dry container and stored in a cool, dark location until needed to synthesize silver nanoparticles.
## Preparation of Alnus nitida extract
The fresh leaves of *Alnus nitida* plant were collected from the upper hilly area of the Ziarat Talash region of District Dir lower, Khyber Pakhtunkhwa Pakistan. The plant was identified by a taxonomist in the Herbarium; Department of Botany Abdul Wali Khan University Mardan, Pakistan. To prepare the extract of *Alnus nitida* leaves, 10 g of fresh leaves were weighed and added to a 500 ml beaker containing 100 ml of distilled water. The beaker was then heated to 60 °C and maintained at this temperature for 10 min before being decanted. The solution was filtered using Whatman Filter Paper No. 1, and the resulting filtrate was centrifuged at 15,000 rpm for 20 min. The resulting pellet was then dispersed in distilled water and kept for further use.
## Chemicals
Silver nitrate (AgNO 3), and all other solvents and chemicals used in the study were of the highest purity and analytical grade. These chemicals were provided by Sigma Aldrich.
## Preparation of silver nitrate solution
The solution of 4 mM silver nitrate was prepared in deionized distilled water. Subsequently, a 0.4 mM silver nitrate solution was prepared through dilution from the aforementioned stock solution and stored in a jar for further synthesis of nanoparticles.
## Biosynthesis of silver nanoparticles using Alnus nitida extract
The silver nitrate solution (0.4 mM) of 30 ml was added to 10 ml of *Alnus nitida* leaves extract and the synthesis of silver nanoparticles was performed by adding 30 ml of 0.4 mM of silver nitrate solution to 10 ml of *Alnus nitida* leaves extract. The reaction was carried out under strict dark conditions to prevent the photoactivation of silver nitrate. After 12 h, the solution changed its color, confirming the formation of silver nanoparticles. The solution containing silver nanoparticles was subjected to centrifugation at 4400 rpm and 25 °C for 15 min. The residues were recovered from the solution and dried at room temperature. The fine powder was obtained and stored for further analysis [[23], [24], [25]].
## Urea-doped silver nanoparticles
The synthesis of urea-doped silver nanoparticles was performed using the protocol of Carmona et al. [ 26]. The silver nanoparticles solution was heated to 37 °C for 5 min. The solution suspension was centrifuged at 4500 rpm for 10 min and washed (300 ml × 2). Urea was poured into the water (1.0 g in 6 ml) and mixed vigorously with slurry (obtained silver nanoparticles) to obtain a homogenate mixture. After homogenization, the sample was frozen and lyophilized. The powder sample was obtained which was stored at 4 °C for further studies [27].
## Plant material and culture conditions
To sterilize the wheat seeds, they were subjected to $0.1\%$ mercuric chloride solution and were kept on filter paper (3–5 min) till dryness to enhance germination speed [28]. Murashige and *Skoog media* (MS Media, M519 Phytotech lab, USA) were used for seeds growth [29]. Therefore, a total of 1.4 g of MS media was carefully measured and combined with 10.5 g of sucrose and 2.8 g of agar in 350 ml of distilled water. The pH of the media was adjusted to 5.2–5.8. The 350 ml of the synthesized growth medium was distributed evenly into seven separate flasks, and each flask was covered with both cotton and aluminum foil. The flasks were then autoclaved to eliminate any potential contaminants that could negatively affect the growth of wheat plants. After the autoclaving process, the flasks were left undisturbed overnight to allow the growth medium to cool to room temperature and attain optimal conditions for subsequent analysis.
The next step involved transferring the sterilized wheat seeds to each of the prepared growth medium-filled flasks, which were then securely plugged with cotton to ensure sterility. The flasks were then thoroughly covered with aluminum foil to prevent any potential contamination of the seeds. These seed-supplemented flasks were subsequently transferred to a growth chamber (BioBase, BJPX-A1500C, Ltd. China) for germination. The growth chamber was maintained at a temperature of 25 ± 2 °C and a humidity of $70\%$, while the illumination was set to 16 h of light followed by 8 h of darkness. The light was provided by Philips TLD 35 fluorescent lamps, with an intensity of 40 μmol m−2 s−1 [30].
## Dispersion and supplementation of media with A.N-AgNPs and urea-doped silver nanoparticles
The dispersion of the synthesized A.N-AgNPs and U-AgNPs was prepared at varying concentrations for the MS media at 0.75 μg/ml, 1.5 μg/ml, 3 μg/ml, 6 μg/ml, and 15 μg/ml per 350 ml and the control with no nanoparticles. A.N-AgNPs and U-AgNPs were dispersed properly in solution using sonication for 1 h. Afterward, the diverse range of A.N-AgNPs and U-AgNPs were supplemented to the cultured wheat seeds on MS media. In this experiment, the flask without A.N-AgNPs and U-AgNPs were designated as the control. The flasks were transferred to a growth chamber to allow the growth of the seeds. Samples from the seed cultures were harvested on the 10th, 20th, 30th, and 40th days after the initial transfer of the seeds. The fresh weight of the wheat plant was accurately measured using an analytical balance (AW 120, Electronic balance) after each harvest. The plant was then dried overnight at 35 °C in an oven, and the dry weight was recorded. Subsequently, the wheat plant was finely ground to a powder and stored for further analysis. The wheat seeds were collected and thoroughly washed to remove any impurities.
## Total phenolic content assay
To prepare the samples for analysis, a multi-channel micropipette was used to load 20 μL of the sample into each of the 96 wells of a microplate. This was followed by the addition of 90 μL of Folin-Ciocalteu reagent to each well, and then 90 μL of sodium carbonate was added to complete the reaction mixture. The microplate was incubated for 30 min. The positive control in this process was different concentrations of gallic acid (25, 20, 15, 10, and 5 μg/ml). A microplate reader was used to record samples absorbance and determine total phenolic content. Each step of the experiment was performed in triplicate.
## Total flavonoids content assay
To determine the total flavonoid content, 20 μL of the sample was added to a 96-well microplate, along with 10 μL of aluminum chloride, 10 μL of potassium acetate, and 160 μL of distilled water. The microplate was incubated for 30 min. The positive control in this process was the final concentrations of Quercetin (40, 20, 10, 5, and 2.5 μg/ml). A microplate reader was used to record the absorbance of samples. Every step in this experiment was performed in triplicate.
## Pigments contents assay
The spectrophotometer at 646.8, 663.2, and 470 nm was used to measure the values of chlorophyll a (Ca), chlorophyll b (Cb), and carotenoids (Cx + c) [31]. To prepare the wheat plant extract, a wheat plant was carefully collected and ground to produce 1.0 ml of juice. The juice was then mixed with 4.0 ml of acetone and 50 mg of CaCO3, after which the mixture was allowed to stand for 5 min. To remove any insoluble particles, the mixture was then centrifuged at 10,000 rpm for 5 min. The absorbances of the resulting supernatant were measured at 646.8 nm, 663.2 nm, and 470 nm, respectively. The quantities of pigments contained in the juice samples were calculated using equations [1], [2], [3], given below:[1]Ca=12.25A663.2−2.79A646.8[2]Cb=21.50A646.8−5.10A663.2[3]Cx+$c = 1000$A470−182Ca−85.02Cb/198
## Determination of antioxidative potential of wheat plants extract in response to A.N-AgNPs and U-AgNPs
The Diphenyl Picryl Hydrazyl (DPPH) assay was used to determine antioxidant potential (free radicle scavenging activity) of wheat plant extract that received diverse treatments of silver and urea-doped silver nanoparticles. About 20 μL samples and a DPPH reagent of 180 μL were loaded into microplate wells. Afterward, samples were incubated for 1 h. Ascorbic acid was used as a positive control with different concentrations (0.75, 1.5, 3, 6, and 15 μg/ml). A microplate reader was used to record the absorbance of samples. Every step of the experiment was performed in triplicate [10,25].
## HPLC-UV characterization
High-performance liquid chromatography with ultraviolet detection (HPLC-UV) was utilized for the characterization and quantification of the sample using established analytical techniques [32]. The wheat plant was crushed with a mortar and pestle, and then approximately 1 g of powder was equally mixed with methanol. The resulting mixture was then subjected to a water bath at a temperature of 70 °C for 1 h. The mixture was centrifuged at 4000 rpm for 10 min and then filtered with Whatman filter paper. HPLC-UV determines the presence of total phenolic and flavonoid content in the plant samples. The wavelength was set to 320 nm for total phenolic content, and the chromatograms were arranged from 190 to 500 nm.
## Statiscial analyses
In our study, we implemented a completely randomized design to perform experiments. These experiments were repeated twice with three replicates in each experiment to ensure statistical accuracy. We employed a linear regression analysis to determine significant mean differences (P \2 0.05). We utilized the latest version of SPSS (version 20) for carrying out the statistical analysis, which helped us to obtain precise and accurate results. All of the figures in this study was prepared by origin 9.0.
## UV–vis spectroscopy
A UV–visible spectrophotometer (UV-2450, Shimadzu, Japan) was used to analyze the UV–*Vis spectra* of silver nanoparticles. A 1 nm resolution was used to examine the reduction of Ag + to Ag0 by leaves extract. The mixture of AgNO3 and leaves extract was incubated for 24 h. The spectra were arranged between 200 and 800 nm. The background accuracy of experiments was corrected using double distilled water.
## Powder X-ray diffraction (PXRD)
The mixture of A.N-AgNPs and U-AgNPs was subjected to centrifugation at 10,000 rpm for 15 min. The resulting pellets were washed with sterile double distilled water and then centrifuged again at 10,000 rpm for 10 min. The purified pellets of A.N-AgNPs and U-AgNPs were dried in an oven at 50 °C and subjected to analysis using an X-ray Diffraction Unit (XRD) (Pan Analytical, X-pert pro, China). The powdered samples were analyzed using a Cu-Kα radiation source with a range of 20–80°, and the presence of phase variety, crystalline nature, and grain size of the samples was evaluated using X-ray diffraction spectroscopy.
The crystalline nature of silver nanoparticles (Fig. 6) and urea-doped A.N-AgNPs (Fig. 7) was confirmed by PXRD analysis. The diffractograms of A.N-AgNPs and U-AgNPs displayed distinct characteristic peaks, providing strong evidence for the crystalline nature of the synthesized nanoparticles. The diffraction peaks of A.N-AgNPs were observed at 32.5°, 38.3°, 44.4°, 64.6°, and 77.8° corresponding the Ag indices [122], [111], [200]; [220] and [311]. The U-AgNPs diffraction peaks at 32.5°, 38.3°, 44.4°, 64.6° and 77.8° corresponds to the [122], [111], [200], [220] and [311] reflection planes, respectively indicating the presence of silver. The presence of Bragg peaks was due to the leaf extract of *Alnus nitida* containing organic compounds. They are responsible to reduce the silver ions and stabilize the resultant nanoparticles. Ibrahim and Roopan et al. have reported similar PXRD results for silver nanoparticles [38,39].Fig. 6PXRD diffractogram of silver nanoparticles. Fig. 6Fig. 7PXRD diffractogram of urea-doped silver nanoparticles. Fig. 7
## Transmission Electron Microscopy (TEM)
TEM images were obtained using a TALOS transmission electron microscope with a voltage of 200 kV. Initially, the samples were dispersed in MiliQ water and then subjected to 15 min of sonication. The resulting samples were drop-cast onto carbon-coated copper grids (300 mesh) and incubated at room temperature for 1 h. Energy-dispersive X-ray analysis (EDX) was conducted on a TALOS machine using a high-energy beam at a voltage of 40 kV.
## Scanning Electron Microscopy (SEM)
The SEM images were obtained using a focused-ion beam scanning electron microscope that was equipped with SE detectors. The voltage used for our samples was approximately 20 kV. Before imaging, the powdered samples were spread onto a black carbon tape surface on steel grids and coated with an ultrathin layer of electrically conducting gold metal [33].
## Formation of dark brown color
The formation of silver nanoparticles through green synthesis was visually observed, as presented in Fig. 1 a & b). The color of the solution altered after adding AgNO3 to the *Alnus nitida* leaves extract. The formation of a precipitate at the bottom showed the presence of silver nanoparticles. The silver nanoparticles were lipolyses to obtain the powder form of nanoparticles. Fig. 1Silver nanoparticles a) color changes confirmed nanoparticle synthesis b) fabricated nanoparticles in dry powder form. Fig. 1
## UV–visible spectroscopy
The confirmation of silver nanoparticles formation was accomplished through the use of a UV-VIS spectrophotometer, and the corresponding spectra are presented in Fig. 2a and b). The wavelength was adjusted in the range of 250–800 nm. The silver nanoparticles showed an absorbance peak at 400 nm which is an important feature of metallic nanoparticles. The UV–vis spectra provided strong evidence that A.N-AgNPs were synthesized rapidly within a few minutes, indicating that *Alnus nitida* significantly enhances the green synthesis of silver nanoparticles. The green synthesis of AgNPs was also observed for *Megaphrynium macrostachyum* leaf extract [34]. The U-AgNPs exhibited a peak at 336 nm, demonstrating the doping effect of urea on the silver nanoparticles. Fig. 2UV–Vis absorption spectra of a) silver nanoparticles (A.N-AgNPs) and b) urea-doped silver nanoparticles (U-AgNPs) using Alnus nitida-based plant extract. Fig. 2
## SEM and TEM
SEM images clearly distinguished that greenly synthesized silver nanoparticles using *Alnus nitida* extract were spherical with a diameter range of less than 100 nm. Further, an insight into the morphology and size details of A.N-AgNPs and U-AgNPs nanoparticles was provided by TEM. The TEM images also suggested the spherical shape of A.N-AgNPs. The urea-doped silver nanoparticles showed a crystalline shape, indicating that urea was successfully doped with silver nanoparticles. The image obtained in this study showed a greater resemblance with the SEM and TEM results of Rautela [35]. The TEM images are presented in Fig, 3 a&b whereas SEM images are given in Fig. 3 c&d. Fig. 3TEM (a, b) and SEM images (c, d) of A.N-AgNPs and U-AgNPs. Fig. 3
## EDX spectrum
Table 1 summarizes its elemental composition of A.N-AgNPs. A pronounced signal of Ag was detected in their characteristic region due to the surface plasmon resonance, which is a typical feature of metallic silver nanoparticles, displaying a strong peak signal at 3.5 keV. Usually, the typical Ag peak is exhibited at 3 KeV [36]. The analysis showed a relative composition (in percent) of various elements as; C ($41.29\%$), O ($36.17\%$), N ($17.49\%$), and Ag ($5.05\%$). Table 2 summarizes its elemental composition urea-doped silver nanoparticles. The EDX spectra recorded at 3 keV depicted an optical absorption peak of silver in the synthesized nanoparticles. Here the depeicted percent elemental composition was; C ($25.95\%$), O ($33.93\%$), N ($39.74\%$), and Ag ($0.38\%$). The surface of silver nanoparticles was bound to other elements which acted as capping agents [37].Table 1The chemical composition of A.N-AgNPs. Table 1ElementsLine typeApparent concentrationk ratiowt%wt% SigmaStandard sample labelManufacturer standardCK line system41.670.4166941.290.48C VitYesNK line system23.100.0411217.490.70BNYesOK line system30.850.1038136.170.47SiO2YesAgL line system6.010.060115.050.23AgYesTotal:100.00Table 2The chemical composition of urea-doped silver nanoparticles. Table 2ElementsLine typeApparent concentrationk ratiowt%wt% SigmaStandard sample labelManufacturer standardCK line system41.450.4145025.950.29C VitYesNK line system144.610.2574639.740.45BNYesOK line system41.290.1389333.930.39SiO2YesAgL line system0.750.007480.380.10AgYesTotal:100.00 Elemental mapping of A.N-AgNPs was performed by FESEM-EDX and the images are displayed in Fig. 4a and b showing the presence of $25\%$ Ag and $75\%$ other metals. The elemental analysis of U-AgNPs was also conducted using FESEM-EDX, which revealed the presence of $5\%$ Ag in U-AgNPs as given in Fig. 5a and b.Fig. 4FESEM-EDX mapping of A.N-AgNPs (a & b) at different magnification. Fig. 4Fig. 5FESEM-EDX mapping of U-AgNPs (a & b) at different magnification. Fig. 5
## Growth and morphological characteristics of wheat seed cultures
A.N-AgNPs exhibited minimal effects on seed germination at a concentration of 0.75 μg/ml. However, a slight increase in fresh weight was observed as the concentration of nanoparticles was increased to 1.5 μg/ml. The seeds demonstrated better growth at higher concentrations of 3 and 6 μg/ml compared to lower dosages. It was found that the wheat seeds gradually adapted to the nanoparticles and induced metabolite production, as compared to the control group (evident from our HPLC analysis). Previousely, the impact of nanoparticles on seed growth has been demonstrated, for instance, Shaikhaldein et al. [ 2020] investigated the impact of silver nanoparticles on Maerua oblongifolia. They have added various concentrations of AgNPs (0, 10, 20, 30, 40, or 50 mgL−1) to the MS medium. The shoots of M. oblongifolia (2–3 cm) were grown in the enriched AgNPs medium, and growth parameters such as length, weight, height, and number were measured after 6 weeks of in vitro shoot regeneration. The study found that treatment with 20 mgL−1 AgNPs significantly enhances shoot length, shoot number, dry weight, fresh weight, and chlorophyll content [40]. In our study, we observed seed inhibition in a few groups during the initial stage, particularly at the highest concentration of 15 μg/ml. After 40 days, these seeds exhibited smaller fresh weight (0.33 ± 0.39 g) and dry weight (0.11 ± 0.21 g) compared to those exposed to lower AgNPs concentrations. The use of nanoparticles in different concentrations can be either effective or harmful for plant biomass and metabolites, as increasing nanoparticle concentration is associated with toxicity in plants. The FW was reduced as the concentration was increased to 15 μg/ml, showing the toxicity of silver nanoparticles at high concentration. A similar study by Yang et al. [ 2018] also demonstrated the toxicity of silver nanoparticles in wheat plants, where different concentrations of AgNPs (20, 200, and 2000 mg kg−1) were applied in the field for wheat growth (*Triticum aestivum* L.). The results showed that increasing AgNPs dosage causes severe phytotoxicity, including shorter plant height, lower biomass, and lower grain weight [41].
In contrast, the treatment using U-AgNPs showed a more significant effect on the growth of seeds, fresh weight, dry weight, and length of roots. The results showed that the seedlings had the highest fresh weight (FW) on day 30th (2.7 ± 0.68 and 2.8 ± 0.69 g; 2.8 ± 0.45 and 3.1 ± 0.45 g) when supplemented with 3 and 6 μg/ml of A.N-AgNPs and U-AgNPs, respectively. The most significant improvement in FW was recorded due to the adaptation of wheat plants with nanoparticles. Ali et al. [ 2019] conducted a study on the production of biomass, antioxidants and secondary metabolites using in vitro callus cultures of Caralluma tuberculata. They used various concentrations of AgNPs and plant growth regulators (PGRs) and found that the combination of AgNPs and PGRs significantly affected the callus proliferation and considerably enhances the callus biomass in the Murashige and Skoog media. The highest dry (0.051 g/L) and fresh (0.78 g/L) callus biomass were reported in vitro cultures at 60 μg/L of AgNPs in combination with 0.5 mg/L 2,4-D plus 3.0 mg/L BA [42]. In our study, the lowest fresh weight on day 10th (FW = 0.24 ± 0.34 and 0.25 ± 0.31) was observed in the group that was supplemented with 0.75 μg/ml of A.N-AgNPs and U-AgNPs, compared to the control group. However, these applications still demonstrated a significant improvement in FW, with the effect of U-AgNPs being considerably higher than that of A.N-AgNPs. On day 20th, the group grown under the effect of 3 and 6 μg/ml of A.N-AgNPs and U-AgNPs, respectively, had the highest FW of 2.5 ± 0.59, 2.7 ± 0.72; and 2.7 ± 0.12, 2.9 ± 0.52 g respectively. The seedlings displayed noticeable improvement in the first few days, after the application of nanoparticles. By day 30th, the wheat plants treated with nanoparticles had exhibited a more significant increase in biomass compared to the control group.
Similarly, U-AgNPs at concentrations of 3 and 6 μg/ml induced maximum growth of roots and shoots, while at 15 μg/L, seed growth was inhibited, resulting in less biomass production. The longest roots and biomass were observed in seedlings treated with 6 μg/ml U-AgNPs, and root length was positively affected with a treatment starting at 1.5 μg/ml. However, shoot length was negatively affected as the concentration was increased to 15 μg/ml. Zia et al. [ 2020] investigated the effect of silver nanoparticles on the number of roots and shoots in cv. Noblessa, cv. Antigua, and cv. Mariposa. They found that supplementing the MS medium with lower concentrations of AgNPs (6 mg/L) increased the number of shoots per plant. High regeneration shoot rates were observed at 8 mg/L, while AgNPs at 12 mg/L increased the number and length of roots per plant compared with the control. Additionally, the fresh and dry weight of regenerated plants was significantly (P \2 0.05) enhanced at 6 mg/L of AgNPs [43].
The dry weight (DW) of the newly emerged seedlings was incredibly improved to 0.38 ± 0.34, 0.24 ± 0.17 g, after U-AgNPs treatment at 3 and 6 μg/ml on day 10th compared to the control group (0.18 g). The maximum DW was observed on days 20th and 30th at a concentration of 6 μg/ml (0.40 ± 0.34, 0.41 ± 0.17 g, respectively). Similarly, A.N-AgNPs showed a DW of 0.37 ± 0.35 b, 0.38 ± 0.43 g at the same concentrations. On day 40th, the maximum DW was observed after applying 6 μg/ml of U-AgNPs (0.27 ± 0.46 g) and A.N-AgNPs (0.25 ± 0.92 g). Nevertheless, increasing the concentration to 15 μg/ml on day 40th resulted in a reduction in the DW of the wheat plants (Table 3). In another study, Vannini et al. [ 2014] investigated the effects of 1 and 10 mg/L AgNPs on germinating *Triticum aestivum* L. seedlings and found that the application of 10 mg L−1 AgNPs had an adverse effect on seedling growth and morphologically modified the root tip of cells [44]. Our findings demonstrated that the wheat seedlings treated with A.N-AgNPs and U-AgNPs at concentrations of 3 and 6 μg/ml, respectively, have the highest fresh and dry weights. Table 3Effect of silver and urea dopped silver nanoparticles on physiological parameters of wheat seedlings. Table 3Doses (μg/ml)FW (g) at dayDW (g) at dayColor at day0102030401020304010203040Control0.90.22 ± 0.12a0.24 ± 0.19a0.33 ± 0.32b0.35 ± 0.37b0.13 ± 0.09a0.15 ± 0.06a0.18 ± 0.12a0.13 ± 0.05aGY/GYY/GA.N-AgNPs0.750.90.24 ± 0.34b2.2 ± 0.37b2.3 ± 0.36b0.34 ± 0.25c0.17 ± 0.11a0.35 ± 0.30b0.15 ± 0.14a0.14 ± 0.06aGY/GYY/G1.50.90.27 ± 0.37b2.4 ± 0.41d2.5 ± 0.52d0.34 ± 0.27c0.25 ± 0.14a0.37 ± 0.34b0.17 ± 0.21a0.16 ± 0.19aGY/GYY/G30.90.27 ± 0.21c2.5 ± 0.59d2.7 ± 0.68e0.36 ± 0.29b0.27 ± 0.19a0.36 ± 0.42d0.16 ± 0.13a0.18 ± 0.22cGGGG60.90.28 ± 0.41d2.7 ± 0.72e2.8 ± 0.69e0.39 ± 0.34b0.28 ± 0.35b0.37 ± 0.35b0.38 ± 0.43d0.25 ± 0.92dGGGG150.90.21 ± 0.27c2.0 ± 0.6d2.1 ± 0.70e0.33 ± 0.39b0.14 ± 0.08a0.16 ± 0.04a0.12 ± 0.06a0.11 ± 0.21cGY/GY/GYU-AgNPs0.750.90.25 ± 0.31b2.3 ± 0.4d2.3 ± 0.45d0.34 ± 0.24c0.17 ± 0.04a0.35 ± 0.45c0.15 ± 0.03a0.14 ± 0.05aGGY/GY/G1.50.90.28 ± 0.32b2.4 ± 0.3b2.5 ± 0.37b0.36 ± 0.26c0.19 ± 0.18a0.37 ± 0.42d0.19 ± 0.21c0.13 ± 0.23cGGY/GY30.90.30 ± 0.37b2.7 ± 0.12a2.8 ± 0.45d0.36 ± 0.67e0.38 ± 0.34b0.18 ± 0.46d0.37 ± 0.27c0.15 ± 0.35bGGGG60.90.31 ± 0.28c2.9 ± 0.52d3.1 ± 0.45d0.43 ± 0.56d0.24 ± 0.17a0.40 ± 0.34b0.41 ± 0.17a0.27 ± 0.46dGGGG150.90.27 ± 0.26a2.1 ± 0.62e2.2 ± 0.45d0.32 ± 0.46d0.15 ± 0.45d0.31 ± 0.27a0.11 ± 0.02a0.12 ± 0.53dY/GY/GY/GYFresh weight, DW dry weight, Y yellow, G green. The superscripted letters (a, b, c, d etc.) shows signifcant diference between variables FW.Values with diferent letters represent signifcant diference (P \2 0.05).
## Effect of A.N-AgNPs and U-AgNPs on physiological parameters of newly germinated wheat plants
Table 3, summarizes the effects of the fabricated silver nanoparticles and urea-doped silver nanoparticles with Thidiazuron supplementation on various physiological parameters on 10, 20,30, and 40 days old plant produced from the wheat seeds in Murashige and Skoog medium.
## Total phenolic content
The total phenolic content (TPC) of wheat plant samples was evaluated on days 10, 20, 30, and 40, and the results are shown in Table 4. TPC was found to be lower on day 10th with the application of 1.5 μg/ml A.N-AgNPs and U-AgNPs. The lowest TPC was recorded after supplementation with 0.75 μg/ml A.N-AgNPs and U-AgNPs. Overall, TPC on days 10, 20, 30, and 40th was higher in plants supplemented with 3 μg/ml A.N-AgNPs and U-AgNPs (251.32 ± 4.97, 260.32 ± 5.16, 265.88 ± 3.62, 268.34 ± 11.44; 261.11 ± 4.57, 265.66 ± 6.17, 273.72 ± 3.62, 271.44 ± 11.63 μg GAE/g of DW) than in the control group (TPC = 211.44 ± 2.82, 223.17 ± 3.75, 231.57 ± 3.62, 240.41 ± 4.16 μg GAE/g of DW). However, the highest TPC was produced in 30, and 40 days old wheat plants after supplementation with 6 μg/ml U-AgNPs and A.N-AgNPs (DW = 289.43 ± 3.62, 295.95 ± 12.62; 274.44 ± 3.62, 277.54 ± 11.92 μg GAE/g of DW respectively. Likewise, at days 20, 30, and 40th, the addition of varying concentrations of A.N-AgNPs and U-AgNPs, ranging from 0.75 μg/ml (low) to 15 μg/ml (lowest), 1.5 μg/ml (slightly higher), 3 μg/ml (higher), and 6 μg/ml (highest) μg/ml, resulted in a notable improvement in the TPC. Among these days, the TPC was highest on day 40 after 6 μg/ml of U-AgNPs were added to the media. The findings unequivocally demonstrate that the addition of U-AgNPs to the media at a concentration of 6 μg/ml has yielded a remarkable enhancement in the phenolic content. This concentration can be deemed as the optimal level for the purposes of this study. The effects of silver nanoparticles (AgNPs) were assessed on the growth of pearl millet (P. glaucum L.), an economically significant crop. In vitro experiments were conducted using MS basal medium with different concentrations of AgNPs (T1 = control, T2 = 20 ppm, T3 = 40 ppm, T4 = 60 ppm, and T5 = 80 ppm). The results of this study demonstrate that the seedlings treated with T3 exhibited significantly higher total phenolic content (0.56 ± 0.0152 μg/mg) compared to the control and T2. As the doses of AgNPs were increased from T3 to T5, there were significant impacts (p ≤ 0.01) on the accumulation of phenolic compounds. These effects caused a notable decrease in total phenolic content in T4 and T5, which amounted to $12.1\%$ and $35.2\%$ respectively, compared to T1. These findings strongly suggest that higher doses of AgNPs induced more stress on the seedlings, resulting in the accumulation of fewer phenolic compounds. It was evident that high doses of AgNPs acted as stressors, hindering the accrual of TPC [45]. The high concentration of U-AgNPs and A.N-AgNPs was inadequate for the wheat plant as excess nanomaterial failed to cope with cells in the media. A study was carried out to investigate the toxic effect of increasing concentrations of AgNPs on wheat plants. The phenolic content in the plants increased when subjected to stripe rust stress (5.6 μg·mg−1 F W); however, the application of AgNPs led to a decline in the phenolic content. The maximum reduction (3.7 μg mg−1 FW) was observed in plants treated with 75 ppm of AgNPs. These findings suggest that the presence of AgNPs may negatively affect the phenolic content in wheat plants, particularly at higher concentrations [46]. On the other hand, lower concentrations of AgNPs have been found to strongly promote the production of secondary metabolites in the growth media of wheat seeds. However, high concentrations of 25 nm AgNPs had a toxic effect when supplemented with the plant Oryza sativa, as they were found to break the cell wall and damage the vacuoles of root cells [47]. Mirzajani et al. [ 2013] reported that root cell penetration did not occur in O. sativa in the presence of low concentrations of AgNPs (30 μg/ml). The lowest TPC content was recorded after increasing AgNPs to 15 μg/ml. Thus, higher concentrations were found to have a toxic effect, destroying the cell structure. Previously, a study showed that a concentration of 30 μg/ml enhanced root growth, while a concentration of 60 μg/ml suppressed cell growth ability. These findings indicate that the effects of AgNPs on plant growth can be highly dependent on the concentration [48]. AgNPs-elicited hairy roots and a higher TPC concentration were produced in the cultures of *Cucumis anguria* [49].Table 4Total phenolic content in wheat plants and control group under the treatment of different concentrations of A.N-AgNPs and U-AgNPs-based nano fertilizers. Table 4Alnus nitida-based nanoparticlesSamples (μg/mL)Total phenolic content (μg GAE/g of DW)Day 10Day 20Day 30Day 40Control211.44 ± 2.82223.17 ± 3.75231.57 ± 3.62240.41 ± 4.16A.N-AgNPs0.75220.67 ± 3.14223.43 ± 2.37225.33 ± 12.67229.77 ± 5.351.5237.98 ± 3.62238.44 ± 4.15240.22 ± 3.62242.61 ± 6.623251.32 ± 4.97260.32 ± 5.16265.88 ± 3.62268.34 ± 11.446269.74 ± 3.22271.66 ± 4.74274.44 ± 3.62277.54 ± 11.9215243.47 ± 4.16239.66 ± 3.64230.35 ± 3.62220.85 ± 2.35U-AgNPs0.75225.13 ± 2.47230.90 ± 2.60232.45 ± 7.85239.97 ± 3.191.5245.45 ± 5.47250.76 ± 4.29255.35 ± 3.62263.44 ± 4.283261.11 ± 4.57265.66 ± 6.17273.72 ± 3.62271.44 ± 11.636279.74 ± 5.39280.41 ± 5.46289.43 ± 3.62295.95 ± 12.6215249.34 ± 4.71244.66 ± 4.17240.12 ± 3.62241.34 ± 3.47DW = Dry weight, GAE = Gallic acid equivalents. Values are means ± SD of three determinations. Means within each column with different letters differ significantly (P \2 0.05).
## Total flavonoids content
In this study, the total flavonoid content (TFC) in wheat plant samples was evaluated on different days (10, 20, 30, and 40) following various concentrations supplementation with A.N-AgNPs and U-AgNPs, and the results are presented in Table 5. At day 10, the application of 1.5 μg/ml A.N-AgNPs and U-AgNPs steered to a lower TFC production. Generally, TFC was higher (80.33 ± 5.15, 86.22 ± 5.35, 90.11 ± 6.67, 93.22 ± 91.14±6.65, 94.55 ± 6.75, 97.22 ± 5.93, 102.43 ± 6.63 μg GAE/g of DW) on day 10, 20, 30, and 40 after being supplemented with 3 μg/ml A.N-AgNPs and U-AgNPs, respectively, compared to the control group (TFC = 62.12 ± 3.35, 66.67 ± 3.75, 68.47 ± 4.14, 71.33 ± 4.32 μg GAE/g of DW). The highest TFC (127.34 ± 7.86 μg GAE/g of DW) was reported in the 40-day-old wheat plant after supplementation with 6 μg/ml U-AgNPs. In the previous study, biogenic silver nanoparticles were synthesized using *Aloe vera* leaf extracts and their effects on the growth characteristics and flavonoid contents of wheat were studied by exposing the plants to different concentrations of Ag nanoparticles and Ag (0, 40, and 80 ppm). Results showed that the application of 40 and 80 ppm of both Ag nanoparticles and Ag significantly increased the flavonoid contents at 300 and 330 nm wavelengths, while no effect was observed on flavonoid contents at the 270 nm wavelength. These findings suggest that the use of Ag nanoparticles can enhance the flavonoid contents of wheat, which could have potential benefits for human health [50]. Similarly, on days 20 and 40, TFC was enhanced following supplementation with different concentrations of A.N-AgNPs and U-AgNPs, such as 0.75 (lowest), 1.5 (slightly high), 3 (high), 6 (highest), and 15 (lowest) μg/ml. Among these days, TFC was highest on day 40 after supplementation with 6 μg/ml U-AgNPs following by AgNPs. The lowest TFC content (78.44 ± 4.75, 71.34 ± 4.32, 69.33 ± 3.57, 66.12 ± 4.75; 82.45 ± 4.68, 79.34 ± 4.13, 79.55 ± 4.42, 75.22 ± 4.55 μg GAE/g of DW was recorded after enhancing the concentration of AgNPs, U-AgNPs to 15 μg/ml compared to other dosages. In a study, different AgNPs concentrations such as 25 ppm and 50 ppm showed remarkably positive impact on the production of chlorophyll content, soluble sugars, and proteins compared to both the control group and AgNO3. However, the study also revealed that a higher concentration of 100 ppm of AgNPs led to a relatively lower level of total phenolic content, flavonoids, compounds, total reducing potential, total antioxidant capacity, and DPPH. The use of high concentrations of AgNPs may result in negative impacts on other parameters [51]. In our study, we reported that high concentration of U-AgNPs and A.N-AgNPs was inadequate for the wheat plant as excess nanomaterial failed to cope with cells in the media. Previously a study reported the toxicity of AgNPs to the plants that enhanced ROS production and altered the plants' anatomical and genetic performances [52]. The excess production of ROS due to AgNPs exposure is causing several toxic effects on plants, including; peroxidation of polyunsaturated fatty acids (referred to as lipid peroxidation), damage to the permeability of cell membranes, and modification of the structure of the cells. Furthermore, it directly damages DNA and protein and causes growth inhibition and potential cell death in plants [53,54]. Typically, plant secondary metabolites accumulate during the late growth phases. However, by providing elicitors to the media, biosynthesis can be stimulated during the initial stages as a counter-defense mechanism. The study reported that the enhancing effects of AgNPs were observed at every growth stage. Providing AgNPs on day 10 resulted in the highest total flavonoid content, with a value of 11.85 mg QUE/g DW [55].Table 5Total flavonoid content in wheat plants and control group under the treatment of different concentrations of A.N-AgNPs and U-AgNPs-based nano fertilizers. Table 5Alnus nitida-based nanoparticlesSamples (μg/mL)Total flavonoids content (μg GAE/g of DW)Day 10Day 20Day 30Day 40Control62.12 ± 3.3566.67 ± 3.7568.47 ± 4.1471.33 ± 4.32A.N-AgNPs0.7573.27 ± 3.9774.77 ± 4.1374.88 ± 4.8575.11 ± 4.761.576.36 ± 4.4878.24 ± 4.8581.44 ± 5.4584.22 ± 5.67380.33 ± 5.1586.22 ± 5.3590.11 ± 6.6793.22 ± 6.446103.23 ± 6.67104.23 ± 6.85107.57 ± 7.46111.47 ± 7.341578.44 ± 4.7571.34 ± 4.3269.33 ± 3.5766.12 ± 4.75U-AgNPs0.7575.22 ± 4.9875.91 ± 4.6776.20 ± 4.8577.19 ± 5.561.581.33 ± 5.5482.13 ± 5.8686.45 ± 5.4589.40 ± 5.83391.14 ± 6.6594.55 ± 6.7597.22 ± 5.93102.43 ± 6.636117.44 ± 7.47121.89 ± 7.56124.57 ± 6.46127.34 ± 7.861582.45 ± 4.6879.34 ± 4.1379.55 ± 4.4275.22 ± 4.55DW = Dry weight, GAE = Gallic acid equivalents. Values are means ± SD of three determinations. Means within each column with different letters differ significantly (P \2 0.05).
## Chlorophyll content analysis
The silver nanoparticles and urea-doped silver nanoparticles showed a significant aspect in the recent pigments study in wheat leaves. The chlorophyll a, chlorophyll b, carotenoids, and total pigments in the leaf extract of wheat at 3 μg/ml were 2.1 ± 0.62, 1.43 ± 0.56, 0.95 ± 0.85, and 2.7 ± 0.73 mg.g-1 FW, respectively. The total chlorophyll content after wheat treatment at 6 μg/ml was remarkably improved such as 2.6 ± 0.63, 1.87 ± 0.33, 0.976, ±0.15 and 2.91 ± 0.86 mg.g-1 FW in the photosynthetic pigment compared to the control and A.N-AgNPs. Therefore, this treatment could be ideal for enhancing chlorophyll content. Previously, Stevia (*Stevia rebaudiana* B.) was exposed to silver nanoparticles at varying concentrations, including 0 mg/L, 12.5 mg/L, 25 mg/L, 50 mg/L, 100 mg/L, and 200 mg/L. The study revealed significant differences in the chlorophyll a, b, and total contents among the different concentrations of AgNPs. Generally, chlorophyll content was increased starting from 25 mg/L. The control treatment and the less concentration of AgNPs (12.5 mg/L) resulted in lower levels of chlorophyll a, b, and total contents [56]. The silver nanoparticles were biosynthesized using *Ochradenus arabicus* and their physiological effects were checked on *Maerua oblongifolia* raised in vitro, Murashige and *Skoog medium* was supplemented with varying concentrations of AgNPs, including 0 mg/L, 10 mg/L, 20 mg/L, 30 mg/L, 40 mg/L, and 50 mg/L, for 6 weeks. The study results revealed that the plants treated with 20 mg/L AgNPs exhibited an increase in shoot length, shoot number, dry weight, fresh weight, and chlorophyll content. Based on these findings, it can be suggested that AgNPs have the potential to serve as a growth-promoting agent for in vitro raised *Maerua oblongifolia* [40]. The treatment U-AgNPs at 15.6 μg/ml highly reduced the pigment content to 0.66 ± 0.19, 0.21 ± 0.06, 0.12 ± 0.04, 0.73 ± 0.25 mg.g-1 FW. The results showed that enhancing silver nanoparticle concentration negatively affects the formation of photosynthetic pigments. A study was conducted to explore the effects of AgNPs or AgNO3 supplementation on in vitro potato plant cultures. The findings demonstrated that the total chlorophyll content increased significantly in cultures treated with 2 mg of both AgNO3 and AgNPs compared to the control group. However, total chlorophyll content was reduced as the concentration of the supplementation increased to 10 and 20 mg. Notably, the AgNPs treatment group exhibited a more distinct reduction in total chlorophyll content compared to the AgNO3 treatment group [57]. The chlorophyll a, chlorophyll b, carotenoids, and total pigments were tremendously increased to 3.2 ± 0.92, 2.17 ± 0.76, 2.46 ± 0.83, and 3.42 ± 0.98 mg.g-1 FW after treatment wheat seeds (U-AgNPs) at 6 μg/ml. At the same time, the urea-doped silver nanoparticles showed fewer adverse effects with less improvement of the chlorophyll contents. The study presented here provided evidence that both A.N-AgNPs and U-AgNPs significantly improved pigment content. Notably, U-AgNPs were found to be highly effective under in vitro conditions, as shown in Fig. 8 a & b. This may be due to the lower silver nanoparticle content in urea-doped silver nanoparticles. Verma et al. have biologically synthesized the silver nanoparticles. The seed treatment method used different concentrations of silver nanoparticles, such as 0, 15, 30, 60, 120, 240, and 480 mg/L. The low concertation of silver nanoparticles enhanced the chlorophyll content, while higher concentrations showed toxicity and reduced the total chlorophyll [58].Fig. 8The effects of A.N-AgNPs (a) and U-AgNPs (b) on chlorophyll content after 20 days of wheat treatment under in vitro conditions. Superscripted letters, such as a, b, c, d, etc., are used to indicate significant differences between variables. Values of different letters showed a statistically significant difference between (P \2 0.05).Fig. 8
## DPPH free radical scavenging activity
The antioxidant activity was performed using the protocol of % DPPH scavenging capacity the results were depicted in Fig. 9. The wheat plant samples treated with A.N-AgNPs were extracted after 10, 20, 30, and 40 days, and the antioxidant activity was measured. At a concentration of 3 μg/ml of A.N-AgNPs, the antioxidant activity on day 10 was $43.98\%$, whereas the maximum antioxidant activity was observed on day 20 at a concentration of 6 μg/ml. A study conducted on *Echium amoenum* plants treated with various concentrations (25 and 50 ppm) of silver nanoparticles revealed a significant improvement in DPPH free radical scavenging in the treatment groups as compared to the control group, thereby demonstrating the efficacy of AgNPs [59]. The lowest activity (47.75, 43.63, 44.26, $45.46\%$) was recorded in 10-, 20-, 30-, and 40-days-old wheat plants derived from media with 15 μg/ml of A.N-AgNPs. In a study, silver nanoparticles were synthesized using pod extract of *Cola nitida* at diverse concentrations (25, 50, 75, 100, and 150 ppm) and tested their DPPH free radical scavenging activity. The results confirmed that various concentrations of AgNPs significantly improved the antioxidant activity of A. caudatus, except for the treatment with 150 ppm AgNPs, which demonstrated lower antioxidant activity than the control group grown with water [60]. These findings support our results, which showed a reduction in activity when the concentration of silver nanoparticles was increased in vitro. In addition, the control group demonstrated the lowest levels of antioxidant activity at $29.51\%$, $31.39\%$, $44.46\%$, and $35.83\%$. However, supplementing with silver nanoparticles (A.N-AgNPs) revealed the potential for A.N-AgNPs to enhance the antioxidant capacity of wheat seeds. The maximum antioxidant activity levels were observed in 20-, 30-, and 40-day-old wheat plants exposed to 3 μg/ml of A.N-AgNPs, reaching values of $43.98\%$, $44.58\%$, $49.54\%$, and $51.42\%$. Furthermore, increasing the concentration of A.N-AgNPs in the media up to 6 μg/ml significantly induced antioxidant potential in wheat plants, resulting in levels of $65.67\%$, $59.76\%$, $61.85\%$, and $66.35\%$. Another study showed that the antioxidant, chloroform fraction was highly effective during the determination of DPPH, ABTS, and FRAP assays, which displayed an IC50 value of 64.99, 69.15, and 268.52 μg/ml. Ethyl acetate extract has also shown potency in tested free radicals. The potency of both types of extracts against lipoxygenase was confirmed with IC50 values of 75.99 and 106.11 μg/ml, respectively. The results of the biological studies conducted on *Ilex dipyrena* showed that it is a good inhibitor of free radicals and lipoxygenase. Thus, it is very important to further investigate the plant for the isolation of significant medicinal compounds [[61], [62], [63]]. The biological activities of inorganic nanoparticles, including silver nanoparticles, can be influenced by various factors such as size distribution, shape, surface charge, coating, composition, aggregation, solubility, and capping agent. These factors have the potential to significantly impact the cellular uptake, biodistribution, stability, toxicity, and potential applications of the nanoparticles [64,65].Fig. 9The effect of different concentrations of silver nanoparticles (A.N-AgNPs) on 10-, 20-, 30-, and 40-day-old wheat plant seed grown in media. The highest % free radical scavenging activity ($67.36\%$) was reported on day 40 at 6 μg/ml A.N-AgNPs compared to the control and diverse group. Superscripted letters, such as a, b, c, d, etc., are used to indicate significant differences between variables. Values of different letters showed a statistically significant difference between (P \2 0.05).Fig. 9
## HPLC analyses of A.N-AgNPs and U-AgNPs -based nanoparticles
The HPLC analysis of our samples (presented in Table 6) revealed that the addition of A.N-AgNPs and U-AgNPs to the growth medium had a significant effect on the production of phenolic compounds during wheat plant growth. The plant extract obtained from *Alnus nitida* was found to contain five different antioxidant phytochemicals. As this plant is known for its greater diversity compared to wheat, it is expected that the potential compounds between the two will also differ from each other. The application of silver nanoparticles (A.N-AgNPs) and urea-doped silver nanoparticles (U-AgNPs) led to significant modifications in the quality and quantity of secondary metabolites. In comparison to the nanoparticle treatments, the control group exhibited only a single compound. The in vitro wheat seeds treatment at 3 and 6 μg/ml produced four significant metabolites namely Dihydroxybenzoic acids (0.6 μg/ml), Apigenin glucosidase (0.03 μg/ml), p-Coumaric acid (1.3 μg/ml), Sinapic acid (0.21 μg/ml); Apigenin glucosidase (0.135 μg/ml), p-Coumaric acid (0.138 μg/ml), Sinapic acid (0.2 μg/ml), and Ferulic acid (0.27 μg/ml) respectively. The results indicated that increasing the concentration of nanoparticles was positively correlated with the production of secondary metabolites. For instance, the effect of U-AgNPs at 3 and 6 μg/ml produced a diverse range of metabolites such as Dihydroxybenzoic acids (1.02 μg/ml), Vanillic acid (0.25 μg/ml), Apigenin glucosidase (0.245 μg/ml), Sinapic acid (0.28 μg/ml), Ferulic acid (0.2 μg/ml); Dihydroxybenzoic acids (0.66 μg/ml), Vanillic acid (0.28 μg/ml), Apigenin glucosidase (0.084 μg/ml), p-Coumaric acid (0.35 μg/ml), Sinapic acid (0.845 μg/ml), and Ferulic acid (0.135 μg/ml) respectively (Table 6). We also investigated that at 6 μg/ml of both A.N-AgNPs and U-AgNPs treatment resulting in six different peaks that highly induced metabolites. In comparison, the U-AgNPs treatment strongly induced the metabolic contents as compared to A.N-AgNPs. Therefore, we concluded that urea-doped silver nanoparticles might be the significant fertilizers for the wheat seed that would enhance the phenolic contents. Amjad et al. [ 2022] have performed HPLC that investigated five possible compounds in Crd-Id such as catechin hydrate, ellagic acid, morin, epigallocatechin gallate, and rutin. In contrast, seven possible compounds were identified in Et-Id such as malic acid, morin, epigallocatechin gallate, ellagic acid, hydrate, pyrogallol, and catechin rutin. Five possible phenolic compounds in Chl-Id were catechin hydrate, ellagic acid, epigallocatechin gallate, rutin, and morin. Previously, a study of HPLC has performed on A. nitida leaves which exhibited six possible phytochemicals in the Met. Ext such as chlorogenic acid, malic acid, quercetin, epigallocatechin gallate, pyrogallol, and ellagic acid. Stem bark in the Met. Ext showed phenolic compounds such as pyrogallol, ellagic acid, and epigallocatechin gallate. Six phenolic compounds were identified in the Met. Ext of seed such as vitamin C, malic acid, quercetin, epigallocatechin gallate, pyrogallol, and ellagic acid. However, five phenolic compounds were detected in the Met. Ext of root such as ellagic acid, epigallocatechin gallate, malic acid, and quercetin [66,67].Table 6High-performance liquid chromatography-based quantification of phenolic compounds induced as a result of A.N-AgNPs and U-AgNPs in the wheat seed cultures. Table 6Sample extractsRetention TimeNo. of peaksWavelengthPossible compound IdentityConcentration (μg/ml)ReferenceAlnus nitida2.2791320Phloroglucinol2.55Standard2.6772320Dihydroxybenzoic acids0.3Standard32.7973320Quercetin0.18Standard35.2914320Rutin0.175Standard36.7585320Morin0.2StandardControl2.6021320Dihydroxybenzoic acids0.64StandardA.N-AgNPs (3 μg/ml)2.24551320Dihydroxybenzoic acids0.6Standard14.4932320Apigenin glucosidase0.03Standard15.2833320p-Coumaric acid1.3Standard16.1834320Sinapic acid0.21StandardA.N-AgNPs (6 μg/ml)14.3501320Apigenin glucosidase0.135Standard15.0842320p-Coumaric acid0.138Standard16.9363320Sinapic acid0.2Standard17.6994320Ferulic acid0.27StandardU-AgNPs (3 μg/ml)2.4341320Dihydroxybenzoic acids1.02Standard13.6622320Vanillic acid0.25Standard14.562320Apigenin glucosidase0.245Standard16.1563320Sinapic acid0.28Standard17.0384320Ferulic acid0.2StandardU-AgNPs (6 μg/ml)2.6291320Dihydroxybenzoic acids0.66Standard13.9562320Vanillic acid0.28Standard14.3193320Apigenin glucosidase0.084Standard15.1684320p-Coumaric acid0.35Standard16.9715320Sinapic acid0.845Standard17.5086320Ferulic acid0.135Standard
## Conclusions and future perspectives
Medicinal plants are highly valuable in both the medical and agricultural sectors, and this study underscores their importance of utilizing leaf extracts from *Alnus nitida* to synthesize silver nanoparticles. The resulting nanoparticles were shown to be spherical, crystalline, and stable, with particle sizes reaching up to 100 nm as evidenced by UV–visible spectroscopy, PXRD, SEM, TEM, and EDX analyses. Moreover, the study highlights the potential of U-AgNPs as a significant elicitor of plant cell metabolism, prompting the production of secondary metabolites. U-AgNPs-based wheat seed cultures demonstrated a marked increase in phenolic, chlorophyll, flavonoid, and antioxidant activity levels when compared to A.N-AgNPs. The application of urea-doped silver nanoparticles could be an effective approach that produced novel secondary metabolites in plant cells. Silver nanoparticles, especially U-AgNPs, have vast potential to boost the yield of total phenolic, chlorophyll, and flavonoid contents in wheat plants. Furthermore, they effectively enhanced antioxidant activity and biomass production in seed cultures of wheat under in vitro conditions, signifying their potential as a tool for enhancing the growth and yield of critical crops. Silver nanoparticles hold great promise in advancing the agricultural sector and promoting sustainable crop growth. However, to fully understand the effects of these nanoparticles on wheat seedlings and their potential impact on the environment, more research is required. Future studies should focus on investigating the optimal concentration, size, and surface coating of these nanoparticles to maximize the benefits for crop growth and development. By continuing to explore the potential of silver nanoparticles and urea-doped silver nanoparticles, we can find innovative ways to improve the efficiency and sustainability of agriculture, benefiting both the environment and society.
## Author contribution statement
Muhamad Zahoor and Sajad Khan; Conceived and designed the experiments: Performed the experiments; Wrote the paper.
Rahm Sher Khan and Muhammad Zahoor, Noor Ul Islam, Tariq Khan, Rahm Sher Khan, and Sikandar Khan: Analyzed and interpreted the data.
Zar Muhammad; Riaz Ullah; Ahmed Bari: Contributed reagents, materials, analysis tools or data.
## Funding statement
The authors extend their appreciation to the researchers supporting Project number (RSP2023R346). King Saud University, Riyadh, Saudi Arabia, for financial support.
## Data availability statement
No data was used for the research described in the article.
## Declaration of interest's statement
The authors declare no conflict of interest.
## Additional information
No additional information is available for this paper.
## Declaration of competing interest
The authors declared that they do not have any conflict in publishing this research article.
## Abbreviation
A.N-AgNPsAlnus nitida based silver nanoparticlesU-AgNPsurea-doped silver nanoparticleFTIRFourier transform infraredSEMScanning electron microscopeTEMTransmission electron microscopyXRDX-Ray DiffractionEDXEnergy dispersive X-RayHPLCHigh-Performance Liquid ChromatographyTDZThidiazuronMS mediaMurashige and Skoog mediumDWDry weightFWFresh weightGAEGallic acid equivalents
## References
1. Sadigov R.. **Rapid growth of the world population and its socioeconomic results**. *Sci. World J.* (2022.0) **2022**. DOI: 10.1155/2022/8110229
2. van Dijk M., Morley T., Rau M.L., Saghai Y.. **A meta-analysis of projected global food demand and population at risk of hunger for the period 2010–2050**. *Nat. Food.* (2021.0) **2** 494-501. DOI: 10.1038/s43016-021-00322-9
3. Dimkpa C.O., Fugice J., Singh U., Lewis T.D.. **Development of fertilizers for enhanced nitrogen use efficiency – trends and perspectives**. *Sci. Total Environ.* (2020.0) **731**. DOI: 10.1016/j.scitotenv.2020.139113
4. Bora K.. **Spatial patterns of fertilizer use and imbalances: evidence from rice cultivation in India**. *Environ. Challenges* (2022.0) **7**. DOI: 10.1016/j.envc.2022.100452
5. Giroto A.S., do Valle S.F., Guimarães G.G.F., Jablonowski N.D., Ribeiro C., Mattoso L.H.C.. **Different Zn loading in Urea–Formaldehyde influences the N controlled release by structure modification**. *Sci. Rep.* (2021.0) **11** 1-12. DOI: 10.1038/s41598-021-87112-2
6. Shukla S.K., Kumar R., Mishra R.K., Pandey A., Pathak A., Zaidi M., Srivastava S.K., Dikshit A.. **Prediction and validation of gold nanoparticles (GNPs) on plant growth promoting rhizobacteria (PGPR): a step toward development of nano-biofertilizers**. *Nanotechnol. Rev.* (2015.0) **4** 439-448. DOI: 10.1515/NTREV-2015-0036/HTML
7. Paul A., Shamim N., Sarkar A., Acharya K., Chakraborty N.. *Biotechnol. Anti-diabetic Med. Plants* (2021.0) 307-340. DOI: 10.1007/978-981-16-3529-8_11
8. Lala S.. **Nanoparticles as elicitors and harvesters of economically important secondary metabolites in higher plants: a review**. *IET Nanobiotechnol.* (2021.0) **15** 28-57. DOI: 10.1049/nbt2.12005
9. Espinosa-Leal C.A., Puente-Garza C.A., García-Lara S.. **In vitro plant tissue culture: means for production of biological active compounds**. *Planta* (2018.0) **248** 1-18. DOI: 10.1007/s00425-018-2910-1
10. Varadharaj V., Ramaswamy A., Sakthivel R., Subbaiya R., Barabadi H., Chandrasekaran M., Saravanan M.. **Antidiabetic and antioxidant activity of green synthesized starch nanoparticles: an in vitro study**. *J. Cluster Sci.* (2020.0) **31** 1257-1266. DOI: 10.1007/s10876-019-01732-3
11. Virmani I., Sasi C., Priyadarshini E., Kumar R., Sharma S.K., Singh G.P., Pachwarya R.B., Paulraj R., Barabadi H., Saravanan M., Meena R.. **Comparative anticancer potential of biologically and chemically synthesized gold nanoparticles**. *J. Cluster Sci.* (2020.0) **31** 867-876. DOI: 10.1007/s10876-019-01695-5
12. Mostafavi E., Zarepour A., Barabadi H., Zarrabi A., Truong L.B., Medina-Cruz D.. **Antineoplastic activity of biogenic silver and gold nanoparticles to combat leukemia: beginning a new era in cancer theragnostic**. *Biotechnol. Reports* (2022.0) **34**. DOI: 10.1016/j.btre.2022.e00714
13. Saravanan M., Vahidi H., Medina Cruz D., Vernet-Crua A., Mostafavi E., Stelmach R., Webster T.J., Ali Mahjoub M., Rashedi M., Barabadi H.. **Emerging antineoplastic biogenic gold nanomaterials for breast cancer therapeutics: a systematic review**. *Int. J. Nanomed.* (2020.0) **15** 3577-3595. DOI: 10.2147/IJN.S240293
14. Nongbet A., Mishra A.K., Mohanta Y.K., Mahanta S., Ray M.K., Khan M., Baek K.H., Chakrabartty I.. **Nanofertilizers: a smart and sustainable attribute to modern agriculture**. *Plants* (2022.0) **11** 2587. DOI: 10.3390/plants11192587
15. Narayanan M., Priya S., Natarajan D., Alahmadi T.A., Alharbi S.A., Krishnan R., Chi N.T.L., Pugazhendhi A.. **Phyto-fabrication of silver nanoparticle using leaf extracts of Aristolochia bracteolata Lam and their mosquito larvicidal potential**. *Process Biochem.* (2022.0) **121** 163-169. DOI: 10.1016/j.procbio.2022.06.022
16. Pandit C., Roy A., Ghotekar S., Khusro A., Islam M.N., Bin Emran T., Lam S.E., Khandaker M.U., Bradley D.A.. **Biological agents for synthesis of nanoparticles and their applications**. *J. King Saud Univ. Sci.* (2022.0) **34**. DOI: 10.1016/j.jksus.2022.101869
17. Sajid M., Khan M.R., Ijaz M.U., Ismail H., Bhatti M.Z., Shah S.A., Ali S., Tareen M.U., Alotaibi S.S., Albogami S.M., Batiha G.E.S.. **Evaluation of phytochemistry and pharmacological properties of alnus nitida**. *Molecules* (2022.0) **27**. DOI: 10.3390/molecules27144582
18. Majeed S., Saravanan M., Danish M., Zakariya N.A., Ibrahim M.N.M., Rizvi E.H., un NisaAndrabi S., Barabadi H., Mohanta Y.K., Mostafavi E.. **Bioengineering of green-synthesized TAT peptide-functionalized silver nanoparticles for apoptotic cell-death mediated therapy of breast adenocarcinoma**. *Talanta* (2023.0) **253**. DOI: 10.1016/j.talanta.2022.124026
19. Barabadi H., Mohammadzadeh A., Vahidi H., Rashedi M., Saravanan M., Talank N., Alizadeh A.. **Penicillium chrysogenum-derived silver nanoparticles: exploration of their antibacterial and biofilm inhibitory activity against the standard and pathogenic acinetobacter baumannii compared to tetracycline**. *J. Cluster Sci.* (2022.0) **33** 1929-1942. DOI: 10.1007/s10876-021-02121-5
20. Patil S., Chandrasekaran R.. **Biogenic nanoparticles: a comprehensive perspective in synthesis, characterization, application and its challenges**. *J. Genet. Eng. Biotechnol.* (2020.0) **18** 1-23. DOI: 10.1186/s43141-020-00081-3
21. Almatroudi A.. **Silver nanoparticles: synthesis, characterisation and biomedical applications**. *Open Life Sci.* (2020.0) **15** 819-839. DOI: 10.1515/biol-2020-0094
22. Barabadi H., Vahidi H., Mahjoub M.A., Kosar Z., Damavandi Kamali K., Ponmurugan K., Hosseini O., Rashedi M., Saravanan M.. **Emerging antineoplastic gold nanomaterials for cervical cancer therapeutics: a systematic review**. *J. Cluster Sci.* (2020.0) **31** 1173-1184. DOI: 10.1007/S10876-019-01733-2/FIGURES/3
23. Jini D., Sharmila S.. **Green synthesis of silver nanoparticles from Allium cepa and its in vitro antidiabetic activity**. *Mater. Today Proc.* (2020.0) **22** 432-438. DOI: 10.1016/J.MATPR.2019.07.672
24. Lan Chi N.T., Narayanan M., Chinnathambi A., Govindasamy C., Subramani B., Brindhadevi K., Pimpimon T., Pikulkaew S.. **Fabrication, characterization, anti-inflammatory, and anti-diabetic activity of silver nanoparticles synthesized from Azadirachta indica kernel aqueous extract**. *Environ. Res.* (2022.0) **208**. DOI: 10.1016/j.envres.2022.112684
25. Sheng Y., Narayanan M., Basha S., Elfasakhany A., Brindhadevi K., Xia C., Pugazhendhi A.. **In vitro and in vivo efficacy of green synthesized AgNPs against Gram negative and Gram positive bacterial pathogens**. *Process Biochem.* (2022.0) **112** 241-247. DOI: 10.1016/j.procbio.2021.12.012
26. Carmona F.J., Dal Sasso G., Ramírez-Rodríguez G.B., Pii Y., Delgado-López J.M., Guagliardi A., Masciocchi N.. **Urea-functionalized amorphous calcium phosphate nanofertilizers: optimizing the synthetic strategy towards environmental sustainability and manufacturing costs**. *Sci. Rep.* (2021.0) **11**. DOI: 10.1038/s41598-021-83048-9
27. Gaiotti F., Lucchetta M., Rodegher G., Lorenzoni D., Longo E., Boselli E., Cesco S., Belfiore N., Lovat L., Delgado-López J.M., Carmona F.J., Guagliardi A., Masciocchi N., Pii Y.. **Urea-doped calcium phosphate nanoparticles as sustainable nitrogen nanofertilizers for viticulture: implications on yield and quality of pinot gris grapevines**. *Agronomy* (2021.0) **11** 1026. DOI: 10.3390/agronomy11061026
28. Davoudpour Y., Schmidt M., Calabrese F., Richnow H.H., Musat N.. **High resolution microscopy to evaluate the efficiency of surface sterilization of Zea Mays seeds**. *PLoS One* (2020.0) **15**. DOI: 10.1371/journal.pone.0242247
29. Murashige T., Skoog F.. **A revised medium for rapid growth and bio assays with tobacco tissue cultures**. *Physiol. Plantarum* (1962.0) **15** 473-497. DOI: 10.1111/j.1399-3054.1962.tb08052.x
30. Khan A.U., Khan T., Khan M.A., Nadhman A., Aasim M., Khan N.Z., Ali W., Nazir N., Zahoor M.. **Iron-doped zinc oxide nanoparticles-triggered elicitation of important phenolic compounds in cell cultures of Fagonia indica**. *Plant Cell Tissue Organ Cult.* (2021.0) **147** 287-296. DOI: 10.1007/s11240-021-02123-1
31. Lichtenthaler H.K., Buschmann C.. **Chlorophylls and carotenoids: measurement and characterization by UV-VIS spectroscopy**. *Curr. Protoc. Food Anal. Chem.* (2001.0) **1** F4.3.1-F4.3.8. DOI: 10.1002/0471142913.faf0403s01
32. Sasidharan S., Chen Y., Saravanan D., Sundram K.M., Yoga Latha L.. **Extraction, isolation and characterization of bioactive compounds from plants' extracts**. *Afr. J. Tradit., Complementary Altern. Med.* (2011.0) **8** 1-10. DOI: 10.4314/ajtcam.v8i1.60483
33. Jemal K., Sandeep B.V., Pola S.. **Synthesis, characterization, and evaluation of the antibacterial activity of allophylus serratus leaf and leaf derived callus extracts mediated silver nanoparticles**. *J. Nanomater.* (2017.0) **2017**. DOI: 10.1155/2017/4213275
34. Eyáane Meva F., Segnou M.L., Ebongue C.O., Ntoumba A.A., Kedi P.B.E., Deli V., Etoh M.A., Mpondo E.M.. **Spectroscopic synthetic optimizations monitoring of silver nanoparticles formation from Megaphrynium macrostachyum leaf extract**. *Rev. Bras. Farmacogn.* (2016.0) **26** 640-646. DOI: 10.1016/j.bjp.2016.06.002
35. Rautela A., Rani J., Debnath (Das) M.. **Green synthesis of silver nanoparticles from Tectona grandis seeds extract: characterization and mechanism of antimicrobial action on different microorganisms**. *J. Anal. Sci. Technol.* (2019.0) **10** 1-10. DOI: 10.1186/s40543-018-0163-z
36. Kaviya S., Santhanalakshmi J., Viswanathan B., Muthumary J., Srinivasan K.. **Biosynthesis of silver nanoparticles using citrus sinensis peel extract and its antibacterial activity**. *Spectrochim. Acta Part A Mol. Biomol. Spectrosc.* (2011.0) **79** 594-598. DOI: 10.1016/j.saa.2011.03.040
37. Naveed M., Batool H., ur Rehman S., Javed A., Makhdoom S.I., Aziz T., Mohamed A.A., Sameeh M.Y., Alruways M.W., Dablool A.S., Almalki A.A., Alamri A.S., Alhomrani M.. **Characterization and evaluation of the antioxidant, antidiabetic, anti-inflammatory, and cytotoxic activities of silver nanoparticles synthesized using brachychiton populneus leaf extract**. *Processes* (2022.0) **10** 1521. DOI: 10.3390/pr10081521
38. Ibrahim H.M.M.. **Green synthesis and characterization of silver nanoparticles using banana peel extract and their antimicrobial activity against representative microorganisms**. *J. Radiat. Res. Appl. Sci.* (2015.0) **8** 265-275. DOI: 10.1016/j.jrras.2015.01.007
39. Roopan S.M., Rohit G., Madhumitha A.A., Rahuman C., Kamaraj A., Bharathi T.V.. **Low-cost and eco-friendly phyto-synthesis of silver nanoparticles using Cocos nucifera coir extract and its larvicidal activity**. *Ind. Crop. Prod.* (2013.0) **43** 631-635. DOI: 10.1016/j.indcrop.2012.08.013
40. Shaikhaldein H.O., Al-Qurainy F., Nadeem M., Khan S., Tarroum M., Salih A.M.. **Biosynthesis and characterization of silver nanoparticles using Ochradenus arabicus and their physiological effect on Maerua oblongifolia raised in vitro**. *Sci. Rep.* (2020.0) **10** 1-8. DOI: 10.1038/s41598-020-74675-9
41. Yang J., Jiang F., Ma C., Rui Y., Rui M., Adeel M., Cao W., Xing B.. **Alteration of crop yield and quality of wheat upon exposure to silver nanoparticles in a life cycle study**. *J. Agric. Food Chem.* (2018.0) **66** 2589-2597. DOI: 10.1021/ACS.JAFC.7B04904/ASSET/IMAGES/ACS.JAFC.7B04904.SOCIAL.JPEG_V03
42. Ali A., Mohammad S., Khan M.A., Raja N.I., Arif M., Kamil A., Mashwani Z. ur R.. **Silver nanoparticles elicited in vitro callus cultures for accumulation of biomass and secondary metabolites in Caralluma tuberculata, Artif. Cells**. *Nanomedicine, Biotechnol.* (2019.0) **47** 715-724. DOI: 10.1080/21691401.2019.1577884
43. Zia M., Yaqoob K., Mannan A., Nisa S., Raza G., ur Rehman R.. **Regeneration response of carnation cultivars in response of silver nanoparticles under in vitro conditions**. *Vegetos* (2020.0) **33** 11-20. DOI: 10.1007/s42535-019-00074-9
44. Vannini C., Domingo G., Onelli E., De Mattia F., Bruni I., Marsoni M., Bracale M.. **Phytotoxic and genotoxic effects of silver nanoparticles exposure on germinating wheat seedlings**. *J. Plant Physiol.* (2014.0) **171** 1142-1148. DOI: 10.1016/j.jplph.2014.05.002
45. Khan I., Raza M., Awan S., Khalid M.. (2019.0). DOI: 10.15666/aeer/1706_1287712892
46. Sabir S., Arshad M., Ilyas N., Naz F., Amjad M.S., Malik N.Z., Chaudhari S.K.. **Protective role of foliar application of green-synthesized silver nanoparticles against wheat stripe rust disease caused by Puccinia striiformis**. *Green Process. Synth.* (2022.0) **11** 29-43. DOI: 10.1515/gps-2022-0004
47. Mazumdar H., Ahmed G.U.. **Phytotoxicity effect of silver nanoparticles on oryza sativa**. *Int. J. ChemTech Res.* (2011.0) **3** 1494-1500
48. Mirzajani F., Askari H., Hamzelou S., Farzaneh M., Ghassempour A.. **Effect of silver nanoparticles on Oryza sativa L. And its rhizosphere bacteria**. *Ecotoxicol. Environ. Saf.* (2013.0) **88** 48-54. DOI: 10.1016/J.ECOENV.2012.10.018
49. Chung I.M., Rajakumar G., Thiruvengadam M.. **Effect of silver nanoparticles on phenolic compounds production and biological activities in hairy root cultures of Cucumis anguria**. *Acta Biol. Hung.* (2018.0) **69** 97-109. DOI: 10.1556/018.68.2018.1.8
50. 50S. Hashemi, Z. Asrar, … S.P.-A.F.C., U. 2016, Effect of biologically synthesized silver nanoparticles on growth characteristics and flavonoid contents of Wheat (Triticum aestivum), Aj.Areeo.Ac.Ir. (n.d.). https://aj.areeo.ac.ir/article_109591_en.html (accessed February 18, 2023).
51. Hasan M., Mehmood K., Mustafa G., Zafar A., Tariq T., Hassan S.G., Loomba S., Zia M., Mazher A., Mahmood N., Shu X.. **Phytotoxic evaluation of phytosynthesized silver nanoparticles on lettuce**. *Coatings* (2021.0) **11** 1-15. DOI: 10.3390/coatings11020225
52. Tripathi D.K., Shweta S., Singh S., Singh R., Pandey V.P., Singh N.C., Sharma S.M., Prasad N.K., Dubey D.K.. **An overview on manufactured nanoparticles in plants: uptake, translocation, accumulation and phytotoxicity**. *Plant Physiol. Biochem.* (2017.0) **110** 2-12. DOI: 10.1016/j.plaphy.2016.07.030
53. Ma C., White J.C., Dhankher O.P., Xing B.. **Metal-based nanotoxicity and detoxification pathways in higher plants**. *Environ. Sci. Technol.* (2015.0) **49** 7109-7122. DOI: 10.1021/acs.est.5b00685
54. Yuan L., Richardson C.J., Ho M., Willis C.W., Colman B.P., Wiesner M.R.. **Stress responses of aquatic plants to silver nanoparticles**. *Environ. Sci. Technol.* (2018.0) **52** 2558-2565. DOI: 10.1021/acs.est.7b05837
55. Zahir A., Nadeem M., Ahmad W., Giglioli-Guivarćh N., Hano C., Abbasi B.H.. **Chemogenic silver nanoparticles enhance lignans and neolignans in cell suspension cultures of Linum usitatissimum L**. *Plant Cell Tissue Organ Cult.* (2019.0) **136** 589-596. DOI: 10.1007/s11240-018-01539-6
56. Castro-González C.G., Sánchez-Segura L., Gómez-Merino F.C., Bello-Bello J.J.. **Exposure of stevia (Stevia rebaudiana B.) to silver nanoparticles in vitro: transport and accumulation**. *Sci. Rep.* (2019.0) **9**. DOI: 10.1038/s41598-019-46828-y
57. Bagherzadeh Homaee M., Ehsanpour A.A.. **Physiological and biochemical responses of potato (Solanum tuberosum) to silver nanoparticles and silver nitrate treatments under in vitro conditions**. *Indian J. Plant Physiol.* (2015.0) **20** 353-359. DOI: 10.1007/s40502-015-0188-x
58. Verma D.K., Patel S., Kushwah K.S.. **Green biosynthesis of silver nanoparticles and impact on growth, chlorophyll, yield and phytotoxicity of Phaseolus vulgaris L**. *Vegetos* (2020.0) **33** 648-657. DOI: 10.1007/s42535-020-00150-5
59. Abbasi F., Jamei R.. **Effects of silver nanoparticles and silver nitrate on antioxidant responses in Echium amoenum**. *Russ. J. Plant Physiol.* (2019.0) **66** 488-494. DOI: 10.1134/S1021443719030026
60. Azeez L., Lateef A., Adebisi S.A.. **Silver nanoparticles (AgNPs) biosynthesized using pod extract of Cola nitida enhances antioxidant activity and phytochemical composition of Amaranthus caudatus Linn**. *Appl. Nanosci.* (2017.0) **7** 59-66. DOI: 10.1007/s13204-017-0546-2
61. Ahmed Z.R., Uddin Z., Shah S.W.A., Zahoor M., Alotaibi A., Shoaib M., Ghias M., Bari W.U.. **Antioxidant, antidiabetic, and anticholinesterase potential of Chenopodium murale L. extracts using in vitro and in vivo approaches**. *Open Chem* (2022.0) **20** 1171-1186. DOI: 10.1515/CHEM-2022-0232/HTML
62. Al-Joufi F.A., Jan M., Zahoor M., Nazir N., Naz S., Talha M., Sadiq A., Nawaz A., Khan F.A.. **Anabasis articulata (forssk.) moq: a good source of phytochemicals with antibacterial, antioxidant, and antidiabetic potential**. *Molecules* (2022.0) **27** 3526. DOI: 10.3390/MOLECULES27113526/S1
63. Ali A., Nisar M., Shah S.W.A., Khalil A.A.K., Zahoor M., Nazir N., Shah S.A., Nasr F.A., Noman O.M., Mothana R.A., Ahmad S., Farooqi H.M.U.. **Anatomical characterization, HPLC analysis, and biological activities of Ilex dipyrena**. *Plants* (2022.0) **11** 617. DOI: 10.3390/PLANTS11050617
64. Brandelli A.. **The interaction of nanostructured antimicrobials with biological systems: cellular uptake, trafficking and potential toxicity**. *Food Sci. Hum. Wellness* (2020.0) **9** 8-20. DOI: 10.1016/j.fshw.2019.12.003
65. Talank N., Morad H., Barabadi H., Mojab F., Amidi S., Kobarfard F., Mahjoub M.A., Jounaki K., Mohammadi N., Salehi G., Ashrafizadeh M., Mostafavi E.. **Bioengineering of green-synthesized silver nanoparticles: in vitro physicochemical, antibacterial, biofilm inhibitory, anticoagulant, and antioxidant performance**. *Talanta* (2022.0) **243**. DOI: 10.1016/j.talanta.2022.123374
66. Ali A., Nisar M., Shah S.W.A., Khalil A.A.K., Zahoor M., Nazir N., Shah S.A., Nasr F.A., Noman O.M., Mothana R.A., Ahmad S., Farooqi H.M.U.. **Anatomical characterization, HPLC analysis, and biological activities of Ilex dipyrena**. *Plants* (2022.0) **11** 617. DOI: 10.3390/plants11050617
67. Khan J., Majid A., Nazir N., Nisar M., Khan Khalil A.A., Zahoor M., Ihsan M., Ullah R., Bari A., Shah A.B.. **Hplc characterization of phytochemicals and antioxidant potential of alnus nitida (Spach) endl**. *Horticulturae* (2021.0) **7**. DOI: 10.3390/horticulturae7080232
|
---
title: Nutrigenomics in the context of evolution
authors:
- Carsten Carlberg
journal: Redox Biology
year: 2023
pmcid: PMC10036735
doi: 10.1016/j.redox.2023.102656
license: CC BY 4.0
---
# Nutrigenomics in the context of evolution
## Abstract
Nutrigenomics describes the interaction between nutrients and our genome. Since the origin of our species most of these nutrient-gene communication pathways have not changed. However, our genome experienced over the past 50,000 years a number of evolutionary pressures, which are based on the migration to new environments concerning geography and climate, the transition from hunter-gatherers to farmers including the zoonotic transfer of many pathogenic microbes and the rather recent change of societies to a preferentially sedentary lifestyle and the dominance of Western diet. Human populations responded to these challenges not only by specific anthropometric adaptations, such as skin color and body stature, but also through diversity in dietary intake and different resistance to complex diseases like the metabolic syndrome, cancer and immune disorders. *The* genetic basis of this adaptation process has been investigated by whole genome genotyping and sequencing including that of DNA extracted from ancient bones. In addition to genomic changes, also the programming of epigenomes in pre- and postnatal phases of life has an important contribution to the response to environmental changes. Thus, insight into the variation of our (epi)genome in the context of our individual's risk for developing complex diseases, helps to understand the evolutionary basis how and why we become ill. This review will discuss the relation of diet, modern environment and our (epi)genome including aspects of redox biology. This has numerous implications for the interpretation of the risks for disease and their prevention.
## Introduction
Nutrition is an essential component of life, since it is composed of molecules that satisfy our body with its requirements of macro- and micronutrients [1]. In addition, these molecules affect our health, since some of them directly communicate with our genome and epigenome by regulating the activity of transcription factors and chromatin modifiers [2]. The complex relationship between nutrition and our (epi)genome is the core of nutrigenomics [3,4]. The daily diet-(epi)genome communication modulates the expression of genes in metabolic organs, such as in adipose tissue, skeletal muscle, liver and pancreas, as well as in the brain and the immune system. The cellular and molecular biology behind these gene regulatory processes maintain homeostasis of our body that prevent the onset of non-communicable diseases, such as obesity, type 2 diabetes (T2D), cardiovascular diseases (CVDs) and cancer.
In a given human population most anthropomorphic properties like height or eye color [5] as well as many physiological characteristics, such as lactase persistence [6] and the risk for developing diseases like T2D [7], exist in many forms. Since these traits are based on the expression and function of genes, the diversity is related to interindividual genomic and epigenomic variations. Accordingly, the members of a population display different levels of biological fitness like mating success, viability and fertility. Traits associated with increased fitness represent adaptations to the environment. This is often caused by evolutionary pressures, such as reduced availability of resources like food or threats like pathogens, and represents the basis of the evolutionary principle of “survival of the fittest”, as first formulated by Darwin [8].
In contrast, modern societies are characterized by intensive medical and social care for the individual. Taking care of the of less well-adapted members of a society is a hallmark of humanity and a significant advance of our species. Therefore, today's consequences of the principles of Darwinism in most cases do not apply anymore to us. However, infections with HIV-1 (human immunodeficiency virus 1) or the pandemic of SARS-CoV-2 (severe acute respiratory syndrome coronavirus 2) may represent exceptions. In any case, each of us is carrying in his/her individual version of the genome a history of the evolutionary past, which largely determines our personal susceptibility for various common diseases. This implies that the evolution of our species needs to be taken into consideration, in order to obtain the full potential of personalized medicine/nutrition [9].
Evolution is often based on positive natural selection, i.e., on a process where an advantageous trait increases its prevalence in a population [10]. For example, individuals with a genetic advantage, such as a single nucleotide variant (SNV) that allows to express the LCT (lactase) gene in adult age can digest lactose and have with milk an important additional source of nutrition [11] (see section below). In the past, this implied that these individuals were more successful in reproduction, since they had a higher number of surviving children, which then also carry the same advantageous SNV. When the selection pressure persists, this increases over a number of generations the prevalence of the respective property until it becomes a major trait. Another prominent example of positive selection is the gene encoding for the CCR5 (C–C chemokine receptor 5) receptor, which is specifically expressed on T cells and essential for the entry of HIV-1. A rather common structural variant of the CCR5 gene is a 32 base pair (bp) deletion, which significantly decreases the functionality of the protein and protects its carriers from HIV-1 infection [12,13]. However, the variant originated from Northern Europe more than 1000 years ago, i.e., far earlier than HIV-1 circulated in humans. Therefore, more likely other infectious diseases, such as smallpox and/or the plague, created a selective pressure for the CCR5 gene variant. In total, some 2000 human genes, i.e., $10\%$ of all protein-coding genes, may have undergone positive natural selection during the past 50,000 years [14]. This affected in particular genes related to the skin, the digestive tract and the immune system, since these organs are in a more intensive contact with the environment than others [15].
In addition to the process of positive selection, there are also randomly occurring mutations and recombination events of the genome, which can cause genetic drifts [16,17]. In humans, this so-called neutral evolution leads to a change of approximately 50 bp per individual and generation [18]. Although the respective genetic drifts are not under evolutionary pressure, they can reach high frequency in a population. Interestingly, in case of small populations or a weak selection pressure even deleterious alleles are able to reach high frequencies [19].
The statement “Nothing in biology makes sense except in the light of evolution” [20] implies that evolution, at least in the past, was also for humans a dominant driver in the development of any kind of biological process and its adaptation to environmental changes. Therefore, the evolutionary history of our species needs to be considered also in the field of nutrigenomics. Accordingly, the focus of this review is to provide an overview on the relation of nutrigenomics and evolution.
## The human genome and its variation
Thousands of complex phenotypic traits determine our physical appearance and what is our risk to develop non-communicable diseases. In addition, each of these traits relates to dozens to hundreds of variants in the genome as well as to environmental influences affecting our epigenomes [21] (see section below). *In* general, the genetic architecture of a trait depends on [22]:•the number of variants that influence the respective phenotype•the relative magnitude of the effects of the different variants on the trait•the frequency of the traits in the population•the interference of the traits with each other and the environment.
Importantly, most of the variants, which are associated with traits, are not found at protein-coding regions but rather affect regulatory genomic regions, e.g., transcription factor binding sites [23]. The latter sites are often referred to as regulatory SNVs, master examples of which will be discussed in the context of lactase persistence.
The Human Genome Project (www.genome.gov/10001772) resulted in 2001 in the release of the first reference haploid sequence of our genome [24,25]. However, it took more than 20 years until all gaps in the sequence could be filled and the telomer-to-telomer (T2T) final version of our genome was published [26]. It comprises 2 × 3.05 billion bp (Gb) on 46 chromosomes (2 × 22 autosomes and XY for males or XX for females). The haploid genome contains 63,494 genes, of which only 19,969 are protein-coding, i.e., the majority of human genes encode for non-coding RNAs with structural and regulatory functions. The human reference genome had been assembled based on individuals from European ancestry, but there are initiatives to include genomes from people with African and Asian ancestry [27,28], in order to have a better representation of all major populations.
Each of us is carrying a number of variants in comparison to the reference genome. They represent in total some $1\%$ of the whole sequence. These are 4–5 million SNVs, where exactly one nucleotide is altered, and some 600,000 structural variants that often affect more than one nucleotide [29]. Most of the latter are insertion-deletions (Indels), where 1–49 bp are added or removed, respectively. In addition, per individual there are some 1000 genomic regions carrying copy number variations (CNVs), where DNA stretches from 50 bp up to 15 million bp (Mb) in length are either inserted or deleted.
Variants to the human genome, which have a minor allele frequency (MAF) of at least $1\%$, are called common, when their MAF is below $1\%$ [30] (www.ncbi.nlm.nih.gov/SNP). Approximately 7 million human SNVs are single nucleotide polymorphisms (SNPs), because they have a MAF higher than $5\%$. The 1000 Genomes Project (www.internationalgenome.org) and other larger whole genome sequencing projects, such as the TOPMed program [31] and Genomics England [32], have provided more than 100,000 whole genome sequences, on the basis of which nearly 500 million SNVs were identified [33]. Concerning SNVs the average difference between two unrelated individuals is $0.1\%$, which in comparison to other species is a very low value and related to the recent origin of non-African homo sapiens populations from a small founding group [34,35] (see section below).
Within the coding sequence the effect of a SNV can either be synonymous, when it does not alter the encoded protein, or non-synonymous, when it causes a change in a single amino acid (missense) or introduces a premature stop codon (nonsense). Each individual's genome contains some 150 SNVs that cause protein truncation and 10,000 SNVs leading to changes in amino acids, while 500,000 SNVs modulate the binding capacity of transcription factor binding sites [33]. Within exons Indels and CNVs can cause frameshift mutations, i.e., a complete change in the protein sequence, and within introns CNVs may affect the splicing process. Importantly, each of us carries 50–100 heterozygous variants in our genome that in a homozygous setting can cause monogenetic disorders.
In the last 20 years the impact of variations of the human genome on the risk for diseases has been primarily investigated by genome-wide association studies (GWASs) using arrays of thousands to millions of SNPs [36]. This method is now replaced by whole exome or whole genome sequencing [37]. The studies aim for a statistically significant association between a genetic variant and the occurrence of a disease. The average SNV density of 1 in 1000 nucleotides suggests that these studies require testing of millions of genetic variants per individual as well as samples from thousands of individuals. GWASs with several thousand individuals could identify odds ratios (ORs) of 1.5, i.e., a $50\%$ increased risk for the tested disease. Larger sample sizes are often achieved by pooling several GWASs through meta-analyses. For example, with more than 50,000 individuals one can determine increased risks by $10\%$, i.e., ORs of 1.1. Disease- and trait-associated genomic loci can be found in the database GWAS Catalog (www.ebi.ac.uk/gwas).
Mendelian disorders, such as cystic fibrosis or Huntington's disease, are monogenetic, i.e., in these cases a single homozygous SNV with a strong effect can explain the occurrence of the rare disease [38] (Fig. 1, left). In contrast, common diseases like T2D, CVDs or cancer have a multigenic basis, i.e., they are based on a multitude of SNVs, each of which having only a minor impact [39] (Fig. 1, right). For example, common traits like body height are determined by at least 180 genomic regions [40]. Interestingly, improved quality and quantity of nutrition had a major environmental influence, which in Europe led within the last few generations to 10 additional centimeters in average height [41].Fig. 1Risk allele frequency of genetic variants. ORs indicate the strength of a genetic effect. Main emphasis on the identification of associations within the diagonal box. Whole genome sequencing of large numbers of individuals identifies far more low frequency SNVs with intermediate ORs (center).Fig. 1 In summary, the variations of our genome represent the genetic component of our risk for diseases as well as for anthropometric traits.
## The impact of epigenetics
Despite the successes in revealing the association of numerous variants of the human genome with complex traits, only some $10\%$ of the heritability of most complex diseases can be explained by genetics [42]. This suggests that based on SNV analysis one cannot reliably estimate the risk of an individual for a particular disease. The only well-known exceptions are age-related macular degeneration [43] and type 1 diabetes [44]. In contrast, the heritability of only $20\%$ of coronary heart disease cases is explained by more than 60 genetic loci [45], $20\%$ of T2D risk by some 300 loci [46] or $20\%$ of breast cancer risk by some 170 loci [47]. It is possible that rare variants with high ORs may explain some of the missing heritability [48], but environmental exposures have the main impact. Many of latter are related to nutritional molecules or their metabolites that affect the epigenome [49].
The physical expression of the epigenome is chromatin, which is a complex of nucleosome-forming histone proteins and genomic DNA [50]. Epigenetics is defined as functionally relevant changes of chromatin that do not affect the sequence of the genome [51]. For example, during embryogenesis, when totipotent embryonic stem cells differentiate via various progenitor cells into all 400 tissues and cell types forming our body [52], there are cascades of changes in DNA and histone methylations, in the context of which the cells are epigenetically programmed. Comparable epigenetic programming takes place when adult stem cells in bone marrow, colon and skin, give rise to new blood cells, enterocytes and keratinocytes, respectively.
Epigenetic changes are mediated by chromatin modifying enzymes. For example, histone acetyltransferases (HATs) add acetyl groups to lysines of histone proteins, while histone deacetylases (HDACs) remove them. Comparably, lysine methyltransferases (KMTs) provide histones with methyl groups and lysines demethylases (KDMs) erase them. DNA can be methylated at cytosines through the action of DNA methyltransferases (DNMTs) and TET (ten-eleven translocation) enzymes, which initiate the removal of the methyl groups. The relative activity of these enzymes determines the level of accessibility of genomic regions, i.e., whether transcription factors can bind to enhancer regions and transcription start sites (TSSs) are accessible to RNA polymerase II (Pol II). Interestingly, many chromatin modifying enzymes use intermediate metabolites of energy metabolism as cofactors, such NAD+ (nicotinamide adenine dinucleotide) for some HDACs, acetyl-CoA by HATs and α-ketoglutarate by KDMs, i.e. the activity of chromatin modifiers largely depends on the redox state [53] and the metabolic status of the cell [54]. Interestingly, the antioxidant vitamin C is a cofactor to dioxygenases, such as some members of the KDM family and TET enzymes [55]. This makes critical steps in epigenetic regulation, such as the demethylation of histones and genomic DNA, dependent on a compound that humans and many other mammals have to take up from diet, because they are incapable to synthesize it themselves.
Some patterns of DNA methylation or histone modifications last for days, months or even years [56]. In this way, the epigenome is able to preserve effects of cellular perturbations, such as the supply with nutrients, as epigenetic drifts, which represent a type of memory [57,58]. When somatic cells divide, these epigenetic drifts may be inherited to daughter cells and via germ cells the epigenetic memory can be transferred even to the next generation [59].
Taken together, many epigenetic variants are the result of responses to environmental changes, such as the supply with nutrients. In this way, the epigenome is able the react faster to evolutionary pressures than the genome and may explain major parts of the missing heritability.
## Migration of homo sapiens and the diversity of human populations
At their origin in Africa, hominins lost some 2 million years ago most of their body hair [60], in order to improve via more efficient sweating their endurance performance [61]. Their initial pale skin developed an intensive pigmentation, in order to protect from sunburn and UV-induced cancer [62]. The pigmentation intensity of skin, eyes and hair can be explained by SNVs in genes encoding for key proteins of melanin synthesis in melanosomes [63]. For example, the pale skin of European populations is explained primarily by SNVs related to the genes OCA2 (OCA2 melanosomal transmembrane protein), SLC45A2 (solute carrier family 45 member A2) and SLC24A5 [64,65]. The SNVs cause a loss of function in the encoded pH regulator, ion transporter and potassium-dependent sodium/calcium exchanger, respectively, and lead to a reduced production of the black/brown eumelanin in melanosomes [63,65]. A reduced skin pigmentation may increase the efficiency of UV-B-induced vitamin D3 production, which is important for populations living at higher latitudes [66]. Vitamin D is, in contrast to vitamins C and E, not a classical antioxidant, i.e., it is not a scavenger for reactive oxygen species (ROS). However, UV-B absorption by the vitamin D3 precursor 7-dehydrocholesterol shields cholesterol-producing animals and plants against radiation damage. Therefore, even simple eukaryotes like phytoplankton produce vitamin D3 although they do not use it for endocrine function purposes [67].
Anatomically modern humans (Homo sapiens) evolved in East Africa some 250,000 years ago [68,69] and spread first over the whole African continent (Fig. 2A). About 75,000 years ago a subgroup of them migrated first to Asia and then further to Oceania, Europe and the Americas [65,70]. The number of modern humans who left Africa at that time and got the ancestors of all homo sapiens at the other continents, was far smaller than those, who stayed in Africa. This created a bottleneck and reduced the genetic diversity of the non-African populations [35]. In addition, after migration a number of human populations became isolated due to geographic, language and political barriers, i.e., over time human genetic variation diverted geographically. Fig. 2Homo sapiens migrations. Anatomically modern humans developed in East Africa and spread first over the whole continent before they started some 75,000 years ago to migrate to Asia. From there they settled in Oceania, Europe and the Americas. Fig. 2 The persistent arrival of homo sapiens in Europe was about 42,000 years ago [71,72]. Both in Europe and in Asia modern humans met ancestral Neanderthal and Denisovan hominins, which they outnumbered by interbreeding [[73], [74], [75]]. Due to this so-called introgression process, 1.5–$2.1\%$ of the genome of modern humans in Europe have Neanderthal origin [76]. In contrast, present Southeast Asians have in average only $0.1\%$ of their genome from Denisovans, while the rate is $3.5\%$ in some Oceanian populations [77]. Many of the genes that we inherited from Neanderthals affect physiological systems all over our body and raise the risk of a number of diseases. However, some *Neanderthal* gene variations also show beneficial effects, e.g., those boosting the immune system. In Europe, hunter-gatherers lived first in the ice-free southwest of the continent [78] and 11–12,000 years ago, after the end of the ice age, some of them followed the migration of animal herds to northern Europe [71]. Archeogenomic data from hundreds of individuals living between 8500 and 2300 years ago [79] suggest that some 8400 to 6000 years ago a wave of people from northwestern Anatolia arrived in southern Europe (Fig. 2B). These so-called Anatolian farmers introduced to the hunter-gatherers the concept of agriculture and started in this way the Neolithic revolution. The latter is characterized by giving up the nomadic lifestyle and the domestication of a number of plant and animal species. Some 5000 years ago, a second wave of migrants, the Yamnaya pastoralists, arrived in Europe. They originated from the Eurasian steppe, introduced the wheel, the horse and their Indo-European languages to the European populations and settled preferentially in northern Europe. Both groups of migrants had lighter skin than the hunter-gatherers [[80], [81], [82]], i.e., the light skin color of today's Europeans became frequent only within the past 5000 years [65,83,84]. This phenotypic adaptation is primarily based on non-synonymous SNVs of the genes SLC24A5 and SLC45A2 [65].
The individual admixtures of the genomic variants originating from hunter-gatherers, Yamnaya pastoralists and Anatolian farmers explain the variation in skin color as well as many other traits of the European population [81,[85], [86], [87]]. This includes also the individual's genetic risk for common diseases [88,89]. The neolithic revolution caused a rapid increase in population size in Europe [90,91], such as the use of milk products as food source for adults and the rise of agriculture [92] (see section below). Some 500 years ago major voluntary and involuntary (slaves) migration started, in particular between Europe, Africa and the Americas. This led to significant population admixtures, in particular in the Americas, but also in other parts of the world.
In summary, after their worldwide expansion the population size of modern humans was growing exponentially [93]. This growth modified the genetic architecture of traits and generated many low-frequency variants of the human genome [94]. Moreover, this caused substantial differences in allele frequency between populations, some of which are relevant to disease risk [95].
## Evolution of human nutrition
Until some 10,000 years ago all members of our species were hunters and gatherers, i.e., they were eating wild animals and plants with an estimated $20\%$/$80\%$ ratio [96] (Table 1). This primarily plant-based diet had rather low energy density, had medium fiber content and was low in fat. Moreover, the paleolithic diet did not contain any sugar and was low in salt but had a high micronutrient density and antioxidant capacity. Since homo sapiens had some 240,000 years, i.e., about 10,000 generations, time to adapt to this type of food, it can be considered as the reference, to which the biochemistry of our body has accommodated. Table 1Evolution of human nutrition. Human diet changed in the shift from hunter and gatherers to farmers. The next change was introduced by the industrial revolution, but the most drastic change in diet was in modern times. Table 1Time periodDietNutritional characteristicsPaleolithic era (more than 10,000 years ago)Wild animals and plantsLow energy density, medium fiber content, no sugar, low glycemic load, low fat and low saltAgricultural revolution (starting 10,000 years ago)Domesticated animals and plants. Use of fermented foods and beveragesMedium energy density, high fiber content, low sugar, medium glycemic load, medium fat and high saltIndustrial revolution (starting 250 years ago)Reliance of refined grains and oils, fatty meat, alcoholic beveragesHigh energy density, medium fiber content, medium sugar, high glycemic load, high fat, and high saltModern era (starting 50 years ago)Industrially produced foods, fatty meat, alcoholic beverages. Consumption of fast foodVery high energy density, low fiber content, high sugar, very high fat, very high glycemic load and high salt Giving up the hunter and gatherer habit and becoming farmers started in different regions of the world as early as 10,000 years ago. This neolithic revolution is characterized by the use of domesticated plants and animals [97]. It shifted the dietary pattern by introducing sugar and using salt for conservation. The more energy-rich diet led to a significant increase in population density. However, the side effect of living in close contact with many other individuals as well as with a number of domesticated animal species was an increased burden of infectious diseases, many of which derive from zoonotic transition [98] (see section below). Thus, the change in diet as well as infectious diseases created evolutionary pressures that pushed the rather rapid adaption of a few key genes [99] (see section below).
The industrial revolution, which started some 250 years ago, resulted in a large number of machines and transport vehicles, so that individuals had to invest less and less physical activity into daily work. In parallel, refined foods, such as oils and grains, were used, so that the fiber content of diet decreased and the sugar and fat load increased. Since some 50 years, a dietary pattern that is characterized by high intakes of pre-packaged foods, refined grains, red meat, processed meat, high-sugar drinks, sweets, fried foods, butter and other high-fat dairy products is referred to as Western diet [100]. It was distributed by US American fast-food and supermarket chains to Europe and has arrived in nearly all countries and human populations. Special impact had high-fructose corn syrup, which is used as a replacement of sucrose as a sweetener [101]. Thus, in modern times the average fiber content of our diet further decreased, while the sugar and fat load increased. Today's diet has a glycemic load higher than ever in our history, while in parallel due to technical revolutions in transportation and computerization physical activity further decreased. Therefore, an increasing proportion of the worldwide population receives a positive energy balance, which is the main reason for the worldwide epidemic of overweight and obesity [102]. Moreover, the raise in life expectancy in all countries of the world increases the percentage of the population with a too high body mass index. In high-income countries this transition, the so-called “energy flipping point”, occurred already more than 50 years ago, but todays it applies to nearly every country on this planet [4].
Hunger and satiety are feelings that are coordinated by numerous hormones signaling to the brain [4]. Satiation hormones control the amount of food intake, while obesity hormones modify these signals [103]. These regulatory circles are modulated by cultural habits, stress and social influences. Taking up less calories than consumed by daily physical activity and the basal metabolic rate, i.e., a negative energy balance, could solve the problem of overweight and prevent obesity. However, neuronal and hormonal control circuits have been evolutionary adapted to make hunger a prime instinct of humans and other animals. This is the main driver of feeding behavior and counteracts strongly with most attempts of reducing body weight [4]. In addition, nutrition-triggered epigenetic programming, which happens in pre- and postnatal phases, can result in an epigenetic predisposition for overweight and obesity [104].
Taken together, our nutritional habits change in the period of only two generations more drastically than in any other time in the history of our species. This time span is too short for expecting any genetic adaptations.
## Genetic adaptation to dietary changes
The evolution of our species is largely driven by change in our nutrition and environment and allowed us to progress and survive. For example, we are the only species who invented, already some 780,000 years ago, the use of fire for cooking [105]. This resulted in a less microbe-burden diet that in addition was easier to digest. In this way, cooking increased the energy yield from diet and triggered the enlargement of our brain, which largely depends on glucose. In addition, we developed receptors for sweet taste [106], which allows us detecting most energy-rich food sources. Nowadays, this survival instinct unfortunately contributes to overweight and obesity [4].
Human diet is majorly composed of starch from different types and forms of grains, potatoes or other root vegetables. Starch is a plant polysaccharide that can be digested to glucose by the enzyme amylase (AMY). The AMY gene family is composed by AMY1, which is expressed and secreted by saliva producing cells, as well as by AMY2A and AMY2B that are expressed in the pancreas [107]. The AMY1 gene copy number increased in populations, such as in Japan, where starch rich diets like rice are favored, while other populations, such as with Siberian Yakut, which primarily eat fish and meat, the copy number stayed low [108]. Today's humans have in average some 16 copies of the AMY1 gene [107], which via higher AMY protein secretion into the saliva improves the digestion of starchy foods and increases the sweet sensation during eating. The AMY1 gene amplification is a master example of positive evolutionary selection in response to an dietary trigger. Interestingly, after the domestication of the wolve some 30–40,000 years ago, the AMY1 gene copy number in dogs also significantly increased, in order to better digest remainders from human food [109].
The alcohol dehydrogenase (ADH) gene cluster is another example positive evolutionary selection in humans. With the invention of agriculture the production and consumption of fermented alcoholic beverages became popular. When humans started to consume alcohol at larger quantities, there was evolutionary pressure for more genes encoding for ethanol metabolizing enzymes. Interestingly, today's populations differ significantly in their sensitivity to alcoholic drinks. However, there is evidence that the adaptation of humans to alcohol consumption already started far earlier than in neolithic times [110]. In addition, alcohol can be oxidized also in microsome by the enzyme CYP2E1 (cytochrome P450 family 2 subfamily E member 1). The actions of both ADHs and CYP2E1 affect the redox status of cells via the NAD/NADH ratio and the generation of ROS, i.e., the evolutionary adaptation to use with alcohol an additional energy-rich food source came with the disadvantage of additional oxidative stress [111].
The introduction of agriculture as well as the migration to new geographic environments positively selected for the genes ADAMTS19 (ADAM metallopeptidase with thrombospondin type 1 motif 19), ADAMTS20, APEH (N-acylaminoacyl-peptide hydrolase), UBR1 (ubiquitin protein ligase E3 component n-recognin 1) and PLAU (plasminogen activator, urokinase) that encode for enzymes of protein metabolism [112]. Furthermore, there are examples of nutrition-related gene variants that are specific for one or several populations, such as MAN2A1 (mannosidase α, class 2A, member 1) in East Asia and West Africa, NCOA1 (nuclear receptor coactivator 1) in West Africa, SLC25A20 in East Asia, SI (sucrase-isomaltase) in East Asia), LEPR (leptin receptor) in East Asia as well as the fatty acid handling SLC27A4 and PPARD (peroxisome proliferator-activated receptor delta) in Europe [112]. Thus, for the best use of local resources human populations genetically adapted to their traditional diet.
The master example of a diet-driven genetic adaptation causes lactase persistence, i.e., the ability to digest milk sugar (lactose) throughout the life and not only as a young breast-fed child [6,11,113]. Lactase persistence is very common in Europe, while it is basically absent in South-East Asia. Lactose is the main carbohydrate energy source for infant mammals. The intestinal enzyme LCT digests the disaccharide into galactose and glucose. The default condition in older children and adults is a significantly reduced expression of the LCT gene after weaning, in order to avoid competition with newborns for breast milk. When these lactase non-persistent individuals consume lactose, they often get intestinal symptoms, such as flatulence, bloating, cramps, diarrhea and nausea, as a consequence of which they avoid drinking milk.
Lactase persistence is based on several regulatory SNVs within introns 9 and 13 of the gene MCM6 (minichromosome maintenance complex component 6), which are part of transcription factor binding sites within enhancer regions 22 and 14 kilo bp (kb) upstream of the TSS of the LCT gene (Fig. 3). Milk is a perfect source of carbohydrates, fat and calcium. When in the past after weaning the input from protective antibodies from breast milk was missing, the mortality of children due to infectious diseases was very high [114]. In contrast, when these children were lactase persistent, they could use milk from domesticated animals as reliable dietary source and had a significant survival advantage [115]. Therefore, regulatory SNVs leading to lactase persistence were under strong positive evolutionary selection. Archeogenomic data indicated that the variant occurred first some 5000 years ago and rapidly rose in frequency in the European population [113,116].Fig. 3Molecular basis of lactase persistence. The genomic region of the genes LCT and MCM6 is shown (A). SNPs located approximately 14 and 22 kb upstream of the TSS of the LCT gene, which are located within introns 13 and 9 of the MCM6 gene, respectively, are associated with lactase persistence. The function of regulatory SNPs is schematically depicted (B). The SNP is part of a transcription factor binding site and provides in one allele (top) high affinity and in the other allele (bottom) low affinity for the transcription factors. In case of rs4988235 at position −13,910 relative to the LCT gene this is POU2F1 (POU class 2 homeobox 1). Moreover, epigenetic effects, such as histone acetylation and methylation as well as DNA methylation can affect the expression of the LCT gene and mediate lactase persistence. Fig. 3 In summary, in addition to prominent examples like the genes LCT, AMY and ADH there are variants of some 100 genomic regions that monitor positive selection [112], in response to diet-driven evolutionary pressure.
## Evolution of human immunity
Together with nutrient deprivation, pathogen infection is the most challenging events for human survival. During the neolithic revolution humans significantly changed their habitat by favoring vector insects, such as mosquitoes, and living together with domesticated animals, such as pigs and chicken, which are reservoirs of pathogenic microbes like bacteria, viruses and parasites. Thus, when comparing with hunters and gatherers societies the burden of infectious diseases of farmers drastically increased. These pathogens represented a strong challenge for the immune system and served as selective pressures in human evolution of the past thousands of years [117]. For example, the pandemics of the “black death”, which were caused by the bacterium Yersinia pestis, killed in the 14th century some $50\%$ of the European population and led a substantial selection in the variations of immune-related genes of the survivors [118].
In order to assure the survival of our species, a sufficiently large number of individuals have to reach their reproductive and child-caring age. Therefore, evolution shaped our immune system in a way that it responds efficiently to acute infections in young people [119]. For example, variants of genes encoding for membrane immune receptors in innate and adaptive immunity were positively selected [120]. This affects not only the fight against microbes but also the control of wound healing, tissue repair, the elimination of dead and cancer cells as well as the formation of a healthy gut microbiome. There is no mechanism of evolutionary pressure beyond the age of reproduction. Therefore, it is likely that genetic traits, which had been selected to ensure fitness in early-life, may lead at older age to immunophenotypes with a high rate of chronic inflammation.
The infection with the intracellular bacterium *Mycobacterium tuberculosis* is a good example for a host-pathogen co-evolution. The first cases of tuberculosis occurred in humans some 70,000 years ago and the disease spread around the world through human migration. Nowadays, only less than $10\%$ of infected persons develop an active form of tuberculosis. This indicates that *Mycobacterium tuberculosis* has adapted well to its host and in most cases does not harm the individual very much. Nevertheless, every year there more than one million people dying from tuberculosis. However, most of these victims are immunocompromised, e.g., by old age, HIV-1 infection or other impairments.
Some 100 years ago a typical treatment of tuberculosis was the exposure with UV-B, which increases the endogenous production of vitamin D3. This is one of multiple examples for the interface between metabolism and immunity, which is often mediated by monocyte-derived macrophages. Another example are metabolic tissues like adipose tissue that attract macrophages and show a combined inflammatory and metabolic response. This is important, since after pathogen invasions the immune system requires lots of energy for rapid cell growth and new protein synthesis. Therefore, inflammatory mediators are able to control energy metabolism, in order to defend most efficiently against pathogens, e.g., through a rapid shift from glucose oxidation to lipid oxidation. Similarly, insulin resistance can be triggered by lipids, so that glucose is preserved for the brain and erythrocytes, which depend on glucose as an energy source.
Taken together, evolutionary adaptation to nutrition and lifestyle changes, as it occurred in neolithic societies, involve immune responses and are mediated via the mutual control of metabolism and inflammation.
## Conclusion
Homo sapiens lived for more than 200,000 years as hunter and gatherer in Africa and had adapted his biochemistry to this type of diet. Migrations to significantly different geographic regions within the past 75,000 years and in particular the shift to a life as farmers some 10,000 years ago created a number of evolutionary pressures. These challenges were not only a change in diet but also the burden of infectious diseases, to both of which our (epi)genome responded by adaptations. However, during the past 50 years changes in our environment and lifestyle were faster than ever in our history. This implies that many of us are not (epi)genetically prepared to the challenges of Western diet paired with the preferential sedentary lifestyle and that most of us will get sooner or later in life the diagnosis of having the metabolic syndrome. Nevertheless, we have the chance to remind our evolutionary history written down in the individual version of our (epi)genome and adapt our lifestyle accordingly.
## Financial support and sponsorship
This publication is part of a project that has received funding from the European Union's Horizon2020 research and innovation program under grant agreement no. 952601.
## Data availability
No data was used for the research described in the article.
## References
1. Chen Y., Michalak M., Agellon L.B.. **Importance of nutrients and nutrient metabolism on human health**. *Yale J. Biol. Med.* (2018) **91** 95-103. PMID: 29955217
2. Ordovas J.M., Ferguson L.R., Tai E.S., Mathers J.C.. **Personalised nutrition and health**. *BMJ* (2018) **361** k2173. DOI: 10.1136/bmj.k2173
3. Müller M., Kersten S.. **Nutrigenomics: goals and strategies**. *Nat. Rev. Genet.* (2003) **4** 315-322. DOI: 10.1038/nrg1047
4. Carlberg C., Ulven S.M., Molnár F.. (2020). DOI: 10.1007/978-3-030-36948-4
5. Sturm R.A., Duffy D.L.. **Human pigmentation genes under environmental selection**. *Genome Biol.* (2012) **13** 248. DOI: 10.1186/gb-2012-13-9-248
6. Gerbault P.. **Evolution of lactase persistence: an example of human niche construction**. *Philos. Trans. R. Soc. Lond. B Biol. Sci.* (2011) **366** 863-877. DOI: 10.1098/rstb.2010.0268
7. Wells J.C.K.. **Body composition and susceptibility to type 2 diabetes: an evolutionary perspective**. *Eur. J. Clin. Nutr.* (2017) **71** 881-889. DOI: 10.1038/ejcn.2017.31
8. Darwin C.. (1859)
9. Livingstone K.M.. **Precision nutrition: a review of current approaches and future endeavors**. *Trends Food Sci. Technol.* (2022) **128** 253-264. DOI: 10.1016/j.tifs.2022.08.017
10. Sabeti P.C.. **Positive natural selection in the human lineage**. *Science* (2006) **312** 1614-1620. PMID: 16778047
11. Segurel L., Bon C.. **On the evolution of Llctase persistence in humans**. *Annu. Rev. Genom. Hum. Genet.* (2017) **18** 297-319. DOI: 10.1146/annurev-genom-091416-035340
12. Sabeti P.C.. **The case for selection at CCR5-Delta32**. *PLoS Biol.* (2005) **3** e378. DOI: 10.1371/journal.pbio.0030378
13. Galvani A.P., Novembre J.. **The evolutionary history of the CCR5-Delta32 HIV-resistance mutation**. *Microb. Infect.* (2005) **7** 302-309. DOI: 10.1016/j.micinf.2004.12.006
14. Bustamante C.D.. **Natural selection on protein-coding genes in the human genome**. *Nature* (2005) **437** 1153-1157. DOI: 10.1038/nature04240
15. Dolgova O., Lao O.. **Evolutionary and medical consequences of archaic introgression into modern human genomes**. *Genes* (2018) **9**. DOI: 10.3390/genes9070358
16. Kimura M.. **The neutral theory of molecular evolution: a review of recent evidence**. *Jpn. J. Genet.* (1991) **66** 367-386. PMID: 1954033
17. Dudley J.T.. **Human genomic disease variants: a neutral evolutionary explanation**. *Genome Res.* (2012) **22** 1383-1394. DOI: 10.1101/gr.133702.111
18. Campbell C.D., Eichler E.E.. **Properties and rates of germline mutations in humans**. *Trends Genet.* (2013) **29** 575-584. DOI: 10.1016/j.tig.2013.04.005
19. Gravel S.. **When Is selection effective?**. *Genetics* (2016) **203** 451-462. DOI: 10.1534/genetics.115.184630
20. Dobzhansky T.. **Nothing in biology makes sense except in the light of evolution**. *Am. Biol. Teach.* (1973) **35** 125-129
21. Consortium G.T.. **Genetic effects on gene expression across human tissues**. *Nature* (2017) **550** 204-213. DOI: 10.1038/nature24277
22. Robinson M.R., Wray N.R., Visscher P.M.. **Explaining additional genetic variation in complex traits**. *Trends Genet.* (2014) **30** 124-132. DOI: 10.1016/j.tig.2014.02.003
23. Choudhuri A.. **Common variants in signaling transcription-factor-binding sites drive phenotypic variability in red blood cell traits**. *Nat. Genet.* (2020) **52** 1333-1345. DOI: 10.1038/s41588-020-00738-2
24. Lander E.S.. **Initial sequencing and analysis of the human genome**. *Nature* (2001) **409** 860-921. PMID: 11237011
25. Venter J.C.. **The sequence of the Hhman genome**. *Science* (2001) **291** 1304-1351. PMID: 11181995
26. Nurk S.. **The complete sequence of a human genome**. *Science* (2022) **376** 44-53. DOI: 10.1126/science.abj6987
27. Bentley A.R., Callier S.L., Rotimi C.N.. **Evaluating the promise of inclusion of African ancestry populations in genomics**. *NPJ Genom Med* (2020) **5** 5. DOI: 10.1038/s41525-019-0111-x
28. GenomeAsia K.C.. **The GenomeAsia 100K Project enables genetic discoveries across Asia**. *Nature* (2019) **576** 106-111. DOI: 10.1038/s41586-019-1793-z
29. Sudmant P.H.. **An integrated map of structural variation in 2,504 human genomes**. *Nature* (2015) **526** 75-81. DOI: 10.1038/nature15394
30. Bomba L., Walter K., Soranzo N.. **The impact of rare and low-frequency genetic variants in common disease**. *Genome Biol.* (2017) **18** 77. DOI: 10.1186/s13059-017-1212-4
31. Taliun D.. **Sequencing of 53,831 diverse genomes from the NHLBI TOPMed Program**. *Nature* (2021) **590** 290-299. DOI: 10.1038/s41586-021-03205-y
32. Investigators G.P.P.. **100,000 genomes pilot on rare-disease diagnosis in health care - preliminary report**. *N. Engl. J. Med.* (2021) **385** 1868-1880. DOI: 10.1056/NEJMoa2035790
33. **A global reference for human genetic variation**. *Nature* (2015) **526** 68-74. DOI: 10.1038/nature15393
34. Relethford J.H.. **Genetic evidence and the modern human origins debate**. *Heredity* (2008) **100** 555-563. DOI: 10.1038/hdy.2008.14
35. Tournebize R., Chu G., Moorjani P.. **Reconstructing the history of founder events using genome-wide patterns of allele sharing across individuals**. *PLoS Genet.* (2022) **18**. DOI: 10.1371/journal.pgen.1010243
36. Uffelmann E.. **Genome-wide association studies**. *Nature Reviews Methods Primers* (2021) **1**. DOI: 10.1038/s43586-021-00056-9
37. Cano-Gamez E., Trynka G.. **From GWAS to function: using functional genomics to identify the mechanisms underlying complex diseases**. *Front. Genet.* (2020) **11** 424. DOI: 10.3389/fgene.2020.00424
38. Jackson M., Marks L., May G.H.W., Wilson J.B.. **The genetic basis of disease**. *Essays Biochem.* (2018) **62** 643-723. DOI: 10.1042/EBC20170053
39. Claussnitzer M.. **A brief history of human disease genetics**. *Nature* (2020) **577** 179-189. DOI: 10.1038/s41586-019-1879-7
40. Jelenkovic A.. **Genetic and environmental influences on height from infancy to early adulthood: an individual-based pooled analysis of 45 twin cohorts**. *Sci. Rep.* (2016) **6**. DOI: 10.1038/srep28496
41. Grasgruber P., Cacek J., Kalina T., Sebera M.. **The role of nutrition and genetics as key determinants of the positive height trend**. *Econ. Hum. Biol.* (2014) **15** 81-100. DOI: 10.1016/j.ehb.2014.07.002
42. Boyle E.A., Li Y.I., Pritchard J.K.. **An expanded view of complex ttaits: from polygenic to omnigenic**. *Cell* (2017) **169** 1177-1186. DOI: 10.1016/j.cell.2017.05.038
43. Gorin M.B.. **Genetic insights into age-related macular degeneration: controversies addressing risk, causality, and therapeutics**. *Mol. Aspect. Med.* (2012) **33** 467-486. DOI: 10.1016/j.mam.2012.04.004
44. Pociot F.. **Type 1 diabetes genome-wide association studies: not to be lost in translation**. *Clin Transl Immunology* (2017) **6** e162. DOI: 10.1038/cti.2017.51
45. Khera A.V., Kathiresan S.. **Genetics of coronary artery disease: discovery, biology and clinical translation**. *Nat. Rev. Genet.* (2017) **18** 331-344. DOI: 10.1038/nrg.2016.160
46. Vujkovic M.. **Discovery of 318 new risk loci for type 2 diabetes and related vascular outcomes among 1.4 million participants in a multi-ancestry meta-analysis**. *Nat. Genet.* (2020) **52** 680-691. DOI: 10.1038/s41588-020-0637-y
47. Ferreira M.A.. **Genome-wide association and transcriptome studies identify target genes and risk loci for breast cancer**. *Nat. Commun.* (2019) **10** 1741. DOI: 10.1038/s41467-018-08053-5
48. Manolio T.A.. **Finding the missing heritability of complex diseases**. *Nature* (2009) **461** 747-753. DOI: 10.1038/nature08494
49. Breton C.V.. **Exploring the evidence for epigenetic regulation of environmental influences on child health across generations**. *Commun Biol* (2021) **4** 769. DOI: 10.1038/s42003-021-02316-6
50. Michael A.K., Thoma N.H.. **Reading the chromatinized genome**. *Cell* (2021) **184** 3599-3611. DOI: 10.1016/j.cell.2021.05.029
51. Carlberg C., Molnár F.. (2019). DOI: 10.1007/978-3-030-22907-8
52. Cai S.. **Nutritional status impacts epigenetic regulation in early embryo development: a scoping review**. *Adv. Nutr.* (2021) **12** 1877-1892. DOI: 10.1093/advances/nmab038
53. Chen C., Wang Z., Qin Y.. **Connections between metabolism and epigenetics: mechanisms and novel anti-cancer strategy**. *Front. Pharmacol.* (2022) **13**. DOI: 10.3389/fphar.2022.935536
54. Dai Z., Ramesh V., Locasale J.W.. **The evolving metabolic landscape of chromatin biology and epigenetics**. *Nat. Rev. Genet.* (2020) **21** 737-753. DOI: 10.1038/s41576-020-0270-8
55. Young J.I., Zuchner S., Wang G.. **Regulation of the epigenome by vitamin C**. *Annu. Rev. Nutr.* (2015) **35** 545-564. DOI: 10.1146/annurev-nutr-071714-034228
56. Li J.. **Metabolic control of histone acetylation for precise and timely regulation of minor ZGA in early mammalian embryos**. *Cell Discov* (2022) **8** 96. DOI: 10.1038/s41421-022-00440-z
57. Su C.. **3D chromatin maps of the human pancreas reveal lineage-specific regulatory architecture of T2D risk**. *Cell Metabol.* (2022) **34** 1394-1409 e1394. DOI: 10.1016/j.cmet.2022.08.014
58. Bruno S., Williams R.J., Del Vecchio D.. **Epigenetic cell memory: the gene's inner chromatin modification circuit**. *PLoS Comput. Biol.* (2022) **18**. DOI: 10.1371/journal.pcbi.1009961
59. Fitz-James M.H., Cavalli G.. **Molecular mechanisms of transgenerational epigenetic inheritance**. *Nat. Rev. Genet.* (2022) **23** 325-341. DOI: 10.1038/s41576-021-00438-5
60. Crawford N.G.. **Loci associated with skin pigmentation identified in African populations**. *Science* (2017) **358**. DOI: 10.1126/science.aan8433
61. David-Barrett T., Dunbar R.I.. **Bipedality and hair loss in human evolution revisited: the impact of altitude and activity scheduling**. *J. Hum. Evol.* (2016) **94** 72-82. DOI: 10.1016/j.jhevol.2016.02.006
62. Greaves M.. **Was skin cancer a selective force for black pigmentation in early hominin evolution?**. *Proc. Biol. Sci.* (2014) **281**. DOI: 10.1098/rspb.2013.2955
63. Pavan W.J., Sturm R.A.. **The genetics of human skin and hair pigmentation**. *Annu. Rev. Genom. Hum. Genet.* (2019) **20** 41-72. DOI: 10.1146/annurev-genom-083118-015230
64. Ju D., Mathieson I.. **The evolution of skin pigmentation associated variation in West Eurasia**. *Proc Nat Acad Sci USA* (2021) **118**. DOI: 10.1073/pnas.2009227118
65. Skoglund P., Mathieson I.. **Ancient genomics of modern humans: the first decade**. *Annu. Rev. Genom. Hum. Genet.* (2018) **19** 381-404. DOI: 10.1146/annurev-genom-083117-021749
66. Jablonski N.G., Chaplin G.. **The roles of vitamin D and cutaneous vitamin D production in human evolution and health**. *International J. Paleopathology* (2018) **23** 54-59. DOI: 10.1016/j.ijpp.2018.01.005
67. Holick M.F.. **Sunlight and vitamin D for bone health and prevention of autoimmune diseases, cancers, and cardiovascular disease**. *Am. J. Clin. Nutr.* (2004) **80** 1678S-1688S. PMID: 15585788
68. Hublin J.J.. **New fossils from Jebel Irhoud, Morocco and the pan-African origin of Homo sapiens**. *Nature* (2017) **546** 289-292. DOI: 10.1038/nature22336
69. Vidal C.M.. **Age of the oldest known Homo sapiens from eastern Africa**. *Nature* (2022) **601** 579-583. DOI: 10.1038/s41586-021-04275-8
70. Pagani L.. **Genomic analyses inform on migration events during the peopling of Eurasia**. *Nature* (2016) **538** 238-242. DOI: 10.1038/nature19792
71. Gunther T.. **Population genomics of Mesolithic Scandinavia: investigating early postglacial migration routes and high-latitude adaptation**. *PLoS Biol.* (2018) **16**. DOI: 10.1371/journal.pbio.2003703
72. Nakagome S.. **Estimating the ages of selection signals from different epochs in human history**. *Mol. Biol. Evol.* (2016) **33** 657-669. DOI: 10.1093/molbev/msv256
73. Fu Q.. **The genetic history of ice age Europe**. *Nature* (2016) **534** 200-205. DOI: 10.1038/nature17993
74. Prufer K.. **The complete genome sequence of a Neanderthal from the Altai Mountains**. *Nature* (2014) **505** 43-49. DOI: 10.1038/nature12886
75. Peyregne S.. **Nuclear DNA from two early Neandertals reveals 80,000 years of genetic continuity in Europe**. *Sci. Adv.* (2019) **5** eaaw5873. DOI: 10.1126/sciadv.aaw5873
76. Petr M., Paabo S., Kelso J., Vernot B.. **Limits of long-term selection against Neandertal introgression**. *Proc. Natl. Acad. Sci. U. S. A.* (2019) **116** 1639-1644. DOI: 10.1073/pnas.1814338116
77. Bergstrom A., Stringer C., Hajdinjak M., Scerri E.M.L., Skoglund P.. **Origins of modern human ancestry**. *Nature* (2021) **590** 229-237. DOI: 10.1038/s41586-021-03244-5
78. Villalba-Mouco V.. **Survival of late Pleistocene hunter-gatherer ancestry in the Iberian peninsula**. *Curr. Biol.* (2019) **29** 1169-1177 e1167. DOI: 10.1016/j.cub.2019.02.006
79. Mathieson I.. **Genome-wide patterns of selection in 230 ancient Eurasians**. *Nature* (2015) **528** 499-503. DOI: 10.1038/nature16152
80. Mathieson I.. **The genomic history of southeastern Europe**. *Nature* (2018) **555** 197-203. DOI: 10.1038/nature25778
81. Haak W.. **Massive migration from the steppe was a source for Indo-European languages in Europe**. *Nature* (2015). DOI: 10.1038/nature14317
82. Damgaard P.B.. **137 ancient human genomes from across the Eurasian steppes**. *Nature* (2018) **557** 369-374. DOI: 10.1038/s41586-018-0094-2
83. Saag L.. **The arrival of siberian ancestry connecting the Eastern Baltic to Uralic speakers further East**. *Curr. Biol.* (2019) **29** 1701-1711 e1716. DOI: 10.1016/j.cub.2019.04.026
84. Schroeder H.. **Unraveling ancestry, kinship, and violence in a Late Neolithic mass grave**. *Proc. Natl. Acad. Sci. U. S. A.* (2019) **116** 10705-10710. DOI: 10.1073/pnas.1820210116
85. Hofmanova Z.. **Early farmers from across Europe directly descended from neolithic aegeans**. *Proc. Natl. Acad. Sci. U. S. A.* (2016) **113** 6886-6891. DOI: 10.1073/pnas.1523951113
86. Slatkin M., Racimo F.. **Ancient DNA and human history**. *Proc. Natl. Acad. Sci. U. S. A.* (2016) **113** 6380-6387. DOI: 10.1073/pnas.1524306113
87. Lamnidis T.C.. **Ancient Fennoscandian genomes reveal origin and spread of Siberian ancestry in Europe**. *Nat. Commun.* (2018) **9** 5018. DOI: 10.1038/s41467-018-07483-5
88. Jukarainen S.. **Genetic risk factors have a substantial impact on healthy life years**. *Nat. Med.* (2022) **28** 1893-1901. DOI: 10.1038/s41591-022-01957-2
89. Visscher P.M., Yengo L., Cox N.J., Wray N.R.. **Discovery and implications of polygenicity of common diseases**. *Science* (2021) **373** 1468-1473. DOI: 10.1126/science.abi8206
90. Beleza S.. **The timing of pigmentation lightning in Europeans**. *Mol. Biol. Evol.* (2013) **30** 24-35. DOI: 10.1093/molbev/mss207
91. Racimo F., Sikora M., Vander Linden M., Schroeder H., Lalueza-Fox C.. **Beyond broad strokes: sociocultural insights from the study of ancient genomes**. *Nat. Rev. Genet.* (2020). DOI: 10.1038/s41576-020-0218-z
92. Charlton S.. **New insights into Neolithic milk consumption through proteomic analysis of dental calculus**. *Archaeol. Anthropol. Sci.* (2019) **11** 6183-6196. DOI: 10.1007/s12520-019-00911-7
93. Henn B.M., Cavalli-Sforza L.L., Feldman M.W.. **The great human expansion**. *Proc. Natl. Acad. Sci. U. S. A.* (2012) **109** 17758-17764. DOI: 10.1073/pnas.1212380109
94. Uricchio L.H., Zaitlen N.A., Ye C.J., Witte J.S., Hernandez R.D.. **Selection and explosive growth alter genetic architecture and hamper the detection of causal rare variants**. *Genome Res.* (2016) **26** 863-873. DOI: 10.1101/gr.202440.115
95. Gurdasani D., Barroso I., Zeggini E., Sandhu M.S.. **Genomics of disease risk in globally diverse populations**. *Nat. Rev. Genet.* (2019) **20** 520-535. DOI: 10.1038/s41576-019-0144-0
96. Milton K.. **Hunter-gatherer diets-a different perspective**. *Am. J. Clin. Nutr.* (2000) **71** 665-667. DOI: 10.1093/ajcn/71.3.665
97. Alt K.W., Al-Ahmad A., Woelber J.P.. **Nutrition and health in human evolution-past to present**. *Nutrients* (2022) **14**. DOI: 10.3390/nu14173594
98. Greger M.. **The human/animal interface: emergence and resurgence of zoonotic infectious diseases**. *Crit. Rev. Microbiol.* (2007) **33** 243-299. DOI: 10.1080/10408410701647594
99. Luca F., Perry G.H., Di Rienzo A.. **Evolutionary adaptations to dietary changes**. *Annu. Rev. Nutr.* (2010) **30** 291-314. DOI: 10.1146/annurev-nutr-080508-141048
100. Carrera-Bastos P., Fontes O'Keefe. **The western diet and lifestyle and diseases of civilization**. *Res. Rep. Clin. Cardiol.* (2011). DOI: 10.2147/rrcc.S16919
101. White J.S.. **Straight talk about high-fructose corn syrup: what it is and what it ain't**. *Am. J. Clin. Nutr.* (2008) **88** 1716S-1721S. DOI: 10.3945/ajcn.2008.25825B
102. Hall K.D.. **The energy balance model of obesity: beyond calories in, calories out**. *Am. J. Clin. Nutr.* (2022) **115** 1243-1254. DOI: 10.1093/ajcn/nqac031
103. Woods S.C., D'Alessio D.A.. **Central control of body weight and appetite**. *J. Clin. Endocrinol. Metab.* (2008) **93** S37-S50. DOI: 10.1210/jc.2008-1630
104. Carlberg C., Velleuer E.. (2023)
105. Zohar I.. **Evidence for the cooking of fish 780,000 years ago at Gesher Benot Ya'aqov, Israel**. *Nat Ecol Evol* (2022) **6** 2016-2028. DOI: 10.1038/s41559-022-01910-z
106. Breslin P.A.. **An evolutionary perspective on food and human taste**. *Curr. Biol.* (2013) **23** R409-R418. DOI: 10.1016/j.cub.2013.04.010
107. Pajic P.. **Independent amylase gene copy number bursts correlate with dietary preferences in mammals**. *Elife* (2019) **8**. DOI: 10.7554/eLife.44628
108. Perry G.H.. **Diet and the evolution of human amylase gene copy number variation**. *Nat. Genet.* (2007) **39** 1256-1260. DOI: 10.1038/ng2123
109. Arendt M., Cairns K.M., Ballard J.W., Savolainen P., Axelsson E.. **Diet adaptation in dog reflects spread of prehistoric agriculture**. *Heredity* (2016) **117** 301-306. DOI: 10.1038/hdy.2016.48
110. Carrigan M.A.. **Hominids adapted to metabolize ethanol long before human-directed fermentation**. *Proc. Natl. Acad. Sci. U. S. A.* (2015) **112** 458-463. DOI: 10.1073/pnas.1404167111
111. Sharifi-Rad M.. **Lifestyle, oxidative stress, and antioxidants: back and forth in the pathophysiology of chronic diseases**. *Front. Physiol.* (2020) **11** 694. DOI: 10.3389/fphys.2020.00694
112. Laland K.N., Odling-Smee J., Myles S.. **How culture shaped the human genome: bringing genetics and the human sciences together**. *Nat. Rev. Genet.* (2010) **11** 137-148. DOI: 10.1038/nrg2734
113. Evershed R.P.. **Dairying, diseases and the evolution of lactase persistence in Europe**. *Nature* (2022) **608** 336-345. DOI: 10.1038/s41586-022-05010-7
114. Lamberti L.M., Fischer Walker C.L., Noiman A., Victora C., Black R.E.. **Breastfeeding and the risk for diarrhea morbidity and mortality**. *BMC Publ. Health* (2011) **11** S15. DOI: 10.1186/1471-2458-11-S3-S15
115. Thorning T.K.. **Milk and dairy products: good or bad for human health? An assessment of the totality of scientific evidence**. *Food Nutr. Res.* (2016) **60**. DOI: 10.3402/fnr.v60.32527
116. Marciniak S., Perry G.H.. **Harnessing ancient genomes to study the history of human adaptation**. *Nat. Rev. Genet.* (2017) **18** 659-674. DOI: 10.1038/nrg.2017.65
117. Karlsson E.K., Kwiatkowski D.P., Sabeti P.C.. **Natural selection and infectious disease in human populations**. *Nat. Rev. Genet.* (2014). DOI: 10.1038/nrg3734
118. Klunk J.. **Evolution of immune genes is associated with the Black Death**. *Nature* (2022) **611** 312-319. DOI: 10.1038/s41586-022-05349-x
119. Liston A., Humblet-Baron S., Duffy D., Goris A.. **Human immune diversity: from evolution to modernity**. *Nat. Immunol.* (2021) **22** 1479-1489. DOI: 10.1038/s41590-021-01058-1
120. Prohaska A.. **Human disease variation in the light of population genomics**. *Cell* (2019) **177** 115-131. DOI: 10.1016/j.cell.2019.01.052
|
---
title: Deficiency of endothelial FGFR1 signaling via upregulation of ROCK2 activity
aggravated ALI/ARDS
authors:
- Yue Deng
- Xingming Huang
- Yan Hu
- Weiting Zhong
- Hua Zhang
- Chunheng Mo
- Hongjun Wang
- Bi-Sen Ding
- Chen Wang
journal: Frontiers in Immunology
year: 2023
pmcid: PMC10036754
doi: 10.3389/fimmu.2023.1041533
license: CC BY 4.0
---
# Deficiency of endothelial FGFR1 signaling via upregulation of ROCK2 activity aggravated ALI/ARDS
## Abstract
Vascular leakage and inflammation are pathological hallmarks of acute lung injury (ALI)/acute respiratory distress syndrome (ARDS). Endothelial cells (ECs) serve as a semipermeable barrier and play a key role in disease progression. It is well known that fibroblast growth factor receptor 1 (FGFR1) is required for maintaining vascular integrity. However, how endothelial FGFR1 functions in ALI/ARDS remains obscure. Here, we revealed that conditional deletion of endothelial FGFR1 aggravated LPS-induced lung injury, including inflammation and vascular leakage. Inhibition of its downstream Rho-associated coiled-coil–forming protein kinase 2 (ROCK2) by AAV Vec-tie-shROCK2 or its selective inhibitor TDI01 effectively attenuated inflammation and vascular leakage in a mouse model. In vitro, TNFα-stimulated human umbilical vein endothelial cells (HUVECs) showed decreased FGFR1 expression and increased ROCK2 activity. Furthermore, knockdown of FGFR1 activated ROCK2 and thus promoted higher adhesive properties to inflammatory cells and higher permeability in HUVECs. TDI01 effectively suppressed ROCK2 activity and rescued the endothelial dysfunction. These data demonstrated that the loss of endothelial FGFR1 signaling mediated an increase in ROCK2 activity, which led to an inflammatory response and vascular leakage in vivo and in vitro. Moreover, inhibition of ROCK2 activity by TDI01 provided great value and shed light on clinical translation.
## Introduction
ALI/ARDS is a severe and dangerous syndrome, clinically characterized by progressive dyspnea and life-threatening arterial hypoxemia, leading to serious mortality ($30\%$-$40\%$) [1]. Due to direct (such as pneumonia) or indirect lung injury (such as sepsis) exposure, the morbidity of ALI/ARDS is considerably high, with approximately $10\%$ of all patients in intensive care units across 50 countries suffering from it [2, 3]. Currently, the treatment options for this disease are still limited with poor effects.
Disruption of endothelial barrier results in increased inflammation and vascular permeability, which are pathological hallmarks of ALI/ARDS. Pulmonary endothelial cells (ECs) lining the innermost layer of blood vessels act as a semipermeable barrier and perform an important role in maintaining vasculature homeostasis with the assistance of intercellular junctions (tight junctions and adherens junctions) [4]. These junctions are facilitated by cytoskeletal microtubules and actin microfilaments under the regulation of multiple signaling pathways [5, 6]. Therefore, it is imperative to explore pathways that precisely reconcile endothelial barrier integrity and inflammation, upon which to seek a promising therapeutic target is of great importance.
The FGFR family member FGFR1, a single-pass transmembrane receptor with tyrosine kinase activity, has diverse ligand-receptor specificity and extensive expression in different tissues [7], which leads to the involvement of FGFR1 in a wide range of fields, such as angiogenesis, wound healing and development [8]. As the most abundant FGFR of endothelial cells [9, 10], FGFR1 is reported to be responsible for maintaining vascular integrity [11, 12]. For instance, FGFR1 activation by recombinant FGF2 treatment decreases permeability in human brain microvascular endothelial cells challenged by oxygen-glucose deprivation/reoxygenation [10] and traumatic brain injury [13]. Moreover, another study reveals that FGFR1-dependent recombinant FGF21 protects against blood brain barrier (BBB) leakage [14]. On the other hand, FGFR1 has been shown to be actively involved in endothelial repair after vascular injury [15]. It serves as a critical mediator in glycocalyx reconstitution of the pulmonary endothelial surface layer during sepsis [10]. Indeed, FGFR1 is suppressed in sepsis [10, 16], and an increase in FGFR1 expression attenuates pulmonary inflammation in ventilator induced lung injury [16]. However, understanding of the mechanism underlying FGFR1 signaling in the regulation of vascular permeability and inflammation in LPS-induced ALI/ARDS remains incomplete.
Rho-associated coiled-coil–forming protein kinases (ROCKs), downstream effectors of the small GTPase RhoA, play a central role in regulating actin cytoskeleton dynamics by phosphorylating multiple downstream substrates, including myosin light chain (MLC), myosin phosphatase-targeting subunit 1 (MYPT-1) and Ezrin/radixin/moesin (ERM) [17]. They function in regulating various cellular functions, such as cellular contraction, proliferation, migration and differentiation [18]. ROCKs have two isoforms, ROCK1 and ROCK2, which share $64\%$ overall identity in their primary amino acid sequences [17]. Enhanced ROCK activity upon stimulation by inflammatory mediators, such as thrombin, LPS and TNFα, strongly disrupts the vascular barrier [19] and is deeply involved in ALI/ARDS [20]. A ROCK2-specific role in monocytic migration, monocyte adhesion toward endothelial cells [21] and vascular permeability [22] has been reported. Considering the potential protective effect of FGFR1 signaling in vascular function, we concluded that deficiency of endothelial FGFR1 by activating ROCK2 aggravated lung injury and that inhibition of ROCK2 activity was considered to be a prospective insight for prevention and treatment of ALI/ARDS.
## Experimental animals
Male wild-type (WT) C57/BL6J mice, 8 weeks of age and weighing 22-24 g, were purchased from the Model Animal Research Center of Nanjing University. Fgfr1 loxP/loxP mice were crossed with VE-cadherin-CreERT2 mice to establish VE-cadherin-CreERT2 Fgfr1 loxP/loxP (Fgfr1 iΔEC/iΔEC) mice, which were identified by genotyping. All mice were housed in a specific pathogen-free (SPF) animal room named the Experimental Animal Center of West China Second Hospital in accordance with the guidelines of the National Institutes of Health. All of the animal experiments were approved by the animal care and were performed in accordance with the guidelines outlined by the committees of West China Second University Hospital, Sichuan University. After the mice were treated intraperitoneally with tamoxifen (100 mg/kg) for 6 days and interrupted for 3 days after the third dose, EC-specific deletion of FGFR1 was induced. WT mice were injected intratracheally with 10 mg/kg LPS or saline, while Fgfr1 iΔEC/iΔEC mice and their control littermates Fgfr1 loxP/loxP mice were injected intratracheally with 2 mg/kg LPS to maintain their survival. Different doses of LPS were dissolved in 50 μl of sterile saline, and all groups received a volume of 50 μl. The lung tissues were harvested after 24 h. The mice were pretreated orally with the selective ROCK2 inhibitor, TDI01, (Beijing Tide Pharmaceutical Co., Ltd.) for 3 days before LPS injection.
## Adeno-associated virus transduction
The adeno-associated virus Vec (AAV Vec) was manufactured by Hanbio Biotechnology (Shanghai, China). To achieve endothelial-selective gene knockdown, sequence encoding shROCK2 or negative control sequence (NC) was constructed under Tie1 promoter and expressed in AAV, resulting in AAV Vec-tie-shROCK2 and AAV Vec-tie-shNC. The virus titer was 1.0×1012 μg/ml. Fgfr1 iΔEC/iΔEC mice were injected intratracheally with AAV Vec expressing ROCK2-specific shRNA and control shRNA (Table S1) and then LPS 1 month later.
## Isolation of mouse pulmonary ECs
Fresh mouse lung tissues were washed twice with cold PBS, minced and incubated in a digestive mixture (1 mg/ml of collagenase I and 1 mg/ml of dispase II in DPBS) on an orbital shaker at 37°C for 30 min. The cell suspension was then diluted with the same volume of cold DPBS, filtered through cell strainers and centrifuged. The cells were then washed once with DPBS and treated with red blood cell lysis reagent for 10 min, and washed and then centrifuged. For EC (CD31+ CD45-) isolation, dynabeads were washed three times with 1 ml of cold MACS wash buffer (2 mM EDTA, $0.1\%$ BSA in DPBS) and incubated respectively with CD45 antibody (BD) and CD31 antibody (BD) at 4°C for 4 h. The dynabeads were then washed three times with MACS wash buffer. The cell deposits were resuspended in 300 μl of MACS wash buffer. Two hundred microliters (200 μl) of Dynabeads-CD45 antibody conjugate (200 μl for 2×107 cells) was added to the cell suspension and then incubated at 4°C for 45 min on a rotator. After incubation, beads bound to CD45+ cells were captured by a magnet, and the supernatant was transferred to a tube containing Dynabeads-CD31. The Dynabeads-CD31 antibody conjugate was incubated with the collected supernatant at 4°C for 45 min on a rotator, and beads with CD31+ CD45- cells were captured by a magnet. Beads with CD31+ CD45- cells were washed three times with cold MACS wash buffer and then prepared for RNA or protein isolation.
## RNA-sequencing
Total endothelial RNA from mouse lungs sorted by dynabeads was isolated with TRIzol reagent according to standard protocols and then sent to Beijing Novo Gene Company for RNA-sequencing analysis. Total amounts and integrity of RNA were assessed using the RNA Nano 6000 Assay Kit of the Bioanalyzer 2100 system (Agilent Technologies, CA, USA). Transcriptome libraries were constructed and quantified by Qubit 2.0 Fluorometer. After the library was qualified, the target amount of data was sequenced by the Illumina NovaSeq 6000. Sequenced reads were aligned to the *Mus musculus* reference genome (GRCm38/mm10) with HISAT2 (v2.0.5). FeatureCounts (v1.5.0-p3) was used to count the read numbers mapped to each gene. Differential expression analysis was performed using the DESeq2 R package (1.20.0). Gene set enrichment analysis (GSEA) was implemented by the clusterProfiler R package (3.8.1).
## In vivo vascular permeability assay
To measure vascular permeability in response to LPS challenge, dextran (70 kDa) and Evans blue dye (EBD) staining were used. In brief, the mice were injected with 50 µl of FITC-dextran or AF555-dextran (50 mg/ml) via tail vein 23.5 h after LPS injection. After 30 min, the whole lung tissues were harvested and perfused through the trachea with 200 μl of $50\%$ OCT and immersed in $4\%$ paraformaldehyde at 4°C for 8 h. Each lobe was embedded in OCT at -80°C for frozen sectioning. Then, 7-µm-thick sections were cut and stained with DAPI. The fluorescence of each group was measured by FV3000 (Olympus) and LSM980 microscope (Zeiss). The mice were injected via tail vein with 100 µl of $1\%$ EBD 23 h after LPS injection. One hour later, the mice were sacrificed and the lung vasculature was flushed with 10 ml of PBS per mouse through the right ventricle to remove intravascular dye. The whole lung tissues were homogenized in 1.5 ml of formamide and incubated at 60°C for 48 h. The homogenates were centrifuged at 12000×g for 30 min, and the absorbance of the supernatant was measured at 620 nm in a 96-well plate.
## Histology
After collection, the lungs were fixed in $4\%$ paraformaldehyde, maintained for 24 h, embedded in paraffin and cut into 5-μm-thick sections. The sections were stained with hematoxylin and eosin before microscopic histological examination. Pictures were taken by Pannoramic MIDI. The severity of lung injury was scored based on the American Thoracic Society [23].
## Measurement of relative mRNA expression
Total RNA was extracted from lung tissues and cells using TRIzol reagent according to the standard protocol and then reverse-transcribed into cDNA. RT-PCR was performed using the CFX96™ Real-Time system (Bio-Rad). Relative IL6, IL1β, IL10, TNFα and Fgfr1 mRNA expression was assessed. The target gene transcript levels were determined and normalized to the housekeeping gene Gapdh using the ΔΔCT method. The primer sequences were shown in Table S2.
## Immunofluorescent staining
For immunofluorescent staining, lung tissues were frozen in OCT and cut into 6-μm-thick sections. Cryopreserved sections were incubated with antibodies against mouse VE-cadherin (R&D Systems), FGFR1 (Sigma-Aldrich), ROCK2 (Abcam), ROCK1 (Abcam), MLC2 (CST), ERM (CST), pROCK2 (Abcam), pROCK1 (Abcam), pMLC2 (CST) and pERM (CST) supplemented with $10\%$ normal donkey serum/$1\%$ BSA/$0.1\%$ Tween 20 at 4°C overnight, followed by incubation with fluorophore-conjugated secondary antibodies at 37°C for 1 h (Jackson ImmunoResearch). Specimens were stained with DAPI and sealed. Images were captured by an LSM980 microscope (Zeiss).
## Cell culture
HUVECs were isolated from the human umbilical cords of healthy volunteers following informed consent and ethics committee approval. They were cultured in endothelial cell basal medium-2 (CC-3156, Lonza) containing 100 U/mL of penicillin and 100 µg/mL of streptomycin (Sigma-Aldrich) at 37°C in a humidified $5\%$ carbon dioxide atmosphere. TNFα (Novoprotein), TDI01 (Beijing Tide Pharmaceutical Co., Ltd.) and azd4547 (MCE) were used to treat HUVECs.
## Monocyte adhesion assay
HUVECs were cultured in 12-well plates. THP-1 cells were suspended (1 × 106 cells/mL) and stained with Hoechst 33342 (Beyotime) at 37°C for 10 min. THP-1 cells were washed three times with PBS to wash off unlabeled THP-1 cells and then plated in HUVECs for a 4-h co-incubation in a $5\%$ CO2 atmosphere at 37°C. To remove the unadhered THP-1 cells, HUVECs were washed with PBS three times. After that, adhered THP-1 cells were captured by Carl Zeiss Microscopy GmbH and counted from three randomly selected areas.
## In vitro detection of permeability in HUVECs
Permeability of endothelial monolayers in vitro was determined by FITC-dextran (3 kDa). HUVECs (1×105 cells/100 μl/well) were plated on the upper chamber of the transwell insert on top of the Matrigel-coated transwell filters (3 μm pore size). After cells adhere (60 min after plating), 200 μl of EBM-2 was added to the upper chamber and 1 ml was added to the lower chamber (in the 24-well plate). After incubating at 37°C for 24 h, the procedure of HUVEC plating was repeated and then incubation was performed for an additional 24 h at 37°C. FITC-dextran was added to the lower chamber to a final concentration of 10 μg/ml. Ten microliters (10 μl) of aliquots of media were removed from the upper chamber and diluted in 90 μl of water/well in a 96-well plate at various time intervals (0, 5, 15, 30, 45, 60 and 120 min after treatment). The fluorescence intensity was measured using a 96-well plate fluorimeter (PE EnVision 2015) with excitation at 485 nm and emission at 535 nm.
## Western blotting
Total protein was extracted from HUVECs and lysed in RIPA buffer containing phosphatase inhibitor cocktail (Bimake) and protease inhibitor cocktail (MCE). Proteins were added to the wells of an SDS-PAGE system, separated by $10\%$ SDS–PAGE and transferred to a PVDF membrane. After blocking in $5\%$ milk for 1 h at room temperature, the membrane was incubated with β-actin (Servicebio), FGFR1 (CST), ROCK2 (Abcam), ROCK1 (Abcam), ERM (CST), MYPT1 (CST), pROCK2 (Abcam), pROCK1 (Abcam), pERM (CST), pMYPT1 (Invitrogen), GAPDH (Servicebio), ICAM1 (Servicebio) and VCAM1 (Abcam) at 4°C overnight and then incubated with horseradish peroxidase (HRP)-conjugated goat anti-rabbit or mouse IgG at room temperature for 1 h. Bands were visualized using the ChemiDoc Imaging System (TOUCH IMAGER™).
## siRNA knockdown
HUVECs in culture were transfected with human FGFR1 siRNA (siFGFR1) and control siRNA (siNC) (GenePharma). The transfection reagent Lipofectamin™ RNAiMAX (Invitrogen) was used. siRNAs were used at a final concentration of 10 nM. The siRNA sequences are shown in Table S3.
## Statistical analysis
Data analysis was carried out using the GraphPad Prism 8 software. Statistical analysis of differences between the two groups was performed using the unpaired Student’s t test. p values < 0.05 were considered statistically significant.
## Pulmonary endothelial FGFR1 expression was decreased in LPS stimulated ALI/ARDS
To investigate the underlying mechanism of ALI/ARDS progression, a mouse model was constructed via intratracheal instillation of LPS (10 mg/kg) in WT mice. Histological analysis of lung tissues showed marked pulmonary edema and patchy neutrophil infiltration in the LPS group (Figure 1A) compared with the saline group. In addition, the lung injury scores (Figure 1A) were quantified and showed significant differences, indicating the presence of severe inflammation. Undoubtedly, LPS triggered a severe inflammatory response supported by excessive accumulation of inflammatory cytokines (IL6, IL1β, IL10, TNFα) mRNA expression (Figure 1B) in lung tissues. Consistent with previous studies [24], we observed a damage to the pulmonary vascular barrier in ALI/ARDS lungs. By measuring and quantifying the interstitial accumulation of the intravenously injected plasma protein tracer FITC-dextran and EBD (Figures 1C, D), mice treated with LPS exhibited significantly higher amounts of FITC-dextran and EBD, suggesting vascular barrier disruption. Overall, we successfully constructed an ALI/ARDS mouse model. Since pulmonary endothelial cells function as a semipermeable cellular barrier and play a critical role in the preservation of vascular integrity and inflammation, we performed RNA-seq on pulmonary ECs isolated from the saline and LPS groups to elucidate the transcriptional changes. Hierarchical clustering (Figure 1E) of 5880 significantly differentially expressed genes (adjusted p ≤ 0.05; absolute fold change ≥ 1) after LPS treatment revealed a huge transcriptional shift, with 2713 upregulated genes and 3167 downregulated genes. To gain further insight into the key genes that maintain pulmonary vascular integrity, we performed a Gene set enrichment analysis and found a significant decrease in FGFR1-associated signaling pathways (Figure 1F). To further verify whether FGFR1 is downregulated in ECs of ALI/ARDS lungs, we confirmed an obvious decrease (Figures 1G, H). Next, we investigated FGFR1 expression in HUVECs challenged with different doses of TNFα. The results showed that TNFα induced a significant reduction in FGFR1 in HUVECs (Figures 1I; S1). Taken together, pulmonary endothelial FGFR1 expression was decreased in LPS-stimulated ALI/ARDS.
**Figure 1:** *An LPS-induced ALI/ARDS mouse model was constructed and endothelial FGFR1 was decreased in the LPS group. Mice were treated with saline and LPS (10 mg/kg) intratracheally for 24 h. Representative lung sections were stained with hematoxylin and eosin and lung injury scores were quantified (A), n=5 per group. Scale bars: 200 µm (left) and 40 µm (right). Total RNA was isolated from saline- and LPS-treated mouse lung tissues. Relative mRNA levels of inflammatory cytokines (IL6, IL1β, IL10, TNFα) were significantly increased in mice treated with LPS compared with saline-treated mice, as determined by RT-PCR analysis (B). n=4 or 5 per group. Images of FITC-dextran were costained with DAPI and quantified by the mean fluorescence intensity of FITC-dextran (C). n=4 per group. Scale bars: 50 µm. Whole lung tissues were stained with EBD-stained lungs (left) and EB content in the lungs (right) was quantified (D). n=6 per group. A gene expression heatmap is shown, and hierarchical clustering was based on 2713 upregulated genes and 3167 downregulated genes between lung ECs intratracheally treated with saline and LPS (E). n=4 or 6 per group. Gene set enrichment analysis of differentially expressed genes in the saline and LPS groups suggested decreased endothelial FGFR1 in the LPS group (F). The relative mRNA expression of pulmonary endothelial FGFR1 in LPS-treated mice was examined (G). n=5 or 6 per group. FGFR1 was costained with VE-cadherin in the lung sections. Red indicates FGFR1, green indicates VE-cadherin (H). n=4 per group. Scale bars: 20 µm. HUVECs were treated with 0, 5, 10, 15 and 20 ng/ml of TNFα for 12 hours. Cell lysates were analyzed by western blotting with FGFR1 and β-actin antibodies (I). Each bar represents the mean ± SD; *p < 0.05, **p < 0.01, ***p < 0.001 and ****p < 0.0001.*
## Deficiency of endothelial FGFR1 aggravated inflammation and pulmonary vascular leakage in ALI/ARDS
To address the role of endothelial FGFR1 in inflammation and pulmonary vascular permeability, we conditionally knocked out FGFR1 in the ECs of adult mice (Fgfr1 iΔEC/iΔEC mice) (Figure 2A) and confirmed the knockdown efficiency (Figures 2B, C; S2). We hypothesized that the deletion of FGFR1 in ECs would aggravate the progression of ALI/ARDS. To reduce the mortality rate in Fgfr1 iΔEC/iΔEC mice, we adjusted the dose of LPS to 2 mg/kg for further experiments (Figure 2A). As expected, significantly more infiltration of neutrophils and red blood cells was observed in Fgfr1 iΔEC/iΔEC mice compared to Fgfr1 loxP/loxP mice, not only in the alveolar space but also in the interstitial space. Moreover, higher lung injury scores reflected aggravation in Fgfr1 iΔEC/iΔEC mice (Figure 2D). We also examined the inflammatory response as assessed by the relative mRNA expression levels of inflammatory cytokines. There was markedly increased IL6, IL1β, IL10 and TNFα mRNA expression in Fgfr1 iΔEC/iΔEC mouse lung tissues, suggesting an escalation of inflammation (Figure 2E). In parallel, Fgfr1 iΔEC/iΔEC mice showed more severe FITC-dextran and EBD extravasation (Figures 2F, G). Consistent with previous studies [11, 12], inhibition of FGFR1 actually resulted in the loss of vascular integrity, leading to inflammation and vascular leakage.
**Figure 2:** *Deletion of FGFR1 in ECs aggravated the severity of inflammation and vascular leakage. Fgfr1
iΔEC/iΔEC mice and Fgfr1
loxP/loxP mice were intraperitoneally injected with tamoxifen. After LPS injection, ALI/ARDS was established (A). Knockout efficiency of endothelial cell-specific deletion of FGFR1 is shown. RT-PCR analysis of endothelial FGFR1 mRNA in Fgfr1
loxP/loxP mice and Fgfr1
iΔEC/iΔEC mice (B). n=4 per group. Western blotting analysis of endothelial FGFR1 in Fgfr1
loxP/loxP mice and Fgfr1
iΔEC/iΔEC mice (C). n=3 per group. Lung sections were examined for signs of inflammation after hematoxylin and eosin staining and were scored for lung injury (D). n=6 per group. Scale bars: 200 µm (left) and 40 µm (right). Significantly higher lung injury scores (D) and higher relative mRNA levels of inflammatory cytokines (IL6, IL1β, IL10, TNFα) (E) suggested aggravated lung inflammation in Fgfr1
iΔEC/iΔEC mice. n=5 per group. Vascular permeability was assessed by extravasation of FITC-dextran (F) and EBD (G). Representative images of immunofluorescent staining and quantification of the mean fluorescence intensity of FITC-dextran are shown (F). n=5 per group. Scale bars: 50 µm. Representative images of the gross appearance of EBD-stained lungs and quantification of EBD content in the lungs are shown (G). n=4 per group. Each bar represents the mean ± SD; *p < 0.05, **p < 0.01, ***p<0.001 and ****p < 0.0001.*
## Endothelial ROCK2 was activated in ALI/ARDS model and functioned as a downstream of FGFR1
Next, we sought to unravel the downstream molecular mechanisms of endothelial FGFR1 involved in vascular leakage and inflammation. As a result, we subjected pulmonary ECs from Fgfr1 iΔEC/iΔEC mice and Fgfr1 loxP/loxP mice to bulk RNA sequencing. By GSEA, the Fgfr1 iΔEC/iΔEC mouse group was enriched for increased GTPase activity-related terms (Figure 3A), consistent with the increased Rho GTPase activity found in our RNA-seq of the saline and LPS groups (Figure 3B) and previous studies [5, 25]. We examined endothelial ROCK activation which correlated with the ratio of pERM/ERM and pMLC2/MLC2 in ECs, suggesting robust ROCK activation in Fgfr1 iΔEC/iΔEC mice (Figure 3C). It is well known that RhoA GTPase has two isoforms downstream effectors, ROCK1 and ROCK2 [26]. According to previous studies, ROCK1 and ROCK2 exert redundant functions in the regulation of actomyosin contractility in mouse embryonic fibroblasts [27]. In order to elucidate the effect of each isoform of ROCKs, the autophosphorylation expression levels of pROCK1 (S1333) and pROCK2 (S1366) in ECs were assessed by immunofluorescent staining. We found that pROCK2 was significantly increased in Fgfr1 iΔEC/iΔEC mice and that pROCK1 showed no obvious change (Figure 3C). Furthermore, knockdown of FGFR1 with siRNA FGFR1 without TNFα treatment significantly increased ROCK2 activity in HUVECs (Figures 3D; S3A) and the effect of siRNA FGFR1 transfection was tested by western blotting (Figures 3D, S3A). Similarly, endothelial ROCK2 was activated in WT mice and HUVECs challenged with different doses of TNFα (Figures 3E, F, S3B). Collectively, these data suggested that endothelial ROCK2 functioned as a downstream of FGFR1 and was activated in the lungs of ALI/ARDS mice and TNFα-stimulated HUVECs.
**Figure 3:** *Endothelial ROCK2 was activated in ALI/ARDS model and functioned as a downstream of FGFR1. RNA-seq analysis was performed on ECs isolated from LPS-treated Fgfr1
iΔEC/iΔEC mice and Fgfr1
loxP/loxP mice. Gene set enrichment analysis of differentially expressed genes showed enrichment of GTPase activity and the GTPase complex (A), which are related to RhoA activation. RhoA GTPase activation was enriched by GSEA in the LPS group compared with the saline group (B). Endothelial ROCK2 activity was assessed by the colocalization of pERM, ERM, pMLC2, MLC2, pROCK2, ROCK2, pROCK1 and ROCK1 (red fluorescent protein) with VE-cadherin (green fluorescent protein) by immunofluorescent staining. Representative images of immunofluorescent staining of endothelial ROCK2 activity in the Fgfr1
iΔEC/iΔEC mice vs the Fgfr1
loxP/loxP mice group (C) and the saline vs LPS group (E) are shown. n=3 per group. Scale bars: 40 µm. HUVECs were transfected with siNC or siFGFR1 for 48 h. The expression of FGFR1 after siRNA transfection was detected by WB with FGFR1 and β-actin antibodies. The phosphorylation of ERM, MYPT1 and ROCK2 was assessed (D). In addition, HUVECs were treated with 0, 5, 10, 15 and 20 ng/ml of TNFα for 12 h, and ROCK2 activity was assessed by WB (F).*
## AAV Vec-tie shROCK2 effectively attenuated inflammation and vascular leakage in Fgfr1
iΔEC/iΔEC mice
To precisely clarify the contribution of endothelial ROCK2 in LPS-treated Fgfr1 iΔEC/iΔEC mice, we intratracheally injected AAV Vec-tie-shROCK2 or AAV Vec-tie-shNC into Fgfr1 iΔEC/iΔEC mice after tamoxifen induction. One month later, LPS (2 mg/kg) was intratracheally injected (Figure 4A). The knockdown efficiency of AAV Vec-tie-shROCK2 was verified by immunofluorescence (Figure 4B). Histology (Figure 4C) and inflammatory cytokines (Figure 4D) showed a significant decline in inflammation in Fgfr1 iΔEC/iΔEC mice treated with AAV Vec-tie-shROCK2. Moreover, decreased extravasation of AF555-dextran (Figure 4E) and EBD (Figure 4F) demonstrated that knockdown of endothelial ROCK2 effectively blocked vascular leakage in Fgfr1 iΔEC/iΔEC mice. Overall, our data indicated that inhibition of endothelial ROCK2 activity exhibited an effective way to relieve the aggravation due to lack of endothelial FGFR1 and provided further evidence that deficiency of endothelial FGFR1 contributed to inflammation and vascular permeability via activation of ROCK2. Furthermore, this finding provided an effective and therapeutic target for ALI/ARDS.
**Figure 4:** *AAV Vec-tie-shROCK2 effectively inhibited inflammation and vascular leakage in Fgfr1
iΔEC/iΔEC mice. After intraperitoneal injection of tamoxifen, Fgfr1
iΔEC/iΔEC mice were intratracheally injected with AAV Vec-tie-shROCK2 or AAV Vec-tie-shNC. One month later, they were intratracheally injected with LPS (2 mg/kg) (A). The efficiency of AAV Vec-tie expressing ROCK2-specific shRNA in pulmonary ECs was verified by immunofluorescence. Representative images of immunofluorescent staining of GFP, ROCK2 and VE-cadherin in lung tissues from Fgfr1
iΔEC/iΔEC mice treated with GFP-AAV Vec expressing either negative control shRNA or ROCK2 specific shRNA are shown (B). n=3 per group. Scale bars: 50 µm. The AAV Vec-tie-shROCK2 group showed significantly decreased inflammation. Representative images stained with hematoxylin and eosin and lung injury scores are shown (C). n=4 or 5 per group. Scale bars: 200 µm (left) and 40 µm (right). Relative mRNA expression levels of inflammatory cytokines (IL6, IL1β, IL10, TNFα) were examined by RT-PCR (D). n=4 or 5 per group. Vascular permeability was also significantly decreased in the AAV Vec-tie-shROCK2 group. Representative images of immunofluorescent staining and quantitative mean fluorescence intensity of AF555-dextran are shown (E). n=5 or 6 per group. Scale bars: 50 µm. Representative images of the gross appearance of EBD-stained lungs and quantification of EBD content in the lungs are shown (F). n=6 per group. Each bar represents the mean ± SD; *p < 0.05, **p < 0.01 and ****p < 0.0001.*
## TDI01 effectively attenuated ALI/ARDS in Fgfr1
iΔEC/iΔEC mice and endothelial dysfunction in FGFR1-deficient HUVECs
After demonstrating the effects of activated ROCK2, we selected ROCK2 rather than ROCK1 as the therapeutic target for ALI/ARDS in Fgfr1 iΔEC/iΔEC mice. TDI01 is a highly selective ROCK2 inhibitor and has shown good efficacy (Beijing Tide Pharmaceutical Co., Ltd.). Therefore, we pretreated Fgfr1 iΔEC/iΔEC mice with TDI01 (200 mg/kg/day) for 3 days before injection of LPS. Subsequently we compared the histology, inflammatory response and pulmonary vascular leakage. The results showed that TDI01 significantly alleviated pulmonary edema and inflammatory response based on significantly decreased lung injury scores (Figure 5A) and inflammatory cytokines (Figure 5B). Moreover, the extravasation of FITC-dextran (Figure 5C) and EBD (Figure 5D) was reduced, suggesting that TDI01 protected against pulmonary vascular leakage. Mechanistically, the TDI01-pretreated Fgfr1 iΔEC/iΔEC mouse group showed decreased pERM, pMLC2 and pROCK2 (Figure 5E). In addition, western blotting analysis indicated that knockdown of FGFR1 by siRNA FGFR1 in HUVECs with or without TNFα stimulation enhanced the ratio of pROCK2/ROCK2, pERM/ERM and pMYPT1/MYPT1 (Figures 5F, S4A). Importantly, TDI01 significantly downregulated the phosphorylation of ROCK2,ERM and MYPT1 (Figures 5F, S4A), demonstrating that TDI01 inhibited ROCK2 activation in vitro. When insulted by inflammatory mediators such as TNFα and LPS, ECs transform into a dysfunctional phenotype with increased adhesion molecules (such as ICAM1, VCAM1) on the cell surface to adhere to leukocytes and increased permeability [2, 28]. To investigate whether knockdown of FGFR1 and ROCK2 activation in HUVECs affect the expression of ICAM1 and VCAM1, HUVECs were transfected with siRNA FGFR1 with or without pretreatment with TDI01 for 24 h before stimulation with TNFα for 12 h. The results suggested that knockdown of FGFR1 enhanced the expression of ICAM1 and VCAM1, which was dampened by inhibition of its downstream ROCK2 activity (Figures 5G, S4B). We also performed a monocyte adhesion assay to assess the recruitment of inflammatory cells to HUVECs. HUVECs were pretreated with TDI01 for 24 h with or without azd4547 (FGFR pan inhibitor) for 12 h, and then challenged by TNFα for 12 h. THP-1 cells were labeled with Hoechst 33342 and then co-incubated with HUVECs. A significant increase in the recruitment of monocytes was observed in the azd4547-treated group, which was blocked by TDI01 (Figures 5H, S4C). To determine the role of decreased FGFR1 and ROCK2 activation in regulating vascular permeability, in vitro TNFα-stimulated HUVECs were pretreated with TDI01 for 24 h and then treated with or without azd4547 for 12 h. The fluorescence intensity of extravasated FITC-dextran was measured (Figure 5I) and the results indicated that loss of FGFR1 signaling promoted vascular leakage, which was blocked by inhibition of ROCK2 activity. Taken together, loss of FGFR1 and activated ROCK2 made HUVECs more adhesive to inflammatory cells and aggravated vascular leakage. Meanwhile, TDI01 inhibited ROCK2 activity and played a therapeutic role.
**Figure 5:** *TDI01 effectively attenuated ALI/ARDS in Fgfr1
iΔEC/iΔEC mice and endothelial dysfunction in FGFR1-deficient HUVECs. Inhibition of ROCK2 by TDI01 suppressed the inflammatory response of Fgfr1
iΔEC/iΔEC mice. Representative images stained with hematoxylin and eosin and lung injury scores are shown (A). n=6 per group. Scale bars: 200 µm (left) and 40 µm (right). The relative mRNA expression of inflammatory cytokines (IL6, IL1β, IL10, TNFα) was reduced in the TDI01-pretreated Fgfr1
iΔEC/iΔEC group (B), n=4 or 5 per group. Vascular barrier dysfunction in Fgfr1
iΔEC/iΔEC mice was rescued by TDI01, as indicated by decreased leakage of FITC-dextran (C), n=4 per group, scale bars: 50 µm and EBD (D), n=5 per group. Endothelial ROCK2 activity was inhibited by TDI01 in Fgfr1
iΔEC/iΔEC mice and was assessed by the colocalization of pERM, ERM, pMLC2, MLC2, pROCK2 and ROCK2 (red fluorescent protein) with VE-cadherin (green fluorescent protein) by immunofluorescent staining. Representative images are shown (E). n=5 per group. Scale bars: 40 µm. HUVECs were transfected with siNC or siFGFR1 and pretreated with TDI01 (0, 10 µM) for 24 h, and then treated with or without 20 ng/ml of TNFα for 12 h. ROCK2 activity was determined by the phosphorylation of ERM, MYPT1 and ROCK2 (F). HUVECs were transfected with siNC or siFGFR1 and pretreated with TDI01 (0, 10 µM) for 24 h, and then treated with 20 ng/ml of TNFα for 12 h. The expression of VCAM1 and ICAM1 was assessed by WB (G). HUVECs were pretreated with TDI01 (0, 1 µM) for 24 h, treated with or without azd4547 (1 µM) for 12 h and then challenged by 20 ng/ml of TNFα for 12 h. Hoechst 33342-labeled monocytes were added to each well and co-incubated for 4 h. Representative images of adhered Hoechst 33342-labeled monocytes are shown (H). n=3 per group, scale bars: 200 µm. HUVECs were pretreated with TDI01 (0, 1 µM) for 24 h, and treated with or without azd4547 (1 µM) for 12 h and then were plated for permeability assay. FITC-dextran and TNFα (20 ng/ml) were simultaneously administered into HUVECs and the fluorescence intensity were recorded at 0, 5, 15, 30, 45, 60, 120 min (I). n=3 per group. Each bar represents the mean ± SD; *p < 0.05, **p < 0.01 and ****p < 0.0001.*
## TDI01 inhibited ROCK2 activity in LPS-induced ALI/ARDS and TNFα-stimulated HUVECs
Previous studies have noted that RhoA GTPase and its effectors (ROCK1, ROCK2) inhibitors offer promising targets for ALI/ARDS [20]. To test whether TDI01 still has a protective role in more severe conditions, we performed pretreatment with TDI01 (200 mg/kg/day) for 3 days before injection of LPS and used a higher dose of LPS (10 mg/kg) to obtain an ALI/ARDS model with more severe inflammation and vascular permeability. Histology (Figure 6A) and inflammatory cytokine expression (Figure 6B) demonstrated an evident reduction in inflammation in the TDI01-pretreated group. In terms of its effect on vascular permeability, we also observed a significant decrease in exosmic FITC-dextran (Figure 6C) and EBD (Figure 6D), which suggested that TDI01 could exert its function in the preservation of vascular integrity. Endothelial ROCK2 activity was significantly inhibited in vivo (Figure 6E). Furthermore, HUVECs were pretreated with TDI01 before TNFα challenge and western blotting analysis showed that ROCK2 activity (Figures 6F, S5A), ICAM1 and VCAM1 (Figures 6G, S5B) were inhibited in a dose-dependent manner, and TDI01 effectively attenuated TNFα-induced monocyte adhesion (Figures 6H, S5C). TDI01 also protected HUVECs against TNFα-induced vascular leakage (Figure 6I). Overall, TDI01 exhibited promising effectiveness for the treatment of ALI/ARDS.
**Figure 6:** *TDI01 could rescue LPS-induced ALI/ARDS and TNFα-stimulated HUVECs. Representative lung sections were stained with hematoxylin and eosin, and lung injury scores were quantified, n=5 per group. Scale bars: 200 µm (left) and 40 µm (right) (A). The relative mRNA expression of inflammatory cytokines (IL6, IL1β, IL10, TNFα) in the TDI01-pretreated WT group was reduced (B). n=4 or 5 per group. Vascular permeability was tested by FITC-dextran (C) and EBD (D), and TDI01 has a protective role in the vascular barrier, as evidenced by decreased extravasation of FITC-dextran (C), n=4 per group, scale bars: 50 µm and Evans blue dye (D), n=5 per group. Endothelial ROCK2 activity was inhibited by TDI01 in WT mice and was assessed by the colocalization of pERM/ERM, pMLC2/MLC2 and pROCK2/ROCK2 (red fluorescent protein) with VE-cadherin (green fluorescent protein) by immunofluorescent staining. Representative images are shown (E). n=3 per group, scale bars: 20 µm. HUVECs were pretreated with 0, 0.3, 1, 3 and 10 µM of TDI01 for 24 h and then treated with 20 ng/ml of TNFα for 12 h. ROCK2 activity was examined by WB (F). The expression of VCAM1 and ICAM1 was assessed by WB (G). HUVECs were pretreated with TDI01 (0, 0.3, 1, 3 µM) for 24 h and then challenged by 20 ng/ml of TNFα for 12 h. Hoechst 33342-labeled monocytes were added to each well and co-incubated for 4 h. Representative images of adhered Hoechst 33342-labeled monocytes are shown (H). n=3 per group, scale bars: 200 µm. HUVECs were pretreated with TDI01 (0, 0.1,0.3,1 µM) for 24 h, stimulated with TNFα (20 ng/ml) for 12 h and then plated for permeability assay. The fluorescence intensity of FITC-dextran was recorded at 0, 5, 15, 30, 45, 60, 120 min (I). n=3 per group. Each bar represents the mean ± SD; *p < 0.05, **p < 0.01 and ***p < 0.001.*
## Discussion
In the present study, we demonstrated that LPS triggered a significant increase in inflammation as well as pulmonary vascular leakage and that endothelial cells acting as a semipermeable barrier played a critical role in the progression of ALI/ARDS. We showed that endothelial FGFR1 signaling was significantly reduced in the LPS-treated WT mouse group. We therefore utilized Fgfr1 iΔEC/iΔEC mice to conditionally knock out endothelial FGFR1 and confirmed that endothelial ROCK2, but not endothelial ROCK1, was activated in Fgfr1 iΔEC/iΔEC mice. Meanwhile, in vitro knockdown of FGFR1 by siRNA activated ROCK2, leading to endothelial dysfunction with increased adhesive properties and permeability in TNFα-treated HUVECs. Furthermore, inhibition of endothelial ROCK2 by AAV Vec-tie-shROCK2 significantly attenuated inflammation and alleviated vascular leakage in Fgfr1 iΔEC/iΔEC mice. In addition, we found that TDI01, a novel selective ROCK2 inhibitor, strongly inhibited ROCK2 activity in LPS-induced ALI/ARDS in vivo and TNFα-treated HUVECs in vitro, suggesting promising therapeutic prospects for ALI/ARDS.
Healthy endothelial cells exhibit an anti-inflammatory phenotype to regulate inflammation. However, when stimulated by LPS, damage-associated molecular patterns (DAMPs) and cytokines, they will transform into an inflammatory phenotype, prone to become vascular leaky and attractive to inflammation [2, 4]. Thus, in this study, we focused on pulmonary endothelial cells and sorted pulmonary ECs from saline or LPS-treated WT mice through magnetic beads to determine transcriptional changes. Our data demonstrated that endothelial FGFR1 was significantly decreased in the LPS-treated WT mouse group. Endothelial FGFR1 has been reported to be an essential element for maintaining vascular homeostasis [11]. However, its mechanism in ALI/ARDS has not yet been elucidated. We next sought to explore the role of endothelial FGFR1 in regulating inflammation and pulmonary vascular permeability. Therefore, we constructed Fgfr1 iΔEC/iΔEC mice with conditional knockout of endothelial FGFR1 for further experiments. A recent study showed that administration of recombinant FGF2 alleviated pulmonary vascular leakage and attenuated the inflammatory response in sepsis-induced ALI by stabilizing adherens junctions [29]. Additionally, FGF10 exhibited its protective function in preservation of alveolar-capillary barrier integrity in high altitude pulmonary edema [30]. The similarity among these studies suggests a protective role of FGFs and is consistent with our finding of aggravated inflammation and pulmonary vascular permeability in Fgfr1 iΔEC/iΔEC mice compared with Fgfr1 loxP/loxP mice. We then sought to identify the downstream targets of FGFR1 and performed RNA-seq analysis of pulmonary ECs from LPS-treated Fgfr1 iΔEC/iΔEC mice and Fgfr1 loxP/loxP mice. Strikingly, our data confirmed that endothelial ROCK2 rather than endothelial ROCK1 was activated not only in Fgfr1 iΔEC/iΔEC mice but also in LPS-treated WT mice, as evidenced by the increased phosphorylation of a variety of substrates, such as the ERM, MYPT1, MLC and autophosphorylation by ROCK2 on S1366, while in vitro knockdown of FGFR1 in HUVECs also showed increased ROCK2 activity. In turn, this finding is consistent with the report that exogenous FGF treatment dependent on FGFR-induced activation of PI3K-Akt-Rac1 signaling inhibits RhoA activity and protects the blood-brain barrier after intracerebral hemorrhage in mice [31]. We also found that decreased FGFR1 and ROCK2 activation made TNFα-treated HUVECs more adhesive to inflammatory cells and more permeable. Overall, our data provided further evidence that deletion of endothelial FGFR1 signaling by upregulating the ROCK2 activity-mediated pathway led to exacerbation of ALI/ARDS.
To determine the role of ROCK2 in LPS-treated Fgfr1 iΔEC/iΔEC mice, we intratracheally injected AAV Vec-tie-shROCK2 for the precise knockdown of ROCK2 in pulmonary ECs. Our data showed that knockdown of ROCK2 significantly alleviated inflammation and vascular leakage in LPS-treated Fgfr1 iΔEC/iΔEC mice and identified endothelial ROCK2 as a promising therapeutic target. Notably, RhoA and its downstream effectors ROCKs (ROCK1 and ROCK2) inhibit dephosphorylation of MLCP and directly phosphorylate MLC to augment the phosphorylation of MLC, resulting in endothelial contraction and subsequent vascular barrier dysfunction [5]. Constructional vascular integrity is supported by tight junctions and adherens junctions, which are linked to the endothelial actin cytoskeleton [4]. Disruption of the endothelial barrier is attributed to the activation of the actin-myosin contractile apparatus, which is regulated by the level of myosin light chain (MLC) phosphorylation. Indeed, phosphorylation is dynamic and regulated by the interaction between calcium/calmodulin-dependent MLC kinase (MLCK, phosphorylation) and MLC phosphatase (MLCP, dephosphorylation) [5, 32]. When MLC is excessively phosphorylated, the tension of actinomyosin contraction pulls endothelial cells apart and contributes to the formation of intercellular gaps.
Due to the extensive expression and complexity of FGF/FGFR1 signaling, we sought a more specific drug for treatment to prevent unwanted side effects. TDI01 is a novel and highly selective ROCK2 inhibitor that is currently being tested in phase I clinical trials for the treatment of idiopathic pulmonary fibrosis and silicosis. Compared with nonselective inhibitors such as Y27632 and fasudil, which target both ROCK1 and ROCK2 [20], TDI01 is expected to have a greater therapeutic effect and fewer side effects. Our data demonstrated that TDI01 protected against inflammation and vascular leakage in LPS-induced ALI/ARDS. In addition, TDI01 rescued endothelial dysfunction by downregulating the expression of VCAM1 and ICAM1, attenuating inflammatory cell adhesion and decreasing permeability. The present study used pretreatment before LPS injection, and TDI01 successfully achieved good preventative and curative effects in alleviating vascular leakage and inflammation, indicating its potential for clinical translation.
In summary, our findings suggest that endothelial FGFR1 signaling plays a key role in preserving the integrity of the pulmonary vascular barrier, and reveal that activated ROCK2 contributing to EC contraction and gap formation of neighboring ECs is responsible for the exacerbation of ALI/ARDS in Fgfr1 iΔEC/iΔEC mice. We also found that TDI01, a new selective ROCK2 inhibitor, prevents and treats ALI/ARDS and provides potential valuable insights for the treatment of this disease.
## Data availability statement
The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found below: GSE216635 (GEO).
## Ethics statement
The animal study was reviewed and approved by the committees of West China Second University Hospital, Sichuan University.
## Author contributions
This study was designed and drafted by YH, HW, B-SD and CW. YD performed all the experiments and prepared the manuscript. XH performed RNA-seq analysis. WZ analyzed the data. HZ and CM revised the manuscript. All authors contributed to the article, read, and approved the final manuscript. All authors contributed to the article and approved the submitted version.
## Conflict of interest
Author WZ and HW were employed by Beijing Tide Pharmaceutical Co., Ltd.
The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
The reviewer MY declared a shared parent affiliation with the author YD, CW to the handling editor at the time of review.
## Publisher’s note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## Supplementary material
The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fimmu.2023.1041533/full#supplementary-material
## References
1. Matthay MA, Zemans RL, Zimmerman GA, Arabi YM, Beitler JR, Mercat A. **Acute respiratory distress syndrome**. *Nat Rev Dis Primers* (2019) **5** 18. DOI: 10.1038/s41572-019-0069-0
2. Meyer NJ, Gattinoni L, Calfee CS. **Acute respiratory distress syndrome**. *Lancet* (2021) **398**. DOI: 10.1016/S0140-6736(21)00439-6
3. Bellani G, Laffey JG, Pham T, Fan E, Brochard L, Esteban A. **Epidemiology, patterns of care, and mortality for patients with acute respiratory distress syndrome in intensive care units in 50 countries**. *Jama* (2016) **315** 788-800. DOI: 10.1001/jama.2016.0291
4. Millar FR, Summers C, Griffiths MJ, Toshner MR, Proudfoot AG. **The pulmonary endothelium in acute respiratory distress syndrome: Insights and therapeutic opportunities**. *Thorax* (2016) **71**. DOI: 10.1136/thoraxjnl-2015-207461
5. Dudek SM, Garcia JGN. **Cytoskeletal regulation of pulmonary vascular permeability**. *J Appl Physiol* (2001) **91**. DOI: 10.1152/jappl.2001.91.4.1487
6. Birukov KG, Zebda N, Birukova AA. **Barrier enhancing signals in pulmonary edema**. *Compr Physiol* (2013) **3**. DOI: 10.1002/cphy.c100066
7. Ornitz DM, Itoh N. **The fibroblast growth factor signaling pathway**. *Wiley Interdiscip Reviews-Developmental Biol* (2015) **4**. DOI: 10.1002/wdev.176
8. Powers CJ, McLeskey SW, Wellstein A. **Fibroblast growth factors, their receptors and signaling**. *Endocrine-Related Cancer* (2000) **7**. DOI: 10.1677/erc.0.0070165
9. Chen P-Y, Qin L, Tellides G, Simons M. **Fibroblast growth factor receptor 1 is a key inhibitor of TGF beta signaling in the endothelium**. *Sci Signaling* (2014) **7**. DOI: 10.1126/scisignal.2005504
10. Ling L, Wang Q, Qian K, Cao Z, Xiao J, Wang X. **bFGF Protects Against Oxygen Glucose Deprivation/Reoxygenation-Induced Endothelial Monolayer Permeability via S1PR1-Dependent Mechanisms**. *Mol Neurobiol* (2017) **55**. DOI: 10.1007/s12035-017-0544-0
11. Murakami M, Nguyen LT, Zhang ZW, Moodie KL, Carmeliet P, Stan RV. **The FGF system has a key role in regulating vascular integrity**. *J Clin Invest* (2008) **118**. DOI: 10.1172/JCI35298
12. De Smet F, Tembuyser B, Lenard A, Claes F, Zhang J, Michielsen C. **Fibroblast growth factor signaling affects vascular outgrowth and is required for the maintenance of blood vessel integrity**. *Chem Biol* (2014) **21**. DOI: 10.1016/j.chembiol.2014.07.018"10.1016/j.chembiol.2014.07.018
13. Chen P, Tang H, Zhang Q, Xu L, Zhou W, Hu X. **Basic fibroblast growth factor (bFGF) protects the blood-brain barrier by binding of FGFR1 and activating the ERK signaling pathway after intra-abdominal hypertension and traumatic brain injury**. *Med Sci Monitor* (2020) **26**. DOI: 10.12659/msm.922009"10.12659/MSM.922009
14. Yu Z, Lin L, Jiang Y, Chin I, Wang X, Li X. **Recombinant FGF21 protects against blood-brain barrier leakage through Nrf2 upregulation in type 2 diabetes mice**. *Mol Neurobiol* (2019) **56**. DOI: 10.1007/s12035-018-1234-2
15. Oladipupo SS, Smith C, Santeford A, Park C, Sene A, Wiley LA. **Endothelial cell FGF signaling is required for injury response but not for vascular homeostasis**. *Proc Natl Acad Sci United States America* (2014) **111**. DOI: 10.1073/pnas.1324235111
16. He K, Han S, An L, Zhang Z. **Inhibition of MicroRNA-214 alleviates lung injury and inflammation via increasing FGFR1 expression in ventilator-induced lung injury**. *Lung* (2021) **199** 63-72. DOI: 10.1007/s00408-020-00415-5
17. Amano M, Nakayama M, Kaibuchi K. **Rho-Kinase/ROCK: A key regulator of the cytoskeleton and cell polarity**. *Cytoskeleton* (2010) **67**. DOI: 10.1002/cm.20472
18. Amin E, Dubey BN, Zhang SC, Gremer L, Dvorsky R, Moll JM. **Rho-kinase: regulation, (dys) function, and inhibition**. *Biol Chem* (2013) **394**. DOI: 10.1515/hsz-2013-0181
19. Mikelis CM, Simaan M, Ando K, Fukuhara S, Sakurai A, Amornphimoltham P. **RhoA and ROCK mediate histamine-induced vascular leakage and anaphylactic shock**. *Nat Commun* (2015) **6**. DOI: 10.1038/ncomms7725
20. Abedi F, Hayes AW, Reiter R, Karimi G. **Acute lung injury: The therapeutic role of rho kinase inhibitors**. *Pharmacol Res* (2020) **155**. DOI: 10.1016/j.phrs.2020.104736
21. Takeda Y, Matoba K, Kawanami D, Nagai Y, Akamine T, Ishizawa S. **ROCK2 regulates monocyte migration and cell to cell adhesion in vascular endothelial cells**. *Int J Mol Sci* (2019) **20**. DOI: 10.3390/ijms20061331
22. Beckers CML, Knezevic N, Valent ET, Tauseef M, Krishnan R, Rajendran K. **ROCK2 primes the endothelium for vascular hyperpermeability responses by raising baseline junctional tension**. *Vasc Pharmacol* (2015) **70** 45-54. DOI: 10.1016/j.vph.2015.03.017
23. Matute-Bello G, Downey G, Moore BB, Groshong SD, Matthay MA, Slutsky AS. **An official American thoracic society workshop report: Features and measurements of experimental acute lung injury in animals**. *Am J Respir Cell Mol Biol* (2011) **44**. DOI: 10.1165/rcmb.2009-0210ST
24. Matthay MA, Ware LB, Zimmerman GA. **The acute respiratory distress syndrome**. *J Clin Invest* (2012) **122**. DOI: 10.1172/JCI60331
25. Birukova AA, Smurova K, Birukov KG, Kaibuchi K, Garcia JGN, and Verin AD. **Role of rho GTPases in thrombin-induced lung vascular endothelial cells barrier dysfunction**. *Microvascular Res* (2004) **67** 64-77. DOI: 10.1016/j.mvr.2003.09.007
26. Riento K, Ridley AJ. **Rocks: Multifunctional kinases in cell behaviour**. *Nat Rev Mol Cell Biol* (2003) **4**. DOI: 10.1038/nrm1128
27. Kuemper S, Mardakheh FK, McCarthy A, Yeo M, Stamp GW, Paul A. **Rho-associated kinase (ROCK) function is essential for cell cycle progression, senescence and tumorigenesis**. *Elife* (2016) **5**. DOI: 10.7554/eLife.12203
28. Laudes IJ, Guo RF, Riedemann NC, Speyer C, Craig R, Sarma JV. **Disturbed homeostasis of lung intercellular adhesion molecule-1 and vascular cell adhesion molecule-1 during sepsis**. *Am J Pathol* (2004) **164**. DOI: 10.1016/S0002-9440(10)63230-0
29. Pan X, Summers C, Griffiths MJ, Toshner MR, Proudfoot AG. **Fibroblast growth factor-2 alleviates the capillary leakage and inflammation in sepsis**. *Mol Med* (2020) **26**. DOI: 10.1186/s10020-020-00221-y
30. She J, Xu S, Zhou Z, Wang F, Mao L, Li H. **KGF-2 targets alveolar epithelia and capillary endothelia to reduce high altitude pulmonary oedema in rats**. *J Cell Mol Med* (2012) **16**. DOI: 10.1111/j.1582-4934.2012.01588.x
31. Huang B, Goolaerts A, Shen J, Bi J, Tong L, Gao L. **Fibroblast growth factors preserve blood-brain barrier integrity through RhoA inhibition after intracerebral hemorrhage in mice**. *Neurobiol Dis* (2012) **46**. DOI: 10.1016/j.nbd.2012.01.008
32. Somlyo AP, Somlyo AV. **Ca2+ sensitivity of smooth muscle and nonmuscle myosin II: Modulated by G proteins, kinases, and myosin phosphatase**. *Physiol Rev* (2003) **83**. DOI: 10.1152/physrev.00023.2003
|
---
title: Leptin/adiponectin ratio as a prognostic factor for increased weight gain in
girls with central precocious puberty
authors:
- Jessie Nallely Zurita-Cruz
- Miguel Angel Villasís-Keever
- Leticia Manuel-Apolinar
- Leticia Damasio-Santana
- Eulalia Garrido-Magaña
- Aleida de Jesús Rivera-Hernández
journal: Frontiers in Endocrinology
year: 2023
pmcid: PMC10036755
doi: 10.3389/fendo.2023.1101399
license: CC BY 4.0
---
# Leptin/adiponectin ratio as a prognostic factor for increased weight gain in girls with central precocious puberty
## Abstract
### Objective
To determine if the leptin, adiponectin, and leptin/adiponectin ratio (LAR) can predict weight gain at the end of GnRH analogs (GnRHa) treatment in girls with central precocious puberty (CPP).
### Material and methods
Study design: prospective cohort. Serum levels of leptin and adiponectin were determined at diagnosis of CPP. Anthropometry was performed at diagnosis of CPP and every six-months, until treatment with GnRHa was discontinued and they presented menarche. Patients were divided according to BMI<94 and BMI>95 percentile at diagnosis of CPP. The outcome was the increased in weight gain (e.g., from normal weight to overweight) at the end of follow-up. Statistical analysis: repeated measures ANOVA test and Student’s t-test were used to compare groups. Logistic regression analysis was used to evaluate the association of leptin and adiponectin levels, as well as LAR values with increased weight gain.
### Results
Fifty-six CPP patients were studied, 18 had BMI >95 percentile and 38 BMI <94 percentile. Of the 18 patients who initially had BMI >95th, two patients went from obesity to overweight, while among the 38 patients who started with BMI <94th, 21 ($55.2\%$) increased their weight gain at the end of follow-up. This last group had higher leptin levels (8.99 ± 0.6 vs 6.14 ± 0.8, $$p \leq 0.005$$) and higher LAR values compared to those who remained in the same weight (1.3 ± 0.5 vs 0.96 ± 0.56, $$p \leq 0.01$$). In the logistic regression analysis, it was found that higher leptin levels and higher LAR values were associated with increased weight gain (RR 1.31, $95\%$CI 1.03-1.66, RR 4.86, $95\%$CI 1.10-21.51, respectively), regardless of birth weight, pubertal stage, age, and bone/chronological age ratio.
### Conclusions
In patients with CPP, leptin levels and higher LAR values appear to be associated with significantly greater weight gain during GhRHa treatment, particularly in girls starting with BMI < 94 percentile.
## Introduction
Central precocious puberty (CPP) is defined as the development of sexual characteristics before 9 years of age in boys and 8 years of age in girls due to activation of the hypothalamic-pituitary-gonadal axis [1, 2]. CPP is a rare disease, with an estimated incidence of 1:10,000, but predominantly occurs in girls in more than $95\%$ of cases [3, 4]. There are many theories regarding the etiology of idiopathic CPP. One of which points out that these girls have a higher percentage of body fat for their chronological age, where leptin has a main role [5]. Leptin acts through its receptor to stimulate the secretion of kisspeptin, a hypothalamic hormone, which in turn promotes the secretion of GnRH in the arcuate nucleus [6, 7].
It is common for girls with CPP to be overweight or obese at the time of diagnosis, but there are conflicting results as to whether weight is increased during or after receiving treatment with GnRH analogs (GnRHa), because there are studies indicating that the BMI increases and in others that it decreases (8–13). However, CPP is considered to be a risk factor for the development of cardiometabolic disorders in adult life, regardless of the presence of obesity [14, 15].
Adipose tissue is considered an endocrine organ that produces a wide variety of biologically active adipokines, such as leptin, adiponectin, tumor necrosis factor-α and interleukin-6 [16].
These adipokines plays a pathophysiological link between dysfunctional adipose tissue and cardiometabolic factors [17]. Leptin is produced proportionally to the amount of body fat, and in patients with obesity there is an association of high leptin levels with cardiometabolic factors and metabolic syndrome [18, 19], while high levels of adiponectin are associated with a better metabolic profile in these patients [20]. However, it has been observed that the leptin/adiponectin ratio (LAR) identify dysfunctional adipose tissue more accurately [21, 22]. Thus, higher LAR values have been considered a better marker of insulin resistance and metabolic syndrome than serum leptin or adiponectin levels [23]. Although the possible role of LAR in girls with CPP has not been studied, it is hypothesized that higher LAR values are a better predictor of weight gain than high leptin levels or low adiponectin levels.
The aim of this study was to determine if the leptin, adiponectin levels, and LAR can predict weight gain at the end of GnRha treatment in girls with CPP.
## Subjects
A prospective cohort study was performed from January 2012 to May 2019 at a tertiary care pediatric center in Mexico City. For 36 months, we followed a cohort of Mexican girls <8 years of age with CPP. All included patients were selected using a consecutive sampling technique. At the time of diagnosis, subjects were classified as Tanner stages II and III. Follow-up started at the time of CPP diagnosis and initiation of leuprolide treatment. We excluded patients with any other disease or therapy associated with weight gain or increased appetite, such as intracranial tumors, Cushing disease, genetic syndromes (e.g., Prader-Willi, Bardet-Biedl, or Alstrom), use of steroids, fluoxetine, insulin sensitizers, hyporexigens, growth hormone, intestinal fat absorption inhibitors, or low birthweight [24].
We found 72 patients who fulfilled the inclusion criteria; but nine patients were excluded: three due to congenital adrenal hyperplasia, four due to low birthweight, and two because the parents did not agree to their daughter participating in the study. Finally, seven patients were eliminated because they were lost to follow-up. Thus, 56 patients were analyzed.
Leuprolide (GnRha) treatment consisted of monthly intramuscular application from diagnosis, discontinuing when patients reached a chronological age of 11 to 12 years, bone age 13 years or older, and growth velocity slowed to <4 cm/year (25–27). Subsequently, the patients continued to be monitored, until they presented menarche, which was the final moment of follow-up.
The research protocol was approved by the hospital ethics and research committee. Parents signed the informed consent and child assent according to the recommendations of the Declaration of Helsinki.
## Definitions
Diagnosis of CPP was made according to the following clinical criteria: Tanner breast stage II or higher, height acceleration, advancement of bone age [28] and confirmed by GnRHa stimulation test. This test is carried out with the application of 3.75 of GnRHa (leuprolide) and, after two hours, luteinizing hormone (LH) levels are measured [29]. CPP is diagnosed with LH levels >7 mU/ml [29, 30]. Adequate suppression of pituitary−gonadal function was defined as a stimulated plasma LH level after GnRH stimulation <6.5 mU/ml at 3, 12, 24 and 36 months, after treatment initiation with GnRHa [29].
Increase in weight gain and in BMIz, until the last evaluation were the primary outcomes measures. A BMI <84 percentile was considered normal weight, while a BMI between 84 and 95 percentile was considered overweight, and obesity when BMI was >95 percentile [31].
## Serum leptin and adiponectin measurements
Twelve-hour fasting serum leptin levels were measured between 7:00 and 8:00 a.m. using venipuncture samples at study onset and at a 12-month follow-up. Plasma samples were frozen at -20°C and analyzed. Leptin and adiponectin levels were measured using an enzyme-linked immunosorbent assay (ELISA) (Human Leptin DuoSet DY 398, R&D Systems, Minneapolis, MN, USA) (Human Adiponectin DuoSet DY 1065, R&D Systems, Minneapolis, MN, USA). All ELISA experiments were determined using Finstruments Multiskan EX (MTX Lab Systems Inc., Vienna, VA, USA) in duplicate per the manufacturer’s recommendations. The intra- and inter-assay coefficients of variation for all measurements were <$7\%$. A standard curve was also included within each assay. The LAR was obtained by dividing the serum concentrations of leptin by those of adiponectin.
## Anthropometry
The patients’ anthropometric measurements were noted by a certified nutritionist and included height, weight, and body fat percentage by bioimpedance (Tanita BC-568 segmental analyzer, Tokyo, Japan). These anthropometric data were assessed every six months until the end of the follow-up.
## Statistics analyses
The Shapiro-Wilk test was applied to the quantitative variables, and a nonparametric distribution was observed. The quantitative variables were normalized by taking the logarithmic, except LAR distribution was normalized by taking the square root. We calculated the mean and standard error (SE) of quantitative variables.
Since patients with normal weight and overweight had increased weight gain compared to those with obesity, at the end of follow-up, two groups were formed according to the baseline BMI percentile (<94 and >95), to carry out all the analyses.
To determine differences in BMIz at study onset and after 12, 24 months, end of treatment and menarche, statistical analysis was performed using repeated measures ANOVA test and Student’s t-test to compare groups.
A logistic regression analysis was used to determine the association of LAR with increased weight gain, adjusted by birthweight, pubertal stage, age, and bone/chronological age ratio.
A p value <0.05 was considered statistically significant. STATA v.14.0 was used for all statistical analyses.
## Baseline
At diagnosis, the mean age of the 56 patients was 7.0 ± 0.18 years, mean bone age was 10.0 ± 0.26 years, and all patients had a bone/chronological age ratio >1. Of the total, 33 patients ($58.9\%$) had puberty onset in Tanner breast stage II, and 23 patients ($41.1\%$) were in breast stage III (see Table 1). The BMIz mean was 1.22 ± 0.12; 25 patients ($44.6\%$) had normal BMIz, 13 patients ($23.2\%$) had overweight, and 18 patients ($32.1\%$) had obesity (Table 1 and Figure 1).
## Follow-up
During the follow-up, 22 patients ($40\%$) discontinued leuprolide at 24 months of treatment, and 34 patients at 36 months. In all patients, mean BMIz increased throughout follow-up (Figure 1A); at the beginning it was 1.22 ± 0.12, at 12 months was 1.35 ± 0.11 ($$p \leq 0.053$$), at 24 months 1.39 ± 0.11 ($$p \leq 0.069$$), while at the end of leuprolide was 1.46 ± 0.12 ($$p \leq 0.006$$), finally, when menarche occurred, between 12 and 18 months after stopping leuprolide, the BMIz was 1.58 ± 0.17 ($$p \leq 0.157$$). However, when comparing those with BMI <94 and >95 percentile at diagnosis, it was observed that the first group had an increase in BMIz (mean 0.73 vs. 1.27, $$p \leq 0.01$$) (Figure 1B), contrary to a decrease in the second group (mean 2.26 vs. 2.10; $$p \leq 0.15$$) (Figure 1C).
Likewise, of the 18 patients who initially had BMI >95th only two patients went from obesity to overweight (Figure 2B). While among the 38 patients who started with BMI <94th, 21 ($55.2\%$) increase in weight gain; as shown in Figure 2A, in the last assessment 15 were classified as obese, 10 as overweight, and 13 as normal weight.
**Figure 2:** *(A) Modification of nutrition status in girls with BMI<94 percentile at diagnosis (B) Modification of nutrition status in girls with BMI>95 percentile at diagnosis.*
## Adipocytokines
At study onset, the mean leptin value was 8.3± 0.47 ng/ml, and levels were similar according to pubertal developmental stage (Tanner stage II, 8.04 ± 0.60 ng/ml vs Tanner stage III, $$p \leq 0.220$$), but as shown in Table 1, in the BMI <94 group the mean was lower than in the BMI >95 group ($$p \leq 0.020$$). And as for adiponectin, the BMI >95 group had lower values than the BMI <94 group, ($$p \leq 0.020$$). When analyzing the patients with BMI <94th at baseline, among the 21 patients who had a increase in weight gain, leptin levels were higher than the group of 17 patients who remained unchanged (8.99 ± 0.6 vs 6.14 ± 0.8, $$p \leq 0.005$$), but the levels of adiponectin were similar between both groups (7.00 ± 0.4 vs 7.16 ± 0.3, $$p \leq 0.61$$).
Regarding LAR at diagnosis, in the 56 patients the mean value was 1.37± 0.11; but in the BMI <94th girls was lower (mean 1.15 ± 0.09), compared to the BMI >95th group (mean 1.84 ± 0.26), $$p \leq 0.012.$$ As shown in Figure 3A, in the 21 patients with increased in weight gain, the values were statistically significantly higher, compared to those of the 17 patients who remained unchanged, 1.30 ± 0.5 vs 0.96 ± 0.5, $$p \leq 0.01.$$ *The data* were similar when comparing the values in those who started with normal weight (Figure 3B) or with overweight (Figure 3C), but the LAR mean value was higher in this last group (1.48 ± 0.16).
**Figure 3:** *(A) Leptin/adiponectin ratio (LAR) values in CCP girls with BMI <94 percentile and who had an increased weight gain (B) LAR values in CPP girls with BMI <84 percentile and who had increased in weight gain (C) LAR values in CPP girls with BMI >85 and <94 percentile and who had an increased in weight gain.*
Finally, multivariate analyses are presented in Tables 2, 3 performed on the 38 patients with BMI<94th; as shown, both leptin levels and LAR were significantly associated with increased in weight gain, regardless of birthweight, pubertal stage, age, and bone/chronological age ratio; but the strength of association of LAR was greater than leptin levels, RR 1.31, $95\%$CI 1.03-1.66 vs RR 4.86, $95\%$CI 1.10-21.51, respectively.
## Discussion
Different studies have shown that girls with CPP have a higher risk of weight gain during treatment with GnRHa, as well as having a more adverse cardiometabolic profile, but information on possible associated factors is lacking. To our knowledge, this is the first study to identify LAR as a prognostic marker associated with increased in weight gain in patients with CPP.
Our findings may be relevant, since there is controversy about whether the greater adiposity that CPP patients have at diagnosis may be the main factor related to in the increased cardiovascular risk. *In* general, these patients tend to increase zBMI, but when analyzed according to their baseline nutritional status, it has been found that CPP girls with normal weight at diagnosis, their zBMI increases more than when they are classified as overweight/obese [8, 9]. Park J et al. included 59 patients, of whom $35.6\%$ were overweight or obese; during follow-up until final height was reached, no change in zBMI was observed, but in patients who started overweight/obese, there was a decrease in zBMI ($p \leq 0.05$) [13]. Similarly, in this study, the increase in zBMI was statistically significant in the 38 patients with BMI < 94th (Figure 1B), and almost all of those with BMI >95th remained in the same zBMI.
Adipose tissue is metabolically active and secretes adipokines, such as leptin and adiponectin; the former causes vascular inflammation and insulin resistance, while the latter inhibits adherence molecules and increases the production of anti-inflammatory cytokines, such as IL-10 (32–34). In addition, leptin has been shown to be useful prognostic factor in weight gain and in the development of type 2 diabetes mellitus [35, 36]. LAR represents a marker of the pathophysiological function of both adipokines and may indicate an imbalance in proinflammatory and anti-inflammatory conditions, as well as adipose tissue dysfunction [37]. Under normal conditions, the ratio of leptin and adiponectin is 1:2, which means LAR values of 0.5; thus, higher values have been associated with an increase in cardiovascular risk [21].
To verify the relationship between leptin and adiponectin, studies have been carried out in human white preadipocytes. Singh et al. reported that leptin may regulate adiponectin mRNA via extracellular signal-regulated kinase (ERK)-dependent activation of signal transducer and activator of transcription 3 (STAT3), but in obese people these pathways are altered [38].
Adipose tissue dysfunction has been regarded as a form of oxidative stress, in which thiobarbituric acid reactive sub-stances (TBARS) are increased, as well as C-reactive protein (CRP), serum amyloid A (SAA) and osteopontin (OPN) levels (39–42). In adult patients with metabolic syndrome, a positive correlation of LAR levels with CRP has been described, suggesting that adipose tissue dysfunction is related to elevated LAR values [23]. This information is consistent with findings from more recent studies in which LAR values could be a more efficient predictor of obesity-related complications compared to leptin or adiponectin levels [41, 43, 44].
LAR values have also been evaluated in children. Some studies in patients without comorbidities with 6-years of follow-up have reported that high leptin levels and LAR are predictors in nonobese subjects. However, it is controversial which of the two is a better predictor; Zhang et al, identified leptin as being better than LAR in change of zBMI (leptin β=0.209 and LAR β=0.146), while Li et al, observed that LAR was better than leptin in change of body fat percentage (leptin β=0.310 and LAR β=0.420) [45, 46]. Among girls with CCP we also found that leptin levels and LAR are associated with increased in weight gain, but LAR seems a better marker (Tables 2, 3). In contrast, Yoo JM et al. reported that neither leptin, adiponectin, nor LAR were predictors of weight gain girls with CPP [47]; this difference could be due to a shorter follow-up time, or because the analysis was not limited to patients with greater weight gain, as in the present study.
Therefore, we consider that, at the diagnosis of CPP, leptin levels and LAR can be used as a prognostic marker to identify those girls who are at greater risk of significantly increasing their weight during treatment with GnRHa. This would help to offer early interventions to reduce the probability of developing comorbidities related to obesity in adulthood, such as metabolic syndrome, polycystic ovarian syndrome, and type 2 diabetes [48, 49].
As for the limitations of the study, we should mention that the sample size was small, so it seems necessary to carry out more studies to verify our findings. Likewise, in other studies, an attempt should be made to establish the LAR cut-off point that identifies patients with a higher risk of increasing their zBMI.
## Conclusions
In patients with CPP, higher leptin levels and LAR values appear to be associated with significantly greater weight gain during GhRHa treatment, particularly in girls starting with BMI < 94 percentile.
## Data availability statement
The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.
## Ethics statement
The studies involving human participants were reviewed and approved by the research protocol by the Hospital de Pediatria National Medical Center XXI Century, Instituto Mexicano del Seguro Social ethics and research committee. Written informed consent to participate in this study was provided by the participants’ legal guardian/next of kin.
## Author contributions
Conceptualization, methodology & formal analysis: MV-K & JZ-C. Investigation: JZ-C, LM-A, LD-S, EG-M, AR-H. Writing, review & editing: MV-K & JZ-C. All authors contributed to the article and approved the submitted version.
## Conflict of interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher’s note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## References
1. Shankar R PO. **Precocious puberty**. *Adv Endocrinol Metab* (1995) **6** 55-89. PMID: 7671102
2. Kaplowitz P, Oberfield S. **Reexamination of the age limit for defining when puberty is precocious in girls in the united states: Implications for evaluation and treatment**. *Pediatrics* (1999) **104**. DOI: 10.1542/peds.104.4.936
3. Bridges NA, Christopher JA, Hindmarsh PC, Brook CGE. **Sexual precocity: Sex incidence and aetiology**. *Arch Dis Child* (1994) **70**. DOI: 10.1136/adc.70.2.116
4. Partsch CJ, Heger S, Sippell WGE. **Management and outcome of central precocious puberty**. *Clin Endocrinol* (2002) **56**. DOI: 10.1046/j.0300-0664.2001.01490.x
5. Friedman JM, Halaas JL. **Leptin and the regulation of body weight in mammals**. *Nature* (1998) **395**. DOI: 10.1038/27376
6. Wakabayashi Y, Nakada T, Murata K, Ohkura S, Mogi K, Navarro VM. **Neurokinin b and dynorphin a in kisspeptin neurons of the arcuate nucleus participate in generation of periodic oscillation of neural activity driving pulsatile gonadotropin-releasing hormone secretion in the goat**. *J Neurosci* (2010) **30**. DOI: 10.1523/JNEUROSCI.5848-09.2010
7. Donato J, Cravo RM, Frazão R, Gautron L, Scott MM, Lachey J. **Leptin’s effect on puberty in mice is relayed by the ventral premammillary nucleus and does not require signaling in Kiss1 neurons**. *J Clin Invest* (2011) **121**. DOI: 10.1172/JCI45106
8. Kim HR, Nam H-K, Rhie Y-J, Lee K-H. **Treatment outcomes of gonadotropin-releasing hormone agonist in obese girls with central precocious puberty**. *Ann Pediatr Endocrinol Metab* (2017) **22**. DOI: 10.6065/apem.2017.22.4.259
9. Yang WJ, Ko KH, Lee KH, Hwang IT, Oh YJ. **The different effects of gonadotropin-releasing hormone agonist therapy on body mass index and growth between normal-weight and overweight girls with central precocious puberty**. *Ann Pediatr Endocrinol Metab* (2017) **22** 49. DOI: 10.6065/apem.2017.22.1.49
10. Yoon JW, Park HA, Lee J, Kim JH. **The influence of gonadotropin-releasing hormone agonists on anthropometric change in girls with central precocious puberty**. *Korean J Pediatr* (2017) **60** 395-402. DOI: 10.3345/kjp.2017.60.12.395
11. Arrigo T, De Luca F, Antoniazzi F, Galluzzi F, Segni M, Rosano M. **Reduction of baseline body mass index under gonadotropin-suppressive therapy in girls with idiopathic precocious puberty**. *Eur J Endocrinol* (2004) **150**. DOI: 10.1530/eje.0.1500533
12. Sinthuprasith P, Dejkhamron P, Wejaphikul K, Unachak KE. **Near final adult height, and body mass index in overweight/obese and normal-weight children with idiopathic central precocious puberty and treated with gonadotropin-releasing hormone analogs**. *J Pediatr Endocrinol Metab* (2019) **32**. DOI: 10.1515/jpem-2019-0136
13. Park J, Hwang TH, Kim YD, Han HS. **Longitudinal follow-up to near final height of auxological changes in girls with idiopathic central precocious puberty treated with gonadotropin-releasing hormone analog and grouped by pretreatment body mass index level**. *Ann Pediatr Endocrinol Metab* (2018) **23** 14-20. DOI: 10.6065/apem.2018.23.1.14
14. Lee HS, Yoon JS, Roh JK, Hwang JS. **Changes in body mass index during gonadotropin-releasing hormone agonist treatment for central precocious puberty and early puberty**. *Endocrine* (2016) **54** 497-503. DOI: 10.1007/s12020-016-1023-5
15. Colmenares A, Gunczler P, Lanes R. **Higher prevalence of obesity and overweight without an adverse metabolic profile in girls with central precocious puberty compared to girls with early puberty, regardless of GnRH analogue treatment**. *Int J Pediatr Endocrinol* (2014) **2014** 5. DOI: 10.1186/1687-9856-2014-5
16. Crewe C, An YA, Scherer PEE. **The ominous triad of adipose tissue dysfunction: Inflammation, fibrosis and impaired angiogenesis**. *J Clin Invest* (2017) **127** 74-82. DOI: 10.1172/JCI88883
17. Unamuno X, Gómez-Ambrosi J, Rodríguez A, Becerril S, Frühbeck G, Catalán VE. **Adipokine dysregulation and adipose tissue inflammation in human obesity**. *Eur J Clin Invest* (2018) **48**. DOI: 10.1111/eci.12997
18. Rodríguez A, Ezquerro S, Méndez-Giménez L, Becerril S, Frühbeck G. **Revisiting the adipocyte: A model for integration of cytokine signaling in the regulation of energy metabolism**. *Am J Physiol Endocrinol Metab* (2015) **309**. DOI: 10.1152/ajpendo.00297.2015
19. Blüher M, Mantzoros CS. **From leptin to other adipokines in health and disease: Facts and expectations at the beginning of the 21st century**. *Metabolism* (2015) **64**. DOI: 10.1016/j.metabol.2014.10.016
20. Stern JH, Rutkowski JM, Scherer PE. **Adiponectin, leptin, and fatty acids in the maintenance of metabolic homeostasis through adipose tissue crosstalk**. *Cell Metab* (2016) **23**. DOI: 10.1016/j.cmet.2016.04.011
21. Frühbeck G, Catalán V, Rodríguez A, Gómez-Ambrosi J. **Adiponectin-leptin ratio: A promising index to estimate adipose tissue dysfunction. relation with obesity-associated cardiometabolic risk**. *Adipocyte* (2018) **7** 57-62. DOI: 10.1080/21623945.2017.1402151
22. Vega GL, Grundy SM. **Metabolic risk susceptibility in men is partially related to adiponectin/leptin ratio**. *J Obes* (2013) **2013** 409679. DOI: 10.1155/2013/409679
23. Frühbeck G, Catalán V, Rodríguez A, Ramírez B, Becerril S, Salvador J. **Involvement of the leptin-adiponectin axis in inflammation and oxidative stress in the metabolic syndrome**. *Sci Rep* (2017) **7** 6619. DOI: 10.1038/s41598-017-06997-0
24. Miras M, Ochetti M, Martín S, Silvano L, Sobrero G, Castro L. **Serum levels of adiponectin and leptin in children born small for gestational age: Relation to insulin sensitivity parameters**. *J Pediatr Endocrinol Metab* (2010) **23**. DOI: 10.1515/jpem.2010.077
25. Heger S, Partsch CJ, Peter M, Blum WF, Kiess W, Sippell WG. **Serum leptin levels in patients with progressive central precocious puberty**. *Pediatr Res* (1999) **46**. DOI: 10.1203/00006450-199907000-00012
26. Pasquino AM, Pucarelli I, Accardo F, Demiraj V, Segni M, Di Nardo R. **Long-term observation of 87 girls with idiopathic central precocious puberty treated with gonadotropin-releasing hormone analogs: Impact on adult height, body mass index, bone mineral content, and reproductive function**. *J Clin Endocrinol Metab* (2008) **93**. DOI: 10.1210/jc.2007-1216
27. Wolters B, Lass N, Reinehr T. **Treatment with gonadotropin-releasing hormone analogues: Different impact on body weight in normal-weight and overweight children**. *Horm Res Paediatr* (2012) **78**. DOI: 10.1159/000346145
28. Tanner JM. **Issues and advances in adolescent growth and development**. *J Adolesc Heal Care* (1987) **8**. DOI: 10.1016/0197-0070(87)90048-9
29. Brito VN, Latronico AC, Arnhold IJP, Mendonca BB. **A single luteinizing hormone determination 2 hours after depot leuprolide is useful for therapy monitoring of gonadotropin-dependent precocious puberty in girls**. *J Clin Endocrinol Metab* (2004) **89**. DOI: 10.1210/jc.2003-031537
30. Carel J-C, Eugster EA, Rogol A, Ghizzoni L, Palmert MR. **Consensus statement on the use of gonadotropin-releasing hormone analogs in children**. *Pediatrics* (2009) **123**. DOI: 10.1542/peds.2008-1783
31. 31
Centers for Disease Control and Prevention. CDC Growth charts (2017). Available at: https://www.cdc.gov/nchs/data/series/sr_11/sr11_246.pdf.. *CDC Growth charts* (2017)
32. Ouchi N WK. **Adiponectin as an anti-inflammatory factor**. *Clin Chim Acta* (2007) **380** 24-30. DOI: 10.1016/j.cca.2007.01.026
33. Candido GZ, Silva ILZ, Martins LT, Koczicki L, Kubo KS, Frigeri HR. **Specific obesity-related adipokines**. *Immunochem Immunopathol* (2015) **01** 108. DOI: 10.4172/2469-9756.1000108
34. Savoye M, Dziura J, Castle J, DiPietro L, Tamborlane WV, Caprio SE. **Importance of plasma leptin in predicting future weight gain in obese children: A two-and-a-half-year longitudinal study**. *Int J Obes Relat Metab Disord* (2002) **26**. DOI: 10.1038/sj.ijo.0802018
35. Chessler SD, Fujimoto WY, Shofer JB, Byoko EJ, Weigle DSE. **Increased plasma leptin levels are associated with fat accumulation in Japanese-americans**. *Diabetes* (1998) **47** 239-243. DOI: 10.2337/diab.47.2.239
36. Weber DR, Leonard MB, Shults J, Zemel BS. **A comparison of fat and lean body mass index to BMI for the identification of metabolic syndrome in children and adolescents**. *J Clin Endocrinol Metab* (2014) **99**. DOI: 10.1210/jc.2014-1684
37. Singh P, Sharma P, Sahakyan KR, Davison DE, Sert-Kuniyoshi FH, Romero-Corral A. **Differential effects of leptin on adiponectin expression with weight gain versus obesity**. *Int J Obes* (2016) **40**. DOI: 10.1038/ijo.2015.181
38. Gómez-Ambrosi J, Salvador J, Rotellar F, Silva C, Catalán V, Rodríguez A. **Increased serum amyloid a concentrations in morbid obesity decrease after gastric bypass**. *Obes Surg* (2006) **16**. DOI: 10.1381/096089206776116525
39. Gómez-Ambrosi J, Catalán V, Ramírez B, Rodríguez A, Colina I, Silva C. **Plasma osteopontin levels and expression in adipose tissue are increased in obesity**. *J Clin Endocrinol Met* (2007) **92**. DOI: 10.1210/jc.2007-0349
40. Frühbeck G, Catalán V, Rodríguez A, Ramírez B, Becerril S, Salvador J. **Adiponectin-leptin ratio is a functional biomarker of adipose tissue inflammation**. *Nutrients* (2019) **11** 454. DOI: 10.3390/nu11020454
41. Gómez-Ambrosi J, Azcona C, Patiño-García A, Frühbeck G. **Serum amyloid a concentration is increased in obese children and adolescents**. *J Pediatr* (2008) **153**. DOI: 10.1016/j.jpeds.2008.01.006
42. Finucane FM, Luan J, Wareham NJ, Sharp SJ, O’Rahilly S, Balkau B. **Correlation of the leptin: Adiponectin ratio with measures of insulin resistance in non-diabetic individuals**. *Diabetologia* (2009) **52**. DOI: 10.1007/s00125-009-1508-3
43. Thorand B, Zierer A, Baumert J, Meisinger C, Herder C, Koenig WE. **Associations between leptin and the leptin/adiponectin ratio and incident type 2 diabetes in middle-aged men and women: Results from the MONICA/KORA augsburg study 1984–2002**. *Diabetes Med* (2010) **27**. DOI: 10.1111/j.1464-5491.2010.03043.x
44. Zhang M, Cheng H, Zhao X, Hou D, Yan Y, Cianflone K. **Leptin and leptin-to-Adiponectin ratio predict adiposity gain in nonobese children over a six-year period**. *Child Obes* (2017) **13**. DOI: 10.1089/chi.2016.0273
45. Li S, Liu R, Arguelles L, Wang G, Zhang J, Shen X. **Adiposity trajectory and its associations with plasma adipokine levels in children and adolescents-a prospective cohort study**. *Obes (Silver Spring)* (2016) **24**. DOI: 10.1002/oby.21378
46. Yoo JW, Song CW, Lim HH. **Leptin and adiponectin levels in girls with central precocious puberty before and during GnRH agonist treatment**. *Ann Pediatr Endocrinol Metab* (2016) **21** 199. DOI: 10.6065/apem.2016.21.4.199
47. Ladeiras-Lopes R, Sampaio F, Bettencourt N, Fontes-Carvalho R, Ferreira N, Leite-Moreira A. **The ratio between visceral and subcutaneous abdominal fat assessed by computed tomography is an independent predictor of mortality and cardiac events**. *Rev Esp Cardiol (Engl Ed)* (2017) **70**. DOI: 10.1016/j.recesp.2016.09.006
48. Chandrasekhar T, Suchitra MM, Pallavi M, Srinivasa Rao PVLN, Sachan A. **Risk factors for cardiovascular disease in obese children**. *Indian Pediatr* (2017) **54**. DOI: 10.1007/s13312-017-1169-0
49. Bray GA, Kim KK. **World Obesity Federation. Obesity: A chronic relapsing progressive disease process. A position statement of the World Obesity Federation**. *Obes Rev* (2017) **18**. DOI: 10.1111/obr.12551
|
---
title: Lower diurnal HPA-axis activity in male hypertensive and coronary heart disease
patients predicts future CHD risk
authors:
- Cathy Degroote
- Roland von Känel
- Livia Thomas
- Claudia Zuccarella-Hackl
- Nadine Messerli-Bürgy
- Hugo Saner
- Roland Wiest
- Petra H. Wirtz
journal: Frontiers in Endocrinology
year: 2023
pmcid: PMC10036761
doi: 10.3389/fendo.2023.1080938
license: CC BY 4.0
---
# Lower diurnal HPA-axis activity in male hypertensive and coronary heart disease patients predicts future CHD risk
## Abstract
### Background
Coronary heart disease (CHD) and its major risk factor hypertension have both been associated with altered activity of the hypothalamus-pituitary-adrenal (HPA)-axis but the biological mechanisms underlying prospective associations with adverse disease outcomes are unclear. We investigated diurnal HPA-axis activity in CHD-patients, hypertensive (HT) and healthy normotensive men (NT) and tested for prospective associations with biological CHD risk factors.
### Methods
Eighty-three male CHD-patients, 54 HT and 54 NT men repeatedly measured salivary cortisol over two consecutive days. Prospective CHD risk was assessed by changes between baseline and follow-up in the prothrombotic factors D-dimer and fibrinogen, the pro-inflammatory measures interleukin (IL)-6, tumor necrosis factor-alpha (TNF-α), and acute phase protein C-reactive protein (CRP), as well as blood lipids in terms of total cholesterol (tChol)/high-density-lipoprotein cholesterol (HDL)-ratio. We aggregated coagulation and inflammatory measures to respective indices.
### Results
The groups differed in repeated daytime cortisol (dayCort) secretion ($$p \leq .005$$,η2 $$p \leq .03$$,$f = 0.18$) and cortisol awakening response (CAR) ($$p \leq .006$$,η2 $$p \leq .03$$,$f = 0.18$), with similarly lower overall dayCort and CAR in CHD-patients and HT, as compared to NT. The groups differed further in cortisol at awakening ($$p \leq .015$$,η2 $$p \leq .04$$,$f = 0.20$) with highest levels in HT (p´s≤.050), and in diurnal slope between waking and evening cortisol ($$p \leq .033$$,η2 $$p \leq .04$$,$f = 0.20$) with steepest slopes in HT (p´s≤.039), although in part not independent of confounders. Lower aggregated dayCort and CAR in terms of area-under-the-curve (AUC) independently predicted increases in future overall CHD risk (AUCdayCort: $$p \leq .021$$,η2 $$p \leq .10$$,$f = 0.33$;AUCCAR: $$p \leq .028$$,η2 $$p \leq .09$$,$f = 0.31$) 3.00 ± 0.06(SEM) years later, with risk prediction most pronounced in fibrinogen (AUCdayCort: $$p \leq .017$$,ΔR 2= 0.12;AUCCAR: $$p \leq .082$$).
### Conclusion
We found evidence for an HPA-axis hypoactivity in CHD and HT with lower diurnal HPA-axis activity predicting increases in cardiovascular risk as evidenced by increases in circulating levels of biomarkers of atherothrombotic risk. Down-regulation of basal HPA-axis activity may contribute to the pathogenesis of atherosclerosis and thrombosis in CHD via effects on coagulation.
## Introduction
Coronary heart disease (CHD) ranks among the leading causes of death in western countries [1]. A major risk factor for CHD is hypertension (HT), a chronic elevation of blood pressure (BP) [2]. Most individuals with hypertension are diagnosed with “essential hypertension” with unknown medical cause [3, 4]. HT and CHD have both been associated with alterations in diurnal activity of the hypothalamus-pituitary-adrenal (HPA)-axis (see below e.g. [5, 6]), but its relevance with respect to mechanisms of disease progression is not fully understood.
The HPA-axis end-product cortisol is important for the integrity of central nervous system function and for maintenance of cardiovascular, metabolic, and immune homeostasis [7]. The basal HPA-axis activity follows a diurnal rhythm characterized by the cortisol awakening response (CAR), a sharp rise by a about $50\%$ to over $100\%$ within the first 30-to-45min after awakening (8–11). This morning peak is followed by a circadian decline with a steady decrease of cortisol throughout the day and minimum levels at midnight [12, 13]. Dysregulated circadian cortisol profiles often comprise either high cortisol levels throughout the day and a flattened diurnal rhythm (HPA-axis hyperactivity), or lower overall cortisol secretion with a flatter diurnal slope and lower morning cortisol levels (HPA-axis hypoactivity) [for review see [14]].
A variety of studies cross-sectionally assessed basal HPA-axis activity in HT and heart patients or individuals with CHD-symptoms, respectively. Basal HPA-axis activity was compared between individuals with HT and normotensives (i.e., a normal blood pressure, NT). Male and female HT who discontinued medication-intake had higher morning cortisol levels compared to NT [15], but not unequivocally [6]. With respect to the CAR, we previously found a blunted CAR in unmedicated HT [6] and also medicated hypertensive men and women showed lower aggregated total CAR compared to NT [16]. Similarly, in predominantly non-hypertensive male and female participants, higher BP related to a lower aggregated total CAR [17]. Regarding diurnal cortisol profiles, we could not previously observe differences between medication-free HT and NT [6]. Evening cortisol was higher in HT compared to NT [15]. In heart disease patients, salivary waking or morning cortisol was not associated with CHD (or markers of subclinical CHD) in most studies (18–20). Patients with cardiovascular disease (CVD) showed flatter CARs than non-CVD participants [21] and hypertensive patients with acute coronary syndrome (ACS) had smaller aggregated CARs than normotensive patients [5], but there are also contradicting findings [18]. With respect to salivary cortisol daytime levels, CHD-patients did not differ in cortisol slopes or total cortisol output over the day from participants without a CHD-diagnosis but symptoms [18]. Regarding associations with markers of subclinical CHD, salivary waking or morning cortisol was not associated with CHD (or markers of subclinical CHD) in most studies (18–20). Various measures reflecting a flatter CAR were associated with higher values of intima-media-thickness [22] and ankle-brachial-index (ABI) [19] in women and coronary artery calcification (CAC) in men and women [19], but not unequivocally [see also [23]]. With respect to diurnal cortisol, a flatter slope over the day was associated with the recurrence of cardiac symptoms in ACS-patients [5] and with higher CAC in population-based studies [24]. Also, higher total cortisol output was related to more carotid plaques [25]. However, there are also diverging diurnal cortisol findings (19, 23–25). Bedtime cortisol levels were higher in ACS-patients compared to healthy controls [20]. Taken together, the above-described cross-sectional findings point to a reduced CAR, flatter slopes over the day, and higher cortisol evening levels in hypertensive individuals as well as in heart patients. However, to the best of our knowledge, it has not yet been investigated whether HT differ from CHD-patients in their basal HPA-axis activity or whether the separately observed dysregulations are comparable in both groups.
Prospective evidence is emerging that a dysregulated basal HPA-axis activity relates to adverse heart disease outcomes, but the underlying biological mechanisms with respect to disease progression are unclear. Lower waking cortisol in patients undergoing coronary artery bypass graft (CABG) surgery predicted cardiac events and death about 3 years later [26]. Moreover, a flatter diurnal cortisol decline predicted cardiac events and death in CABG-patients [26], and CVD-related mortality in Whitehall-II participants [27]. Finally, higher bedtime or evening cortisol levels predicted mortality risk [27, 28], adverse clinical outcomes [26], and incidence of fatal CHD [29], predominantly in heart patients. The mechanisms underlying these prospective associations have rarely been investigated. So far, only two non-patient studies suggest prospective associations that point to the process of atherosclerosis. In policemen, a flatter aggregated CAR predicted a larger 7-year mean decline in brachial artery flow-mediated dilation indicative of endothelial dysfunction [30]. Moreover, in a population-based study, healthy women with flatter diurnal cortisol slopes and higher bedtime levels showed greater progression of aortic stiffness 5 years later [31].
Independent biological CHD risk factors that underly the process of atherosclerosis and thrombosis include markers of coagulation, inflammation, and hyperlipidemia (32–35). To date, despite evidence for cross-sectional associations (20, 36–39), prospective evidence addressing associations between basal HPA-axis activity and independent biological CHD risk factors is lacking so far, not only in healthy participants, but also in hypertension or CHD. A better understanding of the biological mechanisms underlying disease progression may have implications for longer-term therapy in vulnerable populations such as HT and heart patients.
To close the above described gaps in current knowledge, the first objective of our study was to cross-sectionally compare diurnal HPA-axis activity between CHD-patients, HT, and controls with neither HT or CHD (NT). We repeatedly measured salivary cortisol over two consecutive days and hypothesized a blunted CAR, flatter diurnal slopes, and higher evening cortisol levels in both, CHD-patients and HT, as compared to NT. Second, to obtain new mechanistic insights with respect to the clinical relevance of basal HPA-axis activity, we prospectively investigated whether cortisol would predict changes in biological CHD risk factors including markers of coagulation, inflammation, and hyperlipidemia over a mean follow-up of 3 years.
## Study participants
The current investigation is part of a study program assessing psychoneurobiological mechanisms in essential hypertension and CHD (40–43). It was approved by the ethics committee of the Canton of Bern, Switzerland and the study protocol is in accordance with the Declaration of Helsinki. All participants provided written informed consent and were financially compensated for each assessment with CHF 20.
We restricted our study sample to male individuals given the prevalence of cardiovascular diseases at earlier age (e.g. [44, 45]), given sex differences in HPA-axis activity (e.g. [10]), and given the differences between men and women regarding the associations of diurnal cortisol secretion and CHD-symptoms (e.g. [22]). We recruited male participants with either a diagnosis of CHD, apparently healthy individuals with essential hypertension without CHD, or healthy normotensive controls and asked them to provide saliva samples for the assessment of diurnal cortisol profiles. The final cross-sectional sample comprised 191 participants, with 83 CHD-patients, 54 HT (42 medication-free, 12 medicated) as well as 54 NT. All study participants were invited for a follow-up assessment, with 106 subjects (NT:$$n = 32$$, HT:$$n = 31$$, CHD-patients:$$n = 43$$) completing both assessments (after an average of 3yrs (3.00 ± 0.06 SEM). More information is detailed in the Supplemental Material.
Reasons for drop-out at follow-up included could not be reached by phone ($$n = 9$$), lack of time ($$n = 16$$), no interest ($$n = 12$$), excessive demands ($$n = 7$$), severe illness ($$n = 8$$), meanwhile living abroad ($$n = 2$$), discontent with the study management ($$n = 1$$), deceased ($$n = 2$$), or no specific reason given ($$n = 26$$). Further, two participants had to be excluded because of acute infection on the follow-up study day. Notably, due to organizational reasons (relocation of the working group from Bern to Konstanz) the number of available follow-up time-slots per month was substantially reduced compared to the baseline assessment, resulting in a prolonged follow-up time and potentially responsible for the attrition. Attrition was comparable across the study groups (drop-out rate: NT=$40.7\%$; HT=$42.6\%$; CHD=$48.2\%$; Chi2[2]=.85, $$p \leq .66$$).
## Recruitment and general inclusion criteria
CHD-patients. We included patients with stable CHD who had been discharged from the Cardiac Prevention and Rehabilitation Clinic of the Bern University Hospital at least 6 months ago. We asked those patients of the Cardiac Prevention and Rehabilitation Clinic of the Bern University Hospital who volunteered to be contacted for the purpose of scientific studies. All patients were diagnosed with CHD based on coronary angiography and we provide information regarding myocardial infarction (MI), left ventricular ejection fraction (LVEF) ≤ $40\%$, and coronary artery bypass graft surgery (CABG) in Table 1. All patients were under medication based on current guidelines and in stable compensated cardiac conditions [46].
**Table 1**
| Unnamed: 0 | CHD n=83 | HT n=54 | NT n=54 | p | p.1 | p.2 | p.3 |
| --- | --- | --- | --- | --- | --- | --- | --- |
| | CHD n=83 | HT n=54 | NT n=54 | NT, HT, vs. CHD | NT vs. HT | NT vs. CHD | HT vs. CHD |
| Age [years] | 65.02 ± 0.99 (44–85) | 52.74 ± 1.57 (21–74) | 50.80 ± 1.63 (25–78) | <.001 | .39 | <.001 | <.001 |
| BMI [kg/m2] | 27.85 ± 0.43 (21.97–46.44) | 28.51 ± 0.52 (20.35–38.86) | 25.27 ± 0.33 (19.78–30.85) | <.001 | <.001 | <.001 | .31 |
| Study BP [mmHg] | Study BP [mmHg] | Study BP [mmHg] | Study BP [mmHg] | Study BP [mmHg] | Study BP [mmHg] | Study BP [mmHg] | Study BP [mmHg] |
| Study SBP | 141.57 ± 1.64 (111.67–187.33) | 151.96 ± 1.86 (120.67–189.67) | 127.09 ± 1.17 (109.33–139.67) | <.001 | <.001 | <.001 | <.001 |
| Study DBP | 81.20 ± 1.09 (62.33–105.00) | 93.99 ± 1.31 (72.67–115.00) | 78.28 ± 1.03 (58.33–89.50) | <.001 | <.001 | .07 | <.001 |
| Study MAP | 101.33 ± 1.11 (81.89–132.00) | 113.31 ± 1.41 (88.67–139.89) | 94.55 ± 1.01 (75.33–104.83) | <.001 | <.001 | <.001 | <.001 |
| Home BP [mmHg] | Home BP [mmHg] | Home BP [mmHg] | Home BP [mmHg] | Home BP [mmHg] | Home BP [mmHg] | Home BP [mmHg] | Home BP [mmHg] |
| Home SBP | | 143.52 ± 1.28 (119.17–162.33) n=52 | 122.76 ± 0.86 (105.17–134.60) | | <.001 | | |
| Home DBP | | 85.67 ± 1.02 (68.33–103.00) n=52 | 72.23 ± 0.80 (60.00–82.33) | | <.001 | | |
| Medication* | n=83 | n=12 | | | | | |
| LVEF ≤ 40% [%] | 11 (13.3) n=81 | – | – | | | | |
| MI [%] | 47 (56.6) n=81 | – | – | | | | |
| CABG [%] | 23 (27.7) n=81 | – | – | | | | |
| Smoking [%] | 5 (6.0) | – | – | | | | |
| HbA1c [mmol/mol] | 41.49 ± 0.77 (33–72) n=80 | 36.74 ± 0.52 (26–43) n=53 | 36.29 ± 0.49 (26–42) n=51 | <.001 | .58 | <.001 | <.001 |
| Creatinine [μmol/L] | | 80.76 ± 1.41 (64–103) | | | | | |
| Sodium [mmol/L] | | 140.32 ± 0.30 (135–145) n=47 | | | | | |
| Calcium [mmol/L] | | 2.37 ± 0.01 (2.11–2.58) n=46 | | | | | |
| Potassium [mmol/L] | | 4.12 ± 0.04 (3.70–4.90) n=47 | | | | | |
| Cortisol | Cortisol | Cortisol | Cortisol | Cortisol | Cortisol | Cortisol | Cortisol |
| Awakening [nmol/L] | 4.41 ± 0.29 (0.35–18.00) | 5.38 ± 0.41 (0.68–14.83) | 3.94 ± 0.39 (0.02–15.34) | .015 | .006 | .18 | .05 |
| 16:00h [nmol/L] | 1.82 ± 0.14 (0.06–7.38) | 1.77 ± 0.12 (0.14–4.29) | 1.8 ± 0.16 (0.31–5.15) | .87 | .97 | .65 | .68 |
| 22:00h [nmol/L] | 0.85 ± 0.07 (0.09–3.14) | 0.75 ± 0.12 (0.04–6.10) | 0.82 ± 0.10 (0.10–3.77) | .23 | .49 | .35 | .08 |
| SlopeAwake | -0.23 ± 0.02 (-1.00–0.18) | -0.29 ± 0.03 (-0.87–0.17) | -0.20 ± 0.03 (-0.96–0.17) | .033 | .021 | .44 | .039 |
| SlopePeak | -0.46 ± 0.03 (-1.13–0.18) | -0.50 ± 0.03 (-1.11–0.18) | -0.49 ± 0.05 (-1.63–0.04) | .66 | .89 | .52 | .37 |
| M wake-up time [h] | 6:16 ± 0:05 (3:55–8:00) | 6:05 ± 0:06 (4:20–7:27) | 5:55 ± 0:06 (3:27–7:07) | .040 | .27 | .014 | .18 |
| M sleep duration [h] | 7.51 ± 0.10 (4.92–9.95) | 7.28 ± 0.09 (5.94–9.00) | 7.04 ± 0.10 (4.94–8.92) | .004 | .07 | .002 | .15 |
| Coagulation | Coagulation | Coagulation | Coagulation | Coagulation | Coagulation | Coagulation | Coagulation |
| Fibrinogen [g/L] | 2.85 ± 0.06 (1.65–4.46) | 2.62 ± 0.06 (1.47–3.72) | 2.6 ± 0.08 (1.57–4.25) | .010 | .73 | .009 | .013 |
| D-dimer [µg/L] | 622.64 ± 92.37 (45–5177) | 474.04 ± 30.41 (155–1047) | 513.46 ± 47.36 (45–1616) | .94 | .89 | .85 | .74 |
| Inflammation | Inflammation | Inflammation | Inflammation | Inflammation | Inflammation | Inflammation | Inflammation |
| IL-6 [pg/mL] | 0.56 ± 0.04 (0.03–1.57) n=82 | 0.57 ± 0.04 (0.16–1.48) n=53 | 0.50 ± 0.07 (0.22–3.61) n=53 | .18 | .047 | .19 | .54 |
| TNF-α [pg/mL] | 2.09 ± 0.09 (0.71–5.11) n=82 | 1.84 ± 0.06 (0.72–3.28) n=53 | 2.07± 0.10 (0.88–4.91) n=53 | .16 | .07 | .95 | .09 |
| CRP [μg/mL] | 2.15 ± 0.22 (0.07–11.55) n=78 | 2.83 ± 0.26 (0.64–8.65) n=50 | 2.06 ± 0.30 (0.11–9.59) n=40 | <.004 | <.001 | .63 | <.002 |
| tChol/HDL | 3.09 ± 0.08 (1.73–5.86) n=80 | 3.99 ± 0.13 (2.38–6.41) | 3.66 ± 0.13 (2.03–5.72) n=51 | <.001 | .06 | <.001 | <.001 |
Essential hypertension and normotension. We recruited apparently healthy, nonsmoking hypertensive and normotensive men of comparable age by aid of the Swiss-Red-Cross of Bern. Members of our study team accompanied the mobile blood-donation unit that routinely records BP ranges before blood donation. Interested blood donors were given written study information asking for the following inclusion criteria: age between 18-80 years; BP either in the hypertensive or in the normotensive range (see below); smoking less than 5 cigarettes per day; and no alcohol or illicit drug abuse. We accepted intake of antihypertensive medication in a small proportion to increase sample size of hypertensive individuals. Apart from hypertension-related criteria, NT and HT were required to meet the same inclusion and exclusion criteria (alcohol and illicit drug abuse, strenuous exercise, liver and renal diseases, chronic obstructive pulmonary disease, allergies and atopic diathesis, rheumatic diseases, human immunodeficiency virus, cancer, major psychiatric disorders, neurological diseases, and current infectious diseases) as verified by telephone interview using an extensive health questionnaire [42, 43]. Four eligible participants (NT:$$n = 2$$, HT:$$n = 2$$) who reported regular medication intake stopped medication one week prior to participating in the study. To exclude potential cases with secondary hypertension, eligible HT provided blood samples for the routine assessment of serum creatinine, calcium, sodium, and potassium [47]. No eligible HT was diagnosed with secondary hypertension. We measured HbA1c in all participants. Furthermore, we recruited 12 participants previously diagnosed as hypertensives, who were under antihypertensive medication.
## Classification of essential hypertension and normotension
Classification of essential hypertension and normotension of the unmedicated participants was based on a two-step assessment procedure, while medicated hypertensive individuals and CHD-patients were assigned a priori to the study groups.
[1] Home blood pressure measurement. Following written instructions, interested blood donors were asked to measure BP on three days at home using an upper arm digital blood pressure monitor (Omron M6; Omron-Healthcare-Europe B.V., Hoofdorp, Netherlands). Home BP was to be measured twice a day (once in the morning and in the evening) in a seated position after a 15-minute rest. Participants were recruited as hypertensive if the average home systolic BP (SBP) was ≥135mmHg and/or the average home diastolic BP (DBP) was ≥85mmHg according to recommendations for home BP measurements [48]. Correspondingly, participants were recruited as normotensive if the average home SBP was <135mmHg and the average home DBP was <85mmHg. Rendering a minimum of 3 and a maximum of 6 measurements for each participant, we computed the average home BP.
[2] Study blood pressure measurement. To verify the home-measurement based preliminary classification of each medication-free participant as hypertensive or normotensive, trained personnel performed three additional BP measurements during the study session in a seated position after a 15-minute rest by means of sphygmomanometry (Omron M6; Omron-Healthcare-Europe B.V., Hoofdorp, Netherlands). We applied the regular World-Health-Organization/International-Society-of-Hypertension definition of hypertension and classified medication-free participants as hypertensive if their average study SBP was ≥140mmHg and/or their average study DBP was ≥90mmH [49]. Medication-free participants were classified as normotensive if their average study SBP was <140mmHg and their average study DBP was <90mmHg. The final group assignment of medication-free participants was based on congruent home and study BP classification.
## Design and procedure
In anticipation of the experimental session, all participants consumed a semi-standardized breakfast following written instructions and abstained from caffeine and alcohol consumption 24h prior to their arrival at the lab at 8:00h. Questionnaires were administered, and participants´ height and weight were measured. Participants received material and written instructions for saliva collection at home, before resting study BP was assessed.
To assess longitudinal changes in CHD risk factors, blood samples were collected at 11:30h, i.e., after a fasting for 3.5h since arrival. All participants were invited for identical blood sampling procedures scheduled after a minimum of 1.5yrs later (mean ± SEM=3.00 ± 0.06).
## Cortisol sampling protocol
Study participants were asked to obtain saliva samples on two consecutive workdays using salivette collection devices (Sarstedt, Rommelsdorf, Germany). To assess the CAR, five saliva samples were collected immediately after awakening and 15, 30, 45, and 60min (S1-to-S5) thereafter. Further samples were taken at 16:00h and 22:00h (S6-to-S7). Participants were free to wake up in accordance with their normal schedule, but at the latest by 8:00h. They had to remain lying in bed for the first 15min, and to abstain from breakfast during the first 30min, i.e. until after collection of the +30min after awakening salivette. For the breakfast that followed, participants were asked to avoid coffee or juicy drinks. Moreover, participants were instructed not to brush their teeth during the first hour after awakening. They were also told to clean their mouth with water before each saliva collection. In addition, participants were instructed to complete a diary during the sampling period, assessing bed- and wake-up times as well as the accurate sampling times. In addition to self-reports, we used electronic monitoring devices (MEMS Track Cap, Aardex, Switzerland).
A total of 123 participants provided accurate cortisol samples for both consecutive sampling days, whereas 15 participants provided accurate cortisol samples but for two non-consecutive days. Furthermore, CAR data of 53 participants were accurate for only one sampling day because of incomplete ($$n = 33$$) or inaccurate sampling ($$n = 20$$) of the other day. See Supplemental Material for more details.
## Cortisol
Participants were instructed to store their saliva samples in the refrigerator until sampling completion and to then send the collected samples to our laboratory as fast as possible. We stored saliva samples until study completion at –20°C. Biochemical analyses of cortisol [nmol/L] were performed with a competitive time-resolved fluorescence immunoassay (DELFIA) [50] in the Biochemical Laboratory of the University of Trier. Intra- (4.0-$6.7\%$) and inter-assay (7.1-$9.0\%$) coefficients of variation were ≤$9.0\%$.
## Prospective CHD risk assessment
We assessed prospective CHD risk by measuring changes between baseline and follow-up assessment of the following biological risk factors: [1] the prothrombotic factors D-dimer [µg/L] and fibrinogen [g/L], the [2] pro-inflammatory measures interleukin (IL)-6 [pg/mL], tumor necrosis factor alpha (TNF-α) [pg/mL], and the acute phase protein C-reactive protein (CRP) [μg/mL], and [3] blood lipid profiles in terms of total cholesterol (tChol)/high-density lipoprotein cholesterol (HDL)-ratio. Fibrinogen and D-dimer were analyzed at the Center for Laboratory Medicine of the Bern University Hospital (Inselgruppe AG, Bern) applying standard quality procedures following the Clauss method [51] (fibrinogen) and a particle-enhanced immunoturbidimetric assay (INNOVANCE® D-Dimer, Siemens Healthcare GmbH, Erlangen, Germany), respectively. Blood lipids were also analyzed in Bern using in vitro assays (enzymatic colorimetric, Roche, Mannheim, Germany). IL-6, TNF-α, and CRP were analyzed in the biochemical laboratory of the Biological Work and Health Psychology group at the University of Konstanz. Cytokines were determined with a high sensitivity chemiluminescence sandwich immunoassay (Meso Scale Discovery, Rockville, USA), while CRP was determined using a high-sensitive enzyme immunoassay (ELISA, IBL Hamburg, Germany). For more details see Supplemental Methods.
## HbA1c
HbA1c analyses were performed with in vitro assays for the quantitative determination of HbA1c IFCC [mmol/mol] in whole blood (Tina-quant®, Roche, Mannheim, Germany) (see Supplemental Methods).
## Statistical analyses
Statistical analyses were performed using SPSS (Version26.0) statistical software packages for MacIntosh (IBM SPSS Statistics, Chicago IL, USA). All tests were two-tailed with level of significance set at $p \leq .05.$ No outliers were excluded.
We a priori calculated power-analyses using G∗Power3.1. Following our previous findings, we expected an effect size of $f = .35$ with respect to group differences between HT and NT in CAR [6]. Based on our hypotheses we expected comparably smaller differences between CHD-patients and HT. To allow to detect small effects of $f = .10$ in a 3(groups)-by-5(measurement points) repeated measurement ANOVA with a power of $90\%$ and an observed average correlation of the repeated measures of $r = .54$ in cross-sectional analyses, the required total sample size is $$n = 180$.$
A posteriori, we determined f from partial η2 (η2 p) values using G*Power3.1. Effect size parameters f and R 2 changes are reported where appropriate (effect size conventions; small: $f = .10$,ΔR 2=.02; medium: $f = .25$,ΔR 2=.13; large: $f = .40$,ΔR 2=.26) [52].
For all participants, we calculated mean arterial BP (MAP) based on the three BP study measurements by the formula MAP=($\frac{2}{3}$*mean study DBP)+($\frac{1}{3}$*mean study SBP) and body-mass-index (BMI) by the formula BMI=kg/m2.
For data and measures relating to cortisol, we calculated mean values of the two sampling days. To aggregate diurnal cortisol profiles for prospective analyses, we calculated mean total diurnal cortisol released during sampling days computed as area under the curve with respect to ground (AUCdayCort:S1-to-S7). Total CAR´s were calculated accordingly (AUCCAR:S1-to-S5) [53]. Diurnal cortisol slopes were estimated following previous recommendations with one formula anchoring cortisol levels at awakening (SlopeAwake) and the other anchoring the individual morning peak (SlopePeak) [54]. Fibrinogen, D-dimer, tChol/HDL-ratio, IL-6, CRP, and TNF-α changes from baseline to follow-up assessment were computed by subtracting baseline values from follow-up values, with higher change values indicating increases in the respective parameters over time, i.e. over the follow-up period. Building on previous methods [e.g. [37, 55, 56]], we computed an aggregated coagulation index by averaging z-transformed change values of D-dimer and fibrinogen. For an aggregated inflammatory index, we accordingly averaged z-transformed change values of IL-6, CRP, and TNF-α.
All data were tested for normal distribution and homogeneity of variance using Kolmogorov-Smirnov and Levene’s tests prior to statistical analyses. All measures showing a skewed distribution (see Supplemental Material) were log-transformed. While log-transformed data were used for modeling and testing, we depict untransformed data in Tables 1, 2, in Figure 1 and in Supplemental Tables S1, S2 for reasons of clarity. Figure 2 depicts residuals of the dependent and independent variables adjusted for the full set of covariates.
To compute group differences in subject characteristics (Table 1) we used univariate ANOVAs. Cross-sectionally, to analyze whether groups differed in total diurnal cortisol secretion, we calculated repeated measures ANOVAs and ANCOVAs with repeated assessment of cortisol (S1–S7), as the dependent variable and group as the independent variable. Post-hoc testing comprised separate testing for differences between the 3 groups in the CAR (S1–S5), cortisol at awakening (S1), diurnal cortisol slopes (awake-to-last, peak-to-last), as well as evening cortisol (S7). We controlled for possible confounding effects of awakening time, sleep duration the night before saliva sampling, and medication intake in HT, in addition to age and BMI in repeated measures and univariate ANCOVAs [8]. Further post-hoc testing comprised repetition of the previous cortisol analyses but with comparisons of two subject groups instead of three (i.e., HT-vs.-NT, CHD-vs.-NT, and CHD-vs.-HT). We applied Huynh–Feldt correction for repeated measures.
We calculated prospective analyses to shed light on the potential clinical relevance of basal HPA-axis activity. We tested whether diurnal cortisol parameters would predict future changes in CHD risk factors. We calculated multivariate analyses of covariance (MANCOVA) with prospective changes in blood lipids (tChol/HDL-ratio change), as well as the coagulation and inflammatory indexes as dependent variables. As the main independent variable of interest, we entered aggregated cortisol daytime levels (AUCdayCort). To avoid model overfitting given the reduced sample size of $$n = 106$$ in our prospective analyses allowing for a maximum of 11 covariates simultaneously [57], covariates were entered setwise as follows: the default set of covariates comprises age at baseline, time between baseline and follow-up assessments, and medication intake in normotensive and hypertensive individuals at follow-up (model 1). Sleep duration and wake-up time (model 2), study group (model 3), and BMI at baseline in addition to prospective changes in BMI and MAP (model 4) were added successively as covariates to each previous model in complementary analyses (see Table 3). We post-hoc tested significant multivariate effects of aggregated cortisol daytime levels on future CHD risk by repeating the above described MANCOVA procedure while entering the separate parts of daytime cortisol secretion, i.e., CAR, diurnal slopes, and evening cortisol, as independent variable. Post-hoc testing of significant between-subject effects of cortisol parameters on any of the three dependent variables comprised linear regression analyses including changes where appropriate (e.g. D-dimer and fibrinogen change levels if the coagulation index was significant).
**Table 3**
| Cortisol | Model | Multivariate Tests | Multivariate Tests.1 | Tests of Between-Subjects Effects | Tests of Between-Subjects Effects.1 | Tests of Between-Subjects Effects.2 |
| --- | --- | --- | --- | --- | --- | --- |
| Cortisol | Model | F | p | Index | F | p |
| Main analysis | 1 | 3.39 | .021 | Coagulation | 7.56 | .007 |
| AUCdayCort | | | | Inflammation | .44 | .51 |
| AUCdayCort | | | | tChol/HDL | .03 | .85 |
| AUCdayCort | 2 | 3.27 | .025 | Coagulation | 8.27 | .005 |
| AUCdayCort | | | | Inflammation | 0.10 | .75 |
| AUCdayCort | | | | tChol/HDL | 0.009 | .93 |
| AUCdayCort | 3 | 3.22 | .026 | Coagulation | 7.70 | .007 |
| AUCdayCort | | | | Inflammation | 0.32 | .57 |
| AUCdayCort | | | | tChol/HDL | 0.01 | .92 |
| AUCdayCort | 4 | 2.90 | .039 | Coagulation | 7.55 | .007 |
| AUCdayCort | | | | Inflammation | 0.11 | .74 |
| AUCdayCort | | | | tChol/HDL | 0.04 | .85 |
| Post-hoc | 1 | 3.17 | .028 | Coagulation | 8.28 | .005 |
| AUCCAR | | | | Inflammation | .09 | .77 |
| AUCCAR | | | | tChol/HDL | .09 | .77 |
| AUCCAR | 2 | 3.09 | .031 | Coagulation | 8.30 | .005 |
| AUCCAR | | | | Inflammation | 0.04 | .84 |
| AUCCAR | | | | tChol/HDL | 0.09 | .77 |
| AUCCAR | 3 | 2.98 | .035 | Coagulation | 8.10 | .005 |
| AUCCAR | | | | Inflammation | 0.03 | .87 |
| AUCCAR | | | | tChol/HDL | 0.08 | .78 |
| AUCCAR | 4 | 2.70 | .050 | Coagulation | 7.79 | .006 |
| AUCCAR | | | | Inflammation | 0.01 | .94 |
| AUCCAR | | | | tChol/HDL | 0.20 | .66 |
| Waking cortisol | 1 | 0.81 | .49 | Coagulation | 2.17 | .14 |
| Waking cortisol | | | | Inflammation | 0.80 | .37 |
| Waking cortisol | | | | tChol/HDL | 0.14 | .71 |
| Waking cortisol | 2 | 0.83 | .48 | Coagulation | 2.18 | .14 |
| Waking cortisol | | | | Inflammation | 0.91 | .34 |
| Waking cortisol | | | | tChol/HDL | 0.11 | .74 |
| Waking cortisol | 3 | 1.13 | .34 | Coagulation | 2.75 | .10 |
| Waking cortisol | | | | Inflammation | 1.46 | .23 |
| Waking cortisol | | | | tChol/HDL | 0.01 | .91 |
| Waking cortisol | 4 | 1.04 | .38 | Coagulation | 2.38 | .13 |
| Waking cortisol | | | | Inflammation | 1.56 | .22 |
| Waking cortisol | | | | tChol/HDL | 0.06 | .80 |
| SlopePeak | 1 | 1.64 | .18 | Coagulation | 4.70 | .032 |
| SlopePeak | | | | Inflammation | 0.01 | .92 |
| SlopePeak | | | | tChol/HDL | 0.004 | .95 |
| SlopePeak | 2 | 1.62 | .19 | Coagulation | 4.69 | .033 |
| SlopePeak | | | | Inflammation | 0.02 | .90 |
| SlopePeak | | | | tChol/HDL | 0.02 | .90 |
| SlopePeak | 3 | 1.50 | .22 | Coagulation | 4.39 | .039 |
| SlopePeak | | | | Inflammation | 0.02 | .88 |
| SlopePeak | | | | tChol/HDL | 0.04 | .85 |
| SlopePeak | 4 | 1.25 | .30 | Coagulation | 3.77 | .06 |
| SlopePeak | | | | Inflammation | 0.08 | .78 |
| SlopePeak | | | | tChol/HDL | 0.18 | .67 |
| SlopeAwake | 1 | 0.77 | .51 | Coagulation | 1.67 | .20 |
| SlopeAwake | | | | Inflammation | 0.99 | .32 |
| SlopeAwake | | | | tChol/HDL | .08 | .78 |
| SlopeAwake | 2 | 0.78 | .51 | Coagulation | 1.64 | .20 |
| SlopeAwake | | | | Inflammation | 1.01 | .32 |
| SlopeAwake | | | | tChol/HDL | 0.10 | .75 |
| SlopeAwake | 3 | 1.24 | .30 | Coagulation | 2.09 | .15 |
| SlopeAwake | | | | Inflammation | 1.74 | .19 |
| SlopeAwake | | | | tChol/HDL | 0.32 | .57 |
| SlopeAwake | 4 | 1.18 | .32 | Coagulation | 1.82 | .18 |
| SlopeAwake | | | | Inflammation | 1.97 | .16 |
| SlopeAwake | | | | tChol/HDL | 0.19 | .67 |
| Evening cortisol | 1 | 0.76 | .52 | Coagulation | 2.17 | .14 |
| Evening cortisol | | | | Inflammation | 0.002 | .96 |
| Evening cortisol | | | | tChol/HDL | 0.13 | .72 |
| Evening cortisol | 2 | 0.76 | .52 | Coagulation | 2.11 | .15 |
| Evening cortisol | | | | Inflammation | 0.0001 | .99 |
| Evening cortisol | | | | tChol/HDL | 0.15 | .70 |
| Evening cortisol | 3 | 0.80 | .50 | Coagulation | 2.00 | .16 |
| Evening cortisol | | | | Inflammation | 0.04 | .85 |
| Evening cortisol | | | | tChol/HDL | 0.28 | .60 |
| Evening cortisol | 4 | 0.84 | .48 | Coagulation | 1.90 | .17 |
| Evening cortisol | | | | Inflammation | 0.15 | .70 |
| Evening cortisol | | | | tChol/HDL | 0.18 | .67 |
## Group characteristics
Table 1 provides demographic and medical characteristics of CHD-patients, as well as hypertensive and normotensive participants. The three study groups differed in terms of age and BMI: CHD-patients had the highest average age (p ≤.001; mean ± SEM: CHD: 65.02 ± 0.99; HT: 52.74 ± 1.57; NT: 50.80 ± 1.63), whereas HT had a higher BMI than the other groups ($p \leq .001$; mean ± SEM: CHD: 27.85 ± 0.43; HT: 28.51 ± 0.52; NT: 25.27 ± 0.33). As expected, HT showed the highest study values in systolic BP, diastolic BP, and MAP compared with NT and CHD-patients ($p \leq .001$). HT showed the highest CRP ($p \leq .004$) and tChol/HDL-ratios ($p \leq .001$), while CHD-patients had highest fibrinogen and HbA1c levels (p´s ≤.010). On average, HT had serum levels of creatinine, calcium, sodium, and potassium in the normal reference range, thus supporting a diagnosis of essential hypertension. No participant had a diagnosis of a disease affecting the basal activity of the HPA-axis, such as adrenal insufficiency. In addition, all patients were under medication but no participant was treated with glucocorticoid substitution therapy (see Supplemental Material).
## Diurnal HPA-axis activity
Repeated measures AN(C)OVAs with cortisol as repeated dependent variable (S1-S7) revealed that the three groups significantly differed in their total diurnal HPA-axis activity (interaction group-by-time: F(8.16,767.45)=2.73,$$p \leq .005$$,η2 $$p \leq .03$$,$f = 0.18$; with covariates: F(8.43,771.31)=1.95,$$p \leq .047$$,η2 $$p \leq .02$$,$f = 0.14$). As compared to normotensives, both HT and CHD-patients had lower cortisol concentrations during the day (HT-vs.-NT: interaction group-by-time: F(3.83,405.78)=2.85,$$p \leq .026$$,η2 $$p \leq .03$$,$f = 0.18$); with covariates: F(4.04,407.75)=3.31,$$p \leq .011$$,η2 $$p \leq .03$$,$f = 0.18$); CHD-vs.-NT: interaction group-by-time: F(4.13,556.91)=2.75,$$p \leq .026$$,η2 $$p \leq .02$$,$f = 0.14$; with covariates: F(4.25,557.01)=2.29,$$p \leq .055$$,η2 $$p \leq .02$$,$f = 0.14$). Moreover, CHD-patients had lower total diurnal HPA-axis activity as compared to HT, but not independent of covariates (interaction group-by-time: F(3.97,535.52)=2.63,$$p \leq .034$$,η2 $$p \leq .02$$,$f = 0.14$); with covariates: $$p \leq .54$$).
## Cortisol at awakening and cortisol awakening response
Post-hoc testing of total diurnal HPA-axis activity comprised further analysis of cortisol levels within the first hour after awakening (see Figure 1).
Cortisol at awakening. As depicted in Table 1, the three groups differed in their cortisol levels at awakening (F[2,188]=4.28,$$p \leq .015$$,η2 $$p \leq .04$$,$f = 0.20$; with covariates: F[2,183]=3.15,$$p \leq .045$$,η2 $$p \leq .03$$,$f = 0.18$). Cortisol awakening levels were highest in HT, in particular as compared to normotensives who showed lowest awakening levels (HT-vs.-NT: F[1,108]=7.85,$$p \leq .006$$,η2 $$p \leq .07$$,$f = 0.27$; with covariates: F[1,101]=8.38,$$p \leq .005$$,η2 $$p \leq .08$$,$f = 0.29$). Differences between HT and CHD-patients were of borderline significance (CHD-vs.-HT: F[1,137]=3.91,$$p \leq .050$$,η2 $$p \leq .03$$,$f = 0.18$; with covariates: $$p \leq .08$$). However, despite higher cortisol awakening levels, CHD-patients did not significantly differ from normotensives (CHD-vs.-NT: $$p \leq .18$$, with covariates: $$p \leq .20$$).
Cortisol awakening response. Repeated measures AN(C)OVAs with cortisol (S1-S5) as repeated dependent variable showed significant CAR group differences (interaction group-by-time: F(5.67,532.65)=3.10,$$p \leq .006$$,η2 $$p \leq .03$$,$f = 0.18$); with covariates: F(5.82, 532.19)=2.61,$$p \leq .018$$,η2 $$p \leq .03$$,$f = 0.18$). As compared to normotensives, HT and CHD-patients showed a lower CAR (HT-vs.-NT: interaction group-by-time: F(2.61,276.85)=4.89,$$p \leq .004$$,η2 $$p \leq .04$$,$f = 0.20$; with covariates: F(2.75,277.73)=4.59,$$p \leq .005$$,η2 $$p \leq .04$$,$f = 0.20$; CHD-vs.-NT: interaction group-by-time: F(2.70,365.06)=4.34,$$p \leq .007$$,η2 $$p \leq .03$$,$f = 0.18$; with covariates: F(2.80,366.70)=4.23,$$p \leq .007$$,η2 $$p \leq .03$$,$f = 0.18$). However, HT and CHD-patients did not differ in their CAR ($$p \leq .80$$; with covariates: $$p \leq .59$$).
## Diurnal decline and evening cortisol
The groups differed in terms of diurnal cortisol decline from awakening to evening, but not independently of covariates (SlopeAwake: F[2,188]=3.48,$$p \leq .033$$,η2 $$p \leq .04$$,$f = 0.20$; with covariates: $$p \leq .11$$). Diurnal decline from awake to evening was steepest in HT who differed from normotensives with flattest awakening levels (HT-vs.-NT: F[1,106]=5.53,$$p \leq .021$$,η2 $$p \leq .05$$,$f = 0.23$; with covariates: F[1,101]=4.29,$$p \leq .041$$,η2 $$p \leq .04$$,$f = 0.20$). HT and CHD-patients in terms of diurnal decline but not independently of covariates (HT-vs.-CHD: F[1,135]=4.34,$$p \leq .039$$,η2 $$p \leq .03$$,$f = 0.18$: with covariates: $$p \leq .09$$), but CHD-patients did not significantly differ from normotensives (CHD-vs.-NT: $$p \leq .44$$; with covariates: $$p \leq .67$$).
In terms of evening cortisol, HT showed lower levels as compared to CHD-patients (HT-vs.-CHD:$$p \leq .08$$; with covariates: $$p \leq .79$$; HT-vs.-NT: $$p \leq .49$$; CHD-vs.-NT: $$p \leq .35$$; see Table 1).
## Prediction of future CHD risk by diurnal HPA-axis activity
Our main MANCOVA analysis revealed that higher daytime cortisol levels in terms of AUCdayCort significantly related to future overall CHD risk comprised the dependent variables of the MANCOVA: tChol/HDL-ratio change, coagulation and inflammatory indices (MANCOVA multivariate effects: model 1: F[3,99]=3.39,$$p \leq .021$$,η2 $$p \leq .10$$,$f = 0.33$,Wilk’sΛ=.91). Additional controlling for further covariates (models 2-to-4) did not alter the significance of this multivariate effect (p ´s≤.039, see Table 3). AUCdayCort levels were significantly associated with the coagulation index (MANCOVA between-subject effects: model 1: F[1,101]=7.56,$$p \leq .007$$,η2 $$p \leq .07$$;$f = 0.27$; models 2-to-4: p ´s≤.007) but not with the inflammatory index (p ´s≥.51) or prospective changes in tChol/HDL-ratio (p ´s≥.85). Further analysis revealed that AUCdayCort predicted greater increases from baseline to follow-up in fibrinogen (regression analyses: model 1: ß=-.23,$$p \leq .017$$,ΔR 2=0.12; model 2-to-4: ß ´s≥-.26,p ´s≤.012,ΔR 2≥.14) but not D-dimer (p ´s≥.23).
We tested post-hoc the significant multivariate effect of AUCdayCort on future overall CHD risk. Cortisol awakening levels in terms of AUCCAR significantly related to future overall CHD risk (MANCOVA multivariate effects: model 1: F[3,99]=3.17,$$p \leq .028$$,η2 $$p \leq .09$$,$f = 0.31$, WilksΛ=.91; model-2-to-4: p ´s≤.050) with AUCCAR being associated with the coagulation index (MANCOVA between-subject effects: model 1: F[1,101]=8.28,$$p \leq .005$$,η2 $$p \leq .08$$,$f = 0.29$; models 2-to-4: p ´s≤.006) but not with the inflammatory index (p ´s≥.77) or prospective changes in tChol/HDL-ratio (p ´s≥.66). With respect to coagulation measures, AUCCAR predicted greater increases from baseline to follow-up in fibrinogen (regression analyses: model 1: ß=-.17,$$p \leq .082$$,ΔR 2=.10; model 2-to-4: ß ´s≥-.17,p ´s≤.098,ΔR 2≥.10) and in D-dimer (regression analyses: model 1: ß=-.17,$$p \leq .076$$,ΔR 2=.06; model 2-to-4: ß ´s≥-.17,p ´s≤.094,ΔR 2≥.07) towards a trend for significance. Neither cortisol levels at awaking, slopes, nor evening cortisol levels were associated with future CHD risk (p ´s≥.18).
## Discussion
The first objective of our study was to cross-sectionally compare diurnal HPA-axis activity between male CHD-patients, HT, and NT at baseline. The novelty of this study is the comparison between HT and CHD-patients. HT and CHD-patients showed lower overall diurnal cortisol saliva concentrations as compared to healthy controls with lowest concentrations in CHD-patients. We found a reduced CAR in HT and in CHD-patients as compared to NT corroborating previous findings [5, 6, 16, 17, 21]. Moreover, HT and CHD-patients did not differ in their CAR. However, regarding cortisol at awakening, HT showed highest and NT lowest levels of the three study groups, with CHD-patients showing borderline significantly lower awakening levels compared to HT. These results are in line with previous research, with higher early morning salivary cortisol levels in unmedicated HT compared with healthy controls [15], whereas medicated HT showed lower early morning cortisol levels compared to NT [16]. The latter points to a potentially normalizing effect of BP medication on cortisol levels at awakening. In line with this assumption, medicated CHD-patients did not significantly differ from normotensive individuals in their cortisol levels at awakening. Salivary waking or early morning cortisol was not associated with CHD-(measures) in most previous studies including heart patients (18–20). The increased morning cortisol levels in combination with a reduced cortisol response to awakening in our HT may indicate a generally altered HPA-axis activity in the early morning as observed in subjects suffering from a wide range of health problems [54, 58]. Although, HT and CHD-patients did not differ from NT in their evening cortisol levels, patients had significantly higher levels as compared to HT. Evidence from other studies, however, points to cross-sectional associations between higher bedtime [20] or late night [15] cortisol levels in HT and CHD-symptoms. A potential reason for this discrepancy may be that we assessed cortisol at 22:00h, but not at bedtime or late night levels. Nevertheless, the higher evening levels in our patients may add to adverse cardiac outcomes as observed in other studies [e.g. [28]]. Diurnal cortisol decline from waking to evening (SlopeWake) was steepest in HT and flattest in NT, but not independent of covariates, driven by the comparably high awakening levels in HT. Diurnal decline from morning peak to evening (SlopePeak) did not differ between groups. Some previous studies pointed to an association between flatter cortisol slopes and greater cardiovascular risk [5, 24], so our results, with steeper, and thus more normative [59] cortisol declines in HT as compared to healthy controls, seem unexpected. However, in line with our findings, other studies could not detect any association between diurnal cortisol slope and CHD-measures [19, 25]. We offer different explanations for these divergencies: First, the observed result of more normative slopes in hypertensive individuals was not independent of covariates. One of the covariates, later awakening time, was borderline significantly associated with steeper slopes in HT as compared to NT ($$p \leq .054$$). Second the formulas for the calculation of diurnal slopes differ between studies, rendering comparison of effects difficult [54]. Since we calculated the slope using evening cortisol levels instead of levels at bedtime, the dynamics of diurnal HPA-axis activity may have been captured incompletely. Third, the observed group differences in diurnal slopes from waking to evening may to some extent be explained by the elevated morning cortisol levels of HT as compared to CHD-patients and normotensive controls. Taken together, we observed group differences in basal HPA-axis activity, with lower CAR and lower overall diurnal cortisol levels in CHD-patients and HT as compared to healthy controls.
We found evidence for aggregated daytime cortisol and CAR levels in predicting overall CHD risk (i.e. dependent variables in the MANCOVA). In detail, we found lower aggregated cortisol levels to predict higher increases in coagulation markers at follow-up, while inflammation markers and blood lipid profile were not associated with basal HPA-axis activity. Moreover, the prospective association between diurnal cortisol secretion and coagulation was mainly driven by the prediction of fibrinogen increases. So far, only few studies investigated associations between basal HPA-axis and prothrombotic activity: Evidence from cross-sectional studies points to an association between higher cortisol levels [36, 60] or dysregulated diurnal cortisol profiles on the one hand [37] and measures including prothrombotic activity on the other hand, which may explain why circulating cortisol had been associated with atherosclerotic vessel damage [60]. We found a longitudinal association between lower diurnal HPA-axis activity and higher overall CHD risk increase comprising increases in all three aggregated biological risk factor indices, and in particular with fibrinogen increases. Despite evidence for cross-sectional associations between basal HPA-axis activity and inflammation [37] or hyperlipidemia [39, 61], we could not detect prospective associations between daytime cortisol or CAR levels and changes in inflammation markers and blood lipid profiles. These results are in line with a study in 9-to-10 year-old children where baseline cortisol did not predict blood lipid levels 1 year later [61].
The variability between the different measures of cortisol and the outcomes could possibly be attributed to the fact that single cortisol measures may only explain a small proportion of variance and are strongly influenced by situational effects [see [62]]. As a consequence, it requires repeated diurnal cortisol assessments to be able to detect associations with outcomes.
Our results suggest that dysregulation in terms of reduced CAR and lower overall daytime levels may represent an early indicator for increased cardiovascular risk. However, the clinical utility of the different measures of cortisol and the question about whether the monitoring of the HPA-axis activity facilitates the identification of high-risk individuals needs to be clarified in future studies. Further, the mechanisms underlying the observed HPA-axis activity dysregulation in hypertension and CHD are unclear. Since none of our participants had a diagnosis of adrenal insufficiency, we consider it unlikely that the observed lower HPA-axis activity in HT and patients result from adrenal insufficiency. One possible explanation may relate to (former) chronic stress experiences during disease development. Chronic stress has been proposed to play a role in both HT and CHD development that has been associated with altered diurnal HPA-axis activity [63, 64]. According to the Allostatic-Load-Model chronic stress causes repeated activation of stress reactions including HPA-axis and sympathetic-adrenal-medullary axis reactivity which accumulate over time leading to compensatory stress system dysregulations in terms of allostatic load [65]. Given the higher blood pressure and overall SNS activity in hypertension, that notably has been proposed to represent a potential consequence from allostatic load [63], the observed lower HPA-axis activity in HT may represent a compensatory allostatic load system dysregulation. Allostatic dysregulation can lead to allostatic overload with tissue and organ damage, including the cardiovascular, the immune, and the metabolic system [65]. Future studies are needed to further elucidate the role of lifestyle factors (related to allostatic load) in diurnal HPA-axis activity of individuals with elevated CHD risk [see [14]] and whether our findings in salivary cortisol also apply to cumulative measures of cortisol output such as hair cortisol.
In our prospective analyses, we found that lower aggregated cortisol daytime levels and CAR predicted independent biological CHD risk factors and in particular prothrombotic activity about 3 years later. Studies investigating the effects of glucocorticoid excess [e.g. due to Cushing’s syndrome [66]] suggest effects on blood coagulation. However, no clear relationship between hypocortisolism and hypercoagulability has yet been established. It remains to be elucidated whether the observed prospective coagulation increases similarly represent a compensatory allostatic load system dysregulation resulting from the HPA-axis dysregulation [63, 65]. Also, whether the observed lower CAR, either alone or combined with the higher prospective coagulation increases, relates to the higher occurrence of myocardial infarctions in the early morning hours [67] remains to be elucidated.
Limitations of our study include the relatively high drop-out rate and the wide follow-up range that we mainly attribute to logistic reasons. Participants who dropped out did not substantially differ in their characteristics from those completing the follow-up assessment, except for lower TNF-α and higher CRP levels at baseline (for both, see Supplemental Material). Also, apart from the CAR, we measured cortisol only twice and we assessed evening cortisol and not bedtime levels. Moreover, we cannot completely rule out potential effects of repeated thawing during transportation although salivary cortisol measurements have been shown to be quite robust against repeated freeze-and-thaw-cycles [8, 68]. Moreover, the generalizability of our results is limited to middle-aged men of relatively high socioeconomic status and future studies are needed to further elucidate whether our findings also apply to women [19, 22] and participants with differing socioeconomic status [69]. Also, recruitment via blood donor facilities may interfere with generalizability and we cannot rule out that the use of 24-hour automatic BP measurement would have been even more accurate to diagnose hypertension status compared to the applied two-step assessment procedure including repeated home and study BP measurement. Another limitation of our study relates to the medication of the CHD-patients. First, medication in general can affect salivary cortisol assessment at different levels (e.g., with effects on the composition of the saliva or direct effects on the cortisol synthesis) [70]. Further, the effects of CHD-medication on the different parameters of diurnal cortisol secretion (e.g. CAR) have not been investigated systematically to the best of our knowledge [71], so the comparison between our medicated and unmedicated groups are to be interpreted with caution, as potential medication effects cannot be ruled out. Also, it is possible, that CHD-medication prevented substantial increases in CHD risk over time in our drug-treated participants. Finally, despite the prospective nature of our study we cannot draw definite conclusions regarding causality as we cannot exclude potential influences by other factors.
Strengths of our study comprise the use of MEMS caps combined with self-recording of sampling times allowing us to ensure the adherence to the study protocol. Further, we controlled for many potentially confounding variables including waking time and sleep duration and cortisol was assessed on two consecutive days [8].
In conclusion, we found evidence for a downregulation of HPA-axis activity in both, CHD and HT. Our results moreover suggest that lower diurnal HPA-axis activity seems to predict poorer cardiovascular health in HT and CHD by promoting a hypercoagulable state. A down-regulation of basal HPA-axis activity may therefore play a role in the pathogenesis and/or progression of atherosclerosis.
## Data availability statement
The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.
## Ethics statement
The studies involving human participants were reviewed and approved by Ethics committee of the Canton of Bern, Switzerland. The patients/participants provided their written informed consent to participate in this study.
## Author contributions
Conceptualization, PHW. Formal analysis, CD. Funding acquisition, PHW and RvK. Investigation, LT, CZ-H and RvK. Methodology, RvK, RW, HS, NM-B and PHW. Project administration, RvK and PHW. Supervision, RvK and PHW. Visualization, CD and PHW. Writing—original draft, CD and PHW. Writing—review and editing, RvK, RW, HS, NM-B and CZ-H. All authors contributed to the article and approved the submitted version.
## Conflict of interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher’s note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## Supplementary material
The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fendo.2023.1080938/full#supplementary-material
## References
1. Virani SS, Alonso A, Aparicio HJ, Benjamin EJ, Bittencourt MS, Callaway CW. **Heart disease and stroke statistics-2021 update: A report from the American heart association**. *Circulation* (2021) **143**. DOI: 10.1161/CIR.0000000000000950
2. Whelton PK, Carey RM, Aronow WS, Casey DE, Collins KJ, Himmelfarb CD. **2017 ACC/AHA/AAPA/ABC/ACPM/AGS/APhA/ASH/ASPC/NMA/PCNA guideline guideline for the prevention, detection, evaluation, and management of high blood pressure in adults**. *J Am Coll Cardiol* (2018) **71**. DOI: 10.1016/j.jacc.2017.11.006
3. Messerli FH, Williams B, Ritz E. **Essential hypertension**. *Lancet* (2007) **370** 591-603. DOI: 10.1016/S0140-6736(07)61299-9
4. Carretero OA, Oparil S. **Essential hypertension. Part I: Definition and etiology**. *Circulation* (2000) **101**. DOI: 10.1161/01.cir.101.3.329
5. Molloy GJ, Perkins-Porras L, Strike PC, Steptoe A. **Type-D personality and cortisol in survivors of acute coronary syndrome**. *Psychosom Med* (2008) **70**. DOI: 10.1097/PSY.0b013e3181842e0c
6. Wirtz PH, von Känel R, Emini L, Ruedisueli K, Groessbauer S, Maercker A. **Evidence for altered hypothalamus–pituitary–adrenal axis functioning in systemic hypertension: Blunted cortisol response to awakening and lower negative feedback sensitivity**. *Psychoneuroendocrinology* (2007) **32**. DOI: 10.1016/j.psyneuen.2007.02.006
7. Sapolsky RM, Romero LM, Munck AU. **How do glucocorticoids influence stress responses? Integrating permissive, suppressive, stimulatory, and preparative actions***. *Endocr Rev* (2000) **21** 55-89. DOI: 10.1210/edrv.21.1.0389
8. Stalder T, Kirschbaum C, Kudielka BM, Adam EK, Pruessner JC, Wüst S. **Assessment of the cortisol awakening response: Expert consensus guidelines**. *Psychoneuroendocrinology* (2016) **63**. DOI: 10.1016/j.psyneuen.2015.10.010
9. Clow A, Thorn L, Evans P, Hucklebridge F. **The awakening cortisol response: Methodological issues and significance**. *Stress* (2004) **7** 29-37. DOI: 10.1080/10253890410001667205
10. Fries E, Dettenborn L, Kirschbaum C. **The cortisol awakening response (CAR): Facts and future directions**. *Int J Psychophysiol* (2009) **72** 67-73. DOI: 10.1016/j.ijpsycho.2008.03.014
11. Pruessner JC, Wolf OT, Hellhammer DH, Buske-Kirschbaum A, von Auer K, Jobst S. **Free cortisol levels after awakening: A reliable biological marker for the assessment of adrenocortical activity**. *Life Sci* (1997) **61**. DOI: 10.1016/S0024-3205(97)01008-4
12. Ross KM, Murphy MLM, Adam EK, Chen E, Miller GE. **How stable are diurnal cortisol activity indices in healthy individuals? Evidence from three multi-wave studies**. *Psychoneuroendocrinology* (2014) **39**. DOI: 10.1016/j.psyneuen.2013.09.016
13. Stone AA, Schwartz JE, Smyth J, Kirschbaum C, Cohen S, Hellhammer D. **Individual differences in the diurnal cycle of salivary free cortisol: A replication of flattened cycles for some individuals**. *Psychoneuroendocrinology* (2001) **26** 295-306. DOI: 10.1016/S0306-4530(00)00057-3
14. Collomp K, Baillot A, Forget H, Coquerel A, Rieth N, Vibarel-Rebot N. **Altered diurnal pattern of steroid hormones in relation to various behaviors, external factors and pathologies: A review**. *Physiol Behavior* (2016) **164** 68-85. DOI: 10.1016/j.physbeh.2016.05.039
15. Kidambi S, Kotchen JM, Grim CE, Raff H, Mao J, Singh RJ. **Association of adrenal steroids with hypertension and the metabolic syndrome in blacks**. *Hypertension* (2007) **49**. DOI: 10.1161/01.HYP.0000253258.36141.c7
16. Pulopulos MM, Hidalgo V, Puig-Perez S, Salvador A. **Cortisol awakening response and cognitive performance in hypertensive and normotensive older people**. *Horm Behav* (2016) **83** 75-82. DOI: 10.1016/j.yhbeh.2016.05.014
17. Kuehl LK, Hinkelmann K, Muhtz C, Dettenborn L, Wingenfeld K, Spitzer C. **Hair cortisol and cortisol awakening response are associated with criteria of the metabolic syndrome in opposite directions**. *Psychoneuroendocrinology* (2015) **51**. DOI: 10.1016/j.psyneuen.2014.09.012
18. Bhattacharyya MR, Molloy GJ, Steptoe A. **Depression is associated with flatter cortisol rhythms in patients with coronary artery disease**. *J Psychosom Res* (2008) **65**. DOI: 10.1016/j.jpsychores.2008.03.012
19. Hajat A, Diez-Roux AV, Sánchez BN, Holvoet P, Lima JA, Merkin SS. **Examining the association between salivary cortisol levels and subclinical measures of atherosclerosis: The multi-ethnic study of atherosclerosis**. *Psychoneuroendocrinology* (2013) **38**. DOI: 10.1016/j.psyneuen.2012.10.007
20. Nijm J, Kristenson M, Olsson AG, Jonasson L. **Impaired cortisol response to acute stressors in patients with coronary disease. Implications for inflammatory activity**. *J Intern Med* (2007) **262**. DOI: 10.1111/j.1365-2796.2007.01817.x
21. Vreeburg SA, Kruijtzer BP, van Pelt J, van Dyck R, DeRijk RH, Hoogendijk WJ. **Associations between sociodemographic, sampling and health factors and various salivary cortisol indicators in a large sample without psychopathology**. *Psychoneuroendocrinology* (2009) **34**. DOI: 10.1016/j.psyneuen.2009.04.024
22. Hurwitz Eller N, Netterstrøm B, Hansen AM. **Cortisol in urine and saliva: Relations to the intima media thickness, IMT**. *Atherosclerosis* (2001) **159**. DOI: 10.1016/S0021-9150(01)00487-7
23. Whitehead DL, Perkins-Porras L, Strike PC, Magid K, Steptoe A. **Cortisol awakening response is elevated in acute coronary syndrome patients with type-D personality**. *J Psychosom Res* (2007) **62**. DOI: 10.1016/j.jpsychores.2006.11.005
24. Matthews K, Schwartz J, Cohen S, Seeman T. **Diurnal cortisol decline is related to coronary calcification: CARDIA study**. *Psychosom Med* (2006) **68**. DOI: 10.1097/01.psy.0000244071.42939.0e
25. Dekker MJHJ, Koper JW, van Aken MO, Pols HAP, Hofman A, de Jong FH. **Salivary cortisol is related to atherosclerosis of carotid arteries**. *J Clin Endocrinol Metab* (2008) **93**. DOI: 10.1210/jc.2008-0496
26. Ronaldson A, Kidd T, Poole L, Leigh E, Jahangiri M, Steptoe A. **Diurnal cortisol rhythm is associated with adverse cardiac events and mortality in coronary artery bypass patients**. *J Clin Endocrinol Metab* (2015) **100**. DOI: 10.1210/jc.2015-2617
27. Kumari M, Shipley M, Stafford M, Kivimaki M. **Association of diurnal patterns in salivary cortisol with all-cause and cardiovascular mortality: findings from the Whitehall II study**. *J Clin Endocrinol Metab* (2011) **96**. DOI: 10.1210/jc.2010-2137
28. Hammer F, Deutschbein T, Marx A, Güder G, Michalski R, Ertl G. **High evening salivary cortisol is an independent predictor of increased mortality risk in patients with systolic heart failure**. *Int J Cardiol* (2016) **203** 69-73. DOI: 10.1016/j.ijcard.2015.10.084
29. Groffen DA, Bosma H, Koster A, von Bonsdorff MB, Aspelund T, Eiriksdottir G. **A blunted diurnal cortisol response in the lower educated does not explain educational differences in coronary heart disease: Findings from the AGES-Reykjavik study**. *Soc Sci Med* (2015) **127**. DOI: 10.1016/j.socscimed.2014.09.050
30. Violanti JM, Fekedulegn D, Andrew ME, Charles LE, Gu JK, Miller DB. **Subclinical markers of cardiovascular disease among police officers: A longitudinal assessment of the cortisol awakening response and flow mediated artery dilation**. *J Occup Environ Med* (2018) **60**. DOI: 10.1097/JOM.0000000000001358
31. Ikeda A, Steptoe A, Shipley M, Abell J, Kumari M, Tanigawa T. **Diurnal pattern of salivary cortisol and progression of aortic stiffness: Longitudinal study**. *Psychoneuroendocrinology* (2021) **133** 105372. DOI: 10.1016/j.psyneuen.2021.105372
32. Danesh J, Kaptoge S, Mann AG, Sarwar N, Wood A, Angleman SB. **Long-term interleukin-6 levels and subsequent risk of coronary heart disease: Two new prospective studies and a systematic review**. *PloS Med* (2008) **5** e78. DOI: 10.1371/journal.pmed.0050078
33. Ridker PM, Stampfer MJ, Rifai N. **Novel risk factors for systemic Atherosclerosis: A comparison of c-reactive protein, fibrinogen, homocysteine, lipoprotein(a), and standard cholesterol screening as predictors of peripheral arterial disease**. *JAMA* (2001) **285**. DOI: 10.1001/jama.285.19.2481
34. Simes J, Robledo KP, White HD, Espinoza D, Stewart RA, Sullivan DR. **D-dimer predicts long-term cause-specific mortality, cardiovascular events, and cancer in patients with stable coronary heart disease: LIPID study**. *Circulation* (2018) **138**. DOI: 10.1161/CIRCULATIONAHA.117.029901
35. Tuomisto K, Jousilahti P, Sundvall J, Pajunen P, Salomaa V. **C-reactive protein, interleukin-6 and tumor necrosis factor alpha as predictors of incident coronary and cardiovascular events and total mortality. A population-based, prospective study**. *Thromb Haemost* (2006) **95**. DOI: 10.1160/TH05-08-0571
36. Lippi G, Franchini M, Salvagno GL, Montagnana M, Guidi GC. **Higher morning serum cortisol level predicts increased fibrinogen but not shortened APTT**. *J Thromb Thrombolysis* (2008) **26**. DOI: 10.1007/s11239-007-0074-0
37. Piazza JR, Dmitrieva NO, Charles ST, Almeida DM, Orona GA. **Diurnal cortisol profiles, inflammation, and functional limitations in aging: Findings from the MIDUS study**. *Health Psychol* (2018) **37**. DOI: 10.1037/hea0000629
38. DeSantis AS, DiezRoux AV, Hajat A, Aiello AE, Golden SH, Jenny NS. **Associations of salivary cortisol levels with inflammatory markers: The multi-ethnic study of atherosclerosis**. *Psychoneuroendocrinology* (2012) **37**. DOI: 10.1016/j.psyneuen.2011.11.009
39. Rosmond R, Björntorp P. **The hypothalamic-pituitary-adrenal axis activity as a predictor of cardiovascular disease, type 2 diabetes and stroke**. *J Intern Med* (2000) **247**. DOI: 10.1046/j.1365-2796.2000.00603.x
40. Auer A, von Känel R, Lang I, Thomas L, Zuccarella-Hackl C, Degroote C. **Do hypertensive men spy with an angry little eye? Anger recognition in men with essential hypertension - cross-sectional and prospective findings**. *Ann Behav Med* (2022) **56**. DOI: 10.1093/abm/kaab108
41. Degroote C, von Känel R, Thomas L, Zuccarella-Hackl C, Pruessner JC, Wiest R. **Acute stress-induced blood lipid reactivity in hypertensive and normotensive men and prospective associations with future cardiovascular risk**. *J Clin Med* (2021) **10** 3400. DOI: 10.3390/jcm10153400
42. Zuccarella-Hackl C, von Känel R, Thomas L, Hauser M, Kuebler U, Widmer HR. **Macrophage superoxide anion production in essential hypertension: Associations with biological and psychological cardiovascular risk factors**. *Psychosom Med* (2016) **78**. DOI: 10.1097/PSY.0000000000000324
43. Zuccarella-Hackl C, von Känel R, Thomas L, Kuebler P, Schmid J-P, Mattle HP. **Higher macrophage superoxide anion production in coronary artery disease (CAD) patients with type D personality**. *Psychoneuroendocrinology* (2016) **68**. DOI: 10.1016/j.psyneuen.2016.02.031
44. Roth GA, Mensah GA, Johnson CO, Addolorato G, Ammirati E, Baddour LM. **Global burden of cardiovascular diseases and risk factors, 1990–2019: Update from the GBD 2019 study**. *J Am Coll Cardiol* (2020) **76** 2982-3021. DOI: 10.1016/j.jacc.2020.11.010
45. Mosca L, Barrett-Connor E, Wenger NK. **Sex/gender differences in cardiovascular disease prevention: What a difference a decade makes**. *Circulation* (2011) **124**. DOI: 10.1161/CIRCULATIONAHA.110.968792
46. Knuuti J, Wijns W, Saraste A, Capodanno D, Barbato E, Funck-Brentano C. **2019 ESC Guidelines for the diagnosis and management of chronic coronary syndromes**. *Eur Heart J* (2020) **41**. DOI: 10.1093/eurheartj/ehz425
47. Unger T, Borghi C, Charchar F, Khan NA, Poulter NR, Prabhakaran D. **2020 International society of hypertension global hypertension practice guidelines**. *Hypertension* (2020) **75**. DOI: 10.1161/HYPERTENSIONAHA.120.15026
48. Williams B, Mancia G, Spiering W, Agabiti Rosei E, Azizi M, Burnier M. **2018 ESC/ESH guidelines for the management of arterial hypertension**. *Eur Heart J* (2018) **39**. DOI: 10.1093/eurheartj/ehy339
49. Chalmers J, MacMahon S, Mancia G, Whitworth J, Beilin L, Hansson L. **1999 World health organization-international society of hypertension guidelines for the management of hypertension. Guidelines sub-committee of the world health organization**. *Clin Exp Hypertens* (1999) **21**. DOI: 10.3109/10641969909061028
50. Dressendörfer RA, Kirschbaum C, Rohde W, Stahl F, Strasburger CJ. **Synthesis of a cortisol-biotin conjugate and evaluation as a tracer in an immunoassay for salivary cortisol measurement**. *J Steroid Biochem Mol Biol* (1992) **43**. DOI: 10.1016/0960-0760(92)90294-S
51. Clauss A. **[Rapid physiological coagulation method in determination of fibrinogen]**. *Acta Haematol* (1957) **17**. DOI: 10.1159/000205234
52. Cohen J. *Statistical power analysis for the behavioral sciences* (1988)
53. Pruessner JC, Kirschbaum C, Meinlschmid G, Hellhammer DH. **Two formulas for computation of the area under the curve represent measures of total hormone concentration versus time-dependent change**. *Psychoneuroendocrinology* (2003) **28**. DOI: 10.1016/S0306-4530(02)00108-7
54. Adam EK, Quinn ME, Tavernier R, McQuillan MT, Dahlke KA, Gilbert KE. **Diurnal cortisol slopes and mental and physical health outcomes: A systematic review and meta-analysis**. *Psychoneuroendocrinology* (2017) **83** 25-41. DOI: 10.1016/j.psyneuen.2017.05.018
55. von Känel R, Orth-Gomér K. **Autonomic function and prothrombotic activity in women after an acute coronary event**. *J Womens Health (Larchmt)* (2008) **17**. DOI: 10.1089/jwh.2007.0764
56. Eapen DJ, Manocha P, Patel RS, Hammadah M, Veledar E, Wassel C. **Aggregate risk score based on markers of inflammation, cell stress, and coagulation is an independent predictor of adverse cardiovascular outcomes**. *J Am Coll Cardiol* (2013) **62**. DOI: 10.1016/j.jacc.2013.03.072
57. Babyak MA. **What you see may not be what you get: A brief, nontechnical introduction to overfitting in regression-type models**. *Psychosom Med* (2004) **66**. DOI: 10.1097/01.psy.0000127692.23278.a9
58. Kudielka BM, Kirschbaum C. **Awakening cortisol responses are influenced by health status and awakening time but not by menstrual cycle phase**. *Psychoneuroendocrinology* (2003) **28** 35-47. DOI: 10.1016/S0306-4530(02)00008-2
59. Dmitrieva NO, Almeida DM, Dmitrieva J, Loken E, Pieper CF. **A day-centered approach to modeling cortisol: Diurnal cortisol profiles and their associations among U.S. adults**. *Psychoneuroendocrinology* (2013) **38**. DOI: 10.1016/j.psyneuen.2013.05.003
60. von Känel R, Mausbach BT, Kudielka BM, Orth-Gomér K. **Relation of morning serum cortisol to prothrombotic activity in women with stable coronary artery disease**. *J Thromb Thrombolysis* (2008) **25**. DOI: 10.1007/s11239-007-0035-7
61. Dai W, Wagh SA, Chettiar S, Zhou GD, Roy R, Qiao X. **Blunted circadian cortisol in children is associated with poor cardiovascular health and may reflect circadian misalignment**. *Psychoneuroendocrinology* (2021) **129** 105252. DOI: 10.1016/j.psyneuen.2021.105252
62. Kirschbaum C, Steyer R, Eid M, Patalla U, Schwenkmezger P, Hellhammer DH. **Cortisol and behavior: 2. Application of a latent state-trait model to salivary cortisol**. *Psychoneuroendocrinology* (1990) **15** 297-307. DOI: 10.1016/0306-4530(90)90080-s
63. McEwen BS. *Stress, adaptation, and disease: Allostasis and allostatic load. molecular aspects, integrative systems, and clinical advances. Annals of the New York academy of sciences* (1998) 33-44. DOI: 10.1111/j.1749-6632.1998.tb09546.x
64. Yao B-C, Meng L-B, Hao M-L, Zhang Y-M, Gong T, Guo Z-G. **Chronic stress: A critical risk factor for atherosclerosis**. *J Int Med Res* (2019) **47**. DOI: 10.1177/0300060519826820
65. Juster RP, McEwen BS, Lupien SJ. **Allostatic load biomarkers of chronic stress and impact on health and cognition**. *Neurosci Biobehav Rev* (2010) **35** 2-16. DOI: 10.1016/j.neubiorev.2009.10.002
66. Isidori AM, Minnetti M, Sbardella E, Graziadio C, Grossman AB. **Mechanisms in endocrinology: The spectrum of haemostatic abnormalities in glucocorticoid excess and defect**. *Eur J Endocrinol* (2015) **173**. DOI: 10.1530/EJE-15-0308
67. Muller JE, Tofler GH, Stone PH. **Circadian variation and triggers of onset of acute cardiovascular disease**. *Circulation* (1989) **79**. DOI: 10.1161/01.CIR.79.4.733
68. Garde AH, Hansen AM. **Long-term stability of salivary cortisol**. *Scand J Clin Lab Invest* (2005) **65**. DOI: 10.1080/00365510510025773
69. Cohen S, Schwartz JE, Epel E, Kirschbaum C, Sidney S, Seeman T. **Socioeconomic status, race, and diurnal cortisol decline in the coronary artery risk development in young adults (CARDIA) study**. *Psychosom Med* (2006) **68**. DOI: 10.1097/01.psy.0000195967.51768.ea
70. Granger DA, Hibel LC, Fortunato CK, Kapelewski CH. **Medication effects on salivary cortisol: Tactics and strategy to minimize impact in behavioral and developmental science**. *Psychoneuroendocrinology* (2009) **34**. DOI: 10.1016/j.psyneuen.2009.06.017
71. Wolf OT, Fujiwara E, Luwinski G, Kirschbaum C, Markowitsch HJ. **No morning cortisol response in patients with severe global amnesia**. *Psychoneuroendocrinology* (2005) **30**. DOI: 10.1016/j.psyneuen.2004.05.001
|
---
title: Pedal cadence does not affect muscle damage to eccentric cycling performed
at similar mechanical work
authors:
- Hisashi Ueda
- Riki Saegusa
- Yosuke Tsuchiya
- Eisuke Ochi
journal: Frontiers in Physiology
year: 2023
pmcid: PMC10036782
doi: 10.3389/fphys.2023.1140359
license: CC BY 4.0
---
# Pedal cadence does not affect muscle damage to eccentric cycling performed at similar mechanical work
## Abstract
Purpose: This study aimed to investigate muscle damage when performing equal mechanical work of fast and slow pedaling speed by eccentric muscle actions (ECCs) cycling.
Methods: Nineteen young men [mean ± standard deviation (SD) age: 21.0 ± 2.2 years; height: 172.7 ± 5.9 cm; and body mass: 70.2 ± 10.5 kg] performed maximal effort of ECCs cycling exercise with fast speed (Fast) and slow speed trials (Slow). First, subjects performed the Fast for 5 min by one leg. Second, Slow performed until the total mechanical work was equal to that generated during Fast other one leg. Changes in maximal voluntary isometric contraction (MVC) torque of knee extension, isokinetic pedaling peak torque (IPT), range of motion (ROM), muscle soreness, thigh circumference, muscle echo intensity, and muscle stiffness were assessed before exercise, and immediately after exercise, and 1 and 4 days after exercise.
Results: Exercise time was observed in the Slow (1422.0 ± 330.0 s) longer than Fast (300.0 ± 0.0 s). However, a significant difference was not observed in total work (Fast:214.8 ± 42.4 J/kg, Slow: 214.3 ± 42.2 J/kg). A significant interaction effect was not observed in peak values of MVC torque (Fast:1.7 ± 0.4 Nm/kg, Slow: 1.8 ± 0.5 Nm/kg), IPT, muscle soreness (Fast:4.3 ± 1.6 cm, Slow: 4.7 ± 2.9 cm). In addition, ROM, circumference, muscle thickness, muscle echo intensity, and muscle stiffness also showed no significant interaction.
Conclusion: The magnitude of muscle damage is similar for ECCs cycling with equal work regardless of velocity.
## Introduction
Eccentric muscle actions (ECCs), in which the muscles become tensioned while being stretched, could cause by resulting in micro damage to the sarcomere or an inflammatory response, decreased muscle strength, limited flexibility, delayed onset muscle soreness (DOMS), muscle swelling, increased muscle stiffness, creatine kinase (CK), myoglobin (Mb), and interleukin (IL)-6 in blood. ( Clarkson et al., 1992; Morgan and Allen, 1999; Chen et al., 2007; Ochi et al., 2016; Tsuchiya et al., 2019). The degree of muscle damage is caused by ECCs depends on the exercise duration, length, intensity (Nosaka and Sakamoto, 2001; Nosaka and Newton, 2002), repetitions (Hesselink et al., 1996; Chen and Nosaka, 2006) and velocity (Chapman et al., 2006; Chapman et al., 2008a; Ueda et al., 2020).
Several studies have examined the effect of velocities of ECCs on muscle damage (Paddon-Jones et al., 2005; Chapman et al., 2006; Chapman et al., 2008a). In ECCs of elbow flexors, the fast velocity (210°/sec) resulted in greater torque deficit, increased DOMS, upper arm swelling and increased blood CK than the slow velocity (30°/sec) (Chapman et al., 2006). However, the number of contractions in this study differed significantly because the contraction times were matched between the fast and slow conditions (210 contractions in the fast condition and 30 contractions in the slow condition). Therefore, Chapman et al. ( 2008a) compared muscle damage after ECCs in elbow flexion at different contraction velocities under the condition of similar number of contractions. The results showed that the degree of muscle damage was greater in the fast (210°/sec) than in the slow (30°/sec) even when the number of contractions were combined (210 contractions). Similarly, Barreto et al. [ 2019] reported that slow-velocity ECCs in elbow flexion presented faster recovery of muscle strength and less muscle soreness compared with high-velocity. Therefore, it showed that the degree of muscle damage in ECCs due to elbow flexion is greater in the fast condition than in the slow, even when the contraction time and number of contractions were standardized. However, the total work was not exactly the same in the previous studies.
Long-term ECCs cycling training improved muscle strength and increased muscle hypertrophy despite being less demanding on the respiratory circulatory system than concentric cycling training (LaStayo et al., 2000). Therefore, it has been reported to be an effective exercise for the elderly, obese and patients with chronic obstructive pulmonary disease (COPD) (Lastayo et al., 1999; LaStayo et al., 2000; Gonzalez-Bartholin et al., 2019; Julian et al., 2019; Nickel et al., 2020). As acute response to ECCs cycling, we investigated the effects of different velocities on muscle damage in one bout of ECCs cycling (Ueda et al., 2020). Our results showed that muscle damage and delayed onset muscle soreness were significantly greater in the fast (210°/sec) than in the slow (30°/sec) condition (Ueda et al., 2020). Therefore, similar to the above findings in elbow flexors, we concluded that the fast ECCs exercise causes greater muscle damage than the slow. However, a limitation of this study is that the workload (365.7 ± 60.6 W) of the fast was significantly greater than that of the slow (68.3 ± 26.6 W) because the exercise duration was unified to 5 min. Thus, it remains unclear which factor, out of pedaling speed or workload, affected the differences in the degree of muscle damage. Contrarily, previous studies on animals have reported that the torque exerted during ECCs exercise and workload is a factor that determines the magnitude of muscle damage (Lieber and Friden, 1993; Warren et al., 1993; Talbot and Morgan, 1998). However, studies examining the effects of differences in velocity on muscle damage in humans under similar workload have been lacking.
Therefore, the purpose of this study was to compare the muscle damage caused by fast-speed (210°/sec) and slow-speed (30°/sec) ECCs cycling under equal the mechanical work (J) conditions. We hypothesized that ECCs cycling motion with different pedaling speeds had the same degree of muscle damage under uniform workload condition.
## Subjects
Nineteen young men were recruited (mean ± standard deviation (SD) age: 21.0 ± 2.2 years; height: 172.7 ± 5.9 cm; and body mass: 70.2 ± 10.5 kg). None of the subjects had participated in any regular resistance training for at least 1 year prior to this study. The participants were requested to avoid participation in other clinical trials and interventions, such as hot and cold baths, massage, stretching, strenuous exercise, excessive food, or alcohol consumption, and taking any supplement or medication at least 3 months before and during this trial. All subjects were provided with detailed explanations of the study protocol prior to participation and signed an informed consent form in accordance with the Declaration of Helsinki before being enrolled in this study. Written informed consent was obtained from the individual for the publication of any potentially identifiable images or data included in this article. This study was approved by the Ethics Committee for Human Experiments at Teikyo Heisei University (ID: R01-058-2).
## Experimental protocols
The subjects randomly performed maximal-effort ECCs cycling exercise are unilateral by each leg. ECCs cycling exercises were randomly performed on the same day by the non-dominant leg or dominant leg. First, all subjects performed the fast velocity session (Fast) for 5 min. Second, the slow velocity session (Slow) was reached when the total mechanical work done was equal to that generated during Fast (defined as ‘‘mechanical work”), which was also automatically calculated by ECCs cycling. Previous studies have reported that the initial bout of maximal eccentric muscle actions is responsible for conferring protective effects to the contralateral side (Howatson and van Someren, 2007; Xin et al., 2014; Chen et al., 2016; Tsuchiya et al., 2018). We had set the interval between the slow and fast velocities to 15 min, as this effect occurs when the second bout is performed from 1 day to 4 weeks (Chen et al., 2016). The legs were randomly assigned using a table of random numbers to minimize the intergroup differences in terms of age, body weight, and body mass index (BMI). The dependent variables included maximal voluntary isometric contraction (MVC) torque of knee extension, isokinetic pedaling peak torque (IPT) (30° and 210°/s slow and fast velocities, respectively), ROM of the knee joint, muscle soreness assessed using a visual analog scale (VAS), the circumference of the thigh, and echo intensity, muscle thickness, and shear elastic modulus using the ultrasonic scanner. VAS, echo intensity, muscle thickness, and shear elastic modulus were muscles of target the vastus lateralis, vastus medialis, and rectus femoris. These measurements were performed before, immediately after and 1 and 4 days after the ECCs cycling exercise. All subjects attended a familiarization session at least 1 week before the exercise where the subjects were briefed on eccentric exercise protocols. In the familiarization session, the subjects practiced for 3 min with a very light load of cycling exercise in ECCs mode similar to the present experiment.
## Eccentric cycling
The velocities of the ECCs cycling exercise were either 30°/s (5 rpm; Slow) or 210°/s (35 rpm; Fast) using a cycle ergometer (Strength Ergo 240 BK-ERG-003, Mitsubishi Electric Engineering, Tokyo, Japan). The cycling time of 5 min for the Fast was set based on a previous study (Elmer et al., 2010; Ueda et al., 2020). After the Fast, the Slow reached when the total work done was equal to that generated during the Fast. This ergometer was controlled by a servo motor which could be programmed with various exercise programs using a personal computer. For the testing position, the recumbent position was set at a seat angle, i.e., the angle between the backrest and the seat was set to 105°, and the pedal shaft was set at 55 cm from ground level (Kato et al., 2011). The subjects were secured with seat belts for safety. The left and right cranks and pedals of the ergometer were all set to the fixed mode, which enabled the subjects to put their feet on the cleated shoes fitted on the pedals and then generate exercise of the dorsal or plantar flexion of the right ankle joint. The exercise starting positions of the cranks, pedals, and seat were adjusted for enabling the subjects to maintain a comfortable and fixed posture. The subjects were asked to perform all bouts of exercise using either the right or left lower limb (hip and knee joint at 45° of flexion; ankle joint at 0° of plantar/dorsal flexion) and to relax the other lower limb (hip and knee joint at 0° of flexion/extension; relaxed ankle joint) throughout the experiments (Liang et al., 2011; Ueda et al., 2020). The non-exercising leg was secured to a stabilization platform. The range of motion of the knee joint during cycling ranged from about 20° to 120° (0°, full extension). The mechanical work performed during cycling were recorded at a 10-Hz sampling rate in a computer connected to the cycle ergometer (Strength Ergo 240 BK-ERG-003, Mitsubishi Electric Engineering, Tokyo, Japan).
## Maximal voluntary isometric contraction (MVC) torque of knee extension
For the measurement of MVC torque of knee extension, the participants performed the two times 3-s MVCs at knee joint angles of 90° with a 60-s rest between the contractions. The peak torque was considered as the MVC torque of knee extension. The torque signal was amplified using a strain amplifier (LUR-A-100NSA1; Kyowa Electronic Instruments, Tokyo, Japan). The analog torque signal was converted to digital signal using a 16-bit analog-to-digital converter (Power-Lab 16SP; AD Instruments, Bella Vista, Australia). The sampling frequency was set at 10 kHz. The measurement was performed as previously described (Sasaki et al., 2011).
Significant interaction effect was not observed in the MVC torque of knee extension between the Fast and Slow groups are shown in Figure 1A. However, a significant time effect was found at the MVC torque of knee extension. MVC torque of knee extension at the baseline was similar between the two groups (Fast: 1.7 ± 0.4 Nm/kg; Slow: 1.8 ± 0.5 Nm/kg). Compared with the pre-exercise value, MVC torque in both groups significantly decreased immediately after exercise and remained decreased up to 4 days after exercise ($p \leq 0.05$).
**FIGURE 1:** *Changes (mean ± SD) in maximal voluntary isometric contraction (MVC) torque of knee extension (A), and isokinetic pedaling peak torque (IPT) torque of pedaling at 210°/s (B), and 30°/s (C), before (pre), immediately after (post), 1 day, and 4 days after exercise in the slow velocity session (Slow) and fast velocity session (Fast).*
## Isokinetic pedaling peak torque (IPT)
IPT torque of pedaling was applied by measured in a cycle ergometer which is a device similar to the one used for performing eccentric cycling (Strength Ergo 240 BK-ERG-003, Mitsubishi Electric Engineering, Tokyo, Japan). For the measurement of IPT torque of pedaling, the subject performed two three pedals IPT at 30° and 210°/s with a 60-s resting period between contractions. The peak torque of each velocity was used as the IPT.
Significant interaction effect was not observed in IPT torque of pedaling at 210°/s between Fast and Slow groups. IPT of pedaling at 210°/s at the baseline was the same between the two groups (Fast: 1.9 ± 0.35 Nm/kg; Slow: 2.0 ± 0.42 Nm/kg) (Figure 1B). The significant interaction effect was not observed in IPT of pedaling at 30°/s between Fast and Slow groups. However, a significant time effect was found at IPT of pedaling at 30°/s. IPT of pedaling at 30°/s at the baseline was similar between the two groups (Fast: 2.4 ± 0.5 Nm/kg; Slow: 2.5 ± 0.4 Nm/kg). Compared with the pre-exercise value, IPT at 30°/s in the Fast significantly decreased immediately after exercise and remained decreased up to 1 day after exercise, but the Slow significantly decreased immediately after exercise and remained decreased up to 4 days after exercise ($p \leq 0.05$) (Figure 1C).
## Muscle soreness
Muscle soreness was assessed using a 10-cm VAS in which 0 indicated “no pain” and 10 indicated “the worst pain imaginable”; the subject indicated his pain level on this VAS scale. Muscle soreness was assessed by pressure, using a digital muscle stiffness instrument (NEUTONE TDM-NA1, Satou Shouji Inc., Kanagawa, Japan) on vastus lateralis, rectus femoris, and vastus medialis. The pressure was applied perpendicularly to the point on each muscle. The pressures were applied to the vastus lateralis and rectus femoris at the lateral femoral epicondyle and $50\%$ of the greater trochanter, and the vastus medialis at the lateral femoral epicondyle and $30\%$ distal to the greater trochanter. All tests were conducted by the same investigator who had practiced applying the same pressure over time and on different participants.
Significant interaction effect was not observed in muscle soreness (Figure 2). Muscle soreness in three muscle groups at the baseline was similar in the two conditions (vastus lateralis: Fast: 1.8 ± 1.1 cm, Slow: 2.1 ± 1.2 cm; rectus femoris: Fast: 1.8 ± 1.5 cm, Slow: 1.8 ± 1.4 cm; vastus medialis: Fast: 3.3 ± 1.7 cm, Slow: 3.7 ± 1.9 cm). Compared with the pre-exercise values was not significant difference after exercise at any time points in vastus lateralis, rectus femoris, and vastus medialis in both conditions.
**FIGURE 2:** *Changes (mean ± SD) in muscle soreness were recorded using a visual analog scale for the vastus lateralis (A), rectus femoris (B), and vastus medialis (C) immediately after (post), 1 day, and 4 days after exercise in the slow velocity session (Slow) and fast velocity session (Fast).*
## Range of motion
Range of motion was determined as the difference in the joint angles between maximal voluntary flexion and extension of the knee joint using a goniometer (Takase Medical, Tokyo, Japan). The flexion was measured when the subject attempted to maximally flex the knee joint of the exercised leg to touch his hip with his heel while keeping the knee joint aligned to the standing leg and supporting the body by placing both hands on the wall, 30 cm from the foot. The extension was measured when the subject attempted to extend the knee joint of the exercised leg as much as possible. ROM was calculated by subtracting the flexion from extension of the knee joint (Chen et al., 2011; Chen et al., 2013; Ueda et al., 2020).
## Circumference
When each subject stood with his feet approximately 10 cm apart, with his body weight evenly distributed on both feet, the perimeter distance of the thigh perpendicular to the long axis of the femur at the marked mid-trochanterion-tibiale level was measured (Chen et al., 2011). The measurements were performed thrice for each time point, and the average of the three measurements was used for further analysis.
## Muscle stiffness, muscle thickness, and echo intensity
Using ultrasound shear wave elastography, we measured muscle stiffness at vastus lateralis, rectus femoris, and vastus medialis with the probe placed at the position (the vastus lateralis and rectus femoris at the lateral femoral epicondyle and $50\%$ of the greater trochanter, and the vastus medialis at the lateral femoral epicondyle and $30\%$ distal to the greater trochanter) marked for the circumference measurement. An ultrasonic scanner (Aixplorer version 4.2, Supersonic Imagine, France) was used in shear wave elastography mode with a musculoskeletal preset. An electronic linear array probe (SL15-4, Supersonic Imagine France) coated with water soluble transmission gel was placed longitudinally on each muscle head. Muscle shear modulus (μ), a measure of normalized muscle stiffness was calculated using the following equation: μ = ρVs2, where ρ is the density of muscle (assumed to be 1,000 kg/m3) and Vs. is the velocity of shear wave propagation caused by the focused ultrasound beam from the scanner. A 10-mm square map of the muscle shear modulus with a spatial resolution of 1 × 1 mm2 was obtained with each ultrasound image. We calculated the average muscle stiffness by combining the measurements obtained for vastus lateralis, rectus femoris, and vastus medialis (Lacourpaille et al., 2017). A representative value of the shear modulus for each muscle head was then determined via spatial averaging over a 5-mm diameter circle (Ochi et al., 2018). Scanned images of each muscle were transferred to a personal computer and the thicknesses of the vastus lateralis, rectus femoris, and vastus medialis were manually calculated by tracing each muscle using image analysis software (ImageJ, MD, United States). To measure the echo intensity, the gains and contrast were kept consistent over the experimental period. The transverse images were analyzed in a computer, in bitmap (.bmp) format. The average echo intensity for the region of interest (20 × 20 mm) was calculated using ImageJ software that provided a grayscale histogram (0, black; 100, white) for the region, as described in a previous study (Tsuchiya et al., 2019). The echo intensity and muscle thickness were evaluated at the same locations as muscle stiffness.
## Statistical analyses
All analyses were performed using the SPSS software version 27.0 (IBM Corp., Armonk, NY, United States). Values are expressed as means ± SD. Exercise time, energy expenditure, mechanical work, peak torque performed during eccentric cycling, and the baseline data for all outcomes at Fast and Slow were compared using the paired t-test. Time courses of MVC torque of knee extension, IPT of pedaling, ROM, circumference, shear elastic modulus, muscle thickness, and echo intensity of values were calculated based on relative changes from the baseline. MVC torque of knee extension, IPT of pedaling, ROM, muscle soreness, echo intensity, muscle thickness, and shear elastic modulus were compared between the Fast and Slow groups via two-way repeated-measure analysis of variance (ANOVA). When a significant main effect or interaction was detected, Bonferroni’s correction was performed for the post hoc testing. A $p \leq 0.05$ was considered statistically significant.
## During ECCs cycling
As shown in Table 1, a significant difference in exercise time was observed in the Slow longer than Fast. However, a significant difference was not observed in the mechanical work between Slow and Fast.
**TABLE 1**
| Unnamed: 0 | Fast (210 deg/sec) | Slow (30 deg/sec) |
| --- | --- | --- |
| Exercise times (sec) | 300.0 ± 0.0 | 1422.0 ± 330.0 * |
| Mechanical work (J) | 14897.1 ± 2747.1 | 14868.9 ± 2763.0 |
| Mechanical work/body mass (J/kg) | 214.8 ± 42.4 | 214.3 ± 42.2 |
## Range of motion, circumference, and muscle thickness
Figure 3A showed that significant interaction effect was not observed in the range of motion between the Fast and Slow conditions. However, a significant time effect was found in ROM. ROM at the baseline was similar between the two groups (Fast: 112.4° ± 9.7°, Slow: 110.5° ± 8.93°). Compared with the pre-exercise value, ROM in the Slow significantly decreased only immediately after exercise ($p \leq 0.05$). On the other hand, ROM in the Fast was not significant after exercise at any time point compared to before exercise value. A significant interaction effect was not observed in circumference between the Fast and Slow conditions (Figure 3B). The circumference at the baseline was similar between the two groups (Fast: 54.0 ± 6.5 cm, Slow: 54.6 ± 6.8 cm). Compared with the pre-exercise value, circumference in both conditions was not significant difference after exercise at any time point. A significant interaction effect was not observed in muscle thickness between the Fast and Slow conditions. However, a significant time effect was found in muscle thickness (Figure 3C). The muscle thickness at the baseline was similar between the two groups (Fast: 9.2 ± 1.4 cm, Slow: 9.4 ± 1.5 cm). Compared with the pre-exercise value, muscle thickness in both conditions was not significant difference after exercise at any time point.
**FIGURE 3:** *Changes (mean ± SD) in range of motion (A), thigh circumference (B), and muscle thickness (C), immediately after (post), 1 day, and 4 days after exercise in the fast velocity session (Fast) and slow velocity session (Slow).*
## Echo intensity and muscle stiffness
A significant interaction effect was not observed in echo intensity between the Fast and Slow conditions (Figure 4A). Although there was a significant time effect for echo intensity, both conditions were no significant difference at any time point after exercise compared to pre-exercise values. The echo intensity at the baseline was similar between the two groups (Fast: 32.9 ± 17.0, Slow: 34.1 ± 17.3). A significant interaction effect was not observed in muscle stiffness between the Fast and Slow conditions (Figure 4B). However, a significant time effect was found in muscle stiffness. The muscle stiffness at the baseline was the same between the two conditions (Fast: 14.2 ± 6.8 kPa, Slow: 15.6 ± 8.7 kPa). Compared with the pre-exercise value, muscle stiffness in the Slow significantly increased immediately after exercise and remained increased up to 4 days after exercise ($p \leq 0.05$). On the other hand, muscle stiffness in the Fast was not significant after exercise at any time point compared to before exercise value.
**FIGURE 4:** *Changes (mean ± SD) in echo intensity of the quadriceps (A), and muscle stiffness of the knee extensor (vastus laterials, rectus femoris, and vastus medialis) (B), before (pre), immediately after (post), 1 day, and 4 days after eccentric muscle actions in the slow velocity session (Slow) and fast velocity session (Fast).*
## Discussion
The present study compared the muscle damage caused by fast (210°/sec) and slow (30°/sec) ECCs cycling, under equal mechanical work conditions. The results showed absence of difference in muscle damage owing to the difference in pedaling velocity, although torque deficit and delayed onset muscle soreness were observed in both conditions. These results supported our hypothesis.
In the present study, the slow condition was performed until the mechanical work was equal to that obtained in the fast condition, so the slow (1422.0 ± 330.0 s) movement time was much longer than the fast (300.0 ± 0.0 s) (Table 1). On the other hand, the present study monitored the accumulated mechanical work from the torque output, which made it possible to unify the mechanical work of Fast and Slow during the exercise. In a study comparing the degree of muscle damage caused by differences in contraction speed, the contraction time (120 s) of the ECCs exercise with elbow flexion was unified (Chapman et al., 2006), and in another study, the number of contractions was unified to 30 or 210 times (Chapman et al., 2008a). Both studies showed that the muscle damage was greater in the fast-speed condition than in the slow-speed. However, although these previous studies (Chapman et al., 2006; Chapman et al., 2008a) have standardized the contraction time and number of contractions, they have not standardized the mechanical work. Therefore, the present study is the first to examine the effects of different pedaling velocities on muscle damage under equivalent mechanical work conditions using ECCs cycling.
In the present study, there is no significant interaction between the MVC torque of knee extension and IPT of pedaling at 30 and 210 (Figure 1). A previous study comparing the contraction velocity of the ECCs movement with elbow flexors in the fast speed (210°/s) and low speed (30°/s) conditions reported significantly greater torque deficit in the fast (Chapman et al., 2008a) 2). In our previous study, the exercise duration was standardized to 5 min in ECCs cycling, and the results were compared between the fast (210°/s) and slow (30°/s) conditions. The results of this study showed that the torque deficit was significantly greater in the fast (Ueda et al., 2020). Furthermore, in a previous study in which mechanical work was calculated by the number of repetitions and contraction time of ECCs in elbow flexion, it was reported that the mechanical work had no effect on muscle strength loss during ECCs (Chapman et al., 2008b). From these previous studies, it is concluded that contraction velocity during exercise is strongly related to muscle strength loss after ECCs. Contrarily, Mavropalias et al. [ 2020] compared the muscle damage before and after ECCs cycling at high ($20\%$ of peak power for 1 min x five sets) and low intensities ($5\%$ of peak power for 4 min x five sets). They reported that a significant difference in IPT of pedaling at 90°/s in both conditions was absent, despite the 4-fold difference in exercise intensity. Although differences in contraction velocity were not examined, it is suggested that mechanical work may be related to post-exercise muscle weakness in ECCs cycling. In light of these previous studies and the results of the present study, we suggest that the degree of muscle strength loss may be dependent on mechanical work than on velocity during ECCs cycling under equal mechanical work conditions.
The delayed onset of muscle soreness in the present study did not significantly differ between the conditions (Figure 2). In our previous study (Ueda et al., 2020) showed that the delayed onset of muscle soreness of the rectus femoris muscle and medial vastus medialis was significantly higher in the fast (210°/sec) than in slow (30°/sec). However, in this study, the mechanical work during exercise was significantly greater in the fast than in the slow. Regarding the relationship between mechanical work and delayed onset muscle soreness, Paschalis et al. [ 2005] performed ECCs in knee extension in the high intensity condition (10 × 12 sets at maximal effort) followed by the low intensity condition. The results showed that muscle soreness in both conditions was comparable to that in the high-intensity condition. The results reported absence of difference in the muscle soreness between the two conditions, suggesting that mechanical work may be a more important factor for muscle soreness than exercise intensity (Paschalis et al., 2005). Although this study did not examine differences in velocity, the involvement of mechanical work in delayed onset of muscle soreness after ECCs supports the results of the present study. These results suggest that delayed onset muscle soreness after ECCs cycling may be similar when mechanical work is unified even at different contraction velocities and exercise intensities.
It has been reported that the fast velocity (210°/s) ECCs of elbow flexion, i was associated with significantly greater ROM limitation and muscle swelling than the slow velocity (30°/s) (Chapman et al., 2006; Chapman et al., 2008a). Contrarily, in our previous study (Ueda et al., 2020) and present study, differences in ROM and muscle swelling between the fast (210°/s) and slow (30°/s) speed conditions were not observed after the 5 min of ECCs cycling. In the present study, muscle stiffness and echo intensity were also assessed in order to evaluate the condition of skeletal muscle. Changes in muscle stiffness after ECCs have been reported to reflect a rapid disruption of calcium homeostasis after exercise-induced myofibrillar destruction (Lacourpaille et al., 2014; Lacourpaille et al., 2017). It has also been speculated that the increase in echo intensity reflects the influx of water into the muscle (Nosaka and Clarkson, 1996). Similar to the present study, significant differences in changes in muscle stiffness and echo intensity during 5 min of ECCs cycling in the fast (210°/s) and slow (30°/s) conditions were absent in our previous study (Ueda et al., 2020). Future studies should clarify the mechanism in more detail.
The present study has several limitations that should be considered. First, although the mechanical work is matched in this study, it is unclear whether the average power output in ECCs cycling was consistent in both conditions. Further study is necessary to examine the effects of not only the mechanical work and pedaling speed but also the factor of power output on muscle damage. Second, the mechanical work of the Fast condition was smaller than the mechanical work of our previous study under similar exercise conditions (Ueda et al., 2020). The baseline IPT of pedaling at 30°/s was smaller in the present study (169.8 ± 42.9 Nm) than in the previous study (205.4 ± 35.9 Nm). Therefore, it is likely that the present study had different muscular characteristics of the subjects compared to the previous study (Ueda et al., 2020). Third, only two exercise conditions were used in the present study. Future studies should examine the effects of mechanical work and contraction velocity on muscle damage in detail under multiple velocity conditions and exercise loads that cause more severe damage.
## Conclusion
The results of this study suggest that muscle damage by ECCs cycling may be affected by mechanical work more than contraction velocity. The present study may provide useful information for determining the speed conditions for ECCs cycling training. A recent review article (Barreto et al., 2021) has stated that the development of muscle damage should be considered in ECCs cycling training program. In particular, it is paramount to avoid DOMS during the training program in rehabilitation settings. We believe that our results suggest that low pedal cadence may be safer in rehabilitation settings, while high pedal cadence may be more efficient when applying large loads in a short time. Future studies should examine mechanical work, pedaling velocity, and long-term training adaptation.
## Data availability statement
The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.
## Ethics statement
The studies involving human participants were reviewed and approved by the Committee for Human Experimentation of Teikyo Heisei University. The patients/participants provided their written informed consent to participate in this study.
## Author contributions
EO designed the study. HU, YT, and RS measured and collected the data. HU and YT analyzed the data and wrote the main parts of the manuscript. HU, YT, RS, and EO reviewed the manuscript.
## Conflict of interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher’s note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## References
1. Barreto R. V., De Lima L. C. R., Denadai B. S.. **Moving forward with backward pedaling: A review on eccentric cycling**. *Eur. J. Appl. Physiol.* (2021) **121** 381-407. DOI: 10.1007/s00421-020-04548-6
2. Barreto R. V., De Lima L. C. R., Greco C. C., Denadai B. S.. **Protective effect conferred by isometric preconditioning against slow- and fast-velocity eccentric exercise-induced muscle damage**. *Front. Physiol.* (2019) **10** 1203. DOI: 10.3389/fphys.2019.01203
3. Chapman D., Newton M., Sacco P., Nosaka K.. **Greater muscle damage induced by fast versus slow velocity eccentric exercise**. *Int. J. Sports Med.* (2006) **27** 591-598. DOI: 10.1055/s-2005-865920
4. Chapman D. W., Newton M. J., Zainuddin Z., Sacco P., Nosaka K.. **Work and peak torque during eccentric exercise do not predict changes in markers of muscle damage**. *Br. J. Sports Med.* (2008b) **42** 585-591. DOI: 10.1136/bjsm.2007.037929
5. Chapman D. W., Newton M., Mcguigan M., Nosaka K.. **Effect of lengthening contraction velocity on muscle damage of the elbow flexors**. *Med. Sci. Sports Exerc* (2008a) **40** 926-933. DOI: 10.1249/MSS.0b013e318168c82d
6. Chen T. C., Chen H. L., Lin M. J., Yu H. I., Nosaka K.. **Contralateral repeated bout effect of eccentric exercise of the elbow flexors**. *Med. Sci. Sports Exerc* (2016) **48** 2030-2039. DOI: 10.1249/MSS.0000000000000991
7. Chen T. C., Lin K. Y., Chen H. L., Lin M. J., Nosaka K.. **Comparison in eccentric exercise-induced muscle damage among four limb muscles**. *Eur. J. Appl. Physiol.* (2011) **111** 211-223. DOI: 10.1007/s00421-010-1648-7
8. Chen T. C., Nosaka K.. **Effects of number of eccentric muscle actions on first and second bouts of eccentric exercise of the elbow flexors**. *J. Sci. Med. Sport* (2006) **9** 57-66. DOI: 10.1016/j.jsams.2006.03.012
9. Chen T. C., Nosaka K., Sacco P.. **Intensity of eccentric exercise, shift of optimum angle, and the magnitude of repeated-bout effect**. *J. Appl. Physiol. (1985)* (2007) **102** 992-999. DOI: 10.1152/japplphysiol.00425.2006
10. Chen T. C., Tseng W. C., Huang G. L., Chen H. L., Tseng K. W., Nosaka K.. **Low-intensity eccentric contractions attenuate muscle damage induced by subsequent maximal eccentric exercise of the knee extensors in the elderly**. *Eur. J. Appl. Physiol.* (2013) **113** 1005-1015. DOI: 10.1007/s00421-012-2517-3
11. Clarkson P. M., Nosaka K., Braun B.. **Muscle function after exercise-induced muscle damage and rapid adaptation**. *Med. Sci. Sports Exerc* (1992) **24** 512-520. DOI: 10.1249/00005768-199205000-00004
12. Elmer S. J., Mcdaniel J., Martin J. C.. **Alterations in neuromuscular function and perceptual responses following acute eccentric cycling exercise**. *Eur. J. Appl. Physiol.* (2010) **110** 1225-1233. DOI: 10.1007/s00421-010-1619-z
13. Gonzalez-Bartholin R., Mackay K., Valladares D., Zbinden-Foncea H., Nosaka K., Penailillo L.. **Changes in oxidative stress, inflammation and muscle damage markers following eccentric versus concentric cycling in older adults**. *Eur. J. Appl. Physiol.* (2019) **119** 2301-2312. DOI: 10.1007/s00421-019-04213-7
14. Hesselink M. K., Kuipers H., Geurten P., Van Straaten H.. **Structural muscle damage and muscle strength after incremental number of isometric and forced lengthening contractions**. *J. Muscle Res. Cell Motil.* (1996) **17** 335-341. DOI: 10.1007/BF00240930
15. Howatson G., Van Someren K. A.. **Evidence of a contralateral repeated bout effect after maximal eccentric contractions**. *Eur. J. Appl. Physiol.* (2007) **101** 207-214. DOI: 10.1007/s00421-007-0489-5
16. Julian V., Thivel D., Miguet M., Pereira B., Costes F., Coudeyre E.. **Eccentric cycling is more efficient in reducing fat mass than concentric cycling in adolescents with obesity**. *Scand. J. Med. Sci. Sports* (2019) **29** 4-15. DOI: 10.1111/sms.13301
17. Kato M., Tsutsumi T., Yamaguchi T., Kurakane S., Chang H.. **Characteristics of maximum performance of pedaling exercise in recumbent and supine positions**. *J. Sports Sci. Med.* (2011) **10** 491-497. PMID: 24150623
18. Lacourpaille L., Nordez A., Hug F., Couturier A., Dibie C., Guilhem G.. **Time-course effect of exercise-induced muscle damage on localized muscle mechanical properties assessed using elastography**. *Acta Physiol. (Oxf)* (2014) **211** 135-146. DOI: 10.1111/apha.12272
19. Lacourpaille L., Nordez A., Hug F., Doguet V., Andrade R., Guilhem G.. **Early detection of exercise-induced muscle damage using elastography**. *Eur. J. Appl. Physiol.* (2017) **117** 2047-2056. DOI: 10.1007/s00421-017-3695-9
20. Lastayo P. C., Pierotti D. J., Pifer J., Hoppeler H., Lindstedt S. L.. **Eccentric ergometry: Increases in locomotor muscle size and strength at low training intensities**. *Am. J. Physiol. Regul. Integr. Comp. Physiol.* (2000) **278** R1282-R1288. DOI: 10.1152/ajpregu.2000.278.5.R1282
21. Lastayo P. C., Reich T. E., Urquhart M., Hoppeler H., Lindstedt S. L.. **Chronic eccentric exercise: Improvements in muscle strength can occur with little demand for oxygen**. *Am. J. Physiol.* (1999) **276** R611-R615. DOI: 10.1152/ajpregu.1999.276.2.R611
22. Liang N., Nakamoto T., Mochizuki S., Matsukawa K.. **Differential contribution of central command to the cardiovascular responses during static exercise of ankle dorsal and plantar flexion in humans**. *J. Appl. Physiol. (1985)* (2011) **110** 670-680. DOI: 10.1152/japplphysiol.00740.2010
23. Lieber R. L., Friden J.. **Muscle damage is not a function of muscle force but active muscle strain**. *J. Appl. Physiol. (1985)* (1993) **74** 520-526. DOI: 10.1152/jappl.1993.74.2.520
24. Mavropalias G., Koeda T., Barley O. R., Poon W. C. K., Fisher A. J., Blazevich A. J.. **Comparison between high- and low-intensity eccentric cycling of equal mechanical work for muscle damage and the repeated bout effect**. *Eur. J. Appl. Physiol.* (2020) **120** 1015-1025. DOI: 10.1007/s00421-020-04341-5
25. Morgan D. L., Allen D. G.. **Early events in stretch-induced muscle damage**. *J. Appl. Physiol. (1985)* (1999) **87** 2007-2015. DOI: 10.1152/jappl.1999.87.6.2007
26. Nickel R., Troncoso F., Flores O., Gonzalez-Bartholin R., Mackay K., Diaz O.. **Physiological response to eccentric and concentric cycling in patients with chronic obstructive pulmonary disease**. *Appl. Physiol. Nutr. Metab.* (2020) **45** 1232-1237. DOI: 10.1139/apnm-2020-0149
27. Nosaka K., Clarkson P. M.. **Changes in indicators of inflammation after eccentric exercise of the elbow flexors**. *Med. Sci. Sports Exerc* (1996) **28** 953-961. DOI: 10.1097/00005768-199608000-00003
28. Nosaka K., Newton M.. **Difference in the magnitude of muscle damage between maximal and submaximal eccentric loading**. *J. Strength Cond. Res.* (2002) **16** 202-208. DOI: 10.1519/1533-4287(2002)016<0202:ditmom>2.0.co;2
29. Nosaka K., Sakamoto K.. **Effect of elbow joint angle on the magnitude of muscle damage to the elbow flexors**. *Med. Sci. Sports Exerc* (2001) **33** 22-29. DOI: 10.1097/00005768-200101000-00005
30. Ochi E., Maruo M., Tsuchiya Y., Ishii N., Miura K., Sasaki K.. **Higher training frequency is important for gaining muscular strength under volume-matched training**. *Front. Physiol.* (2018) **9** 744. DOI: 10.3389/fphys.2018.00744
31. Ochi E., Tsuchiya Y., Nosaka K.. **Differences in post-exercise T2 relaxation time changes between eccentric and concentric contractions of the elbow flexors**. *Eur. J. Appl. Physiol.* (2016) **116** 2145-2154. DOI: 10.1007/s00421-016-3462-3
32. Paddon-Jones D., Keech A., Lonergan A., Abernethy P.. **Differential expression of muscle damage in humans following acute fast and slow velocity eccentric exercise**. *J. Sci. Med. Sport* (2005) **8** 255-263. DOI: 10.1016/s1440-2440(05)80036-2
33. Paschalis V., Koutedakis Y., Jamurtas A. Z., Mougios V., Baltzopoulos V.. **Equal volumes of high and low intensity of eccentric exercise in relation to muscle damage and performance**. *J. Strength Cond. Res.* (2005) **19** 184-188. DOI: 10.1519/R-14763.1
34. Sasaki Y., Sasaki T., Ishii N.. **Acceleration and force reveal different mechanisms of electromechanical delay**. *Med. Sci. Sports Exerc.* (2011) **43** 1200-1206. DOI: 10.1249/MSS.0b013e318209312c
35. Talbot J. A., Morgan D. L.. **The effects of stretch parameters on eccentric exercise-induced damage to toad skeletal muscle**. *J. Muscle Res. Cell Motil.* (1998) **19** 237-245. DOI: 10.1023/a:1005325032106
36. Tsuchiya Y., Nakazato K., Ochi E.. **Contralateral repeated bout effect after eccentric exercise on muscular activation**. *Eur. J. Appl. Physiol.* (2018) **118** 1997-2005. DOI: 10.1007/s00421-018-3933-9
37. Tsuchiya Y., Yanagimoto K., Ueda H., Ochi E.. **Supplementation of eicosapentaenoic acid-rich fish oil attenuates muscle stiffness after eccentric contractions of human elbow flexors**. *J. Int. Soc. Sports Nutr.* (2019) **16** 19. DOI: 10.1186/s12970-019-0283-x
38. Ueda H., Tsuchiya Y., Ochi E.. **Fast-velocity eccentric cycling exercise causes greater muscle damage than slow eccentric cycling**. *Front. Physiol.* (2020) **11** 596640. DOI: 10.3389/fphys.2020.596640
39. Warren G. L., Hayes D. A., Lowe D. A., Armstrong R. B.. **Mechanical factors in the initiation of eccentric contraction-induced injury in rat soleus muscle**. *J. Physiol.* (1993) **464** 457-475. DOI: 10.1113/jphysiol.1993.sp019645
40. Xin L., Hyldahl R. D., Chipkin S. R., Clarkson P. M.. **A contralateral repeated bout effect attenuates induction of NF-κB DNA binding following eccentric exercise**. *J. Appl. Physiol. (1985)* (2014) **116** 1473-1480. DOI: 10.1152/japplphysiol.00133.2013
|
---
title: Effects of running-based versus body-weight-based high-intensity interval training
on physical fitness in healthy adolescents
authors:
- Zhen Li
- Yang Liu
- Xiaowei Han
- Zhixiong Zhou
journal: Frontiers in Physiology
year: 2023
pmcid: PMC10036788
doi: 10.3389/fphys.2023.1060216
license: CC BY 4.0
---
# Effects of running-based versus body-weight-based high-intensity interval training on physical fitness in healthy adolescents
## Abstract
Objectives: High-intensity interval training improves aerobic endurance, but the effectiveness of different training protocols is unclear. This study compared the effects of running-based high-intensity interval training (R-HIIT) and body weight-based high-intensity interval training (B-HIIT) on physical fitness in adolescents.
Methods: This was a pre-and post-test quasi-experimental design in which a seventh-grade natural class was randomly selected from three homogeneous middle schools, and then the three natural classes were randomly divided into three groups: the R-HIIT group ($$n = 54$$), the B-HIIT group ($$n = 55$$), and the control group (Con, $$n = 57$$). Both intervention groups exercised twice a week for 12 weeks with a 2:1 (1 min:30 s) load-interval ratio and exercise intensity controlled at $70\%$–$85\%$ maximum heart rate. R-HIIT was in the form of running, and B-HIIT was in the form of resistance exercises using the participants’ body weight. The control group was instructed to continue their normal behavior. cardiorespiratory fitness, muscle strength and endurance, and speed were measured before and after the intervention. Statistical differences between and within groups were determined using repeated measures analysis of variance.
Results: Compared to the baseline, both the R-HIIT and B-HIIT intervention groups significantly improved CRF, muscle strength, and speed ($p \leq 0.05$). The B-HIIT group was significantly better than R-HIIT in improving CRF (4.48 mL/kg/min vs 3.34 mL/kg/min, $p \leq 0.05$), and only the B-HIIT improved sit-up muscle endurance (ηp 2 = 0.30, $p \leq 0.05$).
Conclusion: The B-HIIT protocol was significantly more effective than the R-HIIT protocol in developing CRF and improving muscle health indicators.
## 1 Introduction
Participation in regular physical activities (PA) promotes the physical (Lubans et al., 2022), psychological (Lubans et al., 2016), and cognitive (Donnelly et al., 2016) health of children and adolescents. The World Health Organization recommends that children and adolescents should participate in 60 min of moderate to vigorous physical activity every day to obtain health benefits (Chaput et al., 2020). However, data show that less than $20\%$ of children and adolescents worldwide, and only $6\%$ in Asia, meet the recommended levels (Guthold et al., 2020; Loo et al., 2022). Furthermore, the global spread of COVID-19 has exacerbated the trend of declining activity levels among children and adolescents (Farooq et al., 2018; Sa et al., 2020). There is a strong association between a lack of physical activity and high levels of obesity and elevated blood pressure in children, which, if not treated promptly, significantly increases the risk of cardiovascular disease, cancer, and type 2 diabetes in adulthood (*Paulino da* Silva Bento et al., 2021).
School physical education (PE) lessons provide opportunities for children and adolescents to participate in physical activity and promote their health, but, due to curriculum and personal reasons, each student only participates in exercise between $18\%$ and $50\%$ of the total time in a PE lesson (Gehris et al., 2012; Hollis et al., 2017; Breithecker et al., 2021; Bossmann et al., 2022). Therefore, traditional interventions to improve physical health, such as long-term moderate-intensity continuous training (MICT), do not seem suitable for the school environment (*Paulino da* Silva Bento et al., 2021). In addition to the time factor, the intensity of PE lessons is insufficient, weakening their potential health benefits (Hollis et al., 2017; Breithecker et al., 2021; Bossmann et al., 2022). Therefore, it is necessary to explore and develop alternate ways to attract children and adolescents to realize the many health benefits related to regular physical activity.
Researchers and the public have paid much attention to high-intensity interval training (HIIT) in recent years because it can produce physiological benefits similar to MICT in a shorter time (Thompson, 2022). HIIT programs typically include relatively short but vigorous exercise—e.g., exercise achieving >$85\%$ of maximum heart rate (HRmax)—with active rest and recovery periods between intervals (Lubans et al., 2022). The effectiveness of a range of HIIT programs in improving various health indicators, including body composition, cardiorespiratory fitness, and cardiometabolic fitness, has been well established in children and adolescents (Logan et al., 2014; Costigan et al., 2015; Eddolls et al., 2017; Yin et al., 2020). However, considering that the strenuousness of HIIT may cause unhappiness (Ekkekakis et al., 2011), some experts believe that traditional interventions aimed at an increasing activity that rely on motivation and social support are largely ineffective, with the result that most people will not or cannot actively participate (Biddle and Batterham, 2015). In view of these factors, relying on the motivation and self-discipline of young people to participate in HIIT may not be sufficient to carry out health intervention on a large scale.
According to Beets et al. [ 2016], embedding HIIT into an organized, weekly period, located in an environment with the greatest contact with other children (such as school PE class), may overcome the past limitations of HIIT to carry out large-scale intervention and can also make up for the shortcomings of the short participation time and low intensity of traditional PE lessons. In this context, several studies have incorporated HIIT into school PE lessons (Bauer et al., 2022; Cao et al., 2022b) or other school activities (Engel et al., 2019), and the results have shown improved cardiac metabolic health and neuromuscular adaptation (Bauer et al., 2022). Other studies showed improved health-related parameters through shorter intervention time (Cao et al., 2022b; Popowczak et al., 2022), and HIIT exercises were found to be more enjoyable than traditional methods such as MICT (Bond et al., 2015; Malik et al., 2017).
Despite these prospects, many of these studies pay attention to the effect of HIIT on cardiorespiratory fitness (CRF) and cardiovascular metabolism but do not assess the impact on musculoskeletal health. The current international consensus is that it is important to include musculoskeletal health in the adolescent health assessment (Janz et al., 2021). Second, although the effectiveness of HIIT in a school environment has been proven, most studies have only used the traditional strategy of running-based high-intensity interval training (R-HIIT) (Costigan et al., 2015; Bond et al., 2017; Meng et al., 2022), which is a simple and easy assessment, but may not promote the development of other health indicators such as muscle strength (Hulteen et al., 2018). In addition, R-HIIT may require specialized equipment like treadmills and power bicycles, and/or facilities such as gyms and playgrounds, which may be impractical for some schools (Kennedy et al., 2020).
Body weight-based high-intensity interval training (B-HIIT) has received a lot of attention in recent years because it does not require specialized equipment and can be performed with minimal space. Evidence has shown that, under proper supervision, resistance exercises based on one’s own weight, do not negatively impact the growth and development of children and adolescents and can also improve related health indicators and reduce the risk of injury (Behm et al., 2008; Faigenbaum et al., 2009). Some studies have shown that B-HIIT is better than R-HIIT at improving health-related indicators in children and adolescents (Lubans et al., 2022), but others have come to the opposite conclusion (Engel et al., 2019). Therefore, it is unclear whether B-HIIT can be successfully used as a health promotion program with children and adolescents, especially in school settings.
Furthermore, many current health promotion initiatives for children and adolescents emphasize increasing PA levels, and knowledge of the factors influencing PA levels is limited to psychological, social, and environmental aspects (Dobbins et al., 2013; Gorga et al., 2016; Owen et al., 2017; Messing et al., 2019), omitting the potential influence of intrinsic biological control on normal physical activity (Beck et al., 2022). In order to maintain a level of PA or energy expenditure that is generally steady over time, the ActivityStat hypothesis hypothesizes that an imposed increase or decrease in PA in one domain can induce a compensatory adjustment in the opposite direction in another domain (Gomersall et al., 2013). This phenomenon has been confirmed in several intervention studies (Aburto et al., 2011; Haapala et al., 2017): school-based interventions have shown a slight or moderate effect of increasing PA within the school setting, but little change in overall levels of PA due to the use of compensatory mechanisms outside the school setting. However, some studies have also found that children and adolescents who regularly participate in organized sports do not show indicators of compensatory behavior (Spengler et al., 2015). It is therefore suspected that whether PA interventions induce “compensatory behaviors” in adolescents depends on the mode and duration of the intervention (Beck et al., 2022). Little is known about changes in PA in HIIT-influenced adolescents over 1-day or multiple-day intervals (Lee et al., 2021).
Given that adolescent health is directly tied to both increasing daily PA levels and improving adolescent physical fitness levels, this study had two objectives. The primary objective of this was to investigate the effects of two HIIT protocols (R-HIIT and B-HIIT) on adolescent physical fitness in order to establish an intervention framework that would more effectively promote healthy adolescent development. We hypothesize 1 that B-HIIT is better than R-HIIT in improving the physical fitness index of adolescents because it emphasizes using and controlling one’s weight, which can activate more muscles than R-HIIT and impose significant physiological stimulation on the body (Sumpena and Sidik, 2016; Ricci et al., 2020). The secondary objective is to evaluate the changes in daily PA level of adolescents during the intervention period. According to the ActivityStat hypothesis, we hypothesized 2 that the implementation of HIIT intervention in PE class will lead to a decrease in daily PA level of adolescents.
## 2.1 Study design and participants
This was a pre-and post-test study using a quasi-experimental design with the addition of a control group. The 12-week intervention was conducted from September to December 2018. Based on our school-based recruitment through consultations with nearby pediatricians, convenience sampling was used. The following factors guided our decision to choose three middle schools in Beijing, China, as intervention centers: The schools were all coeducational, had comparable teacher-student ratios, teacher qualification standards, and teacher experience levels, were all public schools that followed the same educational curriculum, and each student in each school had similar household income and parental education levels. In addition, all the schools had adequate space for both indoor and outdoor physical activity.
The inclusion criteria for subjects in this study were 1) age 11–13, as this age group was underrepresented in previous studies, 2) no regular physical activity other than physical education classes within 3 months before the experiment, and 3) no contraindication to any exercise. Criteria for exclusion were: 1) exercise contraindications or restrictions; 2) acute illness or sports injuries; 3) psychiatric or other psychological disorders, and 4) congenital or acquired heart disease. All subjects were informed of the risks and requirements of the training program, and we obtained signed informed consent from the students and their guardians before the start of the study. The trial was approved by the Ethics Committee of the Capital Institute of Sports (code: 201712001, approval date: $\frac{2017}{12}$/26) and registered in the Chinese Clinical Trial Registry (registration number: ChiCTR-IOR-17010435).
In this research, a simple randomization method was used and SPSS software was utilized to create random numbers using random sampling. In each of the three intervention centers, one natural class of seventh graders was chosen at random, and the three natural classes were then divided into three groups at random: R-HIIT group ($$n = 54$$), B-HIIT group ($$n = 55$$), and control group (Con, $$n = 57$$). To mitigate potential confounding effects between the intervention and control groups, randomization was not conducted at the individual level within the schools (Weston et al., 2021). The random assignment form was duplicated three times, with one copy each going to the statistician, the project manager, and the principal. After allocation, participants and data analysts were made blind. Participants in the intervention received specialized training for each of the exercise intervention groups. The independent researcher was in charge of statistical data analysis and made sure that he or she was not aware of how the interventions were distributed in particular.
## 2.2.1 Anthropometry
We measured the height and weight of each student before and after the intervention using a body composition analyzer (InBody J20, BIOSPACE, Seoul, Korea), which has been shown to have strong validity in Asian children (Lim et al., 2009) and has been used in large intervention trials with children (Kriemler et al., 2010). Body mass index (BMI) was calculated as weight in kilograms divided by the square of height in meters.
## 2.2.2 Primary outcome: VO2max
The multistage 20-m shuttle run test was used to estimate VO2max (Leger et al., 1988). Subjects were asked to run back and forth on a 20-m track with an audio signal. The test ended when the child failed to reach the finish line at the same time as the audio signal twice in a row or when the child stopped due to physical exhaustion. Finally, VO2max was then estimated using the following formula: VO2max=31.025+3.238×speed−3.248×age+0.1536×age×speed Where speed is that corresponding to that stage (speed = 8 + 0.5 stage no.), age is in years.
## 2.2.3 Secondary outcomes
To test for secondary outcomes, we used the standing long jump, vertical jump, handgrip strength test, 30-s sit-up test, and 50-m sprint. The standing long jump has been shown to be a valid assessment of explosive power (Ruiz et al., 2011). Participants were asked to swing their arms and jump forward onto the floor, where the distance was marked. Children performed three jumps, and the optimal length of the jump, measured in centimeters, was selected as the result.
The vertical jump is a valid measure of muscle strength and is positively correlated with bone structural strength and mineral content (Janz et al., 2021). Subjects were asked to start a downward motion and then jump as high as possible with both legs while touching a signpost with one hand. They performed one practice jump and two test jumps, 30 s apart.
To assess the health and function of adolescents’ musculoskeletal systems, handgrip strength is a trustworthy indicator (Ruiz et al., 2011; Janz et al., 2021). A Baseline pneumatic squeeze bulb dynamometer was used to gauge handgrip strength (Baseline, United States). Subjects were placed in a seated position with their elbows resting on a table. Two consecutive bilateral grip strength measurements were recorded, with the first measurement on the dominant hand. There was a 30-s rest period between each measurement to prevent fatigue. The best value of two trials was chosen for each hand.
The sit-up 30-s test assesses the muscular endurance of the trunk (Popovic et al., 2020). The subject lay on his/her back with knees bent, hands clasped behind the neck then rose to a sitting position, and finally returned to the starting position for one complete sit-up. The outcome was the number of sit-ups correctly completed within 30 s.
The 50-m sprint is used to assess the speed and anaerobic capacity (Fjortoft et al., 2011). On the command “Start”, the child standing behind the starting line had to run as fast as possible to the end of a 50-m track. The experimenter timed the tests with a stopwatch used in the track race (Casio, Japan), which was accurate to 0.01 s.
## 2.2.4 Evaluation measures
An extensive process evaluation was conducted to assess the feasibility and fidelity of the intervention. The intervention center faculty completed daily monitoring reports of the activities to ensure the quality of the interventions. Also, subjects were randomly selected from the intervention and control groups to wear an ActiGraph wGT3X-BT accelerometer (ActiGraph LLC, Pensacola, FL, United States) for 1 week during the intervention, for the purpose of assessing differences in physical activity levels, and patterns between groups.
The accelerometer was attached to a wrist strap and subjects were required to wear the accelerometer on the wrist of the non-dominant hand for 7 consecutive days, including 2 weekend days for at least 10 h a day except when swimming or bathing (Clevenger et al., 2019). The subjects had to wear the accelerometer before the test to ensure that it was being worn correctly. If the participant’s accelerometer collected valid data for more than 3 days (including 1 weekend day and 2 working days), it was included in the final data analysis. A total of 90 subjects, 30 in each group, wore accelerometers. The data for 20 subjects (4 in the B-HIIT group, 9 in the R-HIIT group, and 7 in the control group) were excluded due to incomplete accelerometer data records, resulting in valid data for 70 participants. ActiLife v6.13.3 software was used to initialize, download, and calculate physical activity data. Non-wear time, defined as ≥ 90 consecutive minutes of 0 counts, was removed by the software (Choi et al., 2011). Cumulative activity counts were categorized by intensity into sedentary time (<100 counts/minute), light physical activity (101–2,799 counts/minute), moderate physical activity (2,800–4,000 counts/minute), and vigorous physical activity (>4,000 counts/minute) based on the cutoff point set in ActLife.
## 2.3 Description of intervention
Throughout the 12-week study period, the intervention group participated in two 45-min regular PE classes per week, conducted by the school’s PE teacher and following the regular curriculum. The R-HIIT and B-HIIT groups received the intervention from two trained and qualified physical education teachers, and two topic experts oversaw the procedure to ensure that the intervention was carried out correctly. The intervention group replaced the warm-up exercises in the regular program with the proposed HIIT training session and then continued to complete the planned normal PE session.
HIIT sessions were performed during the first 10–15 min of each class, including a short warm-up, with a total of 8–10 interventions, using a 2:1 (1 min:30 s) work-to-rest ratio, with exercise intensity controlled at $75\%$–$85\%$ of maximum heart rate. A Mio Alpha heart rate monitor (Mio Alpha, United States), was used to monitor heart rate changes per second during exercise. We utilize the computer to track changes in the subjects’ heart rates as reported in real time by the heart rate monitor. When a subject’s heart rate falls short of the target heart rate, the research team members would remind them to intensify their exercise intensity in order to reach the desired heart rate. A work-rest ratio of 2:1 is optimal for male and female participants in HIIT (Laurent et al., 2014), and exercise intensity of $75\%$–$85\%$ HRmax has been shown to generate positive emotions and contribute to adherence to a physical activity regimen (Malik et al., 2019). We chose the heart rate monitor because we considered it a more objective measure of exercise intensity than those used in most previous studies, which used a work-rest ratio of 1:1 and SPRINT as the mode and chose either all-out training or a percentage of maximal aerobic speed as a measure of exercise intensity (*Paulino da* Silva Bento et al., 2021; Solera-Martinez et al., 2021; Cao et al., 2022b; Bossmann et al., 2022; Lubans et al., 2022).
The specific intervention for the R-HIIT group was a “real-world” training program based on a 20-m shuttle run test called “beep training” (Leger et al., 1988; Gripp et al., 2022). A flat surface was first chosen to allow the placement of two cones with a distance of 20 m between them. Each participant’s maximal aerobic speed was used to design sound cues, placing participants with the same maximal aerobic speed in groups so that they could listen to pre-recorded sounds to maintain the correct running speed. Each time they heard a “beep” sound in the music, they turned and completed a full cycle of running from the starting point to the turnaround point and back. After 4 weeks, the frequency of the musical rhythm will be accelerated according to the participants’ improved motor ability Figure 1.
**FIGURE 1:** *Summary of study design. Note: R-HIIT: Running-based high-intensity interval training; B-HIIT: Body weight-based high-intensity interval training.*
The PE teacher led the B-HIIT group in performing the full-body resistance exercises along with the rhythm of the music. Movement patterns included jumping jacks, jumping lunges, burpees, air squats, mountain climbers, and high knees. After 4 weeks of the experiment, as the subjects’ training heart rate improved relative to the target heart rate, movement frequency was increased according to the improvement in the students’ exercise ability Figure 1.
Throughout the 12-week intervention period, the control group participated in regular 45-min PE classes twice a week, following a normal curriculum with the PE teacher.
The present study had high fidelity. Monitoring reports and researcher observations of the intervention groups showed that, with very few exceptions, the participants adhered to the weekly unit plan and a daily instructional plan developed in the study and controlled the intensity of the intervention through heart rate monitoring. The total exercise time and HIIT intervention time remained consistent for each group. Attendance averaged over $90\%$ for each session.
## 2.4 Statistical analysis
We used other relevant studies to estimate the number of subjects needed for the experiment. Based on previous research (Yin et al., 2020), we set the test efficacy (1-β) to 0.80, the incidence of type I error α to 0.05, the correlation between pre-and post-intervention to 0.80, and the effect size to 2.68 (VO2max). The sample size required for each group was calculated to be 20. Considering a $20\%$ dropout rate, we established the minimum number of subjects in each group as 22.
This study adopts the principle of intention-to-treat analysis. The collected data were analyzed using IBM SPSS Statistics 26 software. All data were represented as mean ± SD, and the normal distribution test and variance homogeneity test were carried out by the Shapiro-Wilk test and Levene test, respectively. One-way analysis of variance (age, height, weight, BMI) and Chi-square test (gender) was used to test the differences in demographic variables. Sex, age, and BMI are controlled as covariates. The 3 (group: R-HIIT, B-HIIT, and control) × 2 (time: pre- and post-test) repeated measurement analysis of variance was used to test the difference before and after intervention for the experimental groups. When the interaction of group × time or the main effect of time and group was significant, a simple effect analysis was used to make a statistically significant pairwise comparison (Bonferroni). In addition, the effect size with statistical significance was calculated, which was expressed as ηp 2, in which a small effect was 0.20, a medium effect was 0.50, and a large effect was 0.80. The significance level was set to $p \leq 0.05.$
## 3.1 Study sample characteristics
As shown in Figure 2, 166 of the 197 recruited subjects, or $84.26\%$, were selected to join the study. The other 35 subjects were excluded because they did not sign the informed consent form ($$n = 21$$) or were unwilling to participate ($$n = 14$$). Of the test subjects, 71 were male and 92 were female. No injuries were reported during the 12-week intervention period, but one male and one female dropped out of the R-HIIT group and one male dropped out of the B-HIIT group, for a total of three dropouts ($1.81\%$ rate). No significant differences were found in the baseline demographic characteristics of the two groups, as shown in Table 1. The levels of the other socioeconomic variables in the sample are shown in the Annex.
**FIGURE 2:** *Flowchart of the participant selection process.* TABLE_PLACEHOLDER:TABLE 1
## 3.2 Intervention effects
Regarding the primary outcome, repeated measures variance results showed that there was a significant interaction for VO2max (F[2,163] = 42.91, $p \leq 0.001$, η p 2 = 0.35). A simple effects analysis showed significantly greater improvements in the mean performance of VO2max in both the B-HIIT and R-HIIT groups after intervention ($p \leq 0.01$), as compared to the control, with the highest scores in the B-HIIT group. Regarding the mean within-group changes, both the B-HIIT and R-HIIT significantly increased the mean VO2max from pre-to post-tests ($p \leq 0.01$), and the control group did not ($p \leq 0.05$) (Table 2).
**TABLE 2**
| Variables | Pretest | Post-test | The mean difference (95% CI) | ηp 2 |
| --- | --- | --- | --- | --- |
| VO2max (mL/kg/min) | VO2max (mL/kg/min) | VO2max (mL/kg/min) | VO2max (mL/kg/min) | VO2max (mL/kg/min) |
| B-HIIT | 38.62 ± 3.36 | 43.14 ± 3.80#& | 4.48 [3.89 to 5.07] ** | 0.69 |
| R-HIIT | 37.17 ± 3.43 | 41.32 ± 3.20# | 3.34 [2.74 to 3.94] ** | 0.54 |
| CON | 38.56 ± 2.78 | 41.90 ± 3.59 | 0.75 [−0.173 to 1.33] | 0.04 |
| Standing broad jump (cm) | Standing broad jump (cm) | Standing broad jump (cm) | Standing broad jump (cm) | Standing broad jump (cm) |
| B-HIIT | 168.60 ± 18.53 | 176.45 ± 15.55## | 7.97 [5.38 to 10.55] ** | 0.19 |
| R-HIIT | 167.45 ± 2 5.84 | 175.08 ± 22.47## | 7.86 [5.26 to 10.51] ** | 0.18 |
| CON | 167.08 ± 20.94 | 169.93 ± 24.48 | 2.25 [−0.27 to 4.77] | 0.02 |
| Vertical jump (cm) | Vertical jump (cm) | Vertical jump (cm) | Vertical jump (cm) | Vertical jump (cm) |
| B-HIIT | 239.34 ± 12.89 | 247.41 ± 11.34##&& | 8.10 [6.76 to 9.44] * | 0.58 |
| R-HIIT | 238.28 ± 13.87 | 240.97 ± 14.15 | 2.12 [0.76 to 3.48] * | 0.10 |
| CON | 242.52 ± 8.38 | 242.52 ± 11.13 | 0.05 [−1.32 to 1.33] | 0.00 |
| Handgrip strength (kg) | Handgrip strength (kg) | Handgrip strength (kg) | Handgrip strength (kg) | Handgrip strength (kg) |
| B-HIIT | 43.91 ± 11.57 | 46.76 ± 11.97 | 4.28 [1.76 to 2.80] ** | 0.16 |
| R-HIIT | 42.75 ± 9.10 | 45.75 ± 9.59 | 3.14 [1.88 to 4.39] * | 0.14 |
| CON | 40.14 ± 9.25 | 42.29 ± 10.85 | 2.30 [−1.76 to 4.23] | 0.10 |
| Sit-up (CPM) | Sit-up (CPM) | Sit-up (CPM) | Sit-up (CPM) | Sit-up (CPM) |
| B-HIIT | 40.53 ± 7.67 | 45.43 ± 8.46#& | 4.79 [3.35 to 6.39] * | 0.30 |
| R-HIIT | 42.58 ± 9.13 | 42.15 ± 1.44 | 0.13 [−1.34 to 1.60] | 0.00 |
| CON | 40.77 ± 9.62 | 41.54 ± 9.93 | 0.46 [−0.08 ± 1.73] | 0.00 |
| 50-m sprint (s) | 50-m sprint (s) | 50-m sprint (s) | 50-m sprint (s) | 50-m sprint (s) |
| B-HIIT | 9.12 ± 0.91 | 8.87 ± 0.70 | −0.26 [−0.39 to −0.13] * | 0.13 |
| R-HIIT | 9.30 ± 1.26 | 8.96 ± 0.75 | −0.13 [−0.27 to −0.003] * | 0.04 |
| CON | 9.15 ± 0.78 | 9.17 ± 1.15 | −0.02 [−0.11 to 0.17] | 0.00 |
Regarding secondary outcomes, repeated measures variance results showed that the participant’s standing broad jump (F[2,163] = 6.75, $$p \leq 0.01$$, η p 2 = 0.04), vertical jump (F[2,163] = 38.120, $p \leq 0.001$, η p 2 = 0.33), handgrip strength (F[2,163] = 11.51, $p \leq 0.001$, ηp 2 = 0.21), sit-up (F[2,163] = 16.81, $p \leq 0.001$, η p 2 = 0.18) and 50-m sprint (F[2,163]) = 11.063, $p \leq 0.001$, η p 2 = 0.11) had significant interaction effects. The simple effects analysis showed significantly greater improvements after intervention in the mean performance of standing broad jump ($p \leq 0.01$) in both the B-HIIT and R-HIIT groups, and vertical jump ($p \leq 0.01$) and sit-up ($p \leq 0.05$) performance in the B-HIIT, as compared to the controls. In addition, the mean scores for vertical jump ($p \leq 0.01$) and sit-up ($p \leq 0.05$) were significantly greater in the B-HIIT group than in the R-HIIT after the intervention. Regarding the mean within-group changes from pre-to post-tests, with the exception of sit-ups which had significant improvement ($p \leq 0.01$) only in the B-HIIT group, both the B-HIIT and R-HIIT significantly increased the mean scores for standing broad jump ($p \leq 0.01$), vertical jump ($p \leq 0.05$), handgrip strength ($p \leq 0.05$), and 50-m sprint ($p \leq 0.05$), but not the control group ($p \leq 0.05$) (Table 2).
Measurement of weekly physical activity levels during the intervention showed that adolescents participating in the B-HIIT and R-HIIT groups had significantly lower levels of sedentary behavior and low activity (Figures 3A, B) on school days compared to the control group ($p \leq 0.05$). As shown in (Figures 3C, D), the intervention groups also had significantly longer periods of moderate and vigorous or vigorous physical activity levels than the control group ($p \leq 0.05$). There was no significant difference in physical activity levels between the three groups at the weekend ($p \leq 0.05$).
**FIGURE 3:** *Mean weekly physical activity levels during the intervention. Note: R-HIIT: Running-based high-intensity interval training; B-HIIT: Body weight-based high-intensity interval training; CON: Control group.*
## 4 Discussion
In this study, both HIIT protocols based on the school environment were successful in improving indicators of adolescent cardiorespiratory fitness and muscular strength, and the intervention group engaged in more moderate to vigorous exercise than the control group did throughout the intervention period. However, it was also discovered that the B-HIIT group, which relies on body weight alone, is more efficient than the R-HIIT group, which relies on running, at enhancing CRF, explosive leg strength (vertical leap), and abdominal muscular endurance (Sit-up). Hypothesis 1 of our study has been confirmed, but hypothesis 2 of our study is not true.
Research has shown the ability of R-HIIT in the school setting to improve aerobic endurance (Bauer et al., 2022; Harnish et al., 2022). Our study found that B-HIIT also improved cardiorespiratory fitness in adolescents, which is consistent with other studies (*Paulino da* Silva Bento et al., 2021; Bossmann et al., 2022). Cardiac adaptation through HIIT training can explain why resistance exercises based on body weight can lead to similar aerobic fitness as running-based programs. As long as the intensity is high enough, the type of training can be adjusted according to the sports environment of the school. However, there are contradictory findings. Engel et al. [ 2019] did not find that classroom-based B-HIIT sessions improved variables associated with aerobic endurance. This may be because the intervention time (6.0 ± 1.5 min) and duration (4 weeks) were short and therefore did not elicit sufficient stimulation to improve CRF. Braaksma et al. [ 2018] suggested a physical activity intervention for a minimum duration of 6 weeks to improve CRF in children aged 6–12.
Previous studies of children and adolescents in and out of school have found that when R-HIIT is used, the improvement in aerobic endurance is similar to or slightly lower than that of MICT, depending on the participants’ initial health level and the duration and design of the intervention (Yin et al., 2020; Cao et al., 2022a; Bauer et al., 2022). However, the present study further found that B-HIIT improved CRF better than R-HIIT. The difference value reached 2.32 mL/kg/min. Previous adult data show that each unit reduction in CRF increases the risk of all-cause mortality by 1.73 times, the risk of CVD by 2.27 times, and the risk of cancer by 2.07 times (Imboden et al., 2018). Surveys in adolescent populations have found that CRF is strongly associated with adolescent overweight/obesity, and cognitive control, and is predictive of future CVD health (Perez-Bey et al., 2019; Wisnieski et al., 2019; Shigeta et al., 2021; Gonzalez-Galvez et al., 2022). Therefore, the results of this study have encouraging clinical implications.
Various peripheral and central physiological factors could explain the elevated VO2max after HIIT intervention (Bossmann et al., 2022). In this study, the possible reason why B-HIIT is superior to R-HIIT is that B-HIIT requires the recruitment of more fast muscle fibers as well as the mobilization of all types of muscle during exercise, thus leading to more blood lactate and consequently to an increase in VO2max (Binzen et al., 2001). Some researchers hypothesize that resistance training can indirectly improve endurance by increasing neuromuscular adaptations, improving factors such as running economy and specific muscle output power (Paavolainen et al., 1999).
Traditional HIIT reflects time efficiency in improving several health markers and aerobic/anaerobic performance but ignores the impact on muscle health, a critical factor in the present and future health of children and adolescents (Janz et al., 2021). Costigan et al. found that running-based HIIT had a small overall effect on muscle health in young adults (Costigan et al., 2015). Requirements for a healthy musculoskeletal system include absolute muscle strength, endurance (fatigue tolerance), and the ability to produce power quickly or explosive muscular power (Plowman, 2014). In this study, R-HIIT was able to improve muscle strength in standing, broad jump, vertical jump, and handgrip strength to some extent, but not nearly as much as B-HIIT. Only B-HIIT improved core muscle endurance, as indicated by the sit-up results. This is probably because exercises like mountain climbers in the B-HIIT program improve endurance in the core lumbar and abdominal muscles. Related studies propose that the improvement of core muscle endurance can stabilize the lumbar spine and reduce the risk of lower back pain (Janz et al., 2021).
One study concluded that the increase in muscle strength after interval training could be explained by increased activity of anaerobic glycolytic enzymes in the skeletal muscle (Glaister, 2005). Most B-HIIT exercises are plyometric exercises, which are more effective than running in developing muscle mass and increasing anaerobic glycolytic enzyme activity in skeletal muscle. This, in turn, significantly improves muscle strength and explosive power (Costigan et al., 2015; Usman and Shenoy, 2019), thus explaining why B-HIIT is more effective than R-HIIT. Other research found that increased lower body strength and VO2max facilitate increased stride frequency and length during running (Benedetti et al., 2018). These results may explain improved speed scores in activities such as the 50-m sprint, and they further support the use of resistance training B-HIIT for improved health and endurance. One study found that even a short 4-week B-HIIT intervention was effective in improving health-related indicators in children and adolescents (Engel et al., 2019).
Some studies have suggested that physical activity levels during one part of the day may lead to a compensatory decrease in physical activity levels during another part of the day, a hypothesis known as the “activityStat hypothesis” (Gomersall et al., 2013; Beck et al., 2022). Data from the accelerometers in our study found encouraging results to the contrary: both HIIT groups had significantly higher levels of moderate to vigorous or vigorous physical activity than the control group during the intervention period. Popowczak et al. [ 2022] also found that adolescents were more active on days when they performed HIIT than on other days. Ricci et al. [ 2020] concluded that while maximal heart rate and significant lactate accumulation were induced during HIIT, the affective value and enjoyment of exercise remained positive for most participants. Because enjoyment of exercise is a strong predictor of habitual exercise in children, this suggests that embedding HIIT into school PE classes could also help increase daily activity in children and adolescents.
Our study highlights the potential for embedding HIIT into school PE classes to improve the health of adolescents, and evidence supports the use of B-HIIT as more effective than R-HIIT for improving physical performance. From a practical perspective, B-HIIT training does not require any barbells, equipment, or machines compared to traditional apparatus-based strength training, and all exercises are performed with the children’s own body weight and at fairly high speeds and repetitions. The physical improvement after the intervention represents a good cost-benefit ratio. Furthermore, because B-HIIT has minimal space requirements, it could also be useful for students who are isolated at home due to the pandemic, with online coaching provided by teachers.
## 4.1 Study strengths and limitations
One major strength of the study was that the interventions were implemented in a middle school to compare the effectiveness of two HIIT protocols. The results can serve as a reference for selecting effective HIIT programs, including such factors as load time and intensity, to improve health outcomes. Another strength is that it can be scaled for translation because the HIIT program can be integrated into middle-school PE classes, and the efficiency of HIIT programs used in schools has been observed in improve the students’ physical fitness. Lastly, the present study objectively measured the internal load of HIIT program by monitoring HR.
Our study had some limitations. First, because it was implemented in school PE classes, it was not possible to completely randomize each student. However, as in previous studies in a school setting (Bossmann et al., 2022), minor differences in performance ability between the three groups are expected. Second, the two HIIT protocols were not evaluated in this study for affective (i.e., sensory state) and enjoyment responses in adolescents, and these variables should be evaluated in future studies, as enjoyment is a relevant motivating factor for habitual activity and exercise adherence throughout the lifespan (Ricci et al., 2020). Moreover, the relatively small number of study subjects and the fact that some subjects refused to join the study at the beginning may have led to selection bias in the study results. Finally, the subjects in this study were normal adolescents, and differences in the effects of the two interventions on the body composition of obese adolescents are not yet clear. Future school-based studies with larger sample sizes of students with different characteristics are needed to verify the effects of different HIIT interventions.
## 5 Conclusion
This study compared the efficiency of two HIIT training programs to improve the CRF of adolescents in middle school PE classes. The effect on CRF of the B-HIIT program, which consisted of resistance-based exercises using the participant’s own body weight, was significantly greater than that of the running (R-HIIT) program. This study suggests that just 10 min of HIIT training twice a week in the limited setting of school PE classes can effectively improve the CRF and muscle strength of adolescents.
## Data availability statement
The original contributions presented in the study are included in the article/Supplementary Materials, further inquiries can be directed to the corresponding author.
## Ethics statement
The studies involving human participants were reviewed and approved by the Ethics Committee of the Capital Institute of Sports and registered in the Chinese Clinical Trial Registry (registration number: ChiCTR-IOR-17010435). Written informed consent to participate in this study was provided by the participants’ legal guardian/next of kin.
## Author contributions
ZZ, ZL, XH, and YL conceived and designed this research. YL conducted the experiments and collected the data. ZL and YL analyzed the data. ZL, YL, XH, and ZZ drafted the manuscript. All authors read and approved the final version of the manuscript.
## Conflict of interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher’s note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## References
1. Aburto N. J., Fulton J. E., Safdie M., Duque T., Bonvecchio A., Rivera J. A.. **Effect of a school-based intervention on physical activity: Cluster-randomized trial**. *Med. Sci. Sports Exerc.* (2011) **43** 1898-1906. DOI: 10.1249/MSS.0b013e318217ebec
2. Bauer N., Sperlich B., Holmberg H. C., Engel F. A.. **Effects of high-intensity interval training in school on the physical performance and health of children and adolescents: A systematic review with meta-analysis**. *Sports Med. Open* (2022) **8** 50. DOI: 10.1186/s40798-022-00437-8
3. Beck F., Engel F. A., Reimers A. K.. **Compensation or displacement of physical activity in children and adolescents: A systematic review of empirical studies**. *Child. (Basel)* (2022) **9** 351. DOI: 10.3390/children9030351
4. Beets M. W., Okely A., Weaver R. G., Webster C., Lubans D., Brusseau T.. **The theory of expanded, extended, and enhanced opportunities for youth physical activity promotion**. *Int. J. Behav. Nutr. Phys. Activity* (2016) **13** 120. DOI: 10.1186/s12966-016-0442-2
5. Behm D. G., Faigenbaum A. D., Falk B., Klentrou P.. **Canadian society for exercise Physiology position paper: Resistance training in children and adolescents**. *Appl. Physiology Nutr. Metabolism* (2008) **33** 547-561. DOI: 10.1139/H08-020
6. Benedetti M. G., Furlini G., Zati A., Letizia Mauro G.. **The effectiveness of physical exercise on bone density in osteoporotic patients**. *Biomed. Res. Int.* (2018) **2018** 4840531. DOI: 10.1155/2018/4840531
7. Biddle S. J. H., Batterham A. M.. **High-intensity interval exercise training for public health: A big HIT or shall we HIT it on the head?**. *Int. J. Behav. Nutr. Phys. Activity* (2015) **12** 95. DOI: 10.1186/s12966-015-0254-9
8. Binzen C. A., Swan P. D., Manore M. M.. **Postexercise oxygen consumption and substrate use after resistance exercise in women**. *Med. Sci. Sports Exerc.* (2001) **33** 932-938. DOI: 10.1097/00005768-200106000-00012
9. Bond B., Hind S., Williams C. A., Barker A. R.. **The acute effect of exercise intensity on vascular function in adolescents**. *Med. Sci. Sports Exerc* (2015) **47** 2628-2635. DOI: 10.1249/MSS.0000000000000715
10. Bond B., Weston K. L., Williams C. A., Barker A. R.. **Perspectives on high-intensity interval exercise for health promotion in children and adolescents**. *Open Access J. Sports Med.* (2017) **8** 243-265. DOI: 10.2147/OAJSM.S127395
11. Bossmann T., Woll A., Wagner I.. **Effects of different types of high-intensity interval training (HIIT) on endurance and strength parameters in children and adolescents**. *Int. J. Environ. Res. Public Health* (2022) **19** 6855. DOI: 10.3390/ijerph19116855
12. Braaksma P., Stuive I., Garst R. M. E., Wesselink C. F., van der Sluis C. K., Dekker R.. **Characteristics of physical activity interventions and effects on cardiorespiratory fitness in children aged 6-12 years-A systematic review**. *J. Sci. Med. Sport* (2018) **21** 296-306. DOI: 10.1016/j.jsams.2017.07.015
13. Breithecker J., Jaitner D., Lohmann J., Linke D., Siegle M., Lames M.. **Bewegungszeit, Belastungsintensität und Bewegungsumfang im Sportunterricht**. *Ger. J. Exerc. Sport Res.* (2021) **51** 290-301. DOI: 10.1007/s12662-021-00733-7
14. Cao M., Li S., Tang Y., Zou Y.. **A meta-analysis of high-intensity interval training on glycolipid metabolism in children with metabolic disorders**. *Front. Pediatr.* (2022a) **10** 887852. DOI: 10.3389/fped.2022.887852
15. Cao M., Tang Y., Li S., Zou Y.. **Effects of school-based high-intensity interval training on body composition, cardiorespiratory fitness and cardiometabolic markers in adolescent boys with obesity: A randomized controlled trial**. *Bmc Pediatr.* (2022b) **22** 112. DOI: 10.1186/s12887-021-03079-z
16. Chaput J. P., Willumsen J., Bull F., Chou R., Ekelund U., Firth J.. **2020 WHO guidelines on physical activity and sedentary behaviour for children and adolescents aged 5-17 years: Summary of the evidence**. *Int. J. Behav. Nutr. Phys. Act.* (2020) **17** 141. DOI: 10.1186/s12966-020-01037-z
17. Choi L., Liu Z., Matthews C. E., Buchowski M. S.. **Validation of accelerometer wear and nonwear time classification algorithm**. *Med. Sci. Sports Exerc.* (2011) **43** 357-364. DOI: 10.1249/MSS.0b013e3181ed61a3
18. Clevenger K. A., Pfeiffer K. A., Mackintosh K. A., Mcnarry M. A., Brond J., Arvidsson D.. **Effect of sampling rate on acceleration and counts of hip- and wrist-worn ActiGraph accelerometers in children**. *Physiol. Meas.* (2019) **40** 095008. DOI: 10.1088/1361-6579/ab444b
19. Costigan S. A., Eather N., Plotnikoff R. C., Taaffe D. R., Pollock E., Kennedy S. G.. **Preliminary efficacy and feasibility of embedding high intensity interval training into the school day: A pilot randomized controlled trial**. *Prev. Med. Rep.* (2015) **2** 973-979. DOI: 10.1016/j.pmedr.2015.11.001
20. Dobbins M., Husson H., Decorby K., Larocca R. L.. **School-based physical activity programs for promoting physical activity and fitness in children and adolescents aged 6 to 18**. *Cochrane Database Syst. Rev.* (2013) **9** CD007651. DOI: 10.1002/14651858.cd007651.pub2
21. Donnelly J. E., Hillman C. H., Castelli D., Etnier J. L., Lee S., Tomporowski P.. **Physical activity, fitness, cognitive function, and academic achievement in children: A systematic review**. *Med. Sci. Sports Exerc.* (2016) **48** 1223-1224. DOI: 10.1249/MSS.0000000000000966
22. Eddolls W. T. B., Mcnarry M. A., Stratton G., Winn C. O. N., Mackintosh K. A.. **High-intensity interval training interventions in children and adolescents: A systematic review**. *Sports Med.* (2017) **47** 2363-2374. DOI: 10.1007/s40279-017-0753-8
23. Ekkekakis P., Parfitt G., Petruzzello S. J.. **The pleasure and displeasure people feel when they exercise at different intensities decennial update and progress towards a tripartite rationale for exercise intensity prescription**. *Sports Med.* (2011) **41** 641-671. DOI: 10.2165/11590680-000000000-00000
24. Engel F. A., Wagner M. O., Schelhorn F., Deubert F., Leutzsch S., Stolz A.. **Classroom-based micro-sessions of functional high-intensity circuit training enhances functional strength but not cardiorespiratory fitness in school children-A feasibility study**. *Front. Public Health* (2019) **7** 291. DOI: 10.3389/fpubh.2019.00291
25. Faigenbaum A. D., Kraemer W. J., Blimkie C. J. R., Jeffreys I., Micheli L. J., Nitka M.. **Youth resistance training: Updated position statement paper from the national strength and conditioning association**. *J. Strength Cond. Res.* (2009) **23** S60-S79. DOI: 10.1519/JSC.0b013e31819df407
26. Farooq M. A., Parkinson K. N., Adamson A. J., Pearce M. S., Reilly J. K., Hughes A. R.. **Timing of the decline in physical activity in childhood and adolescence: Gateshead Millennium Cohort Study**. *Br. J. Sports Med.* (2018) **52** 1002-1006. DOI: 10.1136/bjsports-2016-096933
27. Fjortoft I., Pedersen A. V., Sigmundsson H., Vereijken B.. **Measuring physical fitness in children who are 5 to 12 years old with a test battery that is functional and easy to administer**. *Phys. Ther.* (2011) **91** 1087-1095. DOI: 10.2522/ptj.20090350
28. Gehris J., Myers E., Whitaker R.. **Physical activity levels during adventure-physical education lessons**. *Eur. Phys. Educ. Rev.* (2012) **18** 245-257. DOI: 10.1177/1356336x12440365
29. Glaister M.. **Multiple sprint work - physiological responses, mechanisms of fatigue and the influence of aerobic fitness**. *Sports Med.* (2005) **35** 757-777. DOI: 10.2165/00007256-200535090-00003
30. Gomersall S. R., Rowlands A. V., English C., Maher C., Olds T. S.. **The ActivityStat hypothesis: The concept, the evidence and the methodologies**. *Sports Med.* (2013) **43** 135-149. DOI: 10.1007/s40279-012-0008-7
31. Gonzalez-Galvez N., Ribeiro J. C., Mota J.. **Cardiorespiratory fitness, obesity and physical activity in schoolchildren: The effect of mediation**. *Int. J. Environ. Res. Public Health* (2022) **19** 16262. DOI: 10.3390/ijerph192316262
32. Gorga E., Regazzoni V., Bansilal S., Carubelli V., Trichaki E., Gavazzoni M.. **School and family-based interventions for promoting a healthy lifestyle among children and adolescents in Italy: A systematic review**. *J. Cardiovasc. Med.* (2016) **17** 547-555. DOI: 10.2459/JCM.0000000000000404
33. Gripp F., De Jesus Gomes G., De Sousa R. A. L., Alves De Andrade J., Pinheiro Queiroz I., Diniz Magalhaes C. O.. **A real-world high-intensity interval training protocol for cardiorespiratory fitness improvement**. *J. Vis. Exp.* (2022) **180** e63708. DOI: 10.3791/63708
34. Guthold R., Stevens G. A., Riley L. M., Bull F. C.. **Global trends in insufficient physical activity among adolescents: A pooled analysis of 298 population-based surveys with 1·6 million participants**. *Lancet Child Adolesc. Health* (2020) **4** 23-35. DOI: 10.1016/S2352-4642(19)30323-2
35. Haapala H. L., Hirvensalo M. H., Kulmala J., Hakonen H., Kankaanpaa A., Laine K.. **Changes in physical activity and sedentary time in the Finnish schools on the move program: A quasi-experimental study**. *Scand. J. Med. Sci. Sports* (2017) **27** 1442-1453. DOI: 10.1111/sms.12790
36. Harnish C., Duncombe S. L., Barker A. R., Bond B., Earle R., Varley-Campbell J.. **School-based high-intensity interval training programs in children and adolescents: A systematic review and meta-analysis**. *Plos One* (2022) **17** e0266427. DOI: 10.1371/journal.pone.0266427
37. Hollis J. L., Sutherland R., Williams A. J., Campbell E., Nathan N., Wolfenden L.. **A systematic review and meta-analysis of moderate-to-vigorous physical activity levels in secondary school physical education lessons**. *Int. J. Behav. Nutr. Phys. Activity* (2017) **14** 52. DOI: 10.1186/s12966-017-0504-0
38. Hulteen R. M., Morgan P. J., Barnett L. M., Stodden D. F., Lubans D. R.. **Development of foundational movement skills: A conceptual model for physical activity across the lifespan**. *Sports Med.* (2018) **48** 1533-1540. DOI: 10.1007/s40279-018-0892-6
39. Imboden M. T., Harber M. P., Whaley M. H., Finch W. H., Bishop D. L., Kaminsky L. A.. **Cardiorespiratory fitness and mortality in healthy men and women**. *J. Am. Coll. Cardiol.* (2018) **72** 2283-2292. DOI: 10.1016/j.jacc.2018.08.2166
40. Janz K. F., Baptista F., Ren S., Zhu W., Laurson K. R., Mahar M. T.. **Associations among musculoskeletal fitness assessments and health outcomes: The lisbon study for the development and evaluation of musculoskeletal fitness standards in youth**. *Meas. Phys. Educ. Exerc. Sci.* (2021) **26** 297-305. DOI: 10.1080/1091367X.2021.2000414
41. Kennedy S. G., Leahy A. A., Smith J. J., Eather N., Hillman C. H., Morgan P. J.. **Process evaluation of a school-based high-intensity interval training program for older adolescents: The burn 2 learn cluster randomised controlled trial**. *Child. (Basel)* (2020) **7** 299. DOI: 10.3390/children7120299
42. Kriemler S., Zahner L., Schindler C., Meyer U., Hartmann T., Hebestreit H.. **Effect of school based physical activity programme (KISS) on fitness and adiposity in primary schoolchildren: Cluster randomised controlled trial**. *Bmj-British Med. J.* (2010) **340** c785. DOI: 10.1136/bmj.c785
43. Laurent C. M., Vervaecke L. S., Kutz M. R., Green J. M.. **Sex-specific responses to self-paced, high-intensity interval training with variable recovery periods**. *J. Strength Cond. Res.* (2014) **28** 920-927. DOI: 10.1519/JSC.0b013e3182a1f574
44. Lee K. J., Noh B., An K. O.. **Impact of synchronous online physical education classes using tabata training on adolescents during COVID-19: A randomized controlled study**. *Int. J. Environ. Res. Public Health* (2021) **18** 10305. DOI: 10.3390/ijerph181910305
45. Leger L. A., Mercier D., Gadoury C., Lambert J.. **The multistage 20 metre shuttle run test for aerobic fitness**. *J. Sports Sci.* (1988) **6** 93-101. DOI: 10.1080/02640418808729800
46. Lim J. S., Hwang J. S., Lee J. A., Kim D. H., Park K. D., Jeong J. S.. **Cross-calibration of multi-frequency bioelectrical impedance analysis with eight-point tactile electrodes and dual-energy X-ray absorptiometry for assessment of body composition in healthy children aged 6-18 years**. *Pediatr. Int.* (2009) **51** 263-268. DOI: 10.1111/j.1442-200X.2008.02698.x
47. Logan G. R., Harris N., Duncan S., Schofield G.. **A review of adolescent high-intensity interval training**. *Sports Med.* (2014) **44** 1071-1085. DOI: 10.1007/s40279-014-0187-5
48. Loo B. K. G., Okely A. D., Pulungan A., Jalaludin M. Y.. **Asia-Pacific Consensus Statement on integrated 24-hour activity guidelines for children and adolescents**. *Br. J. Sports Med.* (2022) **56** 539-545. DOI: 10.1136/bjsports-2021-104527
49. Lubans D. R., Eather N., Smith J. J., Beets M. W., Harris N. K.. **Scaling-up adolescent high-intensity interval training programs for population health**. *Exerc Sport Sci. Rev.* (2022) **50** 128-136. DOI: 10.1249/JES.0000000000000287
50. Lubans D., Richards J., Hillman C., Faulkner G., Beauchamp M., Nilsson M.. **Physical activity for cognitive and mental health in youth: A systematic review of mechanisms**. *Pediatrics* (2016) **138** e20161642. DOI: 10.1542/peds.2016-1642
51. Malik A. A., Williams C. A., Bond B., Weston K. L., Barker A. R.. **Acute cardiorespiratory, perceptual and enjoyment responses to high-intensity interval exercise in adolescents**. *Eur. J. Sport Sci.* (2017) **17** 1335-1342. DOI: 10.1080/17461391.2017.1364300
52. Malik A. A., Williams C. A., Weston K. L., Barker A. R.. **Perceptual and cardiorespiratory responses to high-intensity interval exercise in adolescents: Does work intensity matter?**. *J. Sports Sci. Med.* (2019) **18** 1-12. PMID: 30787646
53. Meng C., Yucheng T., Shu L., Yu Z.. **Effects of school-based high-intensity interval training on body composition, cardiorespiratory fitness and cardiometabolic markers in adolescent boys with obesity: A randomized controlled trial**. *BMC Pediatr.* (2022) **22** 112. DOI: 10.1186/s12887-021-03079-z
54. Messing S., Rutten A., Abu-Omar K., Ungerer-Rohrich U., Goodwin L., Burlacu I.. **How can physical activity Be promoted among children and adolescents? A systematic review of reviews across settings**. *Front. Public Health* (2019) **7** 55. DOI: 10.3389/fpubh.2019.00055
55. Owen M. B., Curry W. B., Kerner C., Newson L., Fairclough S. J.. **The effectiveness of school-based physical activity interventions for adolescent girls: A systematic review and meta-analysis**. *Prev. Med.* (2017) **105** 237-249. DOI: 10.1016/j.ypmed.2017.09.018
56. Paavolainen L., HäKKINEN K., HäMäLäINEN I., Nummela A., Rusko H.. **Explosive-strength training improves 5-km running time by improving running economy and muscle power**. *J. Appl. Physiol.* (1999) **86** 1527-1533. DOI: 10.1152/jappl.1999.86.5.1527
57. Paulino da Silva Bento A. F., Paez L. C., Manuel De Mendonca Raimundo A.. **High-intensity interval training in high-school physical education classes: Study protocol for a randomized controlled trial**. *Contemp. Clin. Trials Commun.* (2021) **24** 100867. DOI: 10.1016/j.conctc.2021.100867
58. Perez-Bey A., Segura-Jimenez V., Fernandez-Santos J. D., Esteban-Cornejo I., Gomez-Martinez S., Veiga O. L.. **The influence of cardiorespiratory fitness on clustered cardiovascular disease risk factors and the mediator role of body mass index in youth: The UP&DOWN Study**. *Pediatr. Diabetes* (2019) **20** 32-40. DOI: 10.1111/pedi.12800
59. Plowman S. A.. **Top 10 research questions related to musculoskeletal physical fitness testing in children and adolescents**. *Res. Q. Exerc Sport* (2014) **85** 174-187. DOI: 10.1080/02701367.2014.899857
60. Popovic B., Cvetkovic M., Macak D., Scepanovic T., Cokorilo N., Belic A.. **Nine months of a structured multisport program improve physical fitness in preschool children: A quasi-experimental study**. *Int. J. Environ. Res. Public Health* (2020) **17** 4935. DOI: 10.3390/ijerph17144935
61. Popowczak M., Rokita A., Kozlenia D., Domaradzki J.. **Publisher Correction: The high-intensity interval training introduced in physical education lessons decrease systole in high blood pressure adolescents**. *Sci. Rep.* (2022) **12** 3397. DOI: 10.1038/s41598-022-07364-4
62. Ricci J. M., Astorino T. A., Currie K. D., Pfeiffer K. A.. **Cardiometabolic and perceptual responses to body-weight resistance high-intensity interval exercise in boys**. *Pediatr. Exerc Sci.* (2020) **33** 8-15. DOI: 10.1123/pes.2020-0040
63. Ruiz J. R., Castro-Pinero J., Espana-Romero V., Artero E. G., Ortega F. B., Cuenca M. M.. **Field-based fitness assessment in young people: The ALPHA health-related fitness test battery for children and adolescents**. *Br. J. Sports Med.* (2011) **45** 518-524. DOI: 10.1136/bjsm.2010.075341
64. Sa A. S., Miranda T. G., De Paula C. C., Barsanulfo S. R., Teixeira D., Monteiro D.. **COVID-19 and quarantine: Expanding understanding of how to stay physically active at home**. *Front. Psychol.* (2020) **11** 566032. DOI: 10.3389/fpsyg.2020.566032
65. Shigeta T. T., Leahy A. A., Smith J. J., Eather N., Lubans D. R., Hillman C. H.. **Cardiorespiratory and muscular fitness associations with older adolescent cognitive control**. *J. Sport Health Sci.* (2021) **10** 82-90. DOI: 10.1016/j.jshs.2020.05.004
66. Solera-Martinez M., Herraiz-Adillo A., Manzanares-Dominguez I., De la Cruz L. L., Martinez-Vizcaino V., Pozuelo-Carrascosa D. P.. **High-intensity interval training and cardiometabolic risk factors in children: A meta-analysis**. *Pediatrics* (2021) **148** e2021050810. DOI: 10.1542/peds.2021-050810
67. Spengler S., Mess F., Woll A.. **Do media use and physical activity compete in adolescents? Results of the MoMo study**. *Plos One* (2015) **10** e0142544. DOI: 10.1371/journal.pone.0142544
68. Sumpena A., Sidik D. Z.. **The impact of tabata protocol to increase the anaerobic and aerobic capacity**. (2017)
69. Thompson W. R.. **Worldwide survey of fitness trends for 2022**. *Acsms Health and Fit. J.* (2022) **26** 11-20. DOI: 10.1249/fit.0000000000000732
70. Usman T., Shenoy K. B.. **Effects of plyometrics and plyometrics combined with dynamic stretching on vertical jump in male collegiate volleyball players**. *Int. J. Appl. Exerc. Physiology* (2019) **8** 66-73
71. Weston K. L., Innerd A., Azevedo L. B., Bock S., Batterham A. M.. **Process evaluation of project ffab (fun fast activity blasts): A multi-activity school-based high-intensity interval training intervention**. *Front. Sports Act. Living* (2021) **3** 737900. DOI: 10.3389/fspor.2021.737900
72. Wisnieski L., Dalimonte-Merckling D., Robbins L. B.. **Cardiorespiratory fitness as a mediator of the association between physical activity and overweight and obesity in adolescent girls**. *Child. Obes.* (2019) **15** 338-345. DOI: 10.1089/chi.2018.0360
73. Yin J., Zhou Z. X., Lan T. W.. **High-intensity interval training versus moderate-intensity continuous training on health outcomes for children and adolescents: A meta-analysis of randomized controlled trials**. *Biomed Res. Int.* (2020) **2020** 9797439. DOI: 10.1155/2020/9797439
|
---
title: 'Nlrp12 deficiency alters gut microbiota and ameliorates Faslpr
-mediated systemic autoimmunity in male mice'
authors:
- Leila Abdelhamid
- Jiangdi Mao
- Xavier Cabana-Puig
- Jing Zhu
- Brianna K. Swartwout
- Michael R. Edwards
- James C. Testerman
- Jacquelyn S. Michaelis
- Irving Coy Allen
- S. Ansar Ahmed
- Xin M. Luo
journal: Frontiers in Immunology
year: 2023
pmcid: PMC10036793
doi: 10.3389/fimmu.2023.1120958
license: CC BY 4.0
---
# Nlrp12 deficiency alters gut microbiota and ameliorates Faslpr
-mediated systemic autoimmunity in male mice
## Abstract
NLRP12 has dual roles in shaping inflammation. We hypothesized that NLRP12 would modulate myeloid cells and T cell function to control systemic autoimmunity. Contrary to our hypothesis, the deficiency of Nlrp12 in autoimmune-prone B6.Faslpr/lpr mice ameliorated autoimmunity in males but not females. Nlrp12 deficiency dampened B cell terminal differentiation, germinal center reaction, and survival of autoreactive B cells leading to decreased production of autoantibodies and reduced renal deposition of IgG and complement C3. In parallel, Nlrp12 deficiency reduced the expansion of potentially pathogenic T cells, including double-negative T cells and T follicular helper cells. Furthermore, reduced pro-inflammatory innate immunity was observed, where the gene deletion decreased in-vivo expansion of splenic macrophages and mitigated ex-vivo responses of bone marrow-derived macrophages and dendritic cells to LPS stimulation. Interestingly, Nlrp12 deficiency altered the diversity and composition of fecal microbiota in both male and female B6/lpr mice. Notably, however, Nlrp12 deficiency significantly modulated small intestinal microbiota only in male mice, suggesting that the sex differences in disease phenotype might be gut microbiota-dependent. Together, these results suggest a potential pathogenic role of NLRP12 in promoting systemic autoimmunity in males. Future studies will investigate sex-based mechanisms through which NLRP12 differentially modulates autoimmune outcomes.
## Introduction
Mice carrying the Faslpr mutation are models of autoimmune lymphoproliferative syndrome (ALPS) and systemic lupus erythematosus (SLE) [1]. ALPS is a chronic autoimmune disorder characterized by nonmalignant adenopathy and splenomegaly [2], whereas SLE is an autoimmune disease with multisystem involvement [3, 4]. Even though the precise etiology for these autoimmune conditions is still unclear, defective apoptosis and expansion of unusual populations of adaptive immune cells (such as double-negative T cells) leading to aberrant lymphoid hyperplasia contribute to the development of autoimmunity in both ALPS [1, 5, 6] and SLE [7, 8]. ALPS is primarily a disorder of T cell dysregulation (9–12). SLE, on the other hand, involves a complex interplay between disrupted innate immune functions (13–19) and adaptive immune cell abnormalities (20–26) that contributes to the perturbation of tolerance and development of immunopathogenesis [27]. Studies in recent years suggest that microbiota could also modulate autoimmunity and alter disease management outcomes [28, 29]. While the role of gut microbial dysbiosis in ALPS remains unknown, dysregulated gut microbiota is a feature of SLE pathogenesis that is known to interact with both innate [30] and adaptive [31] immune responses. We and others have previously unraveled the dynamic changes of gut microbiota in murine lupus and human SLE (32–36) and delineated the influence of gut microbiota modulation on lupus outcomes in different experimental settings [37, 38].
NACHT, LRR and PYD domains-containing protein 12 (NLRP12) is a cytoplasmic innate sensor that plays dual roles in regulating inflammation [39]. It is a checkpoint inhibitor controlling inflammation but could also form inflammasome in a context-dependent fashion [39]. While the conditions that trigger its regulatory functions are still to be elucidated, NLRP12 has been shown to modulate both innate (40–42) and adaptive [43, 44] immune responses. It is expressed in bone marrow myeloid cells including granulocytes, macrophages and dendritic cells [45] and at a higher level in T cells [43]. Interestingly, NLRP12 has been shown to control the activation and migration of myeloid cells (40–42). NLRP12 negatively regulates monocyte/macrophage activation by suppressing the nuclear factor kappa B (NF-κB) signaling [40, 41]. Impairment of NLRP12 significantly hinders the migration and responsiveness of dendritic cells (DCs) and neutrophils to chemokine stimulation [42]. In parallel, a single missense mutation in Nlrp12 results in defective neutrophil recruitment [46]. In addition, the absence of Nlrp12 impairs CXCL1 production by macrophages and DCs and subsequently hinders neutrophil recruitment in response to various inflammatory stimuli [46] [47]. Moreover, while its role in B cell regulation is still to be determined, NLRP12 could negatively regulate the activation of various T cell subsets including Th1, Th2 and Th17 in a cell-intrinsic manner [43, 44, 48]. Importantly, NLRP12 has been shown to regulate immune responses through modulating the gut microbiota (49–51).
The role of NLRP12 under an autoimmune environment is not fully understood. In fact, NLRP12 has controversial roles in modulating organ-specific inflammatory disorders. For instance, it has been shown to play protective roles in colitis [52]; meanwhile, it exerts dual roles in modulating brain inflammation in experimental autoimmune encephalitis (EAE, a mouse model of multiple sclerosis) [53, 54]. The role of NLRP12 in systemic autoimmune disorders such as ALPS and SLE is unknown. In the current work, we have investigated the role of NLRP12 in a Faslpr -mediated autoimmune mouse model of ALPS and SLE, B6/lpr. We hypothesize that NLRP12 would modulate myeloid cells and T cells to control inflammation under this autoimmune condition. Surprisingly, our data has shown that the deficiency of Nlrp12 ameliorates autoimmunity in our model in a sex-dependent manner. To better understand this observation, we have also delineated the cellular mechanisms through which NLRP12 might shape autoimmune pathogenesis. In addition, we concurrently observed the dynamic changes of gut microbiota upon alteration of NLRP12 that may correlate with disease attenuation in male B6/lpr mice.
## Experimental animals
All experiments were conducted in compliance with the IACUC guidelines of Virginia Tech. Nlrp12-deficient B6/lpr was generated by cross-breeding B6.Nlrp12-/- [42] with B6.Faslpr/lpr mice (The Jackson Laboratory, Bar Harbor, ME). Offspring were genotyped for both the Nlrp12 locus and Faslpr mutation (Figure S1). We monitored the disease progression in both female and male mice housed under specific pathogen-free environment in an AAALAC accredited animal facility at Virginia Tech. All factors including housing, handling, light cycle (12-hour light/dark) were consistent for all mice, which received the hormone-free NIH-31 Modified $6\%$ Mouse/Rat diet. Food and water were provided ad libitum.
## Assessment of renal function
The development of lupus nephritis was assessed through weekly testing of proteinuria levels. Weekly urine samples were collected, and proteinuria levels were measured using a Pierce Coomassie Protein Assay Kit (Thermo Scientific) as we previously described [55]. Additionally, upon euthanasia at 39 weeks of age, kidneys were harvested to determine the deposition of immune complexes in the renal compartments through immunohistochemical staining for IgG as described below. Renal deposition of complement C3 was also determined with immunohistochemical staining.
## Measurement of serum testosterone
Endpoint serum samples were sent to the Ligand Assay & Analysis Core of the Center for Research in Reproduction (CRR) at the University of Virginia for measurement of testosterone levels. Mouse serum testosterone levels were determined using Testosterone Mouse & Rat ELISA (IBL America) following the manufacturer’s recommendations.
## Cell isolation and in vitro stimulation
Total splenocytes and bone marrow (BM) cells were isolated and red blood cell exclusion was achieved following our previously published protocols [55]. Both Splenocytes and BM cells were analyzed using flow cytometry as described below. Furthermore, for in vitro generation of BM-derived myeloid cells, BM cells from femurs were cultured at a density of 106 cells/ml for 6 days in complete RPMI medium (RPMI 1640 supplemented with $10\%$ fetal bovine serum, 1 mM sodium pyruvate, $1\%$ 100 MEM non-essential amino acids, 10 mM HEPES, 55 μM 2-mercaptoethanol, 2 mM L-glutamine, and 100 U/ml penicillin–streptomycin; all from Life Technologies, Grand Island, NY) supplemented with 10 ng/ml recombinant murine GM-CSF (PeproTech) and cultured for 6 days as previously described [56]. For in vitro stimulation of BM-derived myeloid cells, cultures were treated with 50 ng/ml or 1 μg/ml lipopolysaccharide (LPS; eBioscience) for four hours before analysis. At the end of the stimulation period, cells were harvested for both flow cytometry and RT-qPCR analysis whereas the supernatants were collected for ELISA.
## Flow cytometry
Cells were initially blocked with anti-mouse CD$\frac{16}{32}$ (eBioscience) then stained with fluorochrome-conjugated antibodies following our previously published procedures [55]. Zombie Aqua (BioLegend) staining was performed to exclude dead cells. For quantification of B cells in total splenocytes, the following anti-mouse antibodies were used: CD19-Pacific blue, CD27-PE, CD138-APC-Cy7, CD44-PerCP-Cy5.5, IgD-PE-Cy7, GL7-AF647. For splenic T cells, CD3-APC, CD4-FITC, CD8-PE, CD44-PerCP-Cy5.5, CD62L-APC-Cy7, CD69-Pacific blue, CXCR5-PerCP-Cy5.5, and PD-1-APC-Cy7 (BioLegend) were used. For myeloid cell analysis, the following anti-mouse antibodies were used: CD11b-PE, CD11c-PerCP-Cy5, F$\frac{4}{80}$-PE-Cy7 (BioLegend), Gr1-V540 (BD Bioscience). Analysis was performed with a BD FACSAria II flow cytometer (BD Biosciences). Flow cytometry data were analyzed with FlowJo.
## Immunohistochemistry
Spleen and kidney were harvested at the endpoint and embedded in Tissue-Tek OCT Compound (Sakura Finetek) and rapidly frozen in a freezing bath of dry ice and 2-methylbutane. Frozen OCT samples were cryosectioned and unstained slides were stored at −80°C. Immunohistochemical staining procedures were performed as we previously described [55]. Splenic sections were stained for germinal center (GC) formation using the following anti-mouse antibodies: CD3-APC, IgD-PE, GL7-FITC (BioLegend). Renal immune complex deposition was determined using anti-IgG-PE (eBioscience) and anti-C3-FITC (Cedarlanelabs, Burlington, Canada). Slides were mounted with Prolong Gold containing DAPI (Life Technologies). Pictures were visualized with both an EVOSVR FL microscope (Advanced Microscopy Group, Grand Island, NY) and a Zeiss LSM 880 confocal microscope (Zeiss,USA, Fralin Imaging Center, Virginia Tech). Image processing and quantification of the fluorescent intensity were performed with ImageJ and ZEN 2.1 Lite software. Sections from at least 3 mice per group were quantified and the unit used for calculation was “integrated density score.”
## RNA extraction and RT-qPCR
Total RNA extraction was performed from snap-frozen pre-weighed splenic tissue or snap-frozen cultured cells as we previously reported [55, 57]. Tissues or cell pellets were homogenized in *Qiazol lysis* reagent using TissuelyserII homogenizer (Qiagen). Total RNA was isolated using RNeasy Plus Universal Kit (Qiagen) with the elimination of gDNA. Reverse transcription (RT) was carried out using iScript™ Reverse Transcription Supermix (Bio-Rad). Quantitative PCR (qPCR) was performed utilizing the Fast SYBR® Green Master mix and the ABI 7500 Fast Real-Time PCR System (Applied Biosystems). Relative transcript quantities were calculated using the 2−ΔΔCt method and normalized to the level of the 18S rRNA housekeeping gene level. Primer sequences for mouse Bcl6, Prdm1/Blimp1, Tnfsf13b/BAFF, Il21, Tnf, Il1β, Cxcl13, Ccl19/MIP-3β, Ccr7, and androgen receptor are available in Table S1.
## ELISA
Serum samples were obtained at euthanasia, and aliquots were stored at −80°C until processing. Anti-doubles stranded (ds)DNA IgG antibodies were determined following our previously reported procedures [55]. Serum BAFF, IL-6 and IFNγ were determined using ProcartaPlex™ Multiplex Immunoassay (Invitrogen) following manufacturer’s procedures and the data were acquired and analyzed using the Luminex FlexMAP3D™ system (Chicago, USA). For culture supernatants, TNFα was determined using mouse TNFα ELISA MAX kit (BioLegend) following manufacturer’s procedures.
## Microbiota sampling and analyses
Fecal microbiota samples from each mouse at the indicated time points were obtained by taking a mouse out of the cage and collecting a fecal pellet. To avoid cross-contamination, each microbiota sample was collected by using a new pair of sterile tweezers. Samples were stored at −80°C. Similarly, at euthanasia, different intestinal sections (duodenum/jejunum, ileum, and colon) were recovered immediately, and the contents of each section were separately collected by manual extrusion and frozen immediately at −80°C until use. All samples were processed at the same time. Sample homogenization, cell lysis, and DNA extraction were performed as previously described [55, 58]. For 16S rRNA sequencing, the V4 region (ca. 252 bp) of 16S rRNA gene was PCR amplified with 515F and 12-base GoLay barcoded 806R primers [59]. The purified amplicons were sequenced bidirectionally (150 bp PE chemistry) on an Illumina MiSeq at Argonne National Laboratory. Samples were analyzed using the R package phyloseq [60]. Reads were processed and amplicon sequence variants (ASVs) were generated using DADA2 in R. Reads were quality trimmed and filtered using the command fastqPairedFilter with parameters truncLen=c[140,140], maxEE=c[2,2], rm.phix=TRUE, maxN=0, compress=TRUE, multithread=FALSE. DADA2 was used to learn error rates, perform sample inference, dereplicate and merge paired-end reads, and construct a sequence table [61]. Taxonomy was assigned using the SILVA 138 ribosomal RNA (rRNA) database training set [62] using the DADA2 functions, assignTaxonomy and addSpecies. A total of 3327 ASVs were detected in 212 total samples. ASVs seen fewer than three times in at least $20\%$ of samples and samples with fewer than 1000 reads were removed from the dataset, resulting in 205 samples and 187 ASVs used for downstream analyses. ASVs were aggregated at the genus level using the phyloseq function tax_glom. Counts were used for alpha diversity and differential abundance tests, while proportions were used to calculate Bray-Curtis dissimilarity. Differentially abundant and variable taxa between groups were identified using the function differentialTest in corncob [63] and significance was assessed using a Wald test with an FDR cutoff of 0.05. Shannon diversity was calculated using the DivNet [64] functions divnet and testDiversity. Bray-Curtis distances were calculated using the phyloseq function ordinate, specifying “method=“NMDS”, distance=“bray”, trymax=1000”. Significance was assessed using the adonis test in the vegan package with 999 permutations.
## Statistical analysis
Student’s t test was employed for the comparison between two groups. For in vitro culture data involving more than 2 groups, two-way ANOVA with Sidak’s multiple comparison test was employed. Data are shown as mean ± standard error of the mean (SEM). Significant differences were shown as *$P \leq 0.05$, **$P \leq 0.01$, ***$P \leq 0.001$, ****$P \leq 0.0001.$ All analyses were performed with Prism GraphPad.
## Nlrp12 deficiency ameliorates hallmarks of autoimmunity in male B6/lpr mice
Since sex differences exist [65], where females are more generally affected with autoimmune disease (66–68), to investigate the roles of NLRP12 in modulating inflammation in the B6/lpr model of autoimmunity, we monitored the disease progression in both male and female mice. Interestingly, the deficiency of Nlrp12 did not alter disease progression in female mice (Figures S2A, B). In contrast, while splenomegaly was not affected (data not shown), the gene deletion significantly mitigated several hallmarks of lupus disease in male B6/lpr mice, including reduced proteinuria levels (Figure 1A), decreased circulatory levels of anti-dsDNA antibodies (Figure 1B), and reduced deposition of IgG and complement C3 in renal compartments (Figures 1C, D), indicating sex-specific effects of NLRP12 in modulating lupus pathogenesis. Interestingly, we found a trending increase in both serum testosterone level (Figure S2C) and the splenic transcript level of androgen receptor (Figure S2D) in male Nlrp12-/- B6/lpr compared to Nlrp12+/+ (WT) B6/lpr mice, suggesting a potential role for sex hormones.
**Figure 1:** *Nlrp12 deficiency ameliorates hallmarks of autoimmunity in male mice with Faslpr
-mediated systemic autoimmunity. The progression of systemic autoimmunity in a mouse model of ALPS and SLE was assessed in male Nlrp12+/+
(WT) and Nlrp12-/-
(KO) B6/lpr mice. (A) Level of proteinuria over time (n=6 or 8/group). (B–D) Endpoint analyses at 39 weeks of age. (B) Level of anti-double stranded (ds)DNA IgG antibodies. (C) Immunohistochemical stains of kidney sections showing the deposition of IgG (red) and C3 (green) with DAPI staining of nuclei (blue). Pictures were captured with a Zeiss LSM 880 confocal microscope. Bar, 20 μm. (D) Mean intensity scores of IgG-PE and C3-FITC fluorescence as determined by ZEN 2.1 Lite software. Student’s t test was employed for the comparison between two groups. Data are shown as mean ± SEM. Significant differences were shown as *P < 0.05, **P < 0.01, and ***P < 0.001.*
Notably, we monitored male mice from 24 to 39 weeks of age. WT B6/lpr mice generally develop systemic autoimmunity without significant clinical pathology of renal inflammation or nephritis, which was confirmed in our studies. However, Nlrp12-/- B6/lpr mice exhibited even lower proteinuria levels that were significantly different from WT B6/lpr mice during the earlier time window from 24 to 31 weeks of age (Figure 1A), while we could not detect differences in proteinuria level during the later period from 32 to 39 weeks of age (Figure S1E).
Together, these findings indicate that NLRP12 might have pathological roles in modulating systemic autoimmunity in male B6/lpr mice. From now on, we will focus on describing male mice unless noted otherwise.
## Nlrp12 deficiency dampens B cell activation and differentiation
We detected a significantly lower level of autoantibodies and their renal deposition in the absence of NLRP12. Therefore, to delineate the mechanisms through which Nlrp12 deficiency protects against inflammation in our autoimmune model, we investigated its effects on B cell responses. Deficiency of Nlrp12 suppressed B cell responses in male B6/lpr mice (Figure 2). Nlrp12-/- B6/lpr mice had significantly reduced plasma cells (gated as CD19-CD27-CD138+IgD-) to plasmablasts (gated as CD19 +/low CD27 +/low CD138+IgD-) ratio in total splenocytes (Figure 2A), suggesting a blockade right before terminal differentiation of B cells. This is consistent with a significant reduction of the splenic transcript level of Prdm1 (Figure 2B). Interestingly, we also found a nearly significant reduction of the transcript levels of the master regulator of the germinal center (GC) reaction, Bcl6, in splenic tissues of Nlrp12-deficient mice (Figure S3A). Although GC formation shown as GL7 staining in immunohistochemically stained splenic sections was not different (Figure 2C, Figure S3B), there was a significant reduction of the percentage of GL7+ cells in total CD19+ splenic B lymphocytes (Figure 2D; gating strategy is shown in Figure S3C). Moreover, we found that Nlrp12-deficient mice had a significantly reduced percentage of splenic T follicular helper (Tfh) cells (Figure 2E; gated as CXCR5+PD-1+CD4+CD3+ in Figure S3D), as well as significantly reduced staining of CD3+ cells in the GCs (Figures 2C, S3E). Notably, the percentages of Tfh cells were low and highly variable in the WT mice, and the deficiency of NLRP12 further decreased the frequencies of these cells. Furthermore, while we only found a trending reduction of serum IL-21 (Figure S3F), a major cytokine produced by Tfh cells [69], we detected a significant reduction in its splenic transcript level in Nlrp12-deficient mice (Figure 2F). These results indicate that Nlrp12 deficiency might dampen GC reaction by suppressing the functions of Tfh cells. Finally, we found downregulated levels of factors assisting B cells [70] including the splenic transcript levels of the B cell chemoattractant Cxcl13 (Figure 2G) and the circulatory level of the B cell survival factor BAFF (Figure 2H) as well as its splenic transcript level (Figure 2I). These results indicate that Nlrp12 deficiency dampens terminal differentiation, GC reaction, and survival of potentially autoreactive B cells, which might be the reason for decreased production of autoantibodies and ameliorated autoimmune pathologies. Further studies will elucidate whether NLRP12 targets Bcl-6 and/or Blimp-1 to control autoreactive B cell responses.
**Figure 2:** *Nlrp12 deficiency dampens B cell activation and differentiation. Spleens were harvested at the endpoint of 39 weeks of age. (A) The ratio of the frequencies of plasma cells vs. plasmablasts in total splenocytes. (B) Relative transcript level of splenic Prdm1.
(C) Immunohistochemical stains of splenic sections with GL7 (purple), CD3 (red), and IgD (green). Pictures were captured with an EVOSVR FL microscope. (D) GL7+ cells as a percentage of splenic CD19+ B lymphocytes as determined with flow cytometry. (E) Percentage of Tfh cells in total splenocytes. (F) Relative transcript level of splenic Il21. (G) Relative transcript level of splenic Cxcl13. (H) Level of serum BAFF as determined with Luminex assay. (I) Relative transcript level of splenic Tnfsf13b/BAFF. Student’s t test was employed for the comparison between two groups. Data are shown as mean ± SEM. Significant differences were shown as *P < 0.05 and **P < 0.01.*
## Nlrp12 deficiency decreases T cell expansion and responses
Activation of NLR proteins can shape T cell differentiation and responses. For instance, activation of inflammasome-forming NLRs such as NLRP3 often results in the production of proinflammatory cytokines that could drive the differentiation of inflammatory T cells including Th1 and Th17 [71]. However, the exact immunoregulatory functions of NLRP12 in modulating T cell differentiation and responses are still elusive [44, 54]. Since T cells play pivotal roles in amplifying and maintaining inflammation particularly through activating autoreactive B cells [72], producing disease-promoting cytokines, and accumulating autoreactive memory [73], we sought to determine how Nlrp12 deficiency modulates the frequencies and responses of different T cell populations. Deficiency of Nlrp12 significantly reduced percentage of CD3+ T cells in total splenocytes (Figure 3A), consistent with the reduced fluorescence intensity of CD3+ T cells in immunohistochemically stained splenic GCs (Figure S3E). Nlrp12-/- B6/lpr mice also had significantly fewer CD8+ (Figure 3B) and double negative (DN)-T cell (Figure 3C) percentages in total splenocytes, which possibly contributed to the reduction in CD3+ T cells. CD4+ T cell response did not change. Importantly, the generation of DN-T cells is one of the prominent alterations of T cell responses reported in SLE [8] and ALPS (74–76). These DN-T cells could have been generated from activated CD8+ T cells (74, 77–79). Moreover, we found a reduced proportion of CD44+CD62L− effector memory T (TEM) cells in the spleens of Nlrp12-deficient mice (Figure 3D). Together, these results suggest that Nlrp12 deficiency might target T cells to dampen autoimmunity in male B6/lpr mice.
**Figure 3:** *Nlrp12 deficiency decreases T cell expansion and responses. Spleens were harvested at the endpoint of 39 weeks of age. The percentages of total T (A), CD8+ T (B), DN-T (C), and TEM
(D) cells in total splenocytes as determined with flow cytometry are shown. Data are shown as mean ± SEM. Significant differences were determined by Student’s t test and shown as *P < 0.05, **P < 0.01, and ***P < 0.001.*
## Nlrp12 deficiency reduces pro-inflammatory macrophage responses
NLRP12 can modulate the responsiveness of different myeloid cells including neutrophils, dendritic cells (DCs) and macrophages (40–42, 46, 47). We examined the immunophenotypic changes of these populations (see Figure S4A for gating strategies) in different lymphoid compartments including BM and spleen. We found no significant changes in neutrophils (Figure S4B; gated as CD11c-CD11b+Gr1+) or DCs (Figures S4C, D; gated as CD11chigh CD11b+Gr1- or CD11chigh CD11b+Gr1+). However, Nlrp12-/- B6/lpr mice showed a significant reduction of Gr1-F$\frac{4}{80}$+CD11b+CD11c-/low macrophages as the percentage of total splenocytes (Figure 4A). In addition, as the percentage of BM macrophages slightly increased in Nlrp12-/- B6/lpr mice (Figure S5A), the ratio of splenic-to-BM macrophages was significantly reduced with Nlrp12 deficiency (Figure 4B), suggesting decreased migration of these cells from BM to the spleen. Importantly, we also detected significantly reduced splenic transcript levels of Tnf (Figure 4C) and macrophage inflammatory protein 3-β (MIP-3β, gene name Ccl19; Figure 4D). These data suggest that Nlrp12 deficiency might dampen pro-inflammatory responsiveness of splenic macrophages in autoimmune environment. Interestingly, following ex-vivo stimulation of BM-derived myeloid cells with LPS – a potent activator of macrophages [80] that could prime DCs [81] – although there were slightly more BM-derived macrophages with Nlrp12 deficiency regardless of stimulation status (Figure S5B), the percentage of BM-derived DCs in these cultures was significantly reduced with the deficiency (Figure S5C). This suggests decreased priming of DCs and thus reduced functional potential of BM-derived macrophages. In parallel, we detected a significantly reduced level of TNFα in the culture supernatants of BM-derived cells with Nlrp12 deficiency following LPS stimulation (Figure 4E). Similarly, LPS-stimulated BM-derived cells from Nlrp12-/- B6/lpr mice had reduced transcript levels of Tnf (Figure 4F, Figure S5D following 50 ng/ml and 1 μg/ml LPS stimulation, respectively) and Il1β (Figure 4G, Figure S5E), as well as Ccr7 (Figure 4H, Figure S5F), a receptor known to be expressed on DCs following their activation [82]. These ex-vivo findings suggest a potential pathogenic role of NLRP12 in potentiating macrophages and DCs in response to pro-inflammatory triggers. Importantly, the change of IL-1β suggests that NLRP12 inflammasome may facilitate the production of IL-1β that in turn drives the production of other inflammatory mediators including TNF-α [83] and MIP-3β [84]. Together, these results suggest reduced pro-inflammatory innate immunity with Nlrp12 deficiency in male B6/lpr mice.
**Figure 4:** *Nlrp12 deficiency reduces pro-inflammatory macrophage responses. Spleens and BM were harvested at the endpoint of 39 weeks of age. (A) The percentage of Gr1-F4/80+CD11b+CD11c-/low macrophages in total splenocytes. (B) The ratio of splenic to BM macrophages. (C) Relative transcript level of splenic Tnf.
(D) Relative transcript level of splenic Ccl19/MIP-3β. (E-H) BM cells were stimulated ex vivo. (E) Level of TNFα in the culture supernatant as determined with ELISA following 4-h stimulation with 1 μg/ml LPS. (F-H) Transcript levels of Tnf(F), Il1β
(G) and Ccr7
(H) as fold changes over unstimulated controls following 4-h stimulation with 50 ng/ml LPS. Data are shown as mean ± SEM. Significant differences were determined by Student’s t test (A–D, F–H) or two-way ANOVA (E) and shown as *P < 0.05, **P < 0.01, and ****P < 0.0001.*
## Nlrp12 deficiency induces dynamic changes in gut microbiota diversity and composition
Changes of microbiota dynamics have been shown to drive autoimmunity [32, 33] or modulate autoimmunity [28]. This has been established for SLE [37, 38] but not yet for ALPS. Interestingly, NLRP12 could shape inflammatory outcomes through regulating the gut microbiota [49, 85]. Thus, we investigated whether the alteration of Nlrp12 could implicate the gut microbiota in B6/lpr mice. We analyzed the fecal and intestinal microbiotas of both male and female mice seeking to answer the sex-dependent response to Nlrp12 deficiency. We found a clear distinction in the fecal microbiota diversity with or without NLRP12 (Figure 5). Fecal microbiotas had significantly different alpha diversity on the genus level as shown by Shannon diversity estimate in both male (Figure 5A) and female (Figure 5B) B6/lpr mice, where Nlrp12 deficiency led to significantly increased microbiota diversity. However, the difference in alpha diversity was much more pronounced in male than female mice. Similarly, analysis of beta diversity based on Bray Curtis dissimilarity calculation showed significantly different overall taxonomic composition based on the genotype but not the timepoint between WT and Nlrp12-/- B6/lpr male (Figure 5C) and female (Figure 5D) mice. Moreover, the composition of fecal microbiota changed upon alteration of Nlrp12. We detected significant enrichment of various genera in Nlrp12-/- B6/lpr male (Figure 5E) and female mice (Figure 5F). Strikingly, the intestinal microbiota diversity showed clear differences only in male mice that might explain the sex-dependent changes in disease phenotype. Analysis of alpha diversity from different intestinal segments (duodenum/jejunum, ileum, and colon) at 39 weeks of age showed that male (Figure 6A, $$P \leq 0.028$$), but not female (Figure 6B, $$P \leq 0.963$$), mice have distinct microbial composition upon alteration of Nlrp12. Similarly, the overall taxonomic composition was different for genotype and intestinal segment only among males (Figure 6C, $$P \leq 0.001$$) but not females (Figure 6D, $$P \leq 0.065$$). While not many changes were observed as in fecal microbiota, several genera were significantly altered in the intestinal microbiota of WT vs. Nlrp12-/- B6/lpr male (Figure 6E) and female mice (Figure 6F).
**Figure 5:** *Nlrp12 deficiency induces dynamic changes in fecal microbiota diversity and composition. (A, B) Alpha diversity of fecal microbiota based on Shannon diversity estimate in male (A) and female (B) B6/lpr mice upon alteration of NLRP12. (C, D) Non-metric multidimensional scaling (axes NMDS1 vs. NMDS2) showing the segregation of fecal microbiota overtime based on Bray Curtis dissimilarity of beta diversity in male (C) and female (D) mice. (E, F) Differentially abundant bacterial taxa at the genus level in male (E) and female (F) fecal microbiota.* **Figure 6:** *Nlrp12 deficiency induces dynamic changes in intestinal microbiota diversity and composition. Intestinal microbiota was collected at the endpoint of 39 weeks of age. (A, B) Alpha diversity of intestinal microbiota based on Shannon diversity estimate in male (A) and female (B) B6/lpr mice upon alteration of NLRP12. (C, D) Non-metric multidimensional scaling showing the segregation of intestinal microbiota based on Bray Curtis dissimilarity of beta diversity in male (C) and female (D) mice. (E, F) Differentially abundant bacterial taxa at the genus level in male (E) and female (F) intestinal microbiota.*
## Discussion
We investigated the role of NLRP12 in modulating autoimmune-associated inflammation utilizing the Faslpr mutant mice as a model of ALPS and SLE [1]. NLRP12 is an inhibitory checkpoint of inflammation; but at the same time, it can form inflammasome to promote inflammation [39]. So far, triggers that direct the activity of NLRP12 to either way are not fully understood. The findings of this work support the hypothesis that NLRP12 can work towards the inflammasome activation pathway to deteriorate systemic autoimmunity. Inflammasome protein complex including NLRP12 has been proposed to be implicated in ALPS [86]. Similarly, a recent study has shown that the expression of NLRP12 together with other inflammasome-forming innate sensors is increased in the SLE B cells [87]. However, the mechanisms through which NLRP12 could modulate systemic autoimmunity are still elusive.
Here, we show how NLRP12 modulates cellular responses under autoimmune conditions. Nlrp12 deficiency attenuated autoreactive B cell responses in B6/lpr mice, dampening production of autoantibodies and their renal deposition. Mechanistically, Nlrp12 deficiency may have hindered terminal differentiation, GC formation, and survival of autoreactive B cells, suggesting B cells as a potential hub for NLRP12 inflammasome activity in autoimmune conditions. In parallel, NLRP12 is expressed at high levels in T cells [43] and has been shown to modulate the differentiation and responses of different T cell subsets [43, 54]. Specifically, T cells can maintain an inflammatory milieu [88, 89] and potentiate B cell autoreactivity [72], a phenomenon implicated in both SLE [73] and ALPS (74–76). Interestingly, we found that NLRP12 could drive and maintain the accumulation of T cells in the spleen. The deficiency of Nlrp12 reduced the percentage of splenic CD3+ T cells and importantly, the generation of DN-T cells and TEM cells, which are known pathogenic T cell subsets in the autoimmunity (8, 74–76, 90). Moreover, our results suggest that NLRP12 might drive B cell activation through promoting Tfh cells, where mice with intact NLRP12 had an expansion of Tfh cells and upregulated levels of factors associated with B cell help [70]. To this end, our findings warrant further investigation on the cell-specific mechanisms, either intrinsic or extrinsic, through which NLRP12 might target B cell autoreactivity to deteriorate systemic autoimmunity in male B6/lpr mice. Furthermore, we found decreased levels of inflammatory mediators including TNFα, MIP-3β, IL-1β and CCR7 in splenic tissues and/or BM-derived myeloid cell cultures following ex-vivo stimulation for Nlrp12-/- B6/lpr mice, supporting the notion that NLRP12 might trigger inflammasome activation in different immune cell populations to deteriorate systemic inflammation.
Importantly, Nlrp12 deficiency dramatically altered the gut microbiota especially in male mice. Although Nlrp12 alteration significantly changed the diversity and composition of fecal microbiota in both males and females, significant differences in the intestinal microbiota were seen only in male mice. This observation supports the notion that gut microbiota might drive the sex-dependent outcome of Nlrp12 deficiency in our mouse model of systemic autoimmunity. However, future studies are still needed to mechanistically delineate our observations and to demonstrate the potential link between gut microbiota and the sex-dependent outcomes seen in Nlrp12-deficient mice. It is also likely that treating females with testosterone, or gut microbiota from male mice, will restore the male phenotype seen in this study.
In conclusion, the present study provides novel insight into the immunoregulatory role of NLRP12 in systemic autoimmune disorders such as ALPS and SLE. Attenuation of autoreactive cell responses including B, T, and myeloid cells that we have observed in the absence of NLRP12 supports a sex-dependent, pro-inflammatory role of NLRP12 under autoimmune conditions that warrant further investigation to decipher the underlying mechanisms. In addition, the marked differences in microbiota diversity and composition between WT and Nlrp12-/- B6/lpr mice suggest a microbiota-dependent role of NLRP12 in shaping autoimmune pathogenesis. Future studies will reveal a potential gut microbiota-dependent mechanism by which NLRP12 deficiency attenuates autoimmune pathologies in male mice. We will employ antibiotic treatment, co-housing, and gut microbiota transplantation experiments to determine whether changes of the gut microbiota are a cause, or an effect, of the attenuated disease phenotype in male Nlrp12-/- B6/lpr mice. As gut microbiota has been shown to drive autoimmunity in a sex-dependent manner [91], studies with mice deficient in androgen receptors will also reveal a potential role for male hormones that may work in concert with gut microbiota.
## Data availability statement
The data presented in the study are deposited in the NCBI repository, accession number PRJNA805257. https://www.ncbi.nlm.nih.gov/bioproject/PRJNA805257.
## Ethics statement
The animal study was reviewed and approved by IACUC of Virginia Tech.
## Author contributions
XL and SA conceived the study. IA provided B6.Nlrp12-/- mice. LA performed the research. JM, XC-P, JZ, JT and BS contributed to mouse sampling and tissue harvesting. ME contributed to mouse breeding. JSM analyzed the microbiota data. LA and XL analyzed the data and wrote the manuscript. SA edited the manuscript. All authors contributed to the article and approved the submitted version.
## Conflict of interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher’s note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## Supplementary material
The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fimmu.2023.1120958/full#supplementary-material
## References
1. Cohen PL, Eisenberg RA. **Lpr and gld: single gene models of systemic autoimmunity and lymphoproliferative disease**. *Annu Rev Immunol* (1991) **9**. DOI: 10.1146/annurev.iy.09.040191.001331
2. Sneller MC, Dale JK, Straus SE. **Autoimmune lymphoproliferative syndrome**. *Curr Opin Rheumatol* (2003) **15**. DOI: 10.1097/00002281-200307000-00008
3. Lockshin MD, Barbhaiya M, Izmirly P, Buyon JP, Crow MK. **SLE: reconciling heterogeneity**. *Lupus Sci Med* (2019) **6**. DOI: 10.1136/lupus-2018-000280
4. Tsokos GC, Lo MS, Reis PC, Sullivan KE. **New insights into the immunopathogenesis of systemic lupus erythematosus**. *Nat Rev Rheumatol* (2016) **12** 716-730. DOI: 10.1038/nrrheum.2016.186
5. Fisher GH, Rosenberg FJ, Straus SE, Dale JK, Middelton LA, Lin AY. **Dominant interfering fas gene mutations impair apoptosis in a human autoimmune lymphoproliferative syndrome**. *Cell* (1995) **81**. DOI: 10.1016/0092-8674(95)90013-6
6. Rieux-Laucat F, Le Deist F, Hivroz C, Roberts I, Debatin K, Fischer A. **Mutations in fas associated with human lymphoproliferative syndrome and autoimmunity**. *Science* (1995) **268**. DOI: 10.1126/science.7539157
7. White S, Rosen A. **Apoptosis in systemic lupus erythematosus**. *Curr Opin Rheumatol* (2003) **15**. DOI: 10.1097/00002281-200309000-00006
8. Crispín JC, Oukka M, Bayliss G, Cohen RA, Van Beek CA, Stillman IE. **Expanded double negative T cells in patients with systemic lupus erythematosus produce IL-17 and infiltrate the kidneys**. *J Immunol* (2008) **181**. DOI: 10.4049/jimmunol.181.12.8761
9. Mazerolles F, Stolzenberg M-C, Pelle O, Picard C, Neven B, Fischer A. **Autoimmune lymphoproliferative syndrome-FAS patients have an abnormal regulatory T cell (Treg) phenotype but display normal natural treg-suppressive function on T cell proliferation**. *Front Immunol* (2018) **9**. DOI: 10.3389/fimmu.2018.00718
10. Teachey DT, Manno CS, Axsom KM, Andrews T, Choi JK, Greenbaum BH. **Unmasking Evans syndrome: T-cell phenotype and apoptotic response reveal autoimmune lymphoproliferative syndrome (ALPS)**. *Blood* (2005) **105**. DOI: 10.1182/blood-2004-09-3542
11. Fuss IJ, Strober W, Dale JK, Fritz S, Pearlstein GR, Puck JM. **Characteristic T helper 2 T cell cytokine abnormalities in autoimmune lymphoproliferative syndrome, a syndrome marked by defective apoptosis and humoral autoimmunity**. *J Immunol* (1997) **158**. DOI: 10.4049/jimmunol.158.4.1912
12. Jackson CE, Puck JM. **Autoimmune lymphoproliferative syndrome, a disorder of apoptosis**. *Curr Opin Pediatr* (1999) **11**. DOI: 10.1097/00008480-199912000-00009
13. Ma C, Xia Y, Yang Q, Zhao Y. **The contribution of macrophages to systemic lupus erythematosus**. *Clin Immunol* (2019) **207** 1-9. DOI: 10.1016/j.clim.2019.06.009
14. Liao X, Ren J, Reihl A, Pirapakaran T, Sreekumar B, Cecere T. **Renal-infiltrating CD11c+ cells are pathogenic in murine lupus nephritis through promoting CD4+ T cell responses**. *Clin Exp Immunol* (2017) **190** 187-200. DOI: 10.1111/cei.13017
15. Fiore N, Castellano G, Blasi A, Capobianco C, Loverre A, Montinaro V. **Immature myeloid and plasmacytoid dendritic cells infiltrate renal tubulointerstitium in patients with lupus nephritis**. *Mol Immunol* (2008) **45**. DOI: 10.1016/j.molimm.2007.04.029
16. Kaplan MJ. **Neutrophils in the pathogenesis and manifestations of SLE**. *Nat Rev Rheumatol* (2011) **7**. DOI: 10.1038/nrrheum.2011.132
17. Herrada AA, Escobedo N, Iruretagoyena M, Valenzuela RA, Burgos PI, Cuitino L. **Innate immune cells’ contribution to systemic lupus erythematosus**. *Front Immunol* (2019) **10**. DOI: 10.3389/fimmu.2019.00772
18. Waldner H. **The role of innate immune responses in autoimmune disease development**. *Autoimmun Rev* (2009) **8**. DOI: 10.1016/j.autrev.2008.12.019
19. Rönnblom L, Pascual V. **The innate immune system in SLE: Type I interferons and dendritic cells**. *Lupus* (2008) **17**. DOI: 10.1177/0961203308090020
20. Moulton VR, Tsokos GC. **Abnormalities of T cell signaling in systemic lupus erythematosus**. *Arthritis Res Ther* (2011) **13** 1-10. DOI: 10.1186/ar3251
21. Katsuyama T, Tsokos GC, Moulton VR. **Aberrant T cell signaling and subsets in systemic lupus erythematosus**. *Front Immunol* (2018) **9**. DOI: 10.3389/fimmu.2018.01088
22. Kammer GM, Perl A, Richardson BC, Tsokos GC. **Abnormal T cell signal transduction in systemic lupus erythematosus**. *Arthritis Rheumatism* (2002) **46**. DOI: 10.1002/art.10192
23. Morimoto C, Reinherz EL, Schlossman SF, Schur P, Mills J, Steinberg A. **Alterations in immunoregulatory T cell subsets in active systemic lupus erythematosus**. *J Clin Invest* (1980) **66**. DOI: 10.1172/JCI109948
24. Lipsky PE. **Systemic lupus erythematosus: An autoimmune disease of b cell hyperactivity**. *Nat Immunol* (2001) **2**. DOI: 10.1038/ni0901-764
25. Doerner T, Jacobi AM, Lee J, Lipsky PE. **Abnormalities of b cell subsets in patients with systemic lupus erythematosus**. *J Immunol Methods* (2011) **363**. DOI: 10.1016/j.jim.2010.06.009
26. Grammer AC, Lipsky PE. **B cell abnormalities in systemic lupus erythematosus**. *Arthritis Res Ther* (2003) **5** 1-6. DOI: 10.1186/ar1009
27. Zharkova O, Celhar T, Cravens PD, Satterthwaite AB, Fairhurst A-M, Davis LS. **Pathways leading to an immunological disease: Systemic lupus erythematosus**. *Rheumatology* (2017) **56**. DOI: 10.1093/rheumatology/kew427
28. Yurkovetskiy LA, Pickard JM, Chervonsky AV. **Microbiota and autoimmunity: Exploring new avenues**. *Cell Host Microbe* (2015) **17**. DOI: 10.1016/j.chom.2015.04.010
29. Xu H, Liu M, Cao J, Li X, Fan D, Xia Y. **The dynamic interplay between the gut microbiota and autoimmune diseases**. *J Immunol Res* (2019) **2019**. DOI: 10.1155/2019/7546047
30. Jiao Y, Wu L, Huntington ND, Zhang X. **Crosstalk between gut microbiota and innate immunity and its implication in autoimmune diseases**. *Front Immunol* (2020) **11**. DOI: 10.3389/fimmu.2020.00282
31. López P, de Paz B, Rodríguez-Carrio J, Hevia A, Sánchez B, Margolles A. **Th17 responses and natural IgM antibodies are related to gut microbiota composition in systemic lupus erythematosus patients**. *Sci Rep* (2016) **6** 1-12. DOI: 10.1038/srep24072
32. Zhang H, Liao X, Sparks JB, Luo XM. **Dynamics of gut microbiota in autoimmune lupus**. *Appl Environ Microbiol* (2014) **80**. DOI: 10.1128/AEM.02676-14
33. Luo XM, Edwards MR, Mu Q, Yu Y, Vieson MD, Reilly CM. **Gut microbiota in human systemic lupus erythematosus and a mouse model of lupus**. *Appl Environ Microbiol* (2018) **84**. DOI: 10.1128/AEM.02288-17
34. Zhang L, Qing P, Yang H, Wu Y, Liu Y, Luo Y. **Gut microbiome and metabolites in systemic lupus erythematosus: Link, mechanisms and intervention**. *Front Immunol* (2021) **12**. DOI: 10.3389/fimmu.2021.686501
35. Ma Y, Xu X, Li M, Cai J, Wei Q, Niu H. **Gut microbiota promote the inflammatory response in the pathogenesis of systemic lupus erythematosus**. *Mol Med* (2019) **25** 1-16. DOI: 10.1186/s10020-019-0102-5
36. Rosser EC, Mauri C. **A clinical update on the significance of the gut microbiota in systemic autoimmunity**. *J Autoimmun* (2016) **74** 85-93. DOI: 10.1016/j.jaut.2016.06.009
37. Mu Q, Zhang H, Liao X, Lin K, Liu H, Edwards MR. **Control of lupus nephritis by changes of gut microbiota**. *Microbiome* (2017) **5** 1-12. DOI: 10.1186/s40168-017-0300-8
38. Mu Q, Edwards MR, Swartwout BK, Cabana Puig X, Mao J, Zhu J. **Gut microbiota and bacterial DNA suppress autoimmunity by stimulating regulatory b cells in a murine model of lupus**. *Front Immunol* (2020) **11**. DOI: 10.3389/fimmu.2020.593353
39. Tuladhar S, Kanneganti T-D. **NLRP12 in innate immunity and inflammation**. *Mol Aspects Med* (2020) **76** 100887. DOI: 10.1016/j.mam.2020.100887
40. Lich JD, Williams KL, Moore CB, Arthur JC, Davis BK, Taxman DJ. **Cutting edge: Monarch-1 suppresses non-canonical NF-κB activation and p52-dependent chemokine expression in monocytes**. *J Immunol* (2007) **178**. DOI: 10.4049/jimmunol.178.3.1256
41. Zaki MH, Vogel P, Malireddi RS, Body-Malapel M, Anand PK, Bertin J. **The NOD-like receptor NLRP12 attenuates colon inflammation and tumorigenesis**. *Cancer Cell* (2011) **20**. DOI: 10.1016/j.ccr.2011.10.022
42. Arthur JC, Lich JD, Ye Z, Allen IC, Gris D, Wilson JE. **Cutting edge: NLRP12 controls dendritic and myeloid cell migration to affect contact hypersensitivity**. *J Immunol* (2010) **185**. DOI: 10.4049/jimmunol.1002227
43. Lukens JR, Gurung P, Shaw PJ, Barr MJ, Zaki MH, Brown SA. **The NLRP12 sensor negatively regulates autoinflammatory disease by modulating interleukin-4 production in T cells**. *Immunity* (2015) **42**. DOI: 10.1016/j.immuni.2015.03.006
44. Thaiss CA, Elinav E. **NF-κB regulation by NLRs: T cells join the club**. *Immunity* (2015) **42**. DOI: 10.1016/j.immuni.2015.03.010
45. Williams KL, Taxman DJ, Linhoff MW, Reed W, Ting JP-Y. **Cutting edge: Monarch-1: a pyrin/nucleotide-binding domain/leucine-rich repeat protein that controls classical and nonclassical MHC class I genes**. *J Immunol* (2003) **170**. DOI: 10.4049/jimmunol.170.11.5354
46. Ulland TK, Jain N, Hornick EE, Elliott EI, Clay GM, Sadler JJ. **Nlrp12 mutation causes C57BL/6J strain-specific defect in neutrophil recruitment**. *Nat Commun* (2016) **7** 1-13. DOI: 10.1038/ncomms13180
47. Hornick EE, Banoth B, Miller AM, Zacharias ZR, Jain N, Wilson ME. **Nlrp12 mediates adverse neutrophil recruitment during influenza virus infection**. *J Immunol* (2018) **200**. DOI: 10.4049/jimmunol.1700999
48. Prado DS, Veras FP, Ferreira RG, Damasceno LEA, Melo PH, Zamboni DS. **NLRP12 controls arthritis severity by acting as a checkpoint inhibitor of Th17 cell differentiation**. *FASEB J* (2020) **34**. DOI: 10.1096/fj.202000795R
49. Truax AD, Chen L, Tam JW, Cheng N, Guo H, Koblansky AA. **The inhibitory innate immune sensor NLRP12 maintains a threshold against obesity by regulating gut microbiota homeostasis**. *Cell Host Microbe* (2018) **24** 364-78.e6. DOI: 10.1016/j.chom.2018.08.009
50. Sun Z, Pei W, Guo Y, Wang Z, Shi R, Chen X. **Gut microbiota-mediated NLRP12 expression drives the attenuation of dextran sulphate sodium-induced ulcerative colitis by qingchang wenzhong decoction**. *Evidence-Based Complementary Altern Med* (2019) **2019** 9839474. DOI: 10.1155/2019/9839474
51. Chen L, Wilson JE, Koenigsknecht MJ, Chou W-C, Montgomery SA, Truax AD. **NLRP12 attenuates colon inflammation by maintaining colonic microbial diversity and promoting protective commensal bacterial growth**. *Nat Immunol* (2017) **18**. DOI: 10.1038/ni.3690
52. Allen IC, Wilson JE, Schneider M, Lich JD, Roberts RA, Arthur JC. **NLRP12 suppresses colon inflammation and tumorigenesis through the negative regulation of noncanonical NF-κB signaling**. *Immunity* (2012) **36**. DOI: 10.1016/j.immuni.2012.03.012
53. Gharagozloo M, Mahvelati TM, Imbeault E, Gris P, Zerif E, Bobbala D. **The nod-like receptor, Nlrp12, plays an anti-inflammatory role in experimental autoimmune encephalomyelitis**. *J Neuroinflamm* (2015) **12** 1-13. DOI: 10.1186/s12974-015-0414-5
54. Gharagozloo M, Mahmoud S, Simard C, Mahvelati TM, Amrani A, Gris D. **The dual immunoregulatory function of Nlrp12 in T cell-mediated immune response: Lessons from experimental autoimmune encephalomyelitis**. *Cells* (2018) **7** 119. DOI: 10.3390/cells7090119
55. Abdelhamid L, Cabana-Puig X, Swartwout B, Lee J, Li S, Sun S. **Retinoic acid exerts disease stage-dependent effects on pristane-induced lupus**. *Front Immunol* (2020) **11**. DOI: 10.3389/fimmu.2020.00408
56. Goodridge HS, Shimada T, Wolf AJ, Hsu Y-MS, Becker CA, Lin X. **Differential use of CARD9 by dectin-1 in macrophages and dendritic cells**. *J Immunol* (2009) **182**. DOI: 10.4049/jimmunol.182.2.1146
57. Abdelhamid L, Cabana-Puig X, Mu Q, Moarefian M, Swartwout B, Eden K. **Quaternary ammonium compound disinfectants reduce lupus-associated splenomegaly by targeting neutrophil migration and T-cell fate**. *Front Immunol* (2020) **11**. DOI: 10.3389/fimmu.2020.575179
58. Ley RE, Bäckhed F, Turnbaugh P, Lozupone CA, Knight RD, Gordon JI. **Obesity alters gut microbial ecology**. *Proc Natl Acad Sci* (2005) **102**. DOI: 10.1073/pnas.0504978102
59. Caporaso JG, Lauber CL, Walters WA, Berg-Lyons D, Huntley J, Fierer N. **Ultra-high-throughput microbial community analysis on the illumina HiSeq and MiSeq platforms**. *ISME J* (2012) **6**. DOI: 10.1038/ismej.2012.8
60. McMurdie PJ, Holmes S. **Phyloseq: An r package for reproducible interactive analysis and graphics of microbiome census data**. *PloS One* (2013) **8** e61217. DOI: 10.1371/journal.pone.0061217
61. Callahan BJ, McMurdie PJ, Rosen MJ, Han AW, Johnson AJA, Holmes SP. **DADA2: High-resolution sample inference from illumina amplicon data**. *Nat Methods* (2016) **13**. DOI: 10.1038/nmeth.3869
62. McLaren. M. **mikemc/dada2-reference-databases: Silva 138.1 (Version v2)**. *Zenodo* (2021). DOI: 10.5281/zenodo.4587946
63. Martin BD, Witten D, Willis AD. **Modeling microbial abundances and dysbiosis with beta-binomial regression**. *Ann Appl Stat* (2020) **14** 94. DOI: 10.1214/19-AOAS1283
64. Willis AD, Martin BD. **Estimating diversity in networked ecological communities**. *Biostatistics* (2020) **23**. DOI: 10.1093/biostatistics/kxaa015
65. Ngo ST, Steyn FJ, McCombe PA. **Gender differences in autoimmune disease**. *Front Neuroendocrinol* (2014) **35**. DOI: 10.1016/j.yfrne.2014.04.004
66. Munoz-Grajales C, Gonzalez L, Alarcon G, Acosta-Reyes J. **Gender differences in disease activity and clinical features in newly diagnosed systemic lupus erythematosus patients**. *Lupus* (2016) **25**. DOI: 10.1177/0961203316635286
67. Wasef SZY. **Gender differences in systemic lupus erythematosus**. *Gender Med* (2004) **1**. DOI: 10.1016/S1550-8579(04)80006-8
68. Macedo E, Appenzeller S, Costallat L. **Gender differences in systemic lupus erythematosus concerning anxiety, depression and quality of life**. *Lupus* (2016) **25**. DOI: 10.1177/0961203316638934
69. Bélanger S, Crotty S. **Dances with cytokines, featuring TFH cells, IL-21, IL-4 and b cells**. *Nat Immunol* (2016) **17**. DOI: 10.1038/ni.3561
70. Rao DA. **T Cells that help b cells in chronically inflamed tissues**. *Front Immunol* (2018) **9**. DOI: 10.3389/fimmu.2018.01924
71. Gris D, Ye Z, Iocca HA, Wen H, Craven RR, Gris P. **NLRP3 plays a critical role in the development of experimental autoimmune encephalomyelitis by mediating Th1 and Th17 responses**. *J Immunol* (2010) **185**. DOI: 10.4049/jimmunol.0904145
72. Mountz JD, Hsu H-C, Ballesteros-Tato A. **Dysregulation of T follicular helper cells in lupus**. *J Immunol* (2019) **202**. DOI: 10.4049/jimmunol.1801150
73. Suárez-Fueyo A, Bradley SJ, Tsokos GC. **T Cells in systemic lupus erythematosus**. *Curr Opin Immunol* (2016) **43**. DOI: 10.1016/j.coi.2016.09.001
74. Bristeau-Leprince A, Mateo V, Lim A, Magerus-Chatinet A, Solary E, Fischer A. **Human TCR α/β+ CD4– CD8– double-negative T cells in patients with autoimmune lymphoproliferative syndrome express restricted vβ TCR diversity and are clonally related to CD8+ T cells**. *J Immunol* (2008) **181**. DOI: 10.4049/jimmunol.181.1.440
75. Rieux-Laucat F, Magerus-Chatinet A. **Autoimmune lymphoproliferative syndrome: a multifactorial disorder**. *Haematologica* (2010) **95**. DOI: 10.3324/haematol.2010.030395
76. Bleesing JJ, Brown MR, Novicio C, Guarraia D, Dale JK, Straus SE. **A composite picture of TcR alpha/beta(+) CD4(-)CD8(-) T cells (alpha/beta-DNTCs) in humans with autoimmune lymphoproliferative syndrome**. *Clin Immunol* (2002) **104** 21-30. DOI: 10.1006/clim.2002.5225
77. Crispín JC, Tsokos GC. **Human TCR-αβ+ CD4– CD8– T cells can derive from CD8+ T cells and display an inflammatory effector phenotype**. *J Immunol* (2009) **183**. DOI: 10.4049/jimmunol.0901533
78. Mehal WZ, Crispe IN. **TCR ligation on CD8+ T cells creates double-negative cells**. *J Immunol* (1998) **161**. DOI: 10.4049/jimmunol.161.4.1686
79. Rodríguez-Rodríguez N, Apostolidis SA, Penaloza-MacMaster P, Villa JMM, Barouch DH, Tsokos GC. **Programmed cell death 1 and Helios distinguish TCR-αβ+ double-negative (CD4– CD8–) T cells that derive from self-reactive CD8 T cells**. *J Immunol* (2015) **194**. DOI: 10.4049/jimmunol.1402775
80. Meng F, Lowell CA. **Lipopolysaccharide (LPS)-induced macrophage activation and signal transduction in the absence of src-family kinases hck, fgr, and Lyn**. *J Exp Med* (1997) **185**. DOI: 10.1084/jem.185.9.1661
81. Abdi K, Singh NJ, Matzinger P. **Lipopolysaccharide-activated dendritic cells:”exhausted” or alert and waiting**. *J Immunol* (2012) **188**. DOI: 10.4049/jimmunol.1102868
82. Riol-Blanco L, Sánchez-Sánchez N, Torres A, Tejedor A, Narumiya S, Corbí AL. **The chemokine receptor CCR7 activates in dendritic cells two signaling modules that independently regulate chemotaxis and migratory speed**. *J Immunol* (2005) **174**. DOI: 10.4049/jimmunol.174.7.4070
83. Dinarello CA. **A clinical perspective of IL-1β as the gatekeeper of inflammation**. *Eur J Immunol* (2011) **41**. DOI: 10.1002/eji.201141550
84. Maurer M, Von Stebut E. **Macrophage inflammatory protein-1**. *Int J Biochem Cell Biol* (2004) **36**. DOI: 10.1016/j.biocel.2003.10.019
85. Ray K. **NLRP12 regulates gut microbiota to suppress intestinal inflammation**. *Nat Rev Gastroenterol Hepatol* (2017) **14**. DOI: 10.1038/nrgastro.2017.43
86. Palmisani E, Miano M, Grossi A, Lanciotti M, Lupia M, Terranova P. **Autoimmune lymphoproliferative syndrome (ALPS) disease and ALPS phenotype: Are they two distinct entities**. *Hemasphere* (2023) **7**. DOI: 10.1097/HS9.0000000000000845
87. Yang M, Long D, Hu L, Zhao Z, Li Q, Guo Y. **AIM2 deficiency in b cells ameliorates systemic lupus erythematosus by regulating Blimp-1–Bcl-6 axis-mediated b-cell differentiation**. *Signal Transduction Targeted Ther* (2021) **6** 341. DOI: 10.1038/s41392-021-00725-x
88. Koga T, Hedrich CM, Mizui M, Yoshida N, Otomo K, Lieberman LA. **CaMK4-dependent activation of AKT/mTOR and CREM-α underlies autoimmunity-associated Th17 imbalance**. *J Clin Invest* (2014) **124**. DOI: 10.1172/JCI73411
89. Zhou H, Li B, Li J, Wu T, Jin X, Yuan R. **Dysregulated T cell activation and aberrant cytokine expression profile in systemic lupus erythematosus**. *Mediators Inflammation* (2019) **2019** 8450947. DOI: 10.1155/2019/8450947
90. Devarajan P, Chen Z. **Autoimmune effector memory T cells: The bad and the good**. *Immunol Res* (2013) **57** 12-22. DOI: 10.1007/s12026-013-8448-1
91. Markle JG, Frank DN, Mortin-Toth S, Robertson CE, Feazel LM, Rolle-Kampczyk U. **Sex differences in the gut microbiome drive hormone-dependent regulation of autoimmunity**. *Science* (2013) **339**. DOI: 10.1126/science.1233521
|
---
title: 'Association between different metabolic phenotypes and the development of
hypothyroidism: 9 years follow-up of Tehran thyroid study'
authors:
- Behnaz Abiri
- Amirhossein Ramezani Ahmadi
- Maryam Mahdavi
- Farhad Hosseinpanah
- Atieh Amouzegar
- Majid Valizadeh
journal: Frontiers in Endocrinology
year: 2023
pmcid: PMC10036795
doi: 10.3389/fendo.2023.1134983
license: CC BY 4.0
---
# Association between different metabolic phenotypes and the development of hypothyroidism: 9 years follow-up of Tehran thyroid study
## Abstract
### Purpose
The association between metabolic phenotypes and thyroid function has not yet been established; therefore, this study examined whether different metabolic phenotypes are associated with the development of hypothyroidism.
### Methods
Study participants were selected from the Tehran Thyroid Study (TTS). A total of 3338 euthyroid adults were included and categorized into four obesity phenotype groups: metabolically healthy normal weight (MHNW), metabolically healthy obese (MHO), metabolically unhealthy normal weight (MUNW), and metabolically unhealthy obese (MUO). The participants were assessed at baseline and during three follow-up studies at three-year intervals. Multiple logistic regression analysis was used to examine the relationship between metabolic phenotypes and the development of hypothyroidism.
### Results
In the total population, the chi-square test was only significant ($$P \leq 0.008$$) in 3rd year with a higher prevalence of hypothyroidism in the MUNW phenotype, followed by MHO, MUO, and MHNW. Moreover, in the 3rd and 9th years of follow-up, hypothyroidism was more prevalent in MUO only in male subjects ($$P \leq 0.002$$ and 0.035, respectively). In the unadjusted model, the MHO phenotype increased the odds of hypothyroidism compared with the MHNW phenotype (OR=1.51; $95\%$ CI=1.04, 2.18; P-value=0.031). After adjusting for confounding factors, the odds of hypothyroidism were higher in the MUNW (OR=1.86; $95\%$ CI=1.17, 2.96; P-value=0.008), MHO (OR=1.71; $95\%$ CI=1.09, 2.67; P-value=0.018), and MUO (OR=1.64; $95\%$ CI=1.03, 2.62; P-value=0.036) phenotypes than in the MHNW group. The MUNW phenotype increased the risk of hypothyroidism compared to MHNW, only in males. However, in females, the MHO phenotype increased the risk of hypothyroidism compared to MHNW.
### Conclusion
Both obesity and metabolic abnormalities are associated with hyperthyroidism. Healthy metabolic and weight maintenance were associated with a lower risk of hypothyroidism in males and females.
## Introduction
Hypothyroidism is a common pathological condition characterized by a deficiency of thyroid hormones that can be overt or subclinical [1]. There is limited information about the incidence of hypothyroidism in Middle Eastern countries. A systematic review [2] examined the prevalence of thyroid disease in ten Middle Eastern countries; however, the study population was heterogeneous. The incidence rates of subclinical and overt hypothyroidism in Tehran, the capital city of Iran, and an iodine-sufficient region, were 7.62 and 2.0 per 1000 individuals, respectively) [3]. Evidence suggests that hypothyroidism increases the risk of cardiovascular events and mortality [4, 5]. Considering these facts, identifying hypothyroidism risk factors is crucial for preventing its increase.
There is also a growing epidemic of obesity in the global population, with serious adverse health consequences. Recent studies have shown a link between obesity and thyroid dysfunction, and several studies have reported that obesity causes thyroid problems as well as being a result of them (6–8). According to a meta-analysis of 22 studies, obese individuals are more likely to have overt and subclinical hypothyroidism [9]. Typically, obesity is linked to metabolic abnormalities including hypertension, hyperlipidemia, and hyperglycemia. According to a prospective cohort study, participants with metabolic syndrome at baseline are more likely to develop subclinical hypothyroidism in the future [10]. Thus, both obesity and metabolic disorders are closely associated with hypothyroidism.
It is well known that obesity adversely affects metabolic health, but individual responses differ [11]. Some individuals who are obese may have a metabolically healthy obese (MHO) phenotype [12]. Furthermore, metabolically unhealthy normal weight (MUNW) refers to individuals who have abnormal metabolic parameters, but are not obese [13]. It is more accurate to predict cardiovascular disease and mortality from obesity phenotypes that combine obesity with different metabolic profiles [14]. Different types of obesity could also provide insight into whether obesity or coexisting metabolic abnormalities are associated with hypothyroidism.
Until now, the idea that thyroid function could be used to identify obesity phenotypes in individuals with euthyroidism has only been explored in a few studies [15]. However, the relationship between metabolic phenotypes and thyroid function has not been determined. In the present study, we investigated the relationship between different metabolic phenotypes and the development of hypothyroidism, as well as the modulating effect of sex within a 9-year follow-up in a cohort of the Tehran Thyroid Study (TTS).
## Study population
The study participants were recruited from the TTS [16], a cohort study that is conducted within the framework of the Tehran Lipid and Glucose Study (TLGS). The TLGS is a long-term, ongoing community-based research to identify and prevent noncommunicable disorders being carried out in district No. 13, an area of about 13 km2, located in the eastern part of Tehran city, under coverage of Shahid Beheshti University of Medical Sciences and Health Services. In this area, three medical health centers with field data on more than $90\%$ of all covered families were chosen. Baseline measurements were recorded and Three-year follow-up studies were conducted on participants. An initial sample of 15005 participants aged ≥ 3 years was selected by a multistage stratified cluster sampling method for the TLGS [17]. Among 10368 subjects aged ≥20 years, 5786 participants who had thyroid function serum samples at baseline (February 1999– August 2001) and at all follow-up phases (up to March 2011) were chosen to include in the TTS.
In the current study, the inclusion criteria were as follows: [1] adults aged ≥ 20; and [2] individuals with normal thyroid function at baseline. On the other hand, participants were excluded if they had a TSH <0.32 mIU/L or a TSH >5.06 mIU/L in any phase of the study [18]. Patients with genetic disorders, addiction to alcohol and opium, and consumption of some effective drugs (important interfering factors that can impact other parameters). Those with levothyroxine, antithyroid drug, or corticosteroid usage, a history of thyroid surgery, thyroid radiation, or pregnant women were also excluded. Sixty-four individuals lacked data necessary to categorize obesity phenotypes. Finally, 3338 subjects were included. Ultimately, 1533 males and 1805 females were participated in our study (Figure 1). This study was approved by the ethics committee of Research Institute for Endocrine Sciences (RIES) of Shahid Beheshti University of Medical Sciences (code: IR.SBMU.ENDOCRINE.REC.1400.116). Written informed consent was obtained from all participants.
**Figure 1:** *The flowchart of recruitment.*
## Anthropometric measurements
The participants who invited to the TTS were referred to trained physicians after signing an informed consent form. Participants wore light clothing and no shoes during the anthropometric measurements. Weight and height were determined using a digital electronic weighing scale (Seca 707; range 0.1–150 kg; Seca, Hanover, MD) with an accuracy of up to 100 g and a tape meter stadiometer, respectively. In order to calculate body mass index (BMI), weight (kg) was divided by height (meters) squared. We measured waist circumference (WC) in centimeters at the level of the umbilicus.
## Measurements of metabolic indices
Blood samples were taken between 7:00 am and 9:00 am from all study participants, following an overnight fast of 12 to 14 hours. Fasting glucose levels were measured by glucose oxidase and enzymatic colorimetry. Serum total cholesterol (TC) and triglycerides (TGs) levels were determined using the enzymatic calorimetric method with cholesterol esterase, cholesterol oxidase, and glycerol phosphate oxidase, respectively. High-density lipoprotein cholesterol (HDL-C) was measured after the precipitation of apolipoprotein B-containing lipoproteins with phosphotungistic acid. All these biochemical tests were conducted on the day of sampling, using commercial kits (Pars Azmoon, Inc., Tehran, Iran) by the Selectra 2 auto-analyzer (Vital Scientific, Spankeren, The Netherlands). Analyses were performed on all the samples once quality control was achieved. Both inter- and intra-assay coefficients of variation (CVs) were <$2.3\%$ for glucose, <$2\%$ for TC, <$2.1\%$ for TG, and <$3\%$ for HDL-C.
fT4 and TSH concentrations were estimated in -70°C stored serum samples by the electrochemiluminescence immunoassay method using Roche Diagnostics kits and a Roche/Hitachi Cobas e-411 analyzer (Mannheim, Germany). Lyophilized quality control material (Lyphochek *Immunoassay plus* Control; Bio-Rad Laboratories, Hercules, CA) was used to monitor the accuracy of the assay. The intra- and inter-assay CVs were $1.3\%$ and $3.7\%$ for fT4 and $1.5\%$ and $4.5\%$ for TSH measurements, respectively. Thyroid peroxidase antibodies (TPOAb) were assayed by an immunoenzymometric assay kit (IEMA; Monobind, Costa Mesa, CA) and the Sunrise ELISA reader (Tecan Co., Salzburg, Austria); intra- and inter-assay CVs were $3.9\%$ and $4.7\%$, respectively. In the RIES research laboratory, all measurements were performed simultaneously.
Once the subjects had rested for 15 minutes, a qualified physician measured their systolic blood pressure (SBP) and diastolic blood pressure (DBP) twice in a seated position. The first measurement was used to determine the peak inflation level using a mercury sphygmomanometer. In this study, participant’s blood pressure was calculated as the average of two measurements.
## Definition of variables and outcomes
The reference ranges were 0.32–5.06 mIU/L for TSH, and 0.91-1.55 pmol/L for FT4. The reference range for serum TSH and FT4 levels was defined as euthyroidism. Hypothyroidism was defined as TSH > 5.06 mIU/L and FT4 < 0.91 pmol/L (overt hypothyroidism) or TSH > 5.06 mIU/L and FT4 levels within the reference range (subclinical hypothyroidism).
Using a BMI ≥ 25 kg/m2 as a threshold to define overweight/obesity seems to be a more reasonable approach [19]. Abnormal metabolic components were defined based on the Joint Interim Statement (JIS) criteria [18], (i) serum TG ≥150 mg/dL or taking lipid-lowering drugs; (ii) HDL-C <40 mg/dL in men and <50 mg/dL in women, or taking lipid-lowering drugs; (iii) systolic blood pressure (SBP) ≥130 mmHg or diastolic blood pressure (DBP) ≥85 mmHg, or taking antihypertensive drugs; and (iv) fasting blood glucose ≥100 mg/dL or undergoing treatment for diabetes. Participants with < 2 JIS components were considered metabolically healthy, whereas the metabolically unhealthy group included those who met two or more criteria. Since WC is highly correlated with BMI, it was excluded from the definition of metabolically unhealthy status [20].
Subsequently, participants were classified into four groups based on their BMI and metabolic status: [1] metabolically healthy normal weight (MHNW) defined as BMI<25kg/m2 and healthy metabolic status; [2] metabolically healthy overweight/obese (MHO) defined as BMI ≥ 25 kg/m2 and healthy metabolic status; [3] metabolically unhealthy normal weight (MUNW) defined as BMI < 25 kg/m2 and unhealthy metabolic status; [4] metabolically unhealthy overweight/obese (MUO) defined as BMI ≥ 25 kg/m2 and unhealthy metabolic status.
## Statistical analysis
The mean and standard deviation were used when the data had a normal distribution, and the median [25th and 75th percentiles] was used when the data had a skewed distribution. Categorical variables were presented as numbers (percentages). Differences in continuous variables were compared using one-way analysis of variance or Kruskal–Wallis one-way analysis of variance. Comparisons between groups were conducted using the chi-squared test or Fisher’s exact test for categorical variables. The relationship between metabolic phenotypes and hypothyroidism development was examined using a multiple logistic regression analysis. A two-tailed $P \leq 0.05$ was considered statistically significant. All statistical analyses were performed using Stata version 15.1 statistical software (StataCorp LLC, Texas, USA).
## Baseline characteristics
A total of 3338 subjects with a mean age of 39 ± 12.72 years were included in the present study. Males and females constituted 45.9 and 54.1 percent of the study population, respectively. The MUO ($$n = 1354$$) phenotype was the most prevalent at baseline. Table 1 summarizes the baseline characteristics of participants according to their metabolic phenotypes. At baseline, there were significant differences in sex, age, BMI, WC, TC, TG, LDL, HDL, SBP, DBP, FPG, FT4, creatinine (Cr), eGFR, smoking, and physical activity levels among the four groups ($P \leq 0.01$). However, no differences were observed in TSH and TPO-Ab levels at the start of the study ($P \leq 0.05$).
**Table 1**
| Variable | Variable.1 | Total (n= 3338) | MHNW (n= 883) | MUNW (n= 423) | MHO (n= 678) | MUO (n= 1354) | P-value |
| --- | --- | --- | --- | --- | --- | --- | --- |
| Gender, n (%) | Male | 1533 (45.9) | 399 (45.2) | 262 (61.9) | 220 (32.4) | 652 (48.2) | <0.001 |
| Gender, n (%) | Female | 1805 (54.1) | 484 (54.8) | 161 (38.1) | 458 (67.6) | 702 (51.8) | <0.001 |
| Age, year | Age, year | 39.00 (12.72) | 32.35 (11.66) | 41.18 (13.41) | 37.59 (10.96) | 43.35 (12.00) | <0.001 |
| BMI, kg/m2 | BMI, kg/m2 | 26.50 (4.53) | 21.82 (2.11) | 22.97 (1.73) | 28.65 (3.25) | 29.57 (3.42) | <0.001 |
| WC, cm | WC, cm | 87.30 (12.00) | 75.47 (7.25) | 81.22 (6.92) | 90.22 (9.47) | 95.44 (9.36) | <0.001 |
| TC, mg/dL | TC, mg/dL | 201.37 (42.53) | 178.91 (34.53) | 202.88 (42.26) | 199.06 (35.90) | 216.70 (43.76) | <0.001 |
| TG, mg/dL | TG, mg/dL | 142.0 (93.0, 205.0) | 86.0 (65.0, 111.0) | 179.0 (152.75, 232.0) | 106.0 (82.0, 133.0) | 199.0 (160.0, 259.0) | <0.001 |
| HDL, mg/dL | HDL, mg/dL | 41.69 (10.98) | 47.56 (10.54) | 36.40 (8.00) | 46.54 (11.34) | 37.08 (8.74) | <0.001 |
| LDL, mg/dL | LDL, mg/dL | 127.69 (35.83) | 112.92 (31.37) | 127.59 (36.26) | 130.03 (31.20) | 136.79 (37.61) | <0.001 |
| SBP, mmHg | SBP, mmHg | 116.21 (16.47) | 107.70 (11.08) | 117.06 (16.37) | 111.34 (12.34) | 123.98 (17.61) | <0.001 |
| DBP, mmHg | DBP, mmHg | 76.55 (16.47) | 70.79 (8.07) | 76.75 (10.41) | 74.33 (8.47) | 81.37 (10.24) | <0.001 |
| FPG, mg/dL | FPG, mg/dL | 95.22 (27.74) | 86.07 (8.64) | 99.43 (35.90) | 87.87 (11.76) | 103.55 (34.96) | <0.001 |
| TSH, mIU/L | TSH, mIU/L | 1.54 (1.01, 2.34) | 1.55 (1.01, 2.35) | 1.48 (0.96, 2.33) | 1.61 (1.10, 2.38) | 1.51 (0.99, 2.32) | 0.175 |
| FT4, ng/dL | FT4, ng/dL | 1.21 (0.14) | 1.24 (0.14) | 1.22 (0.14) | 1.19 (0.14) | 1.19 (0.15) | <0.001 |
| TPO-Ab, IU/mL | TPO-Ab, IU/mL | 5.27 (3.15, 10.31) | 5.40 (3.13, 10.45) | 5.07 (3.12, 9.55) | 5.21 (3.10, 10.61) | 5.27 (3.20, 10.30) | 0.814 |
| Cr, μmol/L | Cr, μmol/L | 1.03 (0.15) | 1.02 (0.15) | 1.06 (0.14) | 1.02 (0.14) | 1.04 (0.15) | <0.001 |
| eGFR, mL/min/1.73 | eGFR, mL/min/1.73 | 79.34 (12.23) | 84.39 (12.34) | 79.54 (12.70) | 78.07 (10.90) | 76.62 (11.63) | <0.001 |
| Smoking, n (%) | Yes | 429 (12.9) | 118 (13.4) | 70 (16.50) | 64 (9.5) | 177 (13.1) | 0.006 |
| Smoking, n (%) | No | 2902 (87.1) | 761 (86.6) | 353 (83.5) | 613 (90.5) | 1175 (86.9) | 0.006 |
| Physical activity, MET-min/week | Low | 2036 (61.4) | 546 (62.1) | 240 (57.0) | 432 (64.1) | 818 (60.9) | <0.001 |
| Physical activity, MET-min/week | Moderate | 895 (27.0) | 247 (28.1) | 113 (26.8) | 194 (28.8) | 341 (25.4) | <0.001 |
| Physical activity, MET-min/week | High | 386 (11.6) | 86 (9.8) | 68 (16.2) | 48 (7.1) | 184 (13.7) | <0.001 |
## Association between metabolic phenotypes and the development of hypothyroidism
The frequency (%) of hypothyroidism at each measurement time (baseline and 3rd, 6th, and 9th year) is provided in Figure 2 (total population) and Table 2 (according to male and female participants). In the total population, the chi-square test was only significant ($$P \leq 0.008$$) in 3rd year with a higher prevalence of hypothyroidism in the MUNW phenotype followed by MHO, MUO, and MHNW. In later years, the order of prevalence of hypothyroidism (from highest to lowest) was as follows: MHO, MUNW, MUO, and MHNW. However, this difference in the proportion of hypothyroidism was not statistically significant in the 6th ($$P \leq 0.138$$) and 9th ($$P \leq 0.120$$) years. Moreover, in the 3rd and 9th years of follow-up, hypothyroidism was more prevalent in MUO only in male subjects ($$P \leq 0.002$$ and 0.035, respectively).
**Figure 2:** *Prevalence of hypothyroidism according to different metabolic phenotypes in the total population.* TABLE_PLACEHOLDER:Table 2 Table 3 shows multiple logistic regression models of the association between hypothyroidism and metabolic profiles in the total population. In the unadjusted model, the MHO phenotype increased the odds of hypothyroidism compared to the MHNW phenotype (OR=1.51; $95\%$ CI=1.04, 2.18; P-value=0.031). Although a higher odds of hypothyroidism was observed for MUNW and MUO phenotypes compared to MHNW, the difference was not statistically significant ($$P \leq 0.179$$ and 0.563, respectively). In model 1, only MUNW was significantly associated with hypothyroidism compared to the MHNW phenotype after adjustment for the effect of age and sex (OR=1.70; $95\%$ CI=1.08, 2.67; P-value=0.022). The association between MHO and hypothyroidism was not significant in model 1 ($$P \leq 0.062$$). Similarly, in model 2, the odds of hypothyroidism was higher in the MUNW phenotype than in the MHNW phenotype after adjusting for the effects of age, sex, and TPO-Ab (OR=1.76; $95\%$ CI=1.12, 2.79; P-value=0.015). In model 3, the effects of waist circumference, Cr, smoking, and physical activity were adjusted in addition to the previous variables. In this model, the odds of hypothyroidism were higher in MUNW (OR=1.86; $95\%$ CI=1.17, 2.96; P-value=0.008), MHO (OR=1.71; $95\%$ CI=1.09, 2.67; P-value=0.018), and MUO (OR=1.64; $95\%$ CI=1.03, 2.62; P-value=0.036) phenotypes than in MHNW.
**Table 3**
| Unnamed: 0 | Unnamed: 1 | Unadjusted model | Unadjusted model.1 | Model 1 | Model 1.1 | Model 2 | Model 2.1 | Model 3 | Model 3.1 |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| | | OR (95% CI) | P | OR (95% CI) | P | OR (95% CI) | P | OR (95% CI) | P |
| Total | MHNW | Ref. | | Ref. | | Ref. | | Ref. | |
| Total | MUNW | 1.35 (0.87, 2.08) | 0.179 | 1.70 (1.08, 2.67) | 0.022 | 1.76 (1.12, 2.79) | 0.015 | 1.86 (1.17, 2.96) | 0.008 |
| Total | MHO | 1.51 (1.04, 2.18) | 0.031 | 1.44 (0.98, 2.10) | 0.062 | 1.38 (0.94, 2.04) | 0.102 | 1.71 (1.09, 2.67) | 0.018 |
| Total | MUO | 1.10 (0.79, 1.55) | 0.563 | 1.22 (0.84, 1.76) | 0.302 | 1.24 (0.85, 1.81) | 0.258 | 1.64 (1.03, 2.62) | 0.036 |
Table 4 shows the results of multiple logistic regression analysis of the association between metabolic phenotypes and hypothyroidism according to sex. In the unadjusted model, the MUNW phenotype increased the risk of hypothyroidism compared to MHNW only in males (OR=2.53; $95\%$ CI=1.21, 5.31; P-value=0.014). This association remained significant after adjusting for the effect of age in model 1 ($$P \leq 0.019$$), age and TPO-Ab in model 2 ($$P \leq 0.028$$), and age, TPO-Ab, waist circumference, Cr, smoking, and physical activity in model 3 ($$P \leq 0.034$$). Moreover, in females, the MHO phenotype increased the risk of hypothyroidism compared to MHNW in models 1 (OR=1.56; $95\%$ CI=1.02, 2.39; P-value=0.039) and 3 (OR=1.87; $95\%$ CI=1.15, 3.05; P-value=0.012).
**Table 4**
| Unnamed: 0 | Unnamed: 1 | Unadjusted model | Unadjusted model.1 | Model 1 | Model 1.1 | Model 2 | Model 2.1 | Model 3 | Model 3.1 |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| | | OR (95% CI) | P | OR (95% CI) | P | OR (95% CI) | P | OR (95% CI) | P |
| Male | MHNW | Ref. | | Ref. | | Ref. | | Ref. | |
| Male | MUNW | 2.53 (1.21, 5.31) | 0.014 | 2.45 (1.16, 5.19) | 0.019 | 2.38 (1.10, 5.14) | 0.028 | 2.35 (1.07, 5.16) | 0.034 |
| Male | MHO | 0.90 (0.33, 2.44) | 0.843 | 0.88 (0.32, 2.38) | 0.802 | 0.95 (0.35, 2.62) | 0.929 | 0.87 (0.27, 2.77) | 0.810 |
| Male | MUO | 1.66 (0.85, 3.27) | 0.139 | 1.51 (0.75, 3.03) | 0.244 | 1.52 (0.75, 3.11) | 0.246 | 1.36 (0.52, 3.59) | 0.534 |
| Female | MHNW | Ref. | | Ref. | | Ref. | | Ref. | |
| Female | MUNW | 1.14 (0.63, 2.06) | 0.655 | 1.35 (0.73, 2.48) | 0.333 | 1.46 (0.79, 2.69) | 0.227 | 1.51 (0.81, 2.79) | 0.191 |
| Female | MHO | 1.40 (0.93, 2.12) | 0.106 | 1.56 (1.02, 2.39) | 0.039 | 1.50 (0.98, 2.32) | 0.063 | 1.87 (1.15, 3.05) | 0.012 |
| Female | MUO | 0.98 (0.65, 1.46) | 0.914 | 1.18 (0.76, 1.86) | 0.460 | 1.21 (0.77, 1.91) | 0.401 | 1.65 (0.96, 2.83) | 0.070 |
## Discussion
This study examined the relationship between obesity phenotypes and the incidence of hypothyroidism, focusing on differences between males and females. The present cohort study demonstrated a higher prevalence of hypothyroidism in the MUNW phenotype followed by MHO, MUO, and MHNW in 3rd year of follow-up, and in the 3rd and 9th years of follow-up, hypothyroidism was more prevalent among MUO only in males. Males with MUO, MHO, and MUNW phenotypes had a higher risk of hypothyroidism than those with MHNW phenotypes. In females, the MHO phenotype increased the risk of hypothyroidism compared with MHNW.
In a longitudinal study conducted in Shandong, China [21], the authors reported that the non-MHNO group had a significantly higher incidence of hypothyroidism than the MHNO group in males, and the MHO, MUNO, and MUO phenotypes were independent risk factors for the development of hypothyroidism compared with the MHNO phenotype in males but not in females. Previously, in Tehran, Iran [15], researchers explored the relationship between thyroid function and obesity phenotype development, which was different from our goal.
It is currently unclear whether obesity is associated with thyroid autoimmunity [8]. A previous study on subjects without thyroid autoimmunity at baseline found that the abdominal obesity phenotype had no significant impact on the development of TPO-Ab positivity over time [22]. The results of a prospective cohort study indicated a higher likelihood of developing subclinical hypothyroidism in participants with metabolic syndrome (MetS) at baseline [10]. Few studies have examined the association between thyroid autoimmunity and MetS. A cross-sectional study reported that thyroid autoimmunity was associated with high glycated hemoglobin levels, central obesity, dyslipidemia, and MetS among euthyroid individuals [23]. In contrast, a study on postmenopausal euthyroid women found no association between TPOAb positivity and MetS prevalence [24]. In another cross-sectional study [25], using KNHANES VI data of 4775 euthyroid subjects, the researchers found that thyroid autoimmunity is associated with poor physiological health outcomes, such as abdominal obesity, low HDL cholesterol, and hypertension. There could be a number of factors contributing to such inconsistent results, such as ethnicity, diet, lifestyle, age, and sex composition, among others. Therefore, anthropometric state and metabolic disorders are likely linked to hypothyroidism.
Despite the lack of a complete understanding of the mechanisms underlying obesity phenotypes and hypothyroidism, some explanations have been suggested. Chronic low-grade inflammation has been observed in obese individuals. An increase in inflammatory cytokines, including IL-1, IL-6, and TNF-α, inhibits sodium-iodide symporter (NIS) expression, influences iodide uptake activity, and affects thyroid morphology [26, 27]. Leptin may also suppress TSH-induced thyroid function in obese individuals [28]. The deiodinase enzyme may also be modulated by chronic inflammation and may affect thyroid function [29, 30]. It is also possible that lipotoxicity affects the thyroid [31, 32]. According to a recent study, palmitic acid reduces thyroid hormone synthesis by downregulating the expression and activity of NIS, thyroglobulin, and thyroperoxidase [31]. Hypothyroidism caused by high-fat diet may be caused by endoplasmic reticulum stress [32]. There are differences between men and women in the association between obesity/metabolic disorders and thyroid diseases [33, 34]. According to a cohort study, obesity and metabolic conditions may influence thyroid cancer development differently depending on sex [34].
Our study found that males with MHO, MUNO, and MUO phenotypes were independently at risk for hypothyroidism, while females were not. We found a sex difference in the association between obesity phenotypes and hypothyroidism, although the mechanisms underlying this association are unclear. First, testosterone and estradiol affect thyroid function in different ways [35, 36]. Second, visceral adipose tissue accumulates more commonly in men than in women, indicating that obesity poses a greater threat to men [37]. The risk of hypothyroidism may vary depending on the obesity phenotype, which may result in sex-specific alterations in sex hormones. We found that normal weight and a healthy metabolic state reduced the risk of hypothyroidism in both men and women. The validation of our findings and elucidation of the underlying mechanisms require further research.
This study has several limitations that should be considered when interpreting the findings. Owing to its observational nature, this cohort study could not infer causal relationships. A bidirectional and complex relationship exists between obesity and thyroid function [27]. Thyroid organs may be susceptible to lipotoxicity [31, 32], and thyroid hormones play a role in the metabolic control of glucose and lipids [38]. Although we conducted a cohort study with baseline euthyroid participants and adjusted for confounding factors, we cannot exclude reverse causation and unmeasured confounders. Second, our study used BMI as a measure to diagnose obesity. The association between obesity phenotypes and hypothyroidism may be better understood if further studies are performed using body composition and WC data. Thyroid function measurements were performed at each visit to diagnose hypothyroidism. Finally, this cohort included subjects from Tehran, Iran, who underwent regular health examinations. Further research is needed to determine whether our findings apply to other populations with different characteristics.
It is important to note, however, that our study has several strengths despite its limitations. First, this is the first cohort study to examine the sex-specific relationship between obesity phenotypes and hypothyroidism in Iran. Second, our study found that not only the MUO phenotype was an independent risk factor for hypothyroidism in males, but also the MHO and MUNO phenotypes, providing insight into hypothyroidism risk factors in men. As hypothyroidism is more common in females, less attention has been paid to males in the past. Clinical practice should focus on males with unhealthy metabolic phenotypes because they are more likely to develop hypothyroidism.
## Conclusion
Our findings in the TTS cohort showed that obesity and metabolic abnormalities were related to an elevated risk of hypothyroidism, particularly in males. We found that obesity phenotypes and hypothyroidism were not associated in females, in contrast to the findings in males. The results of this study highlight sex differences in the association between metabolic phenotypes and the risk of hypothyroidism in a baseline euthyroid population. Males with unhealthy metabolic phenotypes should be given special attention. Further research is needed to verify and identify the possible mechanisms explaining the sex differences in this association.
## Data availability statement
The original contributions presented in the study are included in the article/supplementary material. Further inquiries can be directed to the corresponding author.
## Ethics statement
The studies involving human participants were reviewed and approved by Research Institute for Endocrine Sciences (RIES) of Shahid Beheshti University of Medical Sciences (code: IR.SBMU.ENDOCRINE.REC.1400.116). The patients/participants provided their written informed consent to participate in this study.
## Author contributions
BA, and MV contributed to the design of study, conducted the searches, drafted and edited the manuscript. ARA and MM contributed to the design of the study, analyzing the data, and revised the manuscript. FH, AA and MV advised and revised the manuscript. All authors have read and approved the final version of the manuscript. MV has primary responsibility for final content. All authors contributed to the article and approved the submitted version.
## Conflict of interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher’s note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## References
1. Taylor PN, Albrecht D, Scholz A, Gutierrez-Buey G, Lazarus JH, Dayan CM. **Global epidemiology of hyperthyroidism and hypothyroidism**. *Nat Rev Endocrinol* (2018) **14**. DOI: 10.1038/nrendo.2018.18
2. Shahrani Al S, El-Metwally A, Al-Surimi K, Salih SB, Saleh Y, Al-Shehri A. **The epidemiology of thyroid diseases in the Arab world: A systematic review**. *J Public Health Epidemiol* (2016) **8**. DOI: 10.5897/JPHE2015.0713
3. Amouzegar A, Ghaemmaghami Z, Beigy M, Gharibzadeh S, Mehran L, Tohidi M. **Natural Course of Euthyroidism and Clues for Early Diagnosis of Thyroid Dysfunction: Tehran Thyroid Study**. *Thyroid* (2017) **27**. DOI: 10.1089/thy.2016.0409
4. Rodondi N, den Elzen WPJ, Bauer DC, Cappola AR, Razvi S, Walsh JP. **Subclinical hypothyroidism and the risk of coronary heart disease and mortality**. *JAMA* (2010) **304**. DOI: 10.1001/jama.2010.1361
5. Moon S, Kim MJ, Yu JM, Yoo HJ, Park YJ. **Subclinical hypothyroidism and the risk of cardiovascular disease and all cause mortality: A meta-analysis of prospective cohort studies**. *Thyroid* (2018) **28**. DOI: 10.1089/thy.2017.0414
6. Licenziati MR, Valerio G, Vetrani I, De Maria G, Liotta F, Radetti G. **Altered thyroid function and structure in children and adolescents who are overweight and obese: Reversal after weight loss**. *J Clin Endocrinol Metab* (2019) **104**. DOI: 10.1210/jc.2018-02399
7. Lee MH, Lee JU, Joung KH, Kim YK, Ryu MJ, Lee SE. **Thyroid dysfunction associated with follicular cell steatosis in obese male mice and humans**. *Endocrinology* (2015) **156**. DOI: 10.1210/en.2014-1670
8. Santini F, Marzullo P, Rotondi M, Ceccarini G, Pagano L, Ippolito S. **Mechanisms in endocrinology: the crosstalk between thyroid gland and adipose tissue: Signal integration in health and disease**. *Eur J Endocrinol* (2014) **171**. DOI: 10.1530/EJE-14-0067
9. Song RH, Wang B, Yao QM, Li Q, Jia X, Zhang JA. **The impact of obesity on thyroid autoimmunity and dysfunction: A systematic review and meta-analysis**. *Front Immunol* (2019) **10**. DOI: 10.3389/fimmu.2019.02349
10. Chang CH, Yeh YC, Caffrey JL, Shih SR, Chuang LM, Tu YK. **Metabolic syndrome is associated with an increased incidence of subclinical hypothyroidism–a cohort study**. *Sci Rep* (2017) **7** 6754. DOI: 10.1038/s41598-017-07004-2
11. Stefan N, Häring H-U, Hu FB, Schulze MB. **Metabolically healthy obesity: epidemiology, mechanisms, and clinical implications**. *Lancet Diabetes Endocrinol* (2013) **1**. DOI: 10.1016/S2213-8587(13)70062-7
12. van Vliet-Ostaptchouk JV, Nuotio ML, Slagter SN, Doiron D, Fischer K, Foco L. **The prevalence of metabolic syndrome and metabolically healthy obesity in Europe: A collaborative analysis of ten large cohort studies**. *BMC Endocr Disord* (2014) **14** 9. DOI: 10.1186/1472-6823-14-9
13. St-Onge MP, Janssen I, Heymsfield SB. **Metabolic syndrome in normal-weight americans: New definition of the metabolically obese, normal-weight individual**. *Diabetes Care* (2004) **27**. DOI: 10.2337/diacare.27.9.2222
14. Arnlov J, Ingelsson E, Sundstrom J, Lind L. **Impact of body mass index and the metabolic syndrome on the risk of cardiovascular disease and death in middle-aged men**. *Circulation* (2010) **121**. DOI: 10.1161/CIRCULATIONAHA.109.887521
15. Amouzegar A, Kazemian E, Abdi H, Mansournia MA, Bakhtiyari M, Hosseini MS. **Association between thyroid function and development of different obesity phenotypes in euthyroid adults: A nine-year follow-up**. *Thyroid* (2018) **28**. DOI: 10.1089/thy.2017.0454
16. Azizi F, Amouzegar A, Delshad H, Tohidi M, Mehran L, Mehrabi Y. **Natural course of thyroid disease profile in a population in nutrition transition: Tehran Thyroid Study**. *Arch Iran Med* (2013) **16**
17. Azizi F, Ghanbarian A, Momenan AA, Hadaegh F, Mirmiran P, Hedayati M. **Prevention of non-communicable disease in a population in nutrition transition: Tehran lipid and glucose study phase II**. *Trials* (2009) **10** 5. DOI: 10.1186/1745-6215-10-5
18. Amouzegar A, Delshad H, Mehran L, Tohidi M, Khafaji F, Azizi F. **Reference limit of thyrotropin (TSH) and free thyroxine (fT4) in thyroperoxidase positive and negative subjects: A population based study**. *J Endocrinol Invest* (2013) **36**
19. Hsu WC, Araneta MR, Kanaya AM, Chiang JL, Fujimoto W. **BMI cut points to identify at-risk Asian americans for type 2 diabetes screening**. *Diabetes Care* (2015) **38**. DOI: 10.2337/dc14-2391
20. Nie X, Ma X, Xu Y, Shen Y, Wang Y, Bao Y. **Characteristics of serum thyroid hormones in different metabolic phenotypes of obesity**. *Front Endocrinol* (2020) **11**. DOI: 10.3389/fendo.2020.00068
21. Wang Y, Lin H, Li Q, Guan L, Zhao M, Zhong F. **Association between different obesity phenotypes and hypothyroidism: A study based on a longitudinal health management cohort**. *Endocrine.* (2021) **72**. DOI: 10.1007/s12020-021-02677-2
22. Amouzegar A, Kazemian E, Abdi H, Gharibzadeh S, Tohidi M, Azizi F. **Abdominal obesity phenotypes and incidence of thyroid autoimmunity: A 9-year follow-up**. *Endocr Res* (2020) **45** 202-2209. DOI: 10.1080/07435800.2020.1749847
23. Chen Y, Zhu C, Chen Y, Wang N, Li Q, Han B. **Are thyroid autoimmune diseases associated with cardiometabolic risks in a population with normal thyroid-stimulating hormone**. *Mediators Inflammation* (2018) **2018** 1856137. DOI: 10.1155/2018/1856137
24. Siemińska L, Wojciechowska C, Walczak K, Borowski A, Marek B, Nowak M. **Associations between metabolic syndrome, serum thyrotropin, and thyroid antibodies status in postmenopausal women, and the role of interleukin-6**. *Endokrynolog Polska* (2015) **66** 394-403
25. Kim HJ, Park SJ, Park HK, Byun DW, Suh K, Yoo MH. **Thyroid autoimmunity and metabolic syndrome: A nationwide population-based study**. *Eur J Endocrinol* (2021) **185**. DOI: 10.1530/EJE-21-0634
26. Longhi S, Radetti G. **Thyroid function and obesity**. *J Clin Res Pediatr Endocrinol* (2013) **5**
27. Fontenelle LC, Feitosa MM, Severo JS, Freitas TE, Morais JB, Torres-Leal FL. **Thyroid function in human obesity: Underlying mechanisms**. *Horm Metab Res* (2016) **48**. DOI: 10.1055/s-0042-121421
28. Isozaki O, Tsushima T, Nozoe Y, Miyakawa M, Takano K. **Leptin regulation of the thyroids: Negative regulation on thyroid hormone levels in euthyroid subjects and inhibitory effects on iodide uptake and Na+/I– symporter mRNA expression in rat FRTL-5 cells**. *Endocr J* (2004) **51**. DOI: 10.1507/endocrj.51.415
29. Boelen A, Fliers E, Stap J, de Vries EM, Surovtseva OV, Kwakkel J. **A novel role for the thyroid hormone-activating enzyme type 2 deiodinase in the inflammatory response of macrophages**. *Endocrinology* (2014) **155**
30. Jakobs TC, Mentrup B, Schmutzler C, Dreher I, Köhrle J. **Proinflammatory cytokines inhibit the expression and function of human type I 5’-deiodinase in HepG2 hepatocarcinoma cells**. *Eur J Endocrinol* (2002) **146**. DOI: 10.1530/eje.0.1460559
31. Zhao M, Zhang X, Gao L, Song Y, Xu C, Yu C. **Palmitic acid downregulates thyroglobulin (Tg), sodium iodide symporter (NIS), and thyroperoxidase (TPO) in human primary thyrocytes: A potential mechanism by which lipotoxicity affects thyroid**. *Int J Endocrinol* (2018) **2018** 4215848. DOI: 10.1155/2018/4215848
32. Zhang X, Shao S, Zhao L, Yang R, Zhao M, Fang L. **ER stress contributes to high-fat diet-induced decrease of thyroglobulin and hypothyroidism**. *Am J Physiol Endocrinol Metab* (2019) **316**. DOI: 10.1152/ajpendo.00194.2018
33. Oh HS, Kwon H, Ahn J, Song E, Park S, Kim M. **Association between thyroid dysfunction and lipid profiles differs according to age and sex: results from the Korean national health and nutrition examination survey**. *Thyroid* (2018) **28**. DOI: 10.1089/thy.2017.0656
34. Kwon H, Chang Y, Cho A, Ahn J, Park SE, Park CY. **Metabolic obesity phenotypes and thyroid cancer risk: A cohort study**. *Thyroid* (2019) **29**. DOI: 10.1089/thy.2018.0327
35. Fortunato RS, Ferreira AC, Hecht F, Dupuy C, Carvalho DP. **Sexual dimorphism and thyroid dysfunction: A matter of oxidative stress**. *J Endocrinol* (2014) **221**. DOI: 10.1530/JOE-13-0588
36. Sosic-Jurjevic B, Filipovic B, Renko K, Miler M, Trifunovic S, Ajdzanovic V. **Testosterone and estradiol treatments differently affect pituitary-thyroid axis and liver deiodinase 1 activity in orchidectomized middle-aged rats**. *Exp Gerontol* (2015) **72** 85-98. DOI: 10.1016/j.exger.2015.09.010
37. Tchernof A, Després J-P. **Pathophysiology of human visceral obesity: an update**. *Physiol Rev* (2013) **93** 359-404. DOI: 10.1152/physrev.00033.2011
38. Mullur R, Liu Y-Y, Brent GA. **Thyroid hormone regulation of metabolism**. *Physiol Rev* (2014) **94**. DOI: 10.1152/physrev.00030.2013
|
---
title: Combined analysis of plasma metabolome and intestinal microbiome sequencing
to explore jiashen prescription and its potential role in changing intestine–heart
axis and effect on chronic heart failure
authors:
- Xialian Cui
- Yangyan Su
- Xiaotong Huang
- Jiaping Chen
- Jiang Ma
- Peiran Liao
- Xin He
journal: Frontiers in Cardiovascular Medicine
year: 2023
pmcid: PMC10036802
doi: 10.3389/fcvm.2023.1147438
license: CC BY 4.0
---
# Combined analysis of plasma metabolome and intestinal microbiome sequencing to explore jiashen prescription and its potential role in changing intestine–heart axis and effect on chronic heart failure
## Abstract
### Background
Heart failure (HF) is a syndrome with global clinical and socioeconomic burden worldwide owing to its poor prognosis. Jiashen Prescription (JSP), a traditional Chinese medicine (TCM) formula, exhibits unambiguous effects on treating HF. Previously, we have reported that underlying mechanisms of JSP by an untargeted metabolomics approach, but the contribution of gut microbiota and metabolic interaction to the cardioprotective efficacy of JSP remains to be elucidated.
### Materials and methods
Firstly, the rat model of heart failure was established by the permanent ligation of the left anterior descending coronary artery. The efficacy evaluation of JSP in treating HF rats was per-formed by left ventricular ejection fraction (LVEF). Then, 16S rRNA gene sequencing and LC/MS-based metabolomic analysis were utilized to explore the characteristics of cecal-contents microecology and plasma metabolic profile, respectively. After that, the correlation between intestinal micro-ecological characteristics and plasma metabolic characteristics was analyzed to explore the potential mechanism of the JSP treatment in HF.
### Results
JSP could improve the cardiac function of heart failure rats and thus ameliorate heart failure via enhancing rat LVEF. Results of intestinal flora analysis revealed that JSP not only adjusted gut microbiota disturbances by enriching species diversity, reducing the abundance of pathogenic bacteria (such as Allobaculum, Brevinema), as well as increasing the abundance of beneficial bacteria (such as Lactobacillus, Lachnospiraceae_NK4A136_group), but also improved metabolic disorders by reversing metabolite plasma levels to normality. Through the conjoint analysis of 8 metabolites and the OTUs relative abundance data in the 16srRNA sequencing results by WGCNA method, 215 floras significantly related to the eight compounds were identified. The results of the correlation analysis demonstrated a significant association between intestinal microbiota and plasma metabolic profile, especially the significant correlation of Ruminococcaceae_UCG-014 and Protoporphyrin IX, Ruminococcaceae_UCG-005, Christensenellaceae_R-7_group and nicotinamide, dihydrofolic acid.
### Conclusion
The present study illustrated the underlying mechanism of JSP to treat heart failure by affecting intestinal flora and plasma metabolites, provide a potential therapeutic strategy against heart failure.
## Introduction
Heart failure (HF) is a chronic progressive disease [1], which is a complex group of clinical signs of impaired ventricular filling or ejection capacity due to any structural or functional abnormality of the heart. It is the end-stage manifestation of various cardiovascular diseases (CVDs), which bring high prevalence and mortality and threaten human health [2]. Thus, discovering novel mechanisms of HF and identifying potential therapeutic targets are extremely important ways of preventing heart failure. Recently, several studies indicated that the intestine microbiota can influence the cell and organ functions of the host and multiple mechanisms and pathways of diseases [3]. Among them, several studies have shown that intestinal microbes may affect the cardiovascular system, and the concept of the “intestine–heart axis” has been gradually applied to explore the treatment of CVD and heart failure [4, 5].
The intestinal microbiome transmits information to the distant organs of the host through various biochemical signals and metabolites. In actuality, the intestinal microbiome can affect the cardiovascular system. First, some studies found that the flora of intestine microbiota in patients with heart failure decreased significantly [6]. This may lead to an imbalance between beneficial and harmful microorganisms. The former can produce many beneficial metabolites, such as short-chain fatty acids, while the latter may bring a large number of harmful metabolites, such as primary bile acid and trimethylamine oxide (TMAO) [7]. These metabolites can directly or indirectly affect the heart after entering the circulatory system. Then, heart failure can lead to structural and functional abnormalities of the heart, which result in decreased cardiac output and tissue perfusion. In turn, it allows bacteria and lipopolysaccharide to translocate into systemic circulation, which induces the inflammation and accelerates the development of heart failure [8]. Therefore, through the study of intestinal microbiota, it is helpful to study the therapeutic mechanism of drugs for heart failure or cardiovascular disease from the perspective of the intestine–heart axis. It was proposed that treatment with *Lactobacillus rhamnosus* GR-1 as a probiotic could delay the development of heart failure after coronary artery occlusion in rats [9]. Methanogens can treat CVD by reducing plasma levels of TMAO [10]. Phenaceglutamide, a metabolite of intestinal microbiome negatively correlated with pulse wave velocity and systolic blood pressure, could also be a potential therapeutic target [11].
Jiashen prescription (JSP) is a clinical prescription used for treating heart failure, which is established under the guidance of the theoretical thought of traditional Chinese medicine (TCM). It is mainly prepared from several Chinese medicinal materials, including Astragali Radix, Salviae Miltiorrhizae Radix et Rhizoma, Periplocae Cortex, Notoginseng Radix et Rhizoma, Leonuri Herba, Citri Reticulatae Pericarpium, Cinnamomi Ramulus, and Descurainiae Semen Lepidii Semen [12]. Previous chemical studies showed that there were at least 68 chemical compounds identified from JSP, mainly including phenolic acids, tanshinones, flavonoids and their glycosides, cardiac glycosides, triterpene saponins, and C21 steroids [13]. From the perspective of modern pharmacological effects, this prescription has the effects of improving heart hemodynamics, enhancing heart function, inhibiting activation of renin–angiotensin, diuresis, and inhibiting ventricular remodeling in animals with heart failure [14]. However, TCM has the characteristics of multi-components and multi-targets. When taking Chinese medicine orally, the most of active ingredients in TCM cannot directly enter the blood system to exert their effects but may be directly or indirectly metabolized by the intestinal microbiome and then enter the blood system to achieve the therapeutic effect on diseases [15].
Thus, intestinal microflora has become a new and important frontier in the understanding of TCM [16]. At present, studies have shown the influence of intestinal microbiota on the metabolism of bioactive ingredients of TCM [17]. Not only polysaccharides [18] but also some macromolecular saponins, terpenes, and alkaloids have been effectively transformed by intestinal microbiome. Then, bioavailability and therapeutic activity are increased as well [19]. In the treatment of chronic syndrome and glycolipid metabolic diseases, which still are global health problems, TCM has its unique advantages [20, 21]. One of the main therapeutic mechanisms is to increase the relative abundance of beneficial bacteria by improving the intestinal environment [22].
Evidently, in this study, the rat model of heart failure was used to investigate the treatment effect of JSP, which was established by the ligation of the left anterior descending coronary artery (LAD) [23]. In the previous study of our research group, Miao et al. found that JSP improved the cardiac function of heart failure rats and thus ameliorated heart failure via enhancing rat LVEF and LVFS and decreasing LVIDd, LVIDs, IVSd, and IVSs. Based on the biochemical analysis and histopathological examination, it was found that JSP could reduce the markers levels of heart failure and myocardial damage that included serum lactate dehydrogenase (LDH) activity and the level of NT-pro BNP and inhibit myocardial fibrosis [24]. In this study, we collected and sequenced the intestinal contents of rats in each treatment group in the previous study. For this purpose, the blood metabolites of rats in different treatment groups were analyzed and screened by the fuzzy C-means clustering method. Then weighted correlation network analysis was used to jointly analyze these metabolites and 16S rRNA high-throughput sequencing data of intestinal microbiome in each treatment group, to find metabolites that may be related to the JSP treatment effect and their significantly related intestinal microorganisms, to clarify the mechanism of JSP treatment through the “intestine–heart axis”.
## Experimental reagents
Jiashen prescription was produced by Tasly Pharmaceutical Group Co., Ltd., Tianjin, China. Captopril (CPT) was purchased from Shanghai Pharmaceutical Group Corp., (Shanghai, China). DNA extraction kits were purchased from Macherey-Nagel (Düren, German), and HPLC-grade methanol and acetonitrile were obtained from Merck (Darmstadt, Germany). Formic acid and 2-chlorophenylalanine were bought from Thermo Fisher Scientific (MA, USA).
## Animals and experimental protocols
All experiments were in accordance with the guidelines for laboratory animal care and use, and the procedures were approved by the Research Ethical Committee of Guangdong Pharmaceutical University (Guangzhou, China). A total of 46 SPF (SD) rats (license NO. SCXK (yue) 2018–0002), weighing from 180 to 220 g, were purchased from the Guangdong Medical Laboratory Animal Center (Guangzhou, China). The animals were housed at 20–25°C, 40–$70\%$ humidity and 12 h dark/light cycle conditions with free access to a standard chow diet, and tap water ad libitum. All rats were exposed to an ‘adaptive feeding' paradigm for a week before the start of experiments.
The experiment included four groups ($$n = 6$$ per group), namely Control group, Model group, JSP group, and CPT (Captopril) group. In the control group ($$n = 6$$), the rats were subjected to the same surgeries except for the ligation. Heart failure symptoms were induced by carrying out ligation of the left anterior descending coronary artery of the rats as previously reported [25], and the chronic congestive heart failure model was established after 4 weeks of operation and normal feeding. The left ventricular ejection fraction (LVEF) of HF rats was < $60\%$, and it has been considered that the chronic HF rat models were successfully established [26]. JSP and CPT were orally administered to the rats at 3 and 0.05 g/kg/day, which was selected based on their human equivalent dose used in clinical practice. In the previous research in our laboratory, this dose of JSP had been used to prove the efficacy of JSP [24].
In total, 40 rats underwent LAD ligation to establish a heart failure model. Among the 40 rats, eight rats died during the operation and the other eight rats did not meet the criteria of chronic heart failure after the operation; the success rate of the heart failure model rats was $60\%$. Therefore, 18 rats ($$n = 18$$) with similar body weight were randomly divided into three groups as follows: [1] HF group, the rats received LAD ligation to induce HF, [2] JSP group, a dose of 3 g/kg/day JSP was given the corresponding drugs by gavage for 4 weeks, and [3] CPT group, a dose of 0.05 g/kg/day captopril was gavage to rats for 4 weeks, and the normal and model groups were fed the same volume of saline via intragastric administration. Echocardiographic studies were performed during the adaption week, before treatment, and after treatment. LVEF was calculated to assess left ventricular systolic function and cardiac function.
## Sample collection
At the end of the experiment, all rats were anesthetized with sodium pentobarbital. Blood was taken from the main abdominal vein into a plasma separator tube, and the samples were centrifuged at 4,000 rpm for 30 min at 4°C. The samples were stored at −80°C until use. The entire intestines were dissected with a sterile scalpel, and the contents of the intestines were collected with sterile lyophilized tubes (100–200 mg/tube). Fresh stool samples from each group were immediately frozen in a liquid nitrogen tank and then stored at −80°C.
## DNA extraction and 16S rRNA sequencing
According to the instructions of the MN NucleoSpin 96 Soi DNA extraction kit, the total genomic DNA was extracted from the samples. 16S rRNA sequencing was performed at the Beijing BMC Biotech Co., Ltd., (Beijing, China). The V3–V4 region of 16S rRNA genes was analyzed. The specific primers 338F (5'-ACTCCTACGGGAGGCAGCAG-3') and 806 R (5'-GGACTACHVGGGTWTCTAAT-3') with the barcodes were applied to amplify the 16S rRNA genes. The PCR mixture contains 5 μl of the purified product of PCR of the target region, MPPI-a of 2.5 μl, MPPI-b of 2.5 μl, and 2 × Q5 HF MM of 10 μl. The PCR products were monitored with $1.8\%$ agarose gel, purified using the OMEGA DNA purification column, and then sequenced on an Illumina Miseq PE 3,000 platform.
## Non-targeted metabolomic analysis
Blood samples were centrifuged at 4,000 rpm for 30 min at 4°C to obtain plasma samples. An aliquot of the plasma sample (100 μl) was mixed with 300 μl of methanol containing 1 ppm of 2-chlorophenylalanine, vortexed for 2 min, and incubated at −20°C for 30 min. The mixture was centrifu, ged at 12,000 rpm for 10 min at 4°C, and the supernatant was obtained for further analysis.
Sample extracts were analyzed using an LC–QTOF–MS/MS. The analytical conditions for the system were as follows: an Aglient 1,290 ultra-high performance liquid chromatography (UPLC) (Agilent Technologies, Inc., USA) connected to an Aglient 6,545 quadrupole time-of-flight (QTOF) mass spectrometer. The chromatographic separation was achieved on an ACQUITY UPLC HSS T3 C18 column (1.8 μm, 2.1 × 100 mm) at 40°C using a mobile phase of $0.1\%$ formic acid (A) and acetonitrile (B). Elution gradients used were as follows: 0–11 min, $95\%$ A; 11–12 min, $10\%$ A; 12–12.1 min, $10\%$ A; and 12.2–14 min, $95\%$ A. The flow rate was 0.40 ml/min and the injection volume was 2 μl. The mass spectrometry was performed in positive and negative ion modes. The parameters of the heated electrospray ionization method were as follows: sheath flow rate of 11 L/min, gas flow rate of 8 L/min, spray voltage of 250 V, positive and negative ionization, fragmentation voltage of 135 V, gas temperature of 325°C, sheath temperature of 325°C, and nebulizer pressure of 40 psi.
## Bioinformatic analysis and statistical analysis
The raw data were quality filtered using Trimmomatic (Version 0.33) [27], the primer sequences were identified and removed using Cutadapt (Version 1.9.1), followed by double-ended reads splicing and chimeric UCHIME (Version 8.1) removal using USEARCH (Version 10.0) [28], resulting in high-quality sequences for subsequent analysis. The sequences were clustered at a $97\%$ similarity level (default) using USEARCH. Operational taxonomic units (OTUs) were filtered by default with a threshold of $0.005\%$ of the number of all sequences sequenced. Raw data were uploaded to NCBI (ID: SUB12078416). The OTU sequences were obtained by splicing the filtered sequencing data with Qiime 2 software, and the abundance of each OTU was calculated and normalized for alpha analysis. Beta analysis was performed using the “FactoMineR” package in R software (Version 4.2.0), and a linear discriminant analysis was performed using the “lefseR” package (alpha analysis was performed using GraphPadPrism9 for statistical analysis and graphing, and linear discriminant analysis (linear discriminant analysis effect size, LEfSe) was performed using Mothur software and LEfSe software for detecting species differences between groups). Species with linear discriminant analysis (LDA) values >4 were considered to have statistically significant differences between groups.
Statistical analysis was performed in R. First, partial least squares discriminant (PLS-DA) analysis was performed using the “mixOmics” package and the “ropls” package, and model substitution tests were performed to screen out metabolites with intra-group differences from the metabolite summary table. The metabolites with RSD of ≤ $10\%$ were subjected to Mfuzz clustering analysis [29]. The subsequent analysis was mainly carried out around the metabolite clusters whose contents were shown trend of first rising and then falling or first falling and then rising in Control, Model and JSP group. These metabolites could be potential biomarkers for the treatment of heart failure.
## Weighted gene co-expression network analysis (WGCNA)
Operational taxonomic units (OTUs) of the intestinal microbiome were divided into modules by using weighted gene co-expression network analysis (WGCNA), then correlated with the metabolome to find metabolites and microbes that performed key functions, identified potential mechanisms involved in specific biological processes, and explored candidate biomarkers [30].
## Establishment of chronic heart failure model in SD rats
After the model rats created by the left anterior descending coronary artery ligation for 4 weeks, the left ventricular of them were dilated and contractility was diminished. Rats showed symptoms of heart failure, such as hair and weight loss, feces rarefaction, reduction in activities, accidie, and extrados. The heart samples of each group were obtained by dissection, and the appearance is shown in Figure 1.
**Figure 1:** *Representative images of the gross appearance of rat hearts.*
The left ventricular ejection fraction (LVEF) of the control group, the model group, and the JSP gastric perfusion group are shown in Table 1. Before being treated with given JSP and CPT, the LVEF of rats in the model group was significantly lower than that in the control group but there was no significant difference among rats in the model group, the JSP group, and the CPT group. After giving JSP and CPT to treatment groups for 4 weeks, the LVEF of rats in the model group was not changed significantly compared to the data from 4 weeks ago, but those in JSP and CPT were higher than that before. The LVEF of the JSP group and the CPT group had no significant difference but were significantly higher than that of the model group. Therefore, as can be seen from this indicator, both JSP and CPT had a good effect on heart failure. Then, the intestinal contents of each treatment group rat were obtained on a sterile bench to study their intestinal microbiome.
**Table 1**
| Group | Control | Model | Model + JSP | Model + CPT |
| --- | --- | --- | --- | --- |
| LVEF (Week 4) | 80.12 ± 4.75a | 56.53 ± 1.34b | 57.07 ± 3.05b | 56.65 ± 2.75b |
| LVEF (Week 8) | 75.39 ± 4.18a | 56.82 ± 2.06b | 74.13 ± 1.69a | 75.13 ± 1.76a |
## Diversity analysis of intestinal microbiome
After extracting the DNA from the intestinal microbiome of the rats in each treatment group, their 16S rRNA sequence data were determined by high-throughput sequencing. After filtering and splicing these sequencing data, the sequence and abundance of 889 OTUs were obtained. The normalized abundance of OTUs was used to perform intestinal microbiota diversity analysis. In the alpha diversity-related indicators shown in Figures 2A, B, the Chao1 index and Shannon index were not significantly different ($P \leq 0.05$) in the species richness and evenness of each group. The between-group variability analysis of the four diversity indexes is detailed in Supplementary Table 1. It can be seen that there was no significant difference in the diversity of gut microbiota species between HF rats and normal rats.
**Figure 2:** *The structure of the intestinal microbiome in the control, model, and JSP groups. (A, B) Shannon and Chao1 indices of α diversity. (C, D) Analysis of β diversity of intestinal microbiome in each group's rats. PCoA analysis of intestinal microbiome based on the OTU data of the control, model, and JPS groups. Each point represents a sample. A clear separation is observed between the samples of control (n = 6), model (n = 6), and JPS (n = 6) groups. (E, F) Percentage of total bacteria presented at phyla and genus levels, respectively.*
The results of the β diversity analysis (Figures 2C, D), based on the principal coordinate analysis of relative abundance (PCoA), showed that the community structures of intestinal microbiome were different among the control group, the model group, and the JSP group. According to the PERMANOVA analysis, the variance contributions to the difference in the bacterial structure were 10.09 and $24.98\%$. The intestinal microbiome structures of rats in each treatment group were significantly different, and the Pr-value (>F) was < 0.001.
The phylum-level species of gut microbes in this study are shown in Figure 2E. The largest average proportion of each bacterial community is Firmicutes, which accounts for 40–$60\%$. Compared with the control group, the Bacteroidetes proportion was decreased in the model group and then increased in the JSP group. However, the proportion of Proteobacteria in the same treatment showed the opposite trend of change.
The genus-level species of gut microbes in this study are shown in Figure 2F. Compared with the control group and the JSP group, the proportions of Prevotella_9, Lactobacillus, and Lachnospiraceae_NK4A136_group were decreased in the model group, whereas the proportion of Allobaculum in the model group was increased compared to that in the control and the model groups.
## Analysis of intestinal microbiome differences
Linear discriminant analysis effect size (LEfSe) was used to identify similar and dominant microbial species in the gut of rats in each treatment group. Compared with the control group, the abundance of Proteobacteria and their derivatives taxa were significantly higher in the model group (Figure 3A). The intestinal microorganisms of rats in the JSP group were rich in Bacteroides and their derivative taxa, including Bacteroidia, Bacteroidales, Prevotella, and Alloprevotella (Figure 3B). In addition, the results of LEfSe also indicated that the floras of Proteobacteria, Erysipelotrichia, and Gammaproteobacteria were key to the isolation of the model group from the other two groups (Figure 3C). The relative abundances of key differentiated bacterial genes in different groups are summarized in Figure 3D.
**Figure 3:** *LEfSe analysis for the gut microbiota alterations in different groups. (A, B) Microbial signatures for the model vs. control groups and JSP vs. model groups, respectively. (C) Overall gut microbiota community with different abundances. (D) Identified gut microbes with significant differences between groups. Data are shown as the means ± SEM (n = 6). Different lowercase letters indicated in the figure depict significant between-group variation (P < 0.05).*
## Determination of rat plasma metabolites and their function
Diseases and drugs often affect the intestinal microbiome, causing changes in their metabolites, which enter the bloodstream through the intestines and have a positive or negative effect on the host's physiology. Therefore, metabolomic analysis of plasma samples from rats of different treatment groups was performed by high-performance liquid chromatography-mass spectrometry (HPLC-MS) in this study. There were 6,565 and 9,203 peaks in total that were identified in negative and positive ion modes, respectively. Next, the peaks were clustered using partial least squares discriminant analysis (PLS-DA) to obtain more reliable metabolites with significant differences between treatment groups and to further test the validity of the method.
The results showed that the plasma metabolism data clusters under different treatments were separated from each other, indicating the presence of many different potential biomarkers (Figures 4A, B). The interpreted and validated values for the differences in the model were R2X = 0.374, R2Y = 0.999, Q2Y = 0.879 (control group vs. model group) and R2X = 0.325, R2Y = 0.990, Q2Y = 0.837 (control group vs. JSP group) in negative ion mode, and R2X = 0.362, R2Y = 0.999, Q2Y = 0.877 (control group vs. model group) and R2X = 0.304, R2Y = 0.993, Q2Y = 0.835 (control group vs. JSP group) in positive ion mode. This indicated that the model adequately explains the source of differences between the samples in each treatment group.
**Figure 4:** *The significant changes in plasma metabolic profiles in HF and treated rats. (A, B) Plots of PLS-DA scores of all peak features in positive and negative ion mode from the untargeted metabolomics analysis of plasma samples of the rats in control vs. model and model vs. JSP, respectively. (C) Clustering diagram of relative abundance patterns of the plasma metabolites based on the Mfuzz algorithm.*
Many metabolites were identified by analyzing the peaks on the tandem mass spectrometry, combining the accurate relative molecular weight, and obtaining the structural information from the compound structure database. Based on the analysis method of fuzzy C-means clustering, we carried out the clustering analysis of metabolites whose relative average deviation within the selected group was < $10\%$. Therefore, these metabolites were clustered into six groups. The change trends of metabolite content in the same clusters were similar in different treatment groups. But there were significant differences in change trends among different cluster groups. As shown in Figure 4C, in Cluster 1, there was no change between control and model groups but was significantly increased in the JSP group. In Cluster 2, the metabolite content was the lowest in the model group. On the contrary, the metabolite content in Clusters 3 and 6 was the highest in the model group. The metabolite content of Clusters 4 and 5 also showed the opposite trend, which was gradually decreasing and gradually increasing in the control, model, and JSP groups, respectively.
Based on the trend of these metabolite content, we speculated that the metabolites in Clusters 2, 3, and 6 were involved in improving the physiological process of heart failure. Therefore, the function of metabolites in these three clusters was performed by metabolic pathway enrichment analysis. Based on ConsensusPathDB (http://ConsensusPathDB.org) analysis, these metabolites were enriched in the following pathways: heme biosynthesis from uroporphyrinogen-III, heme biosynthesis II, pyrimidine deoxyribonucleotides biosynthesis I, M phase, degradation of cysteine and homocysteine heme biosynthesis, cell cycle, metabolism of nucleotides, metabolism of vitamins and cofactors, interconversion of nucleotide di- and triphosphates, sulfur amino acid metabolism, tryptophan metabolism, ferroptosis, and inflammatory mediator regulation of TRP channels (Table 2). At the same time, eight key metabolites were obtained, including ferrous ion (C14818), protoporphyrin IX (C01079), dihydrofolic acid (C00415), deoxyuridine-5'-triphosphate (C00460), nicotinamide (C00153), 3-mercapto-2-oxopropionic acid (C00957), propionylcarnitine (C03017), acyclovir (C06810), and their contents in the control and JSP groups were higher than that in model group. There are seven metabolites that were reduced in the control and JSP groups: kynurenic acid (C01717), reduced L-glutathione (C00051), 5-hydroxyindolepyruvate (C05646), 2-oxohexanedioic acid (C00322), 1-stearoyl-2-hydroxy-sn- glycero-3-phosphoethanolamine (C21484), cinnamaldehyde (C00903), and icilin (C20171) (Table 2).
**Table 2**
| Pathway | Pathway source | Overlapping metabolites | NO. of all pathway metabolites | p-value | q-value | Cluster |
| --- | --- | --- | --- | --- | --- | --- |
| Biosynthesis from uroporphyrinogen-III I | MouseCyc | C01079; C14818 | 11 (11) | 0.001 | 0.012 | 2 |
| Biosynthesis II | MouseCyc | C01079; C14818 | 18 (18) | 0.001 | 0.012 | 2 |
| Pyrimidine deoxyribonucleotides de novo biosynthesis I | MouseCyc | C00415; C00460 | 20 (20) | 0.002 | 0.012 | 2 |
| M Phase | Reactome | C14818; C00153 | 22 (25) | 0.002 | 0.012 | 2 |
| Degradation of cysteine and homocysteine | Reactome | C14818; C00957 | 24 (30) | 0.002 | 0.012 | 2 |
| Biosynthesis | Reactome | C01079; C14818 | 27 (33) | 0.003 | 0.013 | 2 |
| Cell Cycle, Mitotic | Reactome | C14818; C00153 | 34 (39) | 0.005 | 0.014 | 2 |
| Peroxisomal lipid metabolism | Reactome | C14818; C03017 | 35 (59) | 0.005 | 0.014 | 2 |
| Metabolism of porphyrins | Reactome | C01079; C14818 | 36 (42) | 0.006 | 0.014 | 2 |
| DNA Repair | Reactome | C14818; C00153 | 37 (51) | 0.006 | 0.014 | 2 |
| Cell Cycle | Reactome | C14818; C00153 | 38 (43) | 0.006 | 0.014 | 2 |
| Metabolism of nucleotides | Reactome | C14818; C00460; C00415 | 128 (148) | 0.007 | 0.015 | 2 |
| Metabolism of vitamins and cofactors | Reactome | C14818; C00153; C00415 | 132 (172) | 0.007 | 0.015 | 2 |
| Interconversion of nucleotide di- and triphosphates | Reactome | C00415; C00460 | 46 (50) | 0.009 | 0.016 | 2 |
| Sulfur amino acid metabolism | Reactome | C14818; C00957 | 46 (57) | 0.009 | 0.016 | 2 |
| Tryptophan metabolism–Mus musculus (mouse) | KEGG | C05646; C00322; C01717 | 83 (83) | 0.003 | 0.034 | 3 |
| Ferroptosis–Mus musculus (mouse) | KEGG | C00051; C21484 | 31 (31) | 0.005 | 0.034 | 3 |
| Inflammatory mediator regulation of TRP channels–Mus musculus (mouse) | KEGG | C20171; C00903 | 35 (35) | 0.005 | 0.042 | 6 |
## Intestinal microbiome combined with key metabolites WGCNA analysis
In this study, the WGCNA package in R software was used to construct a co-expression network. From the results of 16S rRNA sequencing of the intestinal microbiota of 16 rats, 483 OTUs with average relative abundance >1 and genus-level information were selected to construct the co-expression network. The data obtained earlier were used to construct the scale-free network. First, the adjacency matrix was calculated based on the expression value matrix, and then the topological overlap matrix reflected the similarity of the common expressions was derived. Based on the scale-free topology with R2 = 0.85, the Pearson correlation matrix of the 16S rRNA region OTUs was transformed into a strengthened adjacency matrix according to the power of $r = 11$ and $r = 6$, respectively. Then, the topological overlap matrix was used for hierarchical clustering to draw clustering trees that could jointly characterize the overall distribution of similarity. Finally, the generated clustering trees were cut by the dynamic cut tree algorithm. In this process, OTUs with high similarity in the common expression were clustered into the same branch, and different branches of the clustering tree represented different modules, each of which was assigned a specific color (Figures 5A–C). After constructing the co-expression network, 13 co-expression modules were obtained for intestinal bacterial OTUs (Table 3). Among them, the MEturquoise module had the highest number of OTUs with 128, while the MEtan module had the lowest number of OTUs with 13.
**Figure 5:** *WGCNA analysis of the intestinal microbiome co-expression network based on the relative abundance of OTUs. (A) Relative abundance correlation heat map of intestinal microbiome OTUs. (B) Connectivity and cluster analysis of OTUs' relative abundance in a different module, and the heatmap of connectivity of them. (C) Clustering dendrogram of OTUs, with dissimilarity based on the topological overlap, together with assigned module colors. The clustered branches represent different modules, and each line represents one OTU.* TABLE_PLACEHOLDER:Table 3 The results of the correlation analysis of metabolite content and the relative abundance of OTUs in Clusters 2, 3, and 6 are shown in Figures 6A, B, respectively. Evidently, as shown in Figure 6A, the content of C01079 and C00415 had significant positive correlations with MEblack and MEpurple modules, respectively. The content of C00415 and C00153 had significant negative correlations with MEturquoise and MEgreenyellow modules, respectively. As shown in Figure 6B, the content of C01717, C00051, and C21484 had significant positive correlations with MEturquoise and MEgreenyellow modules, respectively. The content of C01717 and C20171 had significant negative correlations with MEpink and MEpurple, respectively. The OTUs in the five modules were separately constructed as correlation networks, and the results are shown in Figure 6C. The number in each ball was the ID number of the OTU, as detailed in Supplementary Table 1.
**Figure 6:** *(A, B) Associations of different color modules and plasma metabolite from Clusters 2, 3, and 6. In the heatmap, each row corresponds to a module eigengene (ME) and each column to a trait. Each cell contains the corresponding correlation and p-value. The tables were color-coded by correlation according to the color legend. (C) Weighted correlation network of OTUs in five ME modules, including MEturquoise, MEgreenyellow, MEblack, MEpink, and MEpurple.*
As shown in Figure 7, the heat map was constructed by the mean relative abundance of key flora OTUs identified from WGCNA analysis, in which the family and genus information of these OTUs are classified, as shown in Figure 7. Overall, these bacteria were mainly distributed in six families, which were Ruminococcaceae, Chritensenellaceae, Erysipelotrichaceae, Akkermansiaceae, Clostridiaceae-1 and Lachnospiraceae (Figure 7). After correlation analysis of seven key metabolites and identified OTUs (Supplementary Figures 2, 3), it was found that OTU1637, OTU2744, OTU404, OTU5887, and OTU597 at Ruminococcaceae_UCG-014 genus, and had a significantly positive correlation with C01079 content. C00153 and C00415 contents had a significantly negative correlation with OTU2536, OTU4547, OTU6344, OTU915, OTU1568, OTU1063, OTU159, OTU369, OTU374, OTU394, OTU465, and OTU549, which belonged to Ruminococcaceae_UCG-005, Christensenellaceae_R-7_group, and Erysipelotrichaceae. However, C21484 content had a significantly positive correlation with OTU1568, OTU2536, OTU4547, OTU6344, OTU374, OTU394, OTU465, OTU549, OTU159, and OTU369 at the three genera mentioned above. C01717 content had a significantly positive correlation with OTU177, OTU3344, OTU406, and OTU5849 which belonged to Ruminococcaceae (Supplementary Table 1).
**Figure 7:** *The heatmap of intestinal microbiome OTUs related to selected plasma metabolites from Clusters 2, 3, and 6.*
## Discussion
Mounting pieces of evidence have revealed an association between microbial composition and their metabolites in HF. Recent studies had focused on HF with changes in intestinal microbiome structure but ignored its connection with microbial metabolites. Thus, a systematic investigation of the relevance between the intestinal microbiome and its metabolites is greatly warranted. We valued the changing intestinal microbiome and potential metabolites associated with the development and prognosis of HF. Through clustering and metabolite pathway analysis, 15 metabolites with clear physiological functions were identified in the control, the heart failure model, and the JSP treatment group, using the pathways of biosynthesis from uroporphyrinogen-III, biosynthesis II, pyrimidine deoxyribonucleotides biosynthesis I, degradation of cysteine and homocysteine, metabolism of nucleotides, and metabolism of vitamins and cofactors. After combinedly analyzing the contents of these metabolites and the intestinal microorganism OTUs by WGCNA, 116 bacteria that were significantly related to the content change of these metabolites were identified. These associations highlighted potential interactions of microbe and metabolites, which helped to further reveal the effect process by JSP in vivo during the development and prognosis of HF.
There are no significant differences in intestinal microbial diversity among Control, Model, and JSP groups. We have identified some bacteria altered by JSP treatment through LEfSe analysis, such as Erysipelotrichia, Anaplasma, and Gammaproteobacteria. It was found that they were the key differentiators that distinguished the model group from the control and the JSP treatment groups (Figure 3D). The changes in the abundance or ratio of several bacteria could indicate whether the intestinal microbiome was in disorder. For instance, the ratio of Firmicutes to Bacteroides (F/B) was an important indicator of intestinal microbiome disorder [31]. Compared with healthy people, F/B in the intestinal microbiome of patients with hypertension and heart failure was increased [32].
Among them, the bacteria of Erysipelotrichia under Firmicutes had the function of increasing the permeability of the intestinal mucosa and mediating the inflammatory response [33]. In this study, JSP may repair the intestinal barrier by inhibiting the F/B ratio to promote HF prognosis. Some bacteria contained in Lachnospiraceae_NK4A136_group [34, 35], Alloprevotella [36], and Roseburia [37, 38] could produce some short-chain fatty acids in the host intestine to regulate colon movement, immune maintenance, and anti-inflammatory, which may be related to protection from the host's stress reaction, and increased continuously in the prognosis of HF. It was shown that serum TMAO levels were positively associated with an increased abundance of Ruminococcaceae_UCG_005 and Christensenellaceae_R-7_group [39]. TMAO is a molecular metabolite derived from the gut microbiota, which may directly affect the heart by inducing myocardial hypertrophy and fibrosis, endothelial cell and vascular inflammation, as well as cardiac mitochondrial dysfunction, thereby aggravating the progress of HF (40–42). Prevotella-9 was negatively correlated with cardiac ejection fraction in rats with spontaneous hypertensive HF [43]. In this study, we found that the abundance of several bacteria in Prevotella-9 increased after JSP treatment. We speculated that its function was related to improving the ejection fraction, but the specific mechanism was not clear.
To identify plasma metabolites in response to JSP gavage, we attempted to analyze the changes in the content of these metabolites in different treatment groups using fuzzy c-means clustering. Three change clustering modes that respond to JSP processing have been selected, namely Clusters 2, 3, and 6 (Figure 4C). Through the functional analysis of metabolites in these groups, seven metabolites were found, which were distributed in six pathways related to the development and prognosis of HF (Table 1). First, protoporphyrin IX (C01079) is the intermediate product of heme biosynthesis, which is a key metabolite used to bind oxygen and oxidize the guanidine nitrogen of L-arginine to form nitric oxide (NO), playing a role as a vasodilator, as well as the citrulline under the action of nitric oxide synthase [44]. The anti-inflammatory carbon monoxide and bilirubin are produced by processes of heme metabolism, which further enhances the anti-inflammatory effect and ultimately relieves atherosclerosis [45]. Heme is also an important component of antioxidant functional proteins and enzymes in cardiomyocytes [46]. Heme has been shown to be protective against myocardial fibrosis and oxidative stress through inducting of heme oxygenase 1 and the activation of the phosphatidylinositol 3-kinase/AKT signaling pathway [47]. In this study, there were 10 bacteria that were significantly related to protoporphyrin IX, which belong to Ruminococcaceae_UCG-014, Ruminiclostridium_6, Fournierella, Eubacterium xylanophilum_group, Roseburia, and Desulfovibrio. These bacteria may participate in the synthesis of protoporphyrin IX and indirectly affect the change of heme content under the influence of JSP, which could enhance the antioxidant function of myocardial cells, reduce the inflammatory reaction, and ultimately improve the effectiveness of heart failure (Figure 8).
**Figure 8:** *JSP modified heart failure by the effect on the intestinal microbiome and metabolites in the blood.*
Then, in the metabolism pathway of vitamins and cofactors, nicotinamide (C00153) is the precursor of nicotinamide adenine dinucleotide (NAD), which is participated in the wide range of reactions, including regulation of cellular redox status, energy metabolism, and mitochondrial biogenesis [48]. In several models of heart failure, myocardial NAD levels have been depressed disturbed mitochondrial function remodeled metabolism, and occurred inflammation. The emerging evidence has suggested that regulating NAD homeostasis by NAD precursor supplementation has therapeutic efficiency in improving myocardial bioenergetics and function [49]. Under JSP treatment, the abundance of microbiota under 57 genera in rat intestines, including Ruminococcaceae_UCG-005, Christensenellaceae_R-7_group, Akkermansia, and Bacteroides, was significantly correlated with nicotinamide. These bacteria may directly or indirectly participate in the synthesis of nicotinamide so that it could achieve the purpose of alleviating heart failure symptoms (Figure 8). Dihydrofolate (C00415) is an intermediate for the conversion of folic acid to tetrahydrofolic acid, which is a member of the B-vitamin family, and is essential for amino acid metabolism [50]. Folic acid and its active metabolite 5-methyl tetrahydrofolate improved nitric oxide (NO) bioavailability by increasing endothelial NO synthase coupling and NO production as well as by directly scavenging superoxide radicals. By improving NO bioavailability, folic acid may protect or improve endothelial function, thereby preventing or reversing the progression of CVD in those with overt disease or elevated CVD risk [51]. After JSP gavage treatment, it was found that the abundance of 57 bacteria distributed in 29 families and 40 genera in the intestinal microbiome of rats was significantly correlated with the dihydrofolate content in plasma (Figure 8).
Dihydrofolate (C00415) and nicotinamide (C00153) are vitamin B derivatives, which are closely related to a variety of intestinal microbiomes [52, 53] and the development process of heart failure. In these intestinal microflorae related to dihydrofolate (C00415), the secondary metabolites produced by the degradation of Ruminococcaceae_UCG-014 need to be degraded by Bacteroidetes [54]. As the main component of human and animal intestinal microflora, the normal growth of these bacteria depended on porphyrins, including protoporphyrin IX or heme with iron [55] (Figure 8).
Kynurenine (C01717), a metabolite of the L-tryptophan pathway, may mediate immunomodulation, oxidant defense, and apoptosis [56], which are considered pathogenic features in the development of heart failure, and has been shown to predict cardiovascular events [57, 58]. Similarly, PE (18:$\frac{0}{0}$:0) (C21484) participated in the process of lipid peroxidation (LPO) in the pathway of ferroptosis. LPO which is one of their metabolites is involved in immune responses and cell deaths [58, 59]. Kynurenine and PE (18:$\frac{0}{0}$:0) were significantly correlated with the abundance of 10 and 49 species of bacteria, respectively, in this research. It is speculated that these bacteria may, under the influence of JSP, regulate these two metabolites' contents directly or indirectly to alleviate the course of cardiovascular disease and increase the antioxidant capacity of cardiovascular cells (Figure 8).
Results of intestinal microbiome analysis revealed that JSP not only adjusted gut microbiota disturbances by enriching species diversity, reducing the abundance of pathogenic bacteria, such as Allobaculum, and Brevinema, as well as increasing the abundance of beneficial bacteria, including Lactobacillus, and Lachnospiraceae_NK4A136_group but also improved metabolic disorders by reversing metabolite plasma levels to normality. The results of the correlation analysis demonstrated a significant association between intestinal microbiota and plasma metabolic profile. The key relationships in our research would illustrate the underlying mechanism of JSP to treat heart failure by affecting intestinal microbiome and plasma metabolites as well as provide evidence for the interpretation of the mechanism in HF. But the physiological mechanism of these bacteria with the potential function of improving heart failure and how to regulate these endogenous metabolites are still unclear. In the future, we will conduct in-depth research on the function of these intestinal microbiomes to further improve the mechanism of JSP in treating heart failure.
## Data availability statement
The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found below: https://www.ncbi.nlm.nih.gov/, BioProject ID PRJNA881788.
## Ethics statement
The animal study was reviewed and approved by the Research Ethical Committee of Guangdong Pharmaceutical University (Guangzhou, China).
## Author contributions
XHe, PL, and JM contributed to the design concepts of this whole study. XC carried out the animal experiment, performed data analyses, and drafted the manuscript. YS, XHu, and JC carried out the animal experiment and collected data. All authors have read and approved the content of the manuscript.
## Conflict of interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher's note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## Supplementary material
The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fcvm.2023.1147438/full#supplementary-material
## References
1. Rogers C, Bush N. **Heart failure pathophysiology, diagnosis, medical treatment guidelines, and nursing management**. *Nursing Clin North Am.* (2015) **50** 787-99. DOI: 10.1016/j.cnur.2015.07.012
2. Dickstein K. **Diagnosis and assessment of the heart failure patient: the cornerstone of effective management**. *Eur J Heart Fail.* (2005) **7** 303-8. DOI: 10.1016/j.ejheart.2005.01.003
3. Lozupone CA, Stombaugh JI, Gordon JI, Jansson JK, Knight R. **Diversity, stability and resilience of the human gut microbiota**. *Nature.* (2012) **489** 220-30. DOI: 10.1038/nature11550
4. Buglioni A, Burnett JC. **A gut-heart connection in cardiometabolic regulation**. *Nat Med.* (2013) **19** 534-6. DOI: 10.1038/nm.3196
5. Bartolomaeus H, McParland V, Wilck N. **Gut-heart axis. How gut bacteria influence cardiovascular diseases**. *Herz.* (2020) **45** 134-41. DOI: 10.1007/s00059-020-04897-0
6. Pasini E, Aquilani R, Testa C, Baiardi P, Angioletti S, Boschi F. **Pathogenic gut flora in patients with chronic heart failure**. *Jacc-Heart Failure.* (2016) **4** 220-7. DOI: 10.1016/j.jchf.2015.10.009
7. Lin L. **jie ZS, yuan HS, Bin C, Hong Q, xi HZ. Changes of gut microbiome composition and metabolites associated with hypertensive heart failure rats**. *BMC Microbiol.* (2021) **21** 141. DOI: 10.1186/s12866-021-02202-5
8. Zhang Y, Zhang S, Li B, Luo Y, Gong Y, Jin X. **Gut microbiota dysbiosis promotes age-related atrial fibrillation by lipopolysaccharide and glucose-induced activation of NLRP3-inflammasome**. *Cardiovasc Res.* (2022) **118** 785-97. DOI: 10.1093/cvr/cvab114
9. Tracey GX, Grace E. **X HC, P BJ, V HJ, Venkatesh R, et al. Probiotic administration attenuates myocardial hypertrophy and heart failure after myocardial infarction in the rat Circulation**. *Heart Failure.* (2014) **7** 491-9. DOI: 10.1161/CIRCHEARTFAILURE.113.000978
10. Anderson MK, Ferranti EP, Couvillon AE, Emma M, French EC, Miller RC. **The heart and gut relationship: a systematic review of the evaluation of the microbiome and trimethylamine-N-oxide (TMAO) in heart failure**. *Heart Fail Rev.* (2022) **27** 2223-49. DOI: 10.1007/s10741-022-10254-6
11. Cristina M, Massimo M, Marina C, Maria P. **J BM, Jeff T, et al. Metabolomic study of carotid-femoral pulse-wave velocity in women**. *J Hypertension.* (2015) **33** 791-6. DOI: 10.1097/HJH.0000000000000467
12. Zhu M-j, Wang Y-p, Xie S-y, Liu W-h, Li B, Wang Y-x. **Protective effects of Jiashen Prescription on myocardial infarction in rats**. *Chinese J Integrative Med* (2015) **21** 417. DOI: 10.1007/s11655-014-1751-4
13. Sun N, Zhang K, Geng W, Er X, Gao W, He Y. **Analysis of chemical constituents of Jiashen Tablet extract by UPLC-Q-TOF-MS**. *Chinese Herbal Med.* (2018) **49** 293-304. DOI: 10.7501/j.issn.0253-2670.2018.02.006
14. Cui L, Wang Y, Yu R, Li B, Xie S, Gao Y. **Jia-Shen decoction-medicated serum inhibits angiotensin-II induced cardiac fibroblast proliferation**. *Mol Med Rep.* (2016) **14** 1610-6. DOI: 10.3892/mmr.2016.5405
15. Feng W, Ao H, Peng C, Yan D. **Gut microbiota, a new frontier to understand traditional Chinese medicines**. *Pharmacol Res.* (2019) **142** 176-91. DOI: 10.1016/j.phrs.2019.02.024
16. Lin L, Luo L, Zhong M, Xie T, Liu Y, Li H. **Gut microbiota: a new angle for traditional herbal medicine research**. *RSC Adv.* (2019) **9** 17457-72. DOI: 10.1039/C9RA01838G
17. Lin TL, Lu CC, Lai WF, Wu TS, Lu JJ, Chen YM. **Role of gut microbiota in identification of novel TCM-derived active metabolites**. *Protein Cell.* (2021) **12** 394-410. DOI: 10.1007/s13238-020-00784-w
18. Do MH, Seo YS, Park HY. **Polysaccharides: bowel health and gut microbiota**. *Crit Rev Food Sci Nutr.* (2021) **61** 1212-24. DOI: 10.1080/10408398.2020.1755949
19. Gong X, Li X, Bo A, Shi RY Li QY, Lei LJ. **The interactions between gut microbiota and bioactive ingredients of traditional Chinese medicines: a review**. *Pharmacol Res.* (2020) **157** 104824. DOI: 10.1016/j.phrs.2020.104824
20. Yin L, Li XX, Ghosh S, Xie CM, Chen J, Huang H. **Role of gut microbiota-derived metabolites on vascular calcification in CKD**. *J Cell Mol Med.* (2021) **25** 1332-41. DOI: 10.1111/jcmm.16230
21. Nie QX, Chen HH, Hu JL, Fan ST, Nie SP. **Dietary compounds and traditional Chinese medicine ameliorate type 2 diabetes by modulating gut microbiota**. *Crit Rev Food Sci Nutr.* (2019) **59** 848-63. DOI: 10.1080/10408398.2018.1536646
22. Lu YH, Wan HF, Wu YJ, Yang JH Yu L, He Y. **Naoxintong capsule alternates gut microbiota and prevents hyperlipidemia in high-fat-diet fed rats**. *Front Pharmacol.* (2022) **13** 843409. DOI: 10.3389/fphar.2022.843409
23. Ide T, Tsutsui H, Hayashidani S, Kang DC, Suematsu N, Nakamura K. **Mitochondrial DNA damage and dysfunction associated with oxidative stress in failing hearts after myocardial infarction**. *Circ Res.* (2001) **88** 529-35. DOI: 10.1161/01.RES.88.5.529
24. Miao XL, Chen JP, Su YY, Luo JY, He Y, Ma J. **Plasma metabolomic analysis reveals the therapeutic effects of Jiashen tablets on heart failure**. *Front Cardiovasc Med* (2022) **9** 1047322. DOI: 10.3389/fcvm.2022.1047322
25. Ding L, Su XX, Zhang WH, Xu YX, Pan XF. **Gene expressions underlying mishandled calcium clearance and elevated generation of reactive oxygen species in the coronary artery smooth muscle cells of chronic heart failure rats**. *Chinese Med J.* (2017) **130** 460-9. DOI: 10.4103/0366-6999.199825
26. Burrell LM. **R C, A PP, P C, M TA, I JC. Validation of an echocardiographic assessment of cardiac function following moderate size myocardial infarction in the rat**. *Clin Exp Pharmacol Physiol.* (1996) **23** 570-2. DOI: 10.1111/j.1440-1681.1996.tb02782.x
27. Bolger AM, Lohse M, Usadel B. **Trimmomatic: a flexible trimmer for Illumina sequence data**. *Bioinformatics.* (2014) **30** 2114-20. DOI: 10.1093/bioinformatics/btu170
28. Yoon SH, Ha SM, Lim J, Kwon S, Chun J. **A large-scale evaluation of algorithms to calculate average nucleotide identity**. *Antonie Van Leeuwenhoek.* (2017) **110** 1281-6. DOI: 10.1007/s10482-017-0844-4
29. Futschik ME, Carlisle B. **Noise-robust soft clustering of gene expression time-course data**. *J Bioinform Comput Biol* (2005) **3** 965-88. DOI: 10.1142/S0219720005001375
30. Langfelder P, Horvath S. **WGCNA an R package for weighted correlation network analysis**. *BMC Bioinform.* (2008) **9** 559. DOI: 10.1186/1471-2105-9-559
31. Miao X, Chen J, Su Y, Luo J, He Y, Ma J, He X. **Correlation between firmicutes, bacteroidetes and firmicutes/bacteroidetes ratio and lipid profile in severely obese women in Rio De Janeiro - Brazil**. *J Acad Nutr Dietetics* (2022) **9** 122. DOI: 10.1016/j.jand.2022.06.054
32. Marques FZ, Nelson E, Chu PY, Horlock D, Fiedler A, Ziemann M. **High-fiber diet and acetate supplementation change the gut microbiota and prevent the development of hypertension and heart failure in hypertensive mice**. *Circulation.* (2017) **135** 964-77. DOI: 10.1161/CIRCULATIONAHA.116.024545
33. Pan MF, Wan CX, Xie Q, Huang RH, Tao XY, Shah NP. **Changes in gastric microbiota induced by Helicobacter pylori infection and preventive effects of Lactobacillus plantarum ZDY 2013 against such infection**. *J Dairy Sci.* (2016) **99** 970-81. DOI: 10.3168/jds.2015-10510
34. Ma LY Ni YH, Wang Z, Tu WQ Ni LY, Zhuge F. **Spermidine improves gut barrier integrity and gut microbiota function in diet-induced obese mice**. *Gut Microbes.* (2020) **12** 1832857. DOI: 10.1080/19490976.2020.1832857
35. Du SB, Zhou HH, Wang PF, Wang XP, Xue ZP Li J. **Modulation effects of danshen-honghua herb pair on gut microbiota of acute myocardial ischemia model rat**. *FEMS Microbiol Lett.* (2022) **369** 1-10. DOI: 10.1093/femsle/fnac036
36. Li Y, Su XH, Gao Y, Lv CX, Gao ZW, Liu YP. **The potential role of the gut microbiota in modulating renal function in experimental diabetic nephropathy murine models established in same environment**. *Biochim Biophys Acta Mol Basis Dis.* (2020) **1866** 165764. DOI: 10.1016/j.bbadis.2020.165764
37. Tamanai-Shacoori Z, Smida I, Bousarghin L, Loreal O, Meuric V, Fong SB. **et al. Roseburia spp: a marker of health?**. *Future Microbiol.* (2017) **12** 157-70. DOI: 10.2217/fmb-2016-0130
38. Nie K, Ma KJ, Luo WW, Shen ZH, Yang ZY, Xiao MW. **Roseburia intestinalis: a beneficial gut organism from the discoveries in genus and species**. *Front Cell Infect Microbiol.* (2021) **11** 757718. DOI: 10.3389/fcimb.2021.757718
39. Gao J, Yan KT, Wang JX, Dou J, Wang J, Ren M. **Gut microbial taxa as potential predictive biomarkers for acute coronary syndrome and post-STEMI cardiovascular events**. *Sci Rep.* (2020) **10** 26-39. DOI: 10.1038/s41598-020-59235-5
40. Li ZH, Wu ZY, Yan JY, Liu HL, Liu QC, Deng Y. **Gut microbe-derived metabolite trimethylamine N-oxide induces cardiac hypertrophy and fibrosis**. *Lab Invest.* (2019) **99** 346-57. DOI: 10.1038/s41374-018-0091-y
41. Makrecka-Kuka M, Volska K, Antone U, Vilskersts R, Grinberga S, Bandere D. **Trimethylamine N-oxide impairs pyruvate and fatty acid oxidation in cardiac mitochondria**. *Toxicol Lett.* (2017) **267** 32-8. DOI: 10.1016/j.toxlet.2016.12.017
42. Sun XL, Jiao XF, Ma YR, Liu Y, Zhang L, He YZ. **Trimethylamine N-oxide induces inflammation and endothelial dysfunction in human umbilical vein endothelial cells**. *Biochem Biophys Res Commun.* (2016) **481** 63-70. DOI: 10.1016/j.bbrc.2016.11.017
43. Shao MM, Zhu Y. **Long-term metal exposure changes gut microbiota of residents surrounding a mining and smelting area**. *Sci Rep.* (2020) **10** 44-53. DOI: 10.1038/s41598-020-61143-7
44. Jaronczyk K, Bui L, Soong JM, McLaughlin BE, Marks GS, Brien JF. **The source of heme for vascular heme oxygenase - II:**. *Can J Physiol Pharmacol.* (2004) **82** 218-24. DOI: 10.1139/y04-015
45. Wu GH, Zhang JF, Zhao QR, Zhuang WR, Ding JJ, Zhang C. **Molecularly engineered macrophage-derived exosomes with inflammation tropism and intrinsic heme biosynthesis for atherosclerosis treatment**. *Angewandte Chemie Int Edition.* (2020) **59** 4068-74. DOI: 10.1002/anie.201913700
46. Donzelli S, Espey MG, Flores-Santana W, Switzer CH, Yeh GC, Huang JM. **Generation of nitroxyl by heme protein-mediated peroxidation of hydroxylamine but not N-hydroxy-L-arginine**. *Free Radical Biol Med.* (2008) **45** 578-84. DOI: 10.1016/j.freeradbiomed.2008.04.036
47. Mendiburo MJ, Le Blanc S, Espinoza A, Pizarro F, Arredondo M. **Transepithelial heme-iron transport: effect of heme oxygenase overexpression**. *Eur J Nutr.* (2011) **50** 363-71. DOI: 10.1007/s00394-010-0144-5
48. Kumar JS, Subramanian VS, Kapadia R, Kashyap ML, Said HM. **Mammalian colonocytes possess a carrier-mediated mechanism for uptake of vitamin B3 (niacin): studies utilizing human and mouse colonic preparations**. *Am J Physiol Gastrointestinal Liver Physiol.* (2013) **305** G207-G13. DOI: 10.1152/ajpgi.00148.2013
49. Tannous C, Booz GW, Altara R, Muhieddine DH, Mericskay M, Refaat MM. **Nicotinamide adenine dinucleotide: biosynthesis, consumption and therapeutic role in cardiac diseases**. *Acta Physiologica.* (2021) **231** 13551. DOI: 10.1111/apha.13551
50. Stanhewicz AE, Kenney WL. **Role of folic acid in nitric oxide bioavailability and vascular endothelial function**. *Nutr Rev.* (2017) **75** 61-70. DOI: 10.1093/nutrit/nuw053
51. Rezk BM, Haenen G, van der Vijgh WJF, Bast A. **Tetrahydrofolate and 5-methyltetrahydrofolate are folates with high antioxidant activity. Identification of the antioxidant pharmacophore**. *FEBS Lett.* (2003) **555** 601-5. DOI: 10.1016/S0014-5793(03)01358-9
52. Hill MJ. **Intestinal flora and endogenous vitamin synthesis**. *Eur J Cancer Prevent.* (1997) **6** S43-S5. DOI: 10.1097/00008469-199703001-00009
53. Qi Y, Lohman J, Bratlie KM, Peroutka-Bigus N, Bellaire B, Wannemuehler M. **Vitamin C and B-3 as new biomaterials to alter intestinal stem cells**. *J Biomed Mater Res Part A.* (2019) **107** 1886-97. DOI: 10.1002/jbm.a.36715
54. Wang Y, Jiang M, Zhang Z, Sun H. **Effects of over-load iron on nutrient digestibility, haemato-biochemistry, rumen fermentation and bacterial communities in sheep**. *J Anim Physiol Anim Nutr.* (2020) **104** 32-43. DOI: 10.1111/jpn.13225
55. Halpern D, Gruss A. **A sensitive bacterial-growth-based test reveals how intestinal Bacteroides meet their porphyrin requirement**. *BMC Microbiol.* (2015) **15** 282. DOI: 10.1186/s12866-015-0616-0
56. Dschietzig TB, Kellner K-H, Sasse K, Boschann F, Kluesener R, Ruppert J. **Plasma kynurenine predicts severity and complications of heart failure and associates with established biochemical and clinical markers of disease**. *Kidney Blood Press Res.* (2019) **44** 765-76. DOI: 10.1159/000501483
57. Lund A, Nordrehaug JE, Slettom G, Solvang S-EH, Pedersen EKR, Midttun O. **Plasma kynurenines and prognosis in patients with heart failure**. *PLoS ONE.* (2020) **15** 0227365. DOI: 10.1371/journal.pone.0227365
58. Li QH. **Metabolic reprogramming, gut dysbiosis, and nutrition intervention in canine heart disease**. *Front Vet Sci* (2022) **9** 791754. DOI: 10.3389/fvets.2022.791754
59. Chen X, Li X, Xu XD Li LX, Liang NN, Zhang LL. **Ferroptosis and cardiovascular disease: role of free radical-induced lipid peroxidation**. *Free Radic Res.* (2021) **55** 405-15. DOI: 10.1080/10715762.2021.1876856
|
---
title: Addition of metformin for non-small cell lung cancer patients receiving antineoplastic
agents
authors:
- Yan Wang
- Yuanyuan Hu
- Ting Wang
- Guowei Che
- Lu Li
journal: Frontiers in Pharmacology
year: 2023
pmcid: PMC10036803
doi: 10.3389/fphar.2023.1123834
license: CC BY 4.0
---
# Addition of metformin for non-small cell lung cancer patients receiving antineoplastic agents
## Abstract
Background and purpose: Previous studies have found that metformin can inhibit tumor growth and improve outcomes for cancer patients. However, the association between the addition of metformin to the treatment regimen and survival in non-small cell lung cancer (NSCLC) patients receiving antineoplastic agents such as chemotherapy drugs, epidermal growth factor receptor tyrosine kinase inhibitors (EGFR-TKIs), and immune checkpoint inhibitors (ICIs) remains unclear. This study aimed to evaluate the effect of metformin in NSCLC patients who received the aforementioned antineoplastic therapies.
Methods: Several electronic databases were searched for relevant studies published by 10 September 2022. The primary and secondary outcomes were overall survival (OS) and progression-free survival (PFS); eligible studies were those comparing patients with and without the addition of metformin. Hazard ratios (HRs) and $95\%$ confidence intervals (CIs) were combined, with all statistical analyses performed using STATA 15.0.
Results: A total of 19 studies involving 6,419 participants were included, of which six were randomized controlled trials. The overall pooled results indicate that the addition of metformin improved OS (HR = 0.84, $95\%$ CI: 0.71–0.98, $$p \leq 0.029$$) and PFS (HR = 0.85, $95\%$ CI: 0.74–0.99, $$p \leq 0.039$$). However, subgroup analysis based on treatment type and comorbidity of diabetes mellitus demonstrated that improvements in OS and PFS were observed only in diabetic and EGFR-TKI-treated patients (OS: HR = 0.64, $95\%$ CI: 0.45–0.90, $$p \leq 0.011$$; PFS: HR = 0.59, $95\%$ CI: 0.34–1.03, $$p \leq 0.061$$).
Conclusion: Overall, this meta-analysis found that metformin use could improve outcomes for diabetic patients receiving EGFR-TKIs. However, no significant association between the addition of metformin and the survival of non-diabetic NSCLC patients receiving chemotherapy or ICI therapy was identified based on the current evidence.
## Introduction
The survival rate of patients with locally advanced or metastatic non-small cell lung cancer (NSCLC) remains poor, with the opportunity for radical resection having been lost by the time of diagnosis for a significant proportion of NSCLC patients (Kang et al., 2022; Xia et al., 2022; Xiang et al., 2022). Chemotherapy, epidermal growth factor receptor tyrosine kinase inhibitors (EGFR-TKIs), and immune checkpoint inhibitors (ICIs) are currently the main treatments for advanced NSCLC patients. However, the overall therapeutic effects of these treatments are not satisfactory, and their clinical use is usually limited for various reasons, such as primary and acquired resistance. Therefore, other medical measures are needed to improve the therapeutic effects of these antineoplastic agents.
Metformin (1,1-dimethylbiguanide) is the most commonly used drug for treating type 2 diabetes mellitus. In recent decades, substantial evidence has suggested that there is a clear beneficial effect of metformin in cases of malignancies, and that metformin plays a significant role in reducing cancer risk and improving outcomes for cancer patients (Mu et al., 2022; Yao et al., 2022; Zhang Q. et al., 2022; Han et al., 2023). A number of studies have found that metformin could reduce cancer morbidity and mortality rates in diabetic patients (Brancher et al., 2021; Kang et al., 2021). The underlying mechanisms are complicated and involve numerous pathways, such as the liver kinase B1 (LKB1)-dependent AMPK pathway and the GRB/IRS-1/PI3K/AKT/mTOR pathway, and the regulation of certain targets, such as the silent information regulator T1 (SIRT1) and YAP (Han et al., 2023).
In the case of lung cancer, a meta-analysis by Xiao et al. [ 2020] demonstrated that metformin treatment is significantly associated with reduced NSCLC incidence (HR = 0.78, $95\%$ CI: 0.70–0.86). Many studies have verified the beneficial role of metformin in NSCLC patients after surgical resection (Medairos et al., 2016; Yendamuri et al., 2019). However, as mentioned previously, a certain proportion of patients with advanced or inoperable NSCLC receive non-surgical treatments, including chemotherapy, EGFR-TKIs, and ICI therapy, which have become common in recent years. It remains unclear whether the concurrent use of metformin could enhance the efficacy of the aforementioned medications and improve the survival of NSCLC patients.
Luo et al. [ 2021] conducted a meta-analysis to explore the value of metformin as an adjunct treatment alongside antineoplastic agents in lung cancer; they showed that the addition of metformin might improve survival outcomes for lung cancer patients. However, their results were limited in terms of identifying an association between metformin use and the survival of NSCLC patients receiving the aforementioned antineoplastic agents.
Therefore, the aim of this meta-analysis was to identify the value of the concurrent use of metformin during treatment of NSCLC patients with chemotherapy, EGFR-TKIs, and ICIs. The results might contribute to the clinical application of metformin in NSCLC patients receiving the aforementioned antineoplastic agents.
## Materials and methods
This meta-analysis was conducted according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines [2020] (Zhang et al., 2020).
## Literature search
The PubMed, Embase, and Web of Science electronic databases were searched for articles published in the period from their inception to 10 September 2022. The following keywords were used during literature retrieval: metformin, lung, pulmonary, tumor, cancer, carcinoma, neoplasm, survival, prognosis, and prognostic. To avoid omissions, a relatively broad search strategy was developed and implemented, using the expression: metformin AND (lung OR pulmonary) AND (tumor OR cancer OR carcinoma OR neoplasm) AND (survival OR prognosis OR prognostic). In addition, MeSH terms and free words were applied, and references cited in the included publications were also reviewed.
## Inclusion and exclusion criteria
The inclusion criteria were as follows: 1) patients were pathologically diagnosed with primary NSCLC and received chemotherapy, EGFR-TKIs, or ICI therapy; 2) overall survival (OS) and/or progression-free survival (PFS) were compared between patients who did and did not also receive metformin during treatment with the aforementioned therapies; 3) hazard ratios (HRs) with $95\%$ confidence intervals (CIs) were directly reported by the article.
The exclusion criteria were as follows: 1) letters, editorials, meeting abstracts, case reports, and reviews; 2) articles reporting insufficient, overlapping, or duplicated data.
## Data extraction
The following information was collected from each included study: the name of the first author; publication year; country; sample size; study design, including randomized controlled trials (RCTs) and cohort studies; tumor–node–metastasis (TNM) stage; treatment strategy; comorbidity of diabetes mellitus; endpoint; and HR and $95\%$ CI.
## Methodological quality assessment
The quality of the RCTs and cohort studies was assessed using the Jadad scale and the Newcastle–Ottawa scale (NOS), respectively (Bhandari et al., 2001; Wang J. L. et al., 2021). Studies with a Jadad score of 4 or a NOS score of 6 or higher were defined as high-quality studies.
The literature search, selection, data collection, and quality assessment were all performed by two authors independently, and all disagreements were resolved by team discussion.
## Statistical analysis
All statistical analyses carried out in this meta-analysis were conducted using STATA 15.0 software. HRs with $95\%$ CIs were combined to compare OS and PFS between patients who did and did not receive metformin. Heterogeneity among the included studies was evaluated via I2 statistics and Q tests. When significant heterogeneity was observed, in the form of I2 > $50\%$ or $p \leq 0.1$, a random-effects model was applied; otherwise, a fixed-effects model was used (Barili et al., 2018). Subgroup analysis, stratified by treatment type, comorbidity of diabetes mellitus, and study design, was additionally conducted. In addition, a sensitivity analysis was conducted to identify the sources of heterogeneity and evaluate the stability of the pooled results. Furthermore, Begg’s funnel plots were constructed and Egger’s tests were conducted to detect publication bias (Begg and Mazumdar, 1994; Egger et al., 1997). Significant publication bias was defined as $p \leq 0.05.$
## Literature search and retrieval
Initially, 1,342 records were identified from electronic databases, and 252 duplicated records were removed. Next, 1,055 irrelevant publications and 15 unavailable publications were excluded after review of the titles and abstracts, respectively. Eight of the remaining reports were included after review of the full texts, and 11 available studies were included from previous relevant meta-analyses. Thus, a total of 19 studies were included in this meta-analysis (Tan et al., 2011; Ahmed et al., 2015; Chen et al., 2015; Lin et al., 2015; Sayed et al., 2015; Wink et al., 2016; Wen-Xiu et al., 2018; Afzal et al., 2019; Arrieta et al., 2019; Hung et al., 2019; Li et al., 2019; Su et al., 2020; Cortellini et al., 2021; Han et al., 2021; Jacobi et al., 2021; Lee et al., 2021; Skinner et al., 2021; Tsakiridis et al., 2021; Wang Y. et al., 2021). The specific literature retrieval process is presented in Figure 1.
**FIGURE 1:** *Flowchart of literature retrieval and inclusion in this meta-analysis.*
## Basic characteristics of the included studies
The included studies were published between 2011 and 2021, and their sample sizes ranged from 40 to 1,633, for a total of 6,419 patients. Six of the studies were RCTs (Sayed et al., 2015; Arrieta et al., 2019; Li et al., 2019; Lee et al., 2021; Skinner et al., 2021; Tsakiridis et al., 2021); the remainder were cohort studies (Tan et al., 2011; Ahmed et al., 2015; Chen et al., 2015; Lin et al., 2015; Wink et al., 2016; Wen-Xiu et al., 2018; Afzal et al., 2019; Hung et al., 2019; Su et al., 2020; Cortellini et al., 2021; Han et al., 2021; Jacobi et al., 2021; Wang Y. et al., 2021). All included studies were high-quality studies with a Jadad score ≥4 or NOS score ≥6. Detailed information is presented in Table 1.
**TABLE 1**
| Author | Country | Sample size | Study design | TNM stage | Treatment | Diabetes mellitus | Endpoint | Jadad/NOS score |
| --- | --- | --- | --- | --- | --- | --- | --- | --- |
| Tan et al. (2011) | China | 99 | Cohort | II–IV | CT | Yes | OS, PFS | 9 |
| Ahmed et al. (2015) | United States | 40 | Cohort | I–IV | CRT | Yes | OS, PFS | 7 |
| Chen et al. (2015) | China | 90 | Cohort | III–IV | EGFR-TKI | Yes | OS, PFS | 7 |
| Lin et al. (2015) | United States | 340 | Cohort | IV | CT | Yes | OS | 8 |
| Sayed et al. (2015) | Egypt | 30 | RCT | IV | CT | No | OS, PFS | 5 |
| Wink et al. (2016) | Netherlands | 682 | Cohort | II–III | CRT | Yes | OS, PFS | 8 |
| Wen-Xiu et al. (2018) | China | 75 | Cohort | I-IV | CT | Yes | OS | 9 |
| Afzal et al. (2019) | United States | 50 | Cohort | IV | ICI | Mixed | OS, PFS | 8 |
| Arrieta et al. (2019) | Mexico | 139 | RCT | III–IV | EGFR-TKI | No | OS, PFS | 4 |
| Hung et al. (2019) | China | 1633 | Cohort | III–IV | EGFR-TKI | Yes | OS, PFS | 8 |
| Li et al. (2019) | China | 224 | RCT | III–IV | EGFR-TKI | No | OS, PFS | 7 |
| Su et al. (2020) | China | 853 | Cohort | IIIB–IV | EGFR-TKI | NR | OS | 7 |
| Cortellini et al. (2021) | Italy | 950 | Cohort | IV | ICI | NR | OS, PFS | 6 |
| Han et al. (2021) | China | 85 | Cohort | I–IV | EGFR-TKI | Yes | OS, PFS | 9 |
| Jacobi et al. (2021) | Israel | 249 | Cohort | IV | ICI | Yes | OS, PFS | 8 |
| Lee et al. (2021) | South Korea | 164 | RCT | IIIB–IV | CT | Mixed | OS, PFS | 5 |
| Skinner et al. (2021) | Canada | 167 | RCT | III | CRT | No | OS, PFS | 5 |
| Tsakiridis et al. (2021) | Canada | 54 | RCT | III | CRT | No | OS, PFS | 4 |
| Wang J. L. et al. (2021) | China | 495 | Cohort | Advanced | CT | Yes | OS | 8 |
## The association between addition of metformin and OS of NSCLC patients receiving antineoplastic agents
All included studies explored the association between the addition of metformin and the OS of NSCLC patients receiving antineoplastic agents (Afzal et al., 2019; Ahmed et al., 2015; Arrieta et al., 2019; Chen et al., 2015; Cortellini et al., 2021; Han et al., 2021; Hung et al., 2019; Jacobi et al., 2021; Lee et al., 2021; Li et al., 2019; J. J; Lin et al., 2015; Sayed et al., 2015; Skinner et al., 2021; Su et al., 2020; Tan et al., 2011; Tsakiridis et al., 2021; Wang J. L. et al., 2021; Wen-Xiu et al., 2018; Wink et al., 2016). The overall pooled results indicated that patients who were treated with metformin had better rates of OS (HR = 0.84, $95\%$ CI: 0.71–0.98, $$p \leq 0.029$$; I2 = $67.8\%$, $p \leq 0.001$) (Figure 2A). However, subgroup analysis based on treatment type and comorbidity of diabetes mellitus showed that the effect of improved OS was observed only for patients receiving EGFR-TKIs (HR = 0.73, $95\%$ CI: 0.57–0.94, $$p \leq 0.013$$) (Figure 2B) and those with diabetes mellitus (HR = 0.74, $95\%$ CI: 0.62–0.88, $$p \leq 0.001$$) (Figure 2C). In contrast, the association between metformin use and the OS of NSCLC patients receiving chemoradiotherapy was negative (HR = 1.36, $95\%$ CI: 0.79–2.35, $$p \leq 0.261$$). In addition, subgroup analysis based on study design showed that the association between metformin use and improved OS in NSCLC patients was observed only in cohort studies (HR = 0.79, $95\%$ CI: 0.66–0.94, $$p \leq 0.007$$) (Figure 2D).
**FIGURE 2:** *The association between metformin use and overall survival in (A) all NSCLC patients; (B) subgroups by treatment type; (C) subgroups by comorbidity of diabetes mellitus; and (D) subgroups by study design. NSCLC: non-small cell lung cancer.*
To further clarify the nature of the association between metformin use and OS in EGFR-TKI-treated and diabetic NSCLC patients, subgroup analyses were conducted focusing on EGFR-TKI-treated patients (according to comorbidity of diabetes mellitus) and diabetic patients (according to treatment type). The pooled results demonstrated that the addition of metformin could significantly improve the OS of diabetic NSCLC patients receiving EGFR-TKIs (HR = 0.64, $95\%$ CI: 0.45–0.90, $$p \leq 0.011$$) (Table 2).
**TABLE 2**
| Unnamed: 0 | No. of studies | HR | 95% CI | p-value | I2 (%) | Pheterogeneity |
| --- | --- | --- | --- | --- | --- | --- |
| Overall survival | 19 | 0.84 | 0.71–0.98 | 0.029 | 67.8 | <0.001 |
| Treatment | Treatment | Treatment | Treatment | Treatment | Treatment | Treatment |
| Chemotherapy | 10 | 0.86 | 0.67–1.10 | 0.230 | 72.6 | <0.001 |
| With radiotherapy | 4 | 1.36 | 0.79–2.35 | 0.261 | 68.6 | 0.023 |
| EGFR-TKI | 6 | 0.73 | 0.57–0.94 | 0.013 | 52.1 | 0.063 |
| With diabetes mellitus | 3 | 0.64 | 0.45–0.90 | 0.011 | 37.0 | 0.205 |
| Without diabetes mellitus | 2 | 0.82 | 0.41–1.65 | 0.583 | 81.7 | 0.019 |
| Immune checkpoint inhibitor | 3 | 1.12 | 0.90–1.39 | 0.311 | 0.0 | 0.586 |
| Comorbidity of diabetes mellitus | Comorbidity of diabetes mellitus | Comorbidity of diabetes mellitus | Comorbidity of diabetes mellitus | Comorbidity of diabetes mellitus | Comorbidity of diabetes mellitus | Comorbidity of diabetes mellitus |
| Yes | 10 | 0.74 | 0.62–0.88 | 0.001 | 60.6 | 0.007 |
| Chemotherapy | 6 | 0.76 | 0.59–0.97 | 0.030 | 68.9 | 0.007 |
| EGFR-TKI | 3 | 0.64 | 0.45–0.90 | 0.011 | 37.0 | 0.205 |
| Immune checkpoint inhibitor | 1 | 1.29 | 0.69–2.40 | 0.422 | - | - |
| No | 5 | 1.00 | 0.60–1.67 | 0.998 | 75.6 | 0.003 |
| Mixed | 2 | 0.92 | 0.67–1.26 | 0.598 | 0.0 | 0.672 |
| Study design | Study design | Study design | Study design | Study design | Study design | Study design |
| Cohort study | 13 | 0.79 | 0.66–0.94 | 0.007 | 65.7 | <0.001 |
| Randomized controlled trial | 6 | 0.98 | 0.67–1.43 | 0.910 | 69.5 | 0.006 |
| Progression-free survival | 15 | 0.85 | 0.74–0.99 | 0.039 | 60.2 | 0.001 |
| Treatment | Treatment | Treatment | Treatment | Treatment | Treatment | Treatment |
| Chemotherapy | 7 | 0.94 | 0.71–1.24 | 0.643 | 66.4 | 0.007 |
| With radiotherapy | 4 | 1.18 | 0.69–2.00 | 0.551 | 72.8 | 0.011 |
| EGFR-TKI | 5 | 0.70 | 0.52–0.94 | 0.019 | 71.3 | 0.008 |
| With diabetes mellitus | 3 | 0.59 | 0.34–1.03 | 0.061 | 80.3 | 0.006 |
| Without diabetes mellitus | 2 | 0.81 | 0.48–1.37 | 0.431 | 73.4 | 0.053 |
| Immune checkpoint inhibitor | 3 | 1.02 | 0.83–1.24 | 0.864 | 0.0 | 0.859 |
| Comorbidity of diabetes mellitus | Comorbidity of diabetes mellitus | Comorbidity of diabetes mellitus | Comorbidity of diabetes mellitus | Comorbidity of diabetes mellitus | Comorbidity of diabetes mellitus | Comorbidity of diabetes mellitus |
| Yes | 7 | 0.75 | 0.61–0.92 | 0.006 | 60.5 | 0.019 |
| Chemotherapy | 3 | 0.77 | 0.59–1.02 | 0.070 | 37.8 | 0.201 |
| EGFR-TKI | 3 | 0.59 | 0.34–1.03 | 0.061 | 80.3 | 0.006 |
| Immune checkpoint inhibitor | 1 | 1.08 | 0.61–1.92 | 0.793 | - | - |
| No | 5 | 0.96 | 0.64–1.44 | 0.839 | 71.8 | 0.007 |
| Mixed | 2 | 0.98 | 0.72–1.31 | 0.868 | 0.0 | 0.676 |
| Study design | Study design | Study design | Study design | Study design | Study design | Study design |
| Cohort study | 9 | 0.80 | 0.68–0.95 | 0.011 | 57.5 | 0.016 |
| Randomized controlled trial | 6 | 0.97 | 0.71–1.32 | 0.828 | 65.0 | 0.014 |
## The association between addition of metformin and PFS of NSCLC patients receiving antineoplastic agents
Fifteen studies explored the association between the addition of metformin and PFS in NSCLC patients receiving antineoplastic agents (Tan et al., 2011; Ahmed et al., 2015; Chen et al., 2015; Sayed et al., 2015; Wink et al., 2016; Afzal et al., 2019; Arrieta et al., 2019; Hung et al., 2019; Li et al., 2019; Cortellini et al., 2021; Han et al., 2021; Jacobi et al., 2021; Lee et al., 2021; Skinner et al., 2021; Tsakiridis et al., 2021). The overall results showed that metformin use was clearly related to better rates of PFS (HR = 0.85, $95\%$ CI: 0.74–0.99, $$p \leq 0.039$$; I2 = $60.2\%$, $$p \leq 0.001$$) (Figure 3A). However, subgroup analysis stratified by treatment type and comorbidity of diabetes mellitus also indicated that the effect of improved PFS was observed only in patients receiving EGFR-TKIs (HR = 0.70, $95\%$ CI: 0.52–0.94, $$p \leq 0.019$$) (Figure 3B) and those with diabetes mellitus (HR = 0.75, $95\%$ CI: 0.61–0.92, $$p \leq 0.006$$) (Figure 3C). In contrast, the association between metformin use and PFS of NSCLC patients receiving chemoradiotherapy was negative (HR = 1.18, $95\%$ CI: 0.69–2.00, $$p \leq 0.551$$). In addition, subgroup analysis based on study design showed that the association between metformin use and improved PFS in patients with NSCLC was observed only in cohort studies (HR = 0.80, $95\%$ CI: 0.68–0.95, $$p \leq 0.011$$) (Figure 3D).
**FIGURE 3:** *The association between metformin use and progression-free survival in (A) all NSCLC patients; (B) subgroups by treatment type; (C) subgroups by comorbidity of diabetes mellitus; and (D) subgroups by study design. NSCLC: non-small cell lung cancer.*
Similarly, subgroup analyses were performed focusing on patients receiving EGFR-TKI treatment (according to comorbidity of diabetes mellitus) and diabetic patients (according to treatment type). The pooled results demonstrated that the addition of metformin was related to better PFS in diabetic NSCLC patients receiving EGFR-TKIs (HR = 0.59, $95\%$ CI: 0.34–1.03, $$p \leq 0.061$$), although the difference was not significant (Table 2).
## Sensitivity analysis
Sensitivity analyses were conducted for OS and PFS, focusing on all NSCLC patients (Figures 4A, 5A), EGFR-TKI-treated patients (Figures 4B, 5B), and diabetic patients (Figures 4C, 5C). Overall, a small number of the included studies had a clear impact on the results. More RCTs with large samples are needed to verify our findings.
**FIGURE 4:** *Sensitivity analysis for the association between metformin use and overall survival in (A) all NSCLC patients; (B) patients receiving EGFR-TKI therapy; and (C) patients with diabetes mellitus. NSCLC: non-small cell lung cancer; EGFR-TKI: epidermal growth factor receptor-tyrosine kinase inhibitor.* **FIGURE 5:** *Sensitivity analysis for the association between metformin use and progression-free survival in (A) all NSCLC patients; (B) patients receiving EGFR-TKI therapy; and (C) patients with diabetes mellitus. NSCLC: non-small cell lung cancer; EGFR-TKI: epidermal growth factor receptor-tyrosine kinase inhibitor.*
## Publication bias
Begg’s funnel plots for OS and PFS were both symmetrical (Figures 6A, B), and the p-values in Egger’s tests for OS and PFS were 0.218 and 0.900, respectively, indicating that no significant publication bias existed in this meta-analysis.
**FIGURE 6:** *Begg’s funnel plots for (A) overall survival and (B) progression-free survival.*
## Discussion
The current meta-analysis demonstrated that the addition of metformin was beneficial for diabetic NSCLC patients who received EGFR-TKI therapy and that metformin use could significantly improve the survival rate of this group of patients. However, no significant association between the addition of metformin and the survival of non-diabetic patients receiving chemotherapy or ICI was observed. Furthermore, due to the limitations of the included studies, more RCTs with larger samples are needed to verify the beneficial value of metformin use in diabetic and EGFR-TKI-treated NSCLC patients.
In a previous similar meta-analysis, Luo et al. [ 2021] included three RCTs and 11 observational cohort studies involving 3,856 lung cancer patients, and showed that antineoplastic agents combined with metformin significantly improve OS (HR = 0.73, $p \leq 0.001$) and PFS (HR = 0.72, $$p \leq 0.001$$). Similar results were observed when they combined cohort studies, but no significant association between metformin use and survival of lung cancer patients was detected based on limited data from RCTs. The authors conducted additional subgroup analyses based on type of therapy (chemotherapy vs EGFR-TKI), histology (NSCLC vs small cell lung cancer), and stage (III–IV vs I–IV), which produced consistent results, indicating that the addition of metformin could induce clear improvement in the efficacy of antineoplastic agents in lung cancer patients (Luo et al., 2021). However, this conclusion is obviously crude, and several factors—such as therapeutic methods and histology—may affect the objectivity and authenticity of the results.
Similarly, several other meta-analyses have explored the anticancer effect of metformin in cases of lung cancer in recent years, but there are considerable limitations to these analyses (Zhang Q. et al., 2022). Zhang F. et al. [ 2022] explored the anticancer role of metformin in NSCLC patients receiving EGFR-TKIs, but only three studies published before August 2020 were included. Brancher et al. [ 2021] included ten cohort studies and four RCTs and came to the tentative conclusion that metformin use might be associated with improved OS of lung cancer patients; this result was similar to that of several other meta-analyses (Wan et al., 2016; Cao et al., 2017; Zhong et al., 2017; Zhang et al., 2018; Zeng et al., 2019; Xiao et al., 2020; Brancher et al., 2021). Thus, we conducted the current meta-analysis to clarify the value of metformin use in NSCLC patients receiving the aforementioned antineoplastic agents. We demonstrated that benefits of metformin were only experienced by diabetic and EGFR-TKI-treated NSCLC patients.
Several studies have investigated the underlying mechanisms by which metformin affects the therapeutic effects of EGFR-TKIs in NSCLC patients. It has been reported that metformin and EGFR-TKIs have a synergistic therapeutic effect in NSCLC patients with type 2 diabetes (Nguyen et al., 2011). In one study focusing on the treatment of LKB1 wild-type NSCLC cells, it was found that the addition of gefitinib to metformin could inhibit EGFR phosphorylation and its downstream signaling; in addition, increased c-Raf/B-Raf isomerization was found to cause MAPK activation, which induced significant apoptosis in vitro and in vivo (Morgillo et al., 2013). Furthermore, metformin plays a role in overcoming resistance to EGFR-TKIs by inhibiting the PI3K/AKT/mTOR signaling pathway (Tan, 2020). Li et al. [ 2014] also reported on the effects of metformin in inhibiting the IL-6/STAT3 signaling pathway, reversing epithelial–mesenchymal transition (EMT), and overcoming EGFR-TKI drug (gefitinib and erlotinib) resistance in NSCLC cells. However, most relevant studies have been conducted in vitro and in vivo, and the mechanisms are still unclear.
Although we found that metformin was not significantly associated with the survival of NSCLC patients receiving chemotherapy drugs, a number of studies have indicated a role for metformin in increasing the sensitivity of chemotherapy drugs such as doxorubicin, cisplatin, and paclitaxel in NSCLC (Iliopoulos et al., 2011; Tan et al., 2011; Teixeira et al., 2013; Tseng et al., 2013). For example, cisplatin resistance is associated with signal transducer and activator of transcription 3 (STAT3) phosphorylation, production of reactive oxygen species (ROS), and IL-6 secretion, but metformin could inhibit cisplatin-induced STAT3 phosphorylation (through the LKB1-AMPK and mTOR pathway-dependent mechanisms), ROS generation, and autocrine IL-6 secretion (Khan et al., 2013; Lin et al., 2013). Metformin thus plays a role in enhancing cisplatin cytotoxicity and improving the cisplatin resistance of cancer cells (Wang et al., 2014; Wang et al., 2015).
No significant relationship between metformin use and the survival of ICI-treated NSCLC patients was observed in our meta-analysis. A few previous studies have suggested that metformin enhances the efficacy of ICI therapy in NSCLC patients (Yendamuri et al., 2019; Shen et al., 2020; Kim et al., 2021). The expression levels of LKB1 and PD-L1 are closely correlated, and AMPK inhibition reduces PD-L1 levels in NSCLC cells through LKB1 (Shen et al., 2020). Metformin has the ability to enhance the expression of LKB1 and the activation of AMPK, which improves the therapeutic effect of ICIs in NSCLC (Shen et al., 2020). However, most relevant studies investigating the therapeutic role of metformin in NSCLC patients receiving chemotherapy and ICIs have been based on cellular and animal trials. Therefore, more prospective RCTs are needed to clarify the therapeutic role of metformin in NSCLC patients receiving the aforementioned therapies.
There are several limitations to our meta-analysis. First, most of the studies included were retrospective cohort studies with small sample sizes, which might cause some bias. Second, although stratification analysis was performed based specifically on treatment type (EGFR-TKI therapy) and comorbidity of diabetes mellitus and strongly indicated a significant relationship between metformin use and better outcomes for diabetic and EGFR-TKI-treated NSCLC patients, only 1808 patients from three cohort studies were enrolled (Chen et al., 2015; Hung et al., 2019; Han et al., 2021) in this group. Third, due to a lack of original data, we were unable to conduct additional subgroup analyses based on other important parameters, such as metformin dose, TNM stage, and age. Fourth, we were unable to calculate the base and recommended doses of metformin to determine its effect in enhancing the efficacy of EGFR-TKIs in this meta-analysis. Five clear instances of heterogeneity among the included studies were observed in the analysis; however, unfortunately, the sources of heterogeneity were not clarified in the subgroup analyses, and due to the limited amount of evidence, we were unable to conduct more subgroup analyses based on other parameters.
## Conclusion
The addition of metformin to the treatment regimen was beneficial for diabetic NSCLC patients who received EGFR-TKI therapy, and metformin use could significantly improve the survival rate of this group of patients. However, no significant association between the addition of metformin and the survival of non-diabetic NSCLC patients receiving chemotherapy or ICI therapy was identified based on the current evidence. Meanwhile, more RCTs with larger samples are needed to verify the aforementioned findings.
## Data availability statement
The original contributions presented in the study are included in the article/Supplementary Material, further inquiries can be directed to the corresponding authors.
## Ethics statement
The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. All procedures performed in studies that involved human participants were in accordance with the ethical standards of the institutional and/or national research committee and with the 1964 Helsinki Declaration and its later amendments or comparable ethical standards.
## Author contributions
Conceptualization: LL and GC. Literature search: YW and TW. Independent review and risk of bias assessment: YW and TW. Data curation: YH. Data analysis and synthesis: YW, TW, and YH. Initial draft of manuscript: YW and TW. Final manuscript: all authors.
## Conflict of interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher’s note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors, and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## References
1. Afzal M. Z., Dragnev K., Sarwar T., Shirai K.. **Clinical outcomes in non-small-cell lung cancer patients receiving concurrent metformin and immune checkpoint inhibitors**. *Lung Cancer Manag.* (2019) **8** LMT11. DOI: 10.2217/lmt-2018-0016
2. Ahmed I., Ferro A., Cohler A., Langenfeld J., Surakanti S. G., Aisner J.. **Impact of metformin use on survival in locally-advanced, inoperable non-small cell lung cancer treated with definitive chemoradiation**. *J. Of Thorac. Dis.* (2015) **7** 346-355. DOI: 10.3978/j.issn.2072-1439.2014.12.32
3. Arrieta O., Barron F., Padilla M. A. S., Aviles-Salas A., Ramirez-Tirado L. A., Arguelles Jimenez M. J.. **Effect of metformin plus tyrosine kinase inhibitors compared with tyrosine kinase inhibitors alone in patients with epidermal growth factor receptor-mutated lung adenocarcinoma: A phase 2 randomized clinical trial**. *Jama Oncol.* (2019) **5** e192553. DOI: 10.1001/jamaoncol.2019.2553
4. Barili F., Parolari A., Kappetein P. A., Freemantle N.. **Statistical primer: Heterogeneity, random- or fixed-effects model analyses?**. *Interact. Cardiovasc Thorac. Surg.* (2018) **27** 317-321. DOI: 10.1093/icvts/ivy163
5. Begg C. B., Mazumdar M.. **Operating characteristics of a rank correlation test for publication bias**. *Biometrics* (1994) **50** 1088-1101. PMID: 7786990
6. Bhandari M., Richards R. R., Sprague S., Schemitsch E. H.. **Quality in the reporting of randomized trials in surgery: Is the Jadad scale reliable?**. *Control Clin. Trials* (2001) **22** 687-688. DOI: 10.1016/s0197-2456(01)00147-7
7. Brancher S., Ribeiro A. E., Toporcov T. N., Weiderpass E.. **The role of metformin on lung cancer survival: The first systematic review and meta-analysis of observational studies and randomized clinical trials**. *J. Cancer Res. Clin. Oncol.* (2021) **147** 2819-2836. DOI: 10.1007/s00432-021-03728-x
8. Cao X., Wen Z. S., Wang X. D., Li Y., Liu K. Y., Wang X.. **The clinical effect of metformin on the survival of lung cancer patients with diabetes: A comprehensive systematic review and meta-analysis of retrospective studies**. *J. Cancer* (2017) **8** 2532-2541. DOI: 10.7150/jca.19750
9. Chen H., Yao W., Chu Q., Han R., Sun J., Wang D.. **Synergistic effects of metformin in combination with EGFR-TKI in the treatment of patients with advanced non-small cell lung cancer and type 2 diabetes**. *Cancer Lett.* (2015) **369** 97-102. DOI: 10.1016/j.canlet.2015.08.024
10. Cortellini A., Di Maio M., Nigro O., Leonetti A., Cortinovis D. L., Aerts J. G.. **Differential influence of antibiotic therapy and other medications on oncological outcomes of patients with non-small cell lung cancer treated with first-line pembrolizumab versus cytotoxic chemotherapy**. *J. Immunother. Cancer* (2021) **9** e002421. DOI: 10.1136/jitc-2021-002421
11. Egger M., Davey Smith G., Schneider M., Minder C.. **Bias in meta-analysis detected by a simple, graphical test**. *Bmj* (1997) **315** 629-634. DOI: 10.1136/bmj.315.7109.629
12. Han P., Zhou J., Xiang J., Liu Q., Sun K.. **Research progress on the therapeutic effect and mechanism of metformin for lung cancer (Review)**. *Oncol. Rep.* (2023) **49** 3. DOI: 10.3892/or.2022.8440
13. Han R., Jia Y., Li X., Zhao C., Zhao S., Liu S.. **Concurrent use of metformin enhances the efficacy of EGFR-TKIs in patients with advanced EGFR-mutant non-small cell lung cancer-an option for overcoming EGFR-TKI resistance**. *Transl. Lung Cancer Res.* (2021) **10** 1277-1291. DOI: 10.21037/tlcr-20-1153
14. Hung M. S., Chuang M. C., Chen Y. C., Lee C. P., Yang T. M., Chen P. C.. **Metformin prolongs survival in type 2 diabetes lung cancer patients with EGFR-TKIs**. *Integr. Cancer Ther.* (2019) **18** 1534735419869491. DOI: 10.1177/1534735419869491
15. Iliopoulos D., Hirsch H. A., Struhl K.. **Metformin decreases the dose of chemotherapy for prolonging tumor remission in mouse xenografts involving multiple cancer cell types**. *Cancer Res.* (2011) **71** 3196-3201. DOI: 10.1158/0008-5472.CAN-10-3471
16. Jacobi O., Landman Y., Reinhorn D., Icht O., Sternschuss M., Rotem O.. **The relationship of diabetes mellitus to efficacy of immune checkpoint inhibitors in patients with advanced non-small cell lung cancer**. *Oncol. Switz.* (2021) **99** 555-561. DOI: 10.1159/000516671
17. Kang J., Jeong S. M., Shin D. W., Cho M., Cho J. H., Kim J.. **The associations of aspirin, statins, and metformin with lung cancer risk and related mortality: A time-dependent analysis of population-based nationally representative data**. *J. Thorac. Oncol.* (2021) **16** 76-88. DOI: 10.1016/j.jtho.2020.08.021
18. Kang M. J., Won Y. J., Lee J. J., Jung K. W., Kim H. J., Kong H. J.. **Cancer statistics in korea: Incidence, mortality, survival, and prevalence in 2019**. *Cancer Res. Treat.* (2022) **54** 330-344. DOI: 10.4143/crt.2022.128
19. Khan K. H., Yap T. A., Yan L., Cunningham D.. **Targeting the PI3K-AKT-mTOR signaling network in cancer**. *Chin. J. Cancer* (2013) **32** 253-265. DOI: 10.5732/cjc.013.10057
20. Kim Y., Vagia E., Viveiros P., Kang C. Y., Lee J. Y., Gim G.. **Overcoming acquired resistance to PD-1 inhibitor with the addition of metformin in small cell lung cancer (SCLC)**. *Cancer Immunol. Immunother.* (2021) **70** 961-965. DOI: 10.1007/s00262-020-02703-8
21. Lee Y., Joo J., Lee Y. J., Lee E. K., Park S., Kim T. S.. **Randomized phase II study of platinum-based chemotherapy plus controlled diet with or without metformin in patients with advanced non-small cell lung cancer**. *Lung Cancer* (2021) **151** 8-15. DOI: 10.1016/j.lungcan.2020.11.011
22. Li L., Han R., Xiao H., Lin C., Wang Y., Liu H.. **Metformin sensitizes EGFR-TKI-resistant human lung cancer cells**. *Clin. Cancer Res.* (2014) **20** 2714-2726. DOI: 10.1158/1078-0432.CCR-13-2613
23. Li L., Jiang L., Wang Y., Zhao Y., Zhang X. J., Wu G.. **Combination of metformin and gefitinib as first-line therapy for nondiabetic advanced NSCLC patients with EGFR mutations: A randomized, double-blind phase II trial**. *Clin. Cancer Res.* (2019) **25** 6967-6975. DOI: 10.1158/1078-0432.CCR-19-0437
24. Lin C. C., Yeh H. H., Huang W. L., Yan J. J., Lai W. W., Su W. P.. **Metformin enhances cisplatin cytotoxicity by suppressing signal transducer and activator of transcription-3 activity independently of the liver kinase B1-AMP-activated protein kinase pathway**. *Am. J. Respir. Cell Mol. Biol.* (2013) **49** 241-250. DOI: 10.1165/rcmb.2012-0244OC
25. Lin J. J., Gallagher E. J., Sigel K., Mhango G., Galsky M. D., Smith C. B.. **Survival of patients with stage IV lung cancer with diabetes treated with metformin**. *Am. J. Of Respir. And Crit. Care Med.* (2015) **191** 448-454. DOI: 10.1164/rccm.201407-1395OC
26. Luo X., Chen X., Wang L., Yang B., Cai S.. **Metformin adjunct with antineoplastic agents for the treatment of lung cancer: A meta-analysis of randomized controlled trials and observational cohort studies**. *Front. Pharmacol.* (2021) **12** 639016. DOI: 10.3389/fphar.2021.639016
27. Medairos R. A., Clark J., Holoubek S., Kubasiak J. C., Pithadia R., Hamid F.. **Metformin exposure is associated with improved progression-free survival in diabetic patients after resection for early-stage non-small cell lung cancer**. *J. Thorac. Cardiovasc Surg.* (2016) **152** 55-61. DOI: 10.1016/j.jtcvs.2016.03.094
28. Morgillo F., Sasso F. C., Della Corte C. M., Vitagliano D., D'Aiuto E., Troiani T.. **Synergistic effects of metformin treatment in combination with gefitinib, a selective EGFR tyrosine kinase inhibitor, in LKB1 wild-type NSCLC cell lines**. *Clin. Cancer Res.* (2013) **19** 3508-3519. DOI: 10.1158/1078-0432.CCR-12-2777
29. Mu W., Jiang Y., Liang G., Feng Y., Qu F.. **Metformin: A promising antidiabetic medication for cancer treatment**. *Curr. Drug Targets* (2022) **24** 41-54. DOI: 10.2174/1389450124666221104094918
30. Nguyen G. H., Murph M. M., Chang J. Y.. **Cancer stem cell radioresistance and enrichment: Where frontline radiation therapy may fail in lung and esophageal cancers**. *Cancers (Basel)* (2011) **3** 1232-1252. DOI: 10.3390/cancers3011232
31. Sayed R., Saad A. S., El Wakeel L., Elkholy E., Badary O.. **Metformin addition to chemotherapy in stage IV non-small cell lung cancer: An open label randomized controlled study**. *Asian Pac. J. cancer Prev.* (2015) **16** 6621-6626. DOI: 10.7314/apjcp.2015.16.15.6621
32. Shen X., Zhao Y., Liu G., Zhou H. L., Fan J., Zhang L.. **Upregulation of programmed death ligand 1 by liver kinase B1 and its implication in programmed death 1 blockade therapy in non-small cell lung cancer**. *Life Sci.* (2020) **256** 117923. DOI: 10.1016/j.lfs.2020.117923
33. Skinner H., Hu C., Tsakiridis T., Santana-Davila R., Lu B., Erasmus J. J.. **Addition of metformin to concurrent chemoradiation in patients with locally advanced non-small cell lung cancer: The NRG-lu001 phase 2 randomized clinical trial**. *Jama Oncol.* (2021) **7** 1324-1332. DOI: 10.1001/jamaoncol.2021.2318
34. Su V. Y., Yang K. Y., Huang T. Y., Hsu C. C., Chen Y. M., Yen J. C.. **The efficacy of first-line tyrosine kinase inhibitors combined with co-medications in Asian patients with EGFR mutation non-small cell lung cancer**. *Sci. Rep.* (2020) **10** 14965. DOI: 10.1038/s41598-020-71583-w
35. Tan A. C.. **Targeting the PI3K/Akt/mTOR pathway in non-small cell lung cancer (NSCLC)**. *Thorac. Cancer* (2020) **11** 511-518. DOI: 10.1111/1759-7714.13328
36. Tan B. X., Yao W. X., Ge J., Peng X. C., Du X. B., Zhang R.. **Prognostic influence of metformin as first-line chemotherapy for advanced nonsmall cell lung cancer in patients with type 2 diabetes**. *Cancer* (2011) **117** 5103-5111. DOI: 10.1002/cncr.26151
37. Teixeira S. F., Guimarães Idos S., Madeira K. P., Daltoé R. D., Silva I. V., Rangel L. B.. **Metformin synergistically enhances antiproliferative effects of cisplatin and etoposide in NCI-H460 human lung cancer cells**. *J. Bras. Pneumol.* (2013) **39** 644-649. DOI: 10.1590/S1806-37132013000600002
38. Tsakiridis T., Pond G. R., Wright J., Ellis P. M., Ahmed N., Abdulkarim B.. **Metformin in combination with chemoradiotherapy in locally advanced non-small cell lung cancer: The OCOG-ALMERA randomized clinical trial**. *Jama Oncol.* (2021) **7** 1333-1341. DOI: 10.1001/jamaoncol.2021.2328
39. Tseng S. C., Huang Y. C., Chen H. J., Chiu H. C., Huang Y. J., Wo T. Y.. **Metformin-mediated downregulation of p38 mitogen-activated protein kinase-dependent excision repair cross-complementing 1 decreases DNA repair capacity and sensitizes human lung cancer cells to paclitaxel**. *Biochem. Pharmacol.* (2013) **85** 583-594. DOI: 10.1016/j.bcp.2012.12.001
40. Wan G., Yu X., Chen P., Wang X., Pan D., Wang X.. **Metformin therapy associated with survival benefit in lung cancer patients with diabetes**. *Oncotarget* (2016) **7** 35437-35445. DOI: 10.18632/oncotarget.8881
41. Wang J.-L., Tsai Y.-T., Lin C.-H., Cidem A., Staniczek T., Chang G. R.-L.. **Benefits of metformin combined with pemetrexed-based platinum doublets as a first-line therapy for advanced lung adenocarcinoma patients with diabetes**. *Biomolecules* (2021) **11** 11081252. DOI: 10.3390/biom11081252
42. Wang J., Xia S., Zhu Z.. **Synergistic effect of phenformin in non-small cell lung cancer (NSCLC) ionizing radiation treatment**. *Cell Biochem. Biophys.* (2015) **71** 513-518. DOI: 10.1007/s12013-014-0283-z
43. Wang Y., Li J., Chang S., Dong Y., Che G.. **Risk and influencing factors for subsequent primary lung cancer after treatment of breast cancer: A systematic review and two meta-analyses based on four million cases**. *J. Thorac. Oncol.* (2021) **16** 1893-1908. DOI: 10.1016/j.jtho.2021.07.001
44. Wang Y., Lin B., Wu J., Zhang H., Wu B.. **Metformin inhibits the proliferation of A549/CDDP cells by activating p38 mitogen-activated protein kinase**. *Oncol. Lett.* (2014) **8** 1269-1274. DOI: 10.3892/ol.2014.2270
45. Wen-Xiu X., Xiao-Wei Z., Hai-Ying D., Ying-Hui T., Si-Si K., Xiao-Fang Z.. **Impact of metformin use on survival outcomes in non-small cell lung cancer treated with platinum**. *Med. Baltim.* (2018) **97** e13652. DOI: 10.1097/MD.0000000000013652
46. Wink K. C. J., Belderbos J. S. A., Dieleman E. M. T., Rossi M., Rasch C. R. N., Damhuis R. A. M.. **Improved progression free survival for patients with diabetes and locally advanced non-small cell lung cancer (NSCLC) using metformin during concurrent chemoradiotherapy**. *Radiotherapy And Oncol.* (2016) **118** 453-459. DOI: 10.1016/j.radonc.2016.01.012
47. Xia C., Dong X., Li H., Cao M., Sun D., He S.. **Cancer statistics in China and United States, 2022: Profiles, trends, and determinants**. *Chin. Med. J. Engl.* (2022) **135** 584-590. DOI: 10.1097/CM9.0000000000002108
48. Xiang D., Hu S., Mai T., Zhang X., Zhang L., Wang S.. **Worldwide cancer statistics of adults over 75 years old in 2019: A systematic analysis of the global burden of disease study 2019**. *BMC Public Health* (2022) **22** 1979. DOI: 10.1186/s12889-022-14412-1
49. Xiao K., Liu F., Liu J., Xu J., Wu Q., Li X.. **The effect of metformin on lung cancer risk and survival in patients with type 2 diabetes mellitus: A meta-analysis**. *J. Clin. Pharm. Ther.* (2020) **45** 783-792. DOI: 10.1111/jcpt.13167
50. Yao K., Zheng H., Li T.. **Association between metformin use and the risk, prognosis of gynecologic cancer**. *Front. Oncol.* (2022) **12** 942380. DOI: 10.3389/fonc.2022.942380
51. Yendamuri S., Barbi J., Pabla S., Petrucci C., Punnanitinont A., Nesline M.. **Body mass index influences the salutary effects of metformin on survival after lobectomy for stage I NSCLC**. *J. Thorac. Oncol.* (2019) **14** 2181-2187. DOI: 10.1016/j.jtho.2019.07.020
52. Zeng S., Gan H. X., Xu J. X., Liu J. Y.. **Metformin improves survival in lung cancer patients with type 2 diabetes mellitus: A meta-analysis**. *Med. Clin. Barc.* (2019) **152** 291-297. DOI: 10.1016/j.medcli.2018.06.026
53. Zhang F., Han S., Song W.. **Anticancer effects of metformin in experimental animal models of different types of cancer: A systematic review and meta-analysis**. *Lab. Anim. Res.* (2022) **38** 22. DOI: 10.1186/s42826-022-00131-6
54. Zhang J., Wu J., He Q., Liang W., He J.. **The prognostic value of metformin for advanced non-small cell lung cancer: A systematic review and meta-analysis**. *Transl. Lung Cancer Res.* (2018) **7** 389-396. DOI: 10.21037/tlcr.2018.03.14
55. Zhang Q., Zheng J., Wang W., Cornett E. M., Kaye A. D., Urits I.. **The anticancer effect of metformin combined with epidermal growth factor receptor tyrosine kinase inhibitors in non-small cell lung cancer patients with or without type 2 diabetes mellitus: A systematic review and meta-analysis**. *Oncol. Ther.* (2022) **10** 363-375. DOI: 10.1007/s40487-022-00209-0
56. Zhang X., Tan R., Lam W. C., Yao L., Wang X., Cheng C. W.. **PRISMA (preferred reporting Items for systematic reviews and meta-analyses) extension for Chinese herbal medicines 2020 (PRISMA-CHM 2020)**. *Am. J. Chin. Med.* (2020) **48** 1279-1313. DOI: 10.1142/S0192415X20500639
57. Zhong S., Wu Y., Yan X., Tang J., Zhao J.. **Metformin use and survival of lung cancer patients: Meta-analysis findings**. *Indian J. Cancer* (2017) **54** 63-67. DOI: 10.4103/0019-509X.219582
|
---
title: Dysbiotic lung microbial communities of neonates from allergic mothers confer
neonate responsiveness to suboptimal allergen
authors:
- Jeffery C. Bloodworth
- Aki Hoji
- Garen Wolff
- Rabindra K. Mandal
- Nathan W. Schmidt
- Jessy S. Deshane
- Casey D. Morrow
- Kirsten M. Kloepfer
- Joan M. Cook-Mills
journal: Frontiers in Allergy
year: 2023
pmcid: PMC10036811
doi: 10.3389/falgy.2023.1135412
license: CC BY 4.0
---
# Dysbiotic lung microbial communities of neonates from allergic mothers confer neonate responsiveness to suboptimal allergen
## Abstract
In humans and animals, offspring of allergic mothers have increased responsiveness to allergens. This is blocked in mice by maternal supplementation with α-tocopherol (αT). Also, adults and children with allergic asthma have airway microbiome dysbiosis with increased Proteobacteria and may have decreased Bacteroidota. It is not known whether αT alters neonate development of lung microbiome dysbiosis or whether neonate lung dysbiosis modifies development of allergy. To address this, the bronchoalveolar lavage was analyzed by 16S rRNA gene analysis (bacterial microbiome) from pups of allergic and non-allergic mothers with a basal diet or αT-supplemented diet. Before and after allergen challenge, pups of allergic mothers had dysbiosis in lung microbial composition with increased Proteobacteria and decreased Bacteroidota and this was blocked by αT supplementation. We determined whether intratracheal transfer of pup lung dysbiotic microbial communities modifies the development of allergy in recipient pups early in life. Interestingly, transfer of dysbiotic lung microbial communities from neonates of allergic mothers to neonates of non-allergic mothers was sufficient to confer responsiveness to allergen in the recipient pups. In contrast, neonates of allergic mothers were not protected from development of allergy by transfer of donor lung microbial communities from either neonates of non-allergic mothers or neonates of αT-supplemented allergic mothers. These data suggest that the dysbiotic lung microbiota is dominant and sufficient for enhanced neonate responsiveness to allergen. Importantly, infants within the INHANCE cohort with an anti-inflammatory profile of tocopherol isoforms had an altered microbiome composition compared to infants with a pro-inflammatory profile of tocopherol isoforms. These data may inform design of future studies for approaches in the prevention or intervention in asthma and allergic disease early in life.
## Introduction
Allergic asthma is the most common chronic airway disease in children. With the increase in prevalence of allergic diseases, approaches to limit development of allergy early in life are needed (1–3). A maternal history of allergic disease remains the greatest risk factor for development of allergies and asthma in offspring [4]. Although mothers and their children occupy the same homes with the same pollutants and environmental contaminants and allergens, when controlling for these factors, there is increased sensitivity to development of allergy in the offspring of allergic mothers. Moreover, maternal transmission of reactivity to allergen in the offspring is not specific for the type of allergen in patients and animal models (5–11). Consistent with this non-specificity for type of allergen, transfer of splenic dendritic cells (DCs), but not macrophages, from neonatal mice of allergic mothers transfers allergen responsiveness to recipient neonates from non-allergic mothers [4, 12]. Neonates from allergic mothers have increased lung DC subsets, including monocyte-derived DCs (mDCs) and resident DCs (rDCs) but no change in regulatory DCs (pDCs or CD103 + DCs) [13, 14]. The development of responsiveness to allergen results from complex interactions with environmental factors, including allergens and dietary lipids [13, 14].
We have demonstrated that responsiveness to allergen by neonates of allergic mothers is modifiable by the dietary lipids α-tocopherol (αT) and γT in maternal diets [13, 14]. In adult mice, we demonstrated that γ-T elevates allergic responses to chicken egg ovalbumin (OVA) [15, 16] and to house dust mite extract (HDM) [17] and that γT potently ablates the anti-inflammatory benefit of α-T during allergic responses [15, 16]. We also demonstrated that dietary supplementation of allergic mothers during pregnancy and nursing with αT inhibited, whereas γT increased development of allergen (OVA) responsiveness and DC numbers in the offspring of allergic mothers [13, 14]. During allergic inflammation, tocopherols function as anti-oxidants and regulate signal transduction during DC differentiation and leukocyte transendothelial migration (18–20). We demonstrated that tocopherols regulate cell signal transduction by binding to a regulatory domain of protein kinase C; when bound to PKC, αT is an antagonist and γT is an agonist of PKCα (15, 18, 20–23).
In humans, we demonstrated that plasma with low αT & >10 µM γT associates with lower lung function in children [24] and in adults [25]. Based on the prevalence of serum γ-T > 10 µM in adults in the USA and adults in the 2011 USA census, up to 4.5 million U.S. adults had >10 µM serum γ-T and may have had 500 ml lower FEV1 and FVC [25]. We also demonstrated that γ-T associated with increased odds for asthma in China [17]. It is reported that patients with asthma or allergy have low levels of α-tocopherol (26–29), suggesting that an increase in α-tocopherol may be necessary, in combination with other regimens, to decrease allergic disease. We also reported that higher plasma αT and lower γT concentrations in children at 3 years of age associate with better lung function at 6–19 years of age in project VIVA [24]. Together the preclinical studies in mice and humans indicate that a higher plasma γT concentration is a pro-inflammatory tocopherol profile and a higher plasma αT with lower γT concentration is an anti-inflammatory tocopherol profile with regards to allergic inflammation and lung function. Thus, the preclinical studies and data for clinical associations suggest that tocopherols may modify mediators that regulate the development of allergic disease.
It is suggested that risk for allergic disease in humans is associated with in utero and early exposures to environmental factors [30]. Microbiota are acquired from the environment in utero and as neonates. Since the advent of non-culture-dependent microbe characterization, the lungs have been known to harbor commensal and pathogenic microbes [31]. The airways are colonized by a diverse range of bacterial, archaeal, protozoal and fungal microorganisms known collectively as the airway microbiome. Microbes occupy the airways during health and disease, but the abundance and diversity of microbes is altered during lung diseases including asthma. Briefly, microbes of the *Proteobacterium phylum* are elevated during allergic asthma, including members of the Streptococcus, Haemophilus, and *Moraxella* genera (32–35). Streptococcus colonization during early life is a strong predictor of allergic asthma development [35]. Alterations in the microbiota have a profound association with allergic asthma development during childhood [35]. It is not known whether αT supplementation modifies the lung microbiome in offspring of allergic mothers. Moreover, it is not known whether lung microbiome of offspring of allergic mothers affects the development of responsiveness to allergen in the offspring lung.
We report that in mice, the lung microbiome of offspring of allergic mothers is altered before allergen challenge of the neonate. This is blocked by maternal supplementation with αT. Moreover, transfer of the lung microbial communities of offspring of allergic mothers to offspring of non-allergic mothers confer responsiveness to allergen. In infants, a pro-inflammatory profile of tocopherol isoforms associates with an altered airway microbiome.
## Animals
Adult C57BL/6 female and male mice were from Jackson Laboratory, Bar Harbor, Maine and maintained under specific pathogen free (SPF) conditions at Indiana University Lab Animal Resource Center. C57BL/6 mice are used in this study because C57BL/6 mice have been vital for our studies of mediators that regulate the development of allergic responses by offspring of allergic mothers [13, 14, 36]. The studies were approved by the Indiana University Institutional Review Committee for animals.
## Tocopherol and basal diets
αT is necessary for mouse and human placental development [37, 38]. Standard basal mouse chow diet contains 45 mg αT/kg of diet and 45 mg γT/kg of diet and supports fetal development in mice. Supplemented αT diets contain 250 mg αT/kg of diet and 45 mg γT/kg of diet [13, 14]. Translation of mouse basal α-T doses to humans is calculated as we previously described (page 173 of 39). Taking into account differences in food consumption and metabolism [39], a 45 mg αT/kg of diet for mice is 57 mg αT/day for human adults. For healthy adult humans, 15 mg αT/day is recommended, but asthmatics have low plasma αT (26–29). The 250 mg αT/kg of diet for mice is 285 mg αT/day for human adults, which is well below upper safety limits of 1,000 mg αT/day in human pregnancy and near the 268 mg (400IU) d-αT in pre-eclampsia pregnancy trials (40–45). A relevant dose is a dose that achieves similar fold changes in tissues in mice and humans.
For supplementation of diets with tocopherol, D-α-tocopherol (>$98\%$ pure) from Sigma was sent to Dyets, Inc (Bethlehem, PA) to produce the diets with 250 mg αT/kg of diet (catalog#103373) [13, 46]. The purity of these tocopherols that were used to make the diets and the tocopherol concentrations in the diets were confirmed by HPLC with electrochemical detection as previously described [13, 46]. These αT supplemented diets increase tocopherols 3-fold in mothers and pups (13, 14, 47–49). This is similar to fold tissue changes achievable in humans (15, 16, 21–23, 46).
## Allergens
In our studies, pups received the same allergen or different allergen than the mother because in mice (5–11) and humans [4], allergen responses by offspring are not specific to the allergen to which the mother had been exposed. We induced allergic lung inflammation in mothers or pups with OVA or HDM. OVA is a well-characterized model purified allergen that, in humans, can also be inhaled when exposed to powered egg or when gasping during egg allergic reactions. HDM extract is a model environmental allergen from Greer and has been used in allergy shot induction of tolerance in humans. Mothers with allergic responses and allergic inflammation at the time of mating [15, 46, 50] are mated to non-allergic fathers [13, 14]. On gestational day 18 (GD18) [during time of fetal hematopoiesis], we collect mother plasma, placentas, and fetal livers (site of hematopoiesis in the fetus) [13, 14]. To assess offspring development of allergy, pups receive a suboptimal allergen sensitization/challenge protocol (Figure 1) [13, 14]. There are no differences by sex so data include both sexes [13, 14].
**Figure 1:** *Enhanced responsiveness to challenge with HDM or OVA by pups of mothers with allergy was inhibited by maternal supplementation with α-tocopherol. The allergen of the mother and offspring can differ. (A,C) Allergic and non-allergic mothers received basal diet or diet supplemented with αT (250 mg αT/kg of diet) during pregnancy and nursing. Timeline for allergen-sensitization and allergen-challenge of mothers and offspring. (B,D) Pup BAL eosinophils, monocytes, lymphocytes and neutrophils. (E) Relative IL-5 mRNA expression in lungs of HDM-challenged pups of allergic and non-allergic mothers with basal or αT-supplemented diets. BAL, bronchoalveolar lavage. n = 8–10 mice/group. Saline treated pups did not have allergic inflammation (data not shown). *p < 0.05.*
## Separation of microbiota from mouse BAL
For lung microbiota transfers and microbiota taxa analyses, the microbes were separated from the BAL by differential centrifugation as follows: the BAL was centrifuged for 10 min at 1200 rpm to pellet host cells and supernatant collected. This supernatant was centrifuged at 10 min at 15,000 rpm (500×g) to pellet the microbiome. The supernatant, which contains soluble proteins and mucins [51], was removed and pellet contains the microbiota. The microbiota were 96 ± $0.7\%$ viable as determined by the Biotium Bacterial Viability and Gram Stain Kit with analyses by flow cytometry. Also for visualization of microbiota, separated BAL microbial communities from allergen-challenged pups of allergic and from non-allergic mothers were suspended in a minimal volume, fixed in a small spot on a glass slide, stained with a gram stain for bacteria, and analyzed by microscopy.
## OVA administration, lung microbiota transfers and analysis of inflammation
C57BL/6 female mice were maintained on chow diet. The mice were sensitized by intraperitoneal injection (200 μl) of OVA grade V (Sigma-Aldrich Co.) (5 μg)/alum (1 mg) or saline/alum (1 mg) on days 0 and 7 [13, 14]. The mice were exposed to nebulized saline or $3\%$ (w/v) OVA in saline for 15 min on 3 consecutive days at 8, 12, and 16 weeks of age and then mated. The pregnant and nursing dams received basal diet (45 mg αT/kg of chow) or αT-supplemented diet (250 mg αT/kg of chow) as indicated in the figures.
In experiments with transfer of lung microbial communities, BAL was collected from neonates at PND5. The BAL microbial communities were separated by differential centrifugation For transfer of microbial communities, the pelleted donor microbiota were suspended in saline and immediately administered to lungs of recipient pups to most closely represent the donor microbiome levels and abundance. The few hours between collection of BAL microbiota from donor pups to administration to recipient pups was within the timeframe described for survival of aerobic microbes, as these aerobic microbes can survive for days in PBS [52]. The BAL microbial communities of 2 donors were combined for each recipient and administered in 10 µl intranasally to each PND4 recipient pup for adequate inoculation, similarly to studies by others with viral inoculation of neonates [53].
For allergen challenge of the pups, six-day old pups were sub-optimally sensitized by treating with only one 50 µl i.p. injection (rather than two injections) of 5 µg OVA/1 mg alum [13, 14]. At 13, 14, and 15 days old, the pups were challenged for 15 min with $3\%$ OVA. At 16 days old, the pups were weighed, euthanized and tissues collected. Pup bronchoalveolar lavage (BAL) cells were stained and counted as previously described [13, 14]. OVA-specific IgE was determined by ELISA as previously described [13, 14].
## 16S rRNA gene analysis of mouse BAL
For microbiome analyses, the microbes were separated from the BAL by differential centrifugation. To limit confounding contributions from contaminant bacteria during collection and sequencing reagents [54, 55], data from the BAL microbiome of pups of allergic mothers were compared to control BAL groups and $$n = 8$$–10 pups from 3 to 4 mothers per group. The same sterile reagents were used within an experiment. 16S rRNA gene amplicons were generated via PCR amplification using primers 5′TATGGTAATTGTGTGCCAGCMGCCGCGGTAA3′ and 5′AGTCAGTCAGCCGGACTACHVGGGTWGCTAAT3′ [56]. The full sequencing protocol is published by Kumar et al. [ 56] The 16S rRNA gene sequencing was performed on the Illumina MiSeq platform. The ASV table for the mouse microbiome studies was generated by an analysis pipeline using CLC Genomic Workbench Microbial Module (CLCGW-MM). This includes the preprocessing of V4 16S amplicon (250 bp) reads, mapping to SILVA 16S v.132 SSURef, and filtering of initial ASVs (relative abundance > 1 × 10−5) as described in the University of Alabama at Birmingham protocol [56, 57]. The nomenclature for Bacteroidetes has recently been updated to Bacteroidota [58], therefore we are using these synonymously in this manuscript. Differential abundant analysis was done by a built-in function in the CLCGW-MM, which generated FDR and log2fold differences in the taxa between two comparison groups. Data are shown as % abundance of 16S rRNA gene amplicon counts of total counts within specific taxonomic levels. Alpha and beta diversity analyses were performed using Quantitative Insights into Microbial Ecology v2 2022.8 [59, 60]. Principal component analysis was performed using EMPeror [61].
## qPCR analysis of cytokines and chemokines
Total RNA was isolated from 50 to 100 mg lung tissue using the QIAGEN RNeasy Mini Kit (catalog #74136). cDNA was prepared using a MMLV Reverse Transcriptase kit (QuantaBio, catalog #95047) and analyzed by PCR on an ABI 7300 Thermal Cycler (Applied Biosystems). Taqman probes and Taqman Universal Master Mix were used as directed (Applied Biosystems, catalog #4304437). Taqmanprobes used were GAPDH (catalog #4331182) and MUC5AC (catalog#4331182). IL-5, IL-13, IL-33, and CCL11 expression levels were quantified using SsoAdvanced Universal SYBR Green (Biorad catalog# 1725271) with the following primers obtained from Integrated DNA Technologies. TargetForward primer sequenceReverse primer sequenceCCL11TGTAGCTCTTCAGTAGTGTGTTGCTTCTATTCCTGCTGCTCACGGAPDHGTGGAGTCATACTGGAACATGTAGAATGGTGAAGGTCGGTGTGIL-33AATCACGGCAGAATCATCGAGAAAGGAGCCAGAGGATCTCCGATTIL-13CCAGGGCTACACAGAACCCGGCTCTTGCTTGCCTTGGTGGIL-5ACTGTCCGTGGGGGTACTGTCCTCGCACACTTCTCTTTTTGG
## INHANCE cohort
INHANCE cohort [62, 63] is an urban cohort (birth to 18 months of age) in the *Indianapolis area* ($$n = 180$$, $70\%$ Black or mixed-race Black; NIH K23 AI135094-01 PI Kloepfer). Of these 180 infants, 43 of the 3–5 months of age infants and 50 of the 12–18 months of age infants had nasal 16S microbiome data [62, 63] and sufficient plasma volume available for tocopherol analyses by HPLC.Human 16S rRNA gene sequencing data were analyzed using Quantitative Insights into Microbial Ecology v2 2022.8 [59, 60]. Human sequences were aligned with the SILVA 138.1 taxonomy database [57]. Also serum tocopherol concentrations at 3–5 months or 12–18 months of age were measured by HPLC with electrochemical detection [64] as previously described [24]. Because we have demonstrated that in children and adults that better lung function associates with increasing αT when gamma-tocopherol (γT) concentrations are lower (24, 25, 65–67), the INHANCE cohort infants were placed in groups based on below or above median αT and median γT concentrations (64 and manuscript in preparation) withQ1 (high γT, low αT), Q2 (high γT, high αT), Q3 (low γT, low αT) and Q4 (low γT, high αT). Q4 has an anti-inflammatory profile of tocopherols and Q2 has a pro-inflammatory profile for tocopherol isoforms for allergic lung inflammation and lung function as in our previous reports in children and adults [13, 14, 24, 25, 66, 68].
## Data availability
The raw fastq files of the 16S rRNA analysis from mouse BAL in (Figures 2–5) are deposited as NCBI BioProject repository, accession number ID PRJNA925891. The raw fastq files of the 16S rRNA analysis in INHANCE cohort in (Figure 8) are deposited as NCBI BioProject repository, accession number ID PRJNA928382.
**Figure 2:** *Pups of allergic mothers have altered lung bacteria microbial composition. Mouse treatments were as in (Figure 1A). BAL microbiota from pups at (A) PND4 and (B,C) 24 h after OVA-challenge (PND16) was separated and analyzed by 16S rRNA gene sequencing and a microbiome analysis pipeline. (A) At PND4, before allergen exposure, there was increased Proteobacteria and decreased Bacteroidota in the lungs of offspring of allergic mothers (log2FC > 0.6, FDR < 0.1) as compared to offspring of non-allergic mothers with basal diet. (B) ASV table of the relative abundance of phyla within the total pup BAL bacterial microbiome. *p < 0.05 compared to other groups. (C) BAL microbiota from pup BAL PND16 were separated and concentrated by differential centrifugation as in the methods, suspended in minimal volume for fixation in a small spot on a glass slide and stained with gram stain from bacteria. Representative images of lung microbiota are shown.*
## Statistics
Data in the figures were analyzed by a one-way ANOVA followed by Tukey's or Dunnett's multiple comparisons test (JMP software, SAS Institute). Data in figures are presented as the means ± the standard errors. Data include both genders because there were no differences in outcomes by gender (data not shown). For analyses of the 16S microbiome from INHANCE cohort infants 3–5 months and 12–18 months of age, cutoffs were set for the data, including removal of 3 samples with insufficient ASV detection (<$4\%$ total ASV reads/sample compared to other samples), removal of ASV's that had less than $\frac{6}{43}$ samples with reads, and to address extension of findings in mice to human, phyla were included in analyses for phyla observed in the mouse models. These participants were placed in 4 groups based on the median αT and γT concentrations [64]. Then an abundance cutoff was set at >$0.003\%$ abundance for the sum of the ASV averages for the groups. This yielded 181 ASV for the 3–5 months age and 217 ASV for the 12–18 months age. Based on predetermined results in mice with αT supplementation and low γT, analyses were made in comparison to the group Q4 which had a serum αT concentration above the median and a serum γT below the median concentration. There was no formal adjustment for multiple testing because the analyses were selected based on preclinical mechanistic microbiome outcomes. Furthermore, the associations tested were established a priori at the onset of the project with microbiome as the primary analysis with tocopherol isoforms.
## Enhanced responsiveness to challenge with HDM or OVA by pups of mothers allergic to OVA is inhibited by dietary supplementation of the mother with α-tocopherol
Pups of allergic mothers respond to suboptimal OVA sensitization with allergen and this allergen responsiveness of the offspring is reduced by dietary supplementation of the mother with αT during pregnancy and nursing [13, 14]. It is not known whether the pups of allergic mothers also respond to HDM and whether this is modified by αT. Pups of OVA-challenged mothers were responsive to suboptimal sensitization and challenge with OVA (Figures 1A,B) or HDM (Figures 1C,D) with increased numbers of leukocytes (Figures 1B,D) and this was blocked by maternal dietary supplementation with α-tocopherol as compared to pups from allergic mothers with a basal diet (Figures 1B,D). OVA increased numbers of eosinophils, monocytes and lymphocytes in the BAL (Figures 1B,D), OVA-specific IgE [13, 14], and lung cytokines [13, 14] in pups of allergic mothers as compared to pups of non-allergic mothers. HDM challenge of pups of allergic mothers (Figure 1C) also increased numbers of eosinophils, monocytes and lymphocytes in the BAL (Figure 1D) and lung IL-5 expression (Figure 1E). It has been demonstrated that pups of allergic mothers that are challenged with saline do not have allergic lung inflammation and that BAL cell numbers are similar to allergen-challenged pups of non-allergic mothers.[5] There were no sex differences in pup weight or eosinophilia as we previously reported [13, 14], so data include both sexes.
## Pups of allergic mothers but not pups of non-allergic mothers exhibited lung bacterial microbiome dysbiosis
The airway microbiome is altered in adult humans with allergic asthma and in adult mice with allergic lung inflammation [69]. This airway microbiome dysbiosis in adults has an increased abundance of Proteobacteria and decreased Bacteroidota (69–71). It is not known whether the lung microbiome is altered in pups of allergic mothers. It is also not known whether the lung microbiome plays a role in regulation of airway response to allergen. Interestingly, at PND4 before pup exposure to allergen, the BAL of pups of allergic mothers with a basal diet had increased abundance of Proteobacteria and decreased abundance of Bacteroidota (log2FC > 0.6, FDR < 0.1) as compared to pups of non-allergic mothers with a basal diet (Figure 2A). To assess whether allergen alters the bacterial microbiome of pups, the pups of allergic mothers and pups of non-allergic mothers were challenged with a purified allergen. OVA was used as a purified protein allergen, thereby avoiding contaminant bacterial 16S in the extracts from HDM. The microbiota, that was separated from the BAL of allergen-challenged PND16 pups, contained gram negative and gram positive microbiota as determined by gram-staining of BAL bacteria fixed to glass slides (Figure 2C). The 16S analyses of the PND16 BAL microbiota of allergen-challenged pups of allergic mothers demonstrated an increase in abundance of Bacteriodota and decrease in abundance of Proteobacteria (Figure 2B). There were increases in several taxa within the Proteobacteria, Firmicutes, Fusobacteria and Verrucomicrobia, but decreases in taxa with the Bacteroidota and several other Firmicutes, Proteobacteria and Archaea as compared to pups of non-allergic mothers and as compared to pups of allergic mothers supplemented with αT (Figure 3).
**Figure 3:** *Pups of allergic mothers have altered bacteria microbiome. Mouse treatments were as in (Figure 1A). Pup BAL microbiota were separated and analyzed by 16S rRNA gene sequencing at PND16. Shown are the % abundance for pup BAL bacteria with a significant difference in the OVA/Basal group compared to the other groups. *p < 0.05.*
## Transfer of the BAL microbial community of pups of allergic mothers to pups of non-allergic mothers sustained the donor microbiome in the recipient pups
Neonate bacterial load increases over PND0-14 [72]. To address a potential function for the microbiome dysbiosis in allergen responsiveness, the microbial community was obtained from the BAL of PND4 pups without allergen exposure. The donor PND4 BAL microbial community was separated from the BAL and transferred intranasally to recipient PND4 pups. Then, the PND4 pups without donor microbiota and the PND4 pups that received the microbiota transfers were challenged with allergen (Figure 4A). For the transfers, the pup groups are designated as maternal treatment of the donor pups → maternal treatment of the recipient pups. The donor sample 16S microbiome had increased Proteobacteria and decreased *Bacteroidota taxa* in the lungs of offspring of allergic mothers (log2FC > 0.6, FDR < 0.1) (Figure 2A).
**Figure 4:** *After intranasal microbiome transfers and airway allergen challenge, there was pup BAL microbiota with increased Proteobacteria and decreased Bacteroidota taxa for pups that were either recipient pups of mothers in the OVA,basal group or were pups receiving microbiome from pups of mothers in the OVA,basal group. (A) Timeline for treatment of mothers and pups. (B) Donor BAL microbiome was administered intranasally in 10 µl to PND4 recipient pups (as indicated in figures as the group of pups providing donor BAL microbiome for transfer into a recipient group of pups, i.e., donor → recipient group). Yellow arrows on the x-axis are those groups with donor and recipients within the same group. In RED BOX are groups with recipient or donor microbiota of PND16 pups of allergic mothers (OVA/basal). Blue arrows within panel B indicate that Bacteroidota are decreased and Gamma-Proteobacteria are increased in groups in red box. N = 8/group. In panels B,C only, the OVA was in 0.09% saline; nevertheless, it did not alter the fold effect on BAL cell inflammation which is included in (Figure 5) with data from 7 microbiome transfer experiments. (C) In RED BOX are recipient or donor microbiota of PND16 pups from allergic mothers (OVA/basal). Data are presented as percent abundance of bacteria taxa. *, p < 0.05 as compared to Saline,basal → Saline,basal group (yellow arrow in graphs in C). Sal/B, saline-treated mother with basal diet. Sal/αT, saline-treated mother with αT-supplemented diet. OVA/B, OVA allergen-treated mother with a basal diet. OVA/αT, OVA allergen-treated mother with αT-supplemented diet.*
The PND4 donor microbiota were also analyzed for alpha-diversity and beta-diversity. The PND4 donor microbiota groups had a similar Shannon within-group alpha-diversity index; the Shannon Index incorporates total number of bacterial species and relative differences in the abundance of various species in the microbiota community of a group (Supplementary Figure S1A). For beta-diversity analysis, the donor groups did not separate in the Principal Component Analysis (PCA) of the Unweighted Unifrac and Weighted Unifrac between-group beta-diversity analysis of bacterial microbiota (Supplementary Figures S2B,C, left panels); the weighted-*Unifrac analysis* incorporates only the relative abundance of taxa shared between samples and the unweighted-*Unifrac analysis* incorporates only the presence/absence of taxa between groups. In contrast, when incorporating both overall abundance per sample and abundance of each taxa of the microbiota communities by the Bray-Curtis beta-diversity distance analyses, there was clustering by PCA for the donor saline groups and for the donor OVA groups, which was unaffected by αT (Supplementary Figure S2A, left panel).
**Figure 5:** *Recipient or donor microbiota from pups of allergic mothers (OVA,basal) conferred responsiveness to allergen in the recipient pups (red box). Mice were treated as in timeline in (Figure 4A) BAL (A) eosinophils, (B) monocytes, (C) lymphocytes, and (D) neutrophils are presented as mean ± SEM. Data are from 7 experiments. N = 10–36/group. Sal/B, saline-treated mother with basal diet. Sal/αT, saline-treated mother with αT-supplemented diet. OVA/B, OVA allergen-treated mother with a basal diet. OVA/αT, OVA allergen-treated mother with αT-supplemented diet. *p < 0.05 as compared to the saline,basal → saline,basal group., +p < 0.1 as compared to no donor → Saline/basal group.*
Alpha-diversity was also assessed for the PND16 pup microbiota from the BAL of allergen-challenged pups with and without microbiota transfer. Without the microbiota transfers, the BAL of allergen-challenged PND16 pups from allergic mothers (the no donor → OVA/B group in Supplementary Figure S1B) had decreased alpha-diversity as compared to the saline groups (the no donor → Sal/B group and the no donor → Sal/αT group) (Supplementary Figure S1B). With the microbiota transfers, the BAL of allergen-challenged PND16 pups of allergic mothers (OVA) as either donor or recipients (designated as microbiota donor → recipient pairs of pups) had reduced alpha-diversity (Supplementary Figure S1B, red box) as compared to several control groups, including the no donor → Sal/B, the Sal/B → Sal/B, the no donor → Sal/αT or the OVA/αT → OVA/αT (Supplementary Figure S1B).
Beta-diversity was assessed for the PND16 pup microbiota from the BAL of allergen-challenged pups with and without microbiota transfer. There was minimal separation of the PND16 groups in the PCA plot of the Unweighted Unifrac and Weighted Unifrac between-group beta-diversity bacterial microbiota analyses (Supplementary Figures S2B,C, right panels). In the PCA plot of the Bray-Curtis beta-diversity distance analyses of the allergen-challenged pups without donor microbiota transfers (Supplementary Figure S2A, right panel with cone-shaped symbols), there was some separation of the no donor → OVA/B group as compared to the other no donor groups. In the PCA plot of the Bray-Curtis beta-diversity distance analyses of the allergen-challenged pups that received donor microbiota transfers, there was unique clustering of microbiota from pups of allergic mothers with basal diet (OVA/B) as either donors or recipients (Supplementary Figure S2A, right panel with sphere-shaped symbols); these are the groups with allergic inflammation in Figure 5.
Notably, when either the BAL microbial community of the donor pup or the recipient pup was from an allergic mother with basal diet (OVA/B), the recipient pup BAL had an increase in abundance of the class Gamma-proteobacteria and decrease in abundance of the class Bacteroidia (Figure 4B), as compared to the saline/B → saline/B group of pups (Figure 4B). In Figure 4C, when the BAL microbial community was from a group with a donor pup or the recipient pup from an allergic mother with basal diet (OVA/B), there was an increase in a Proteobacteria and a Fusobacteria and a decrease in several *Bacteriodota taxa* and a Firmicute. These data suggest that the BAL microbial community of the pups of allergic mothers with basal diet was dominant.
## Transfer of the dysbiotic BAL microbial community of pups of allergic mothers to pups of non-allergic mothers conferred enhanced responsiveness to allergen in the recipient pups, demonstrating a functional role for the lung microbiome
The BAL cells were assessed for the pups in (Figure 4). Without microbiota transfers, the pups of allergic mothers had increases in BAL eosinophils, monocytes and lymphocytes (Figure 5), no-donor groups). After intranasal microbiome transfers and airway allergen challenge, there were increased numbers of BAL eosinophils, monocytes and lymphocytes in the pups that were either recipient pups of mothers in the OVA,basal group or were pups receiving microbiome from pups of mothers in the OVA,basal group (i.e. OVA/B as donor or recipient) as compared with the pups of the control saline/B → saline/B group (Figure 5). The donor → recipient pup groups without an OVA/B group in the donor or the recipients did not develop lung eosinophilia after allergen exposure (Figure 5). Interestingly, the mice were in a specific-pathogen-free facility and dysbiosis of the transferred microbial communities was sustained in recipient pups (Figure 4) and only exhibited in pups with allergic lung responses (Figure 5). These novel transfer studies demonstrate that the dysbiotic microbiome of pups of allergic mothers enhances pup responsiveness to allergen.
## The transferred BAL microbial community influenced induction of allergen-specific IgE and the allergen-induced expression of cytokines
Mediators of allergic inflammation were measured including serum allergen-specific antibodies, the chemokines and cytokines that mediate eosinophilia, and the mucin Muc5ac. We have reported that anti-OVA IgE is increased in the OVA/B group and this is reduced by OVA/αT [13]. The serum of pups in the OVA/B → OVA/B group and OVA/B → Sal/B group had elevated anti-OVA IgE after allergen exposure (Figure 6), suggesting that the transfer of microbial communities of pups of allergic mothers with basal diet is sufficient to mediate enhanced induction of anti-OVA IgE in these pups. In contrast, there were no increases in anti-OVA IgG2b and anti-OVA IgG1 (Figure 6). The chemokine CCL11, which mediates recruitment of eosinophils, and IL-33, which is important in induction of allergic inflammation, was increased in the groups with OVA/B as donor or recipient and in the no donor → OVA/B group (Figure 7). Similarly, IL-5 and IL-13 had a significant increase in the no donor → OVA/B group and had either a trend or significant increase in most of the microbiota transfer groups with pups of OVA/B-treated moms that were either the donor or recipient of the microbe transfers (Figure 7). Muc5ac was increased in several groups that had OVA/B as donor or recipient (Figure 7). The pups with transfers of microbial communities from pups of saline-treated mothers did not have an increase in CCL11, IL-13, IL-5 or Muc5ac (Figure 7). There was also no increase in IL-33 for the recipient pups with saline-treated mothers, except a small increase for the OVA/αT → saline/B group (Figure 7). Thus, transfer of microbial communities with the OVA/B group as the donor or recipient regulated these mediators of allergic inflammation.
**Figure 6:** *Recipient or donor microbiota from pups of allergic mothers (OVA,basal) conferred allergen sensitization with increased IgE but not increased IgG2b or IgG1 (red box). Mice were treated as in timeline in (Figure 4A). Serum (A) anti-OVA IgE, (B) anti-OVA IgG2b, and (C) anti-OVA IgG1 as determined by ELISA. Data are presented as mean ± SEM. N = 6–9/group. Sal/B, saline-treated mother with basal diet. Sal/αT, saline-treated mother with αT-supplemented diet. OVA/B, OVA allergen-treated mother with a basal diet. OVA/αT, OVA allergen-treated mother with αT-supplemented diet. *p < 0.05 as compared to the saline,basal → saline,basal group. +p < 0.1 as compared to no donor → Saline/basal group.* **Figure 7:** *Recipient or donor microbiome from pups of allergic mothers (OVA,basal) conferred allergen-induced increases in CCL11, IL-13, IL-5, IL-33, and Muc5ac(red box). Mice were treated as in timeline in (Figure 4A). Lung cytokine expression was determined by qPCR. (A) CCL11. (B) IL-13. (C) IL-5. (D) IL-33. (E) Muc5ac. N = 6–9/group. Data are presented as mean ± SEM. N = 6–9/group. Sal/B, saline-treated mother with basal diet. Sal/αT, saline-treated mother with αT-supplemented diet. OVA/B, OVA allergen-treated mother with a basal diet. OVA/αT, OVA allergen-treated mother with αT-supplemented diet. *p < 0.05 as compared to the saline,basal → saline,basal group. **p < 0.05 as compared to saline/αT → saline/αT group.*
## A human infant plasma pro-inflammatory tocopherol isoform profile associated with altered lung microbiome
We have demonstrated that in children and adults that better lung function associates with increasing αT concentrations when the gamma-tocopherol (γT) concentration is lower (24, 25, 65–67). To extend our microbiota studies in mice to humans, it was determined whether infants within the INHANCE cohort with an anti-inflammatory tocopherol isoform profile (high αT with low γT levels) had an altered microbiota composition compared to infants with a pro-inflammatory tocopherol isoform profile (high γT levels). To assess infant microbiome associations with an anti-inflammatory tocopherol profile, the INHANCE cohort infants that had 16S microbiota data and sufficient plasma volume for tocopherol analysis were placed in groups based on median αT and γT concentrations [64]. The median serum tocopherol concentrations at 3–5 months of age were 28 µM αT and 2.6 µM γT and at 12–18 months of age were 19 µM αT and 2 µM γT (Table 1). The higher medians for 3–5 months infants are consistent with increased tocopherol concentrations during pregnancy [24] that will influence early life tocopherol concentrations in infants. The four groups are Q1 (high γT, low αT), Q2 (high γT, high αT), Q3 (low γT, low αT) and Q4 (low γT, high αT) were defined using the median serum tocopherol concentrations (Table 1). Thus, the microbiome of groups Q1, Q2 and Q3 groups were compared to Q4 because Q4 had the anti-inflammatory profile (low γT, high αT) for allergic lung inflammation, lung function and wheeze [13, 14, 24, 25, 66, 68] and had the highest lung function [64]. To examine the associations of αT without elevated γT, as this was the condition in the mouse studies in (Figures 1–7), Q4 was compared to Q3. In infants 3–5 months of age, there was a significance or trend for higher % abundance in some Firmicutes and *Bacteroidota taxa* in Q4 compared to Q3 (Figure 8A). As infants, the airway microbiome matures from birth to 1 year of life (73–75). In INHANCE infants at 12–18 months of age, there was significantly lower % abundance in a Firmicute in Q4 compared to Q3 (Figure 8B). Moreover, for the group Q2, which has a pro-inflammatory tocopherol isoform profile with allergic lung inflammation and function [13, 14, 24, 25, 66, 68], there was a significantly higher % abundance in taxa of a Firmicute and an Proteobacteria (Figure 8B). These data suggest that tocopherol profiles associate with altered microbiome abundance of several taxa in infants.
**Figure 8:** *Infants with an anti-inflammatory tocopherol isoform profile (higher αT, lower γT) for allergic responses had a different abundance of bacterial microbiota compared to other tocopherol isoform profiles. Serum αT and γT for infants in the INHANCE cohort were measured by HPLC. Four groups of infants (Q1, Q2, Q3, Q4) for 3–5 months and for 12–18 months infants were generated using high and low αT and γT concentrations in (Table 1) that were defined as higher or lower than the median concentration for the tocopherol isoform for the age group. (A) 3–5 months and (B) 12–18 months of life. p values are given for significant differences or trends in taxa compared to group Q4.* TABLE_PLACEHOLDER:Table 1
## Discussion
We report that pups of allergic mothers had allergic lung inflammation and lung microbial community dysbiosis with increased Proteobacteria and decreased Bacteroidota before and after allergen sensitization. This indicated a sustained dysbiosis in the pups of allergic mothers. *The* generation of lung microbe community dysbiosis was blocked by supplementation of the mothers with αT during gestation and nursing. We also demonstrated a functional effect of the microbiome dysbiosis. Fascinatingly, in studies with BAL microbe community transfers, the lung microbiome dysbiosis of neonates of allergic mothers mediated enhanced neonate responsiveness to allergen with increases in eosinophils, monocytes, lymphocytes, CCL11, IL-5, IL-13, IL-33 in the lungs of recipient pups. There was also increased serum anti-OVA IgE. In contrast, neonates of allergic mothers were not protected from development of allergy by the transfer of non-dysbiotic microbial communities from either neonates of non-allergic mothers or neonates of αT-supplemented allergic mothers. These data suggest that the dysbiotic lung microbiome is dominant and sufficient for enhanced neonate responsiveness to allergen. Furthermore, human infants with an anti-inflammatory tocopherol profile compared to a pro-inflammatory tocopherol profile had an altered abundance of several Proteobacteria, Firmicutes and Bacteriodota taxa.
In humans, the onset of atopy and asthma correlates with home environment, viral infections, and antibiotic exposures [76]. It has been suggested that the diversity of overall bacterial environmental exposures rather than any one exposure may be contribute to immune skewing and allergen responses [77]. Human adults with allergic asthma have airway microbe community dysbiosis with decreased Bacteroides and increased Firmicutes and Proteobacteria (78–81). This includes an increase in the Proteobacteria Haemophilus in asthmatics [82]. The increase in abundance of Proteobacteria, including Haemophilus and Moraxella, is also observed in adult patients with neutrophilic asthma and the abundance of these microbiota was associated with asthma severity [83]. In another study of induced sputum from patients with severe asthma, an enrichment of Moraxella, Haemophilus, or Streptococcea associated with severe airway obstruction and airway neutrophilia [84]. Because multiple bacterial taxa associate with regulation of allergic or neutrophilic asthma, it suggests that individual taxa of the bacterial microbial community in the lung may be less important than the interactions of bacteria in general or shared features of the bacteria. A similar airway microbial community is present in infants with wheeze. In mice and human infants, the nasal and lung airway differ in bacteria species, but are similar at the family level [85], suggesting that there may be a similar regulation or function of the species in allergic airway inflammation. It has been reported that PND6 neonatal mice have a predominance of Firmicutes and gamma-Proteobacteria in the lung but then as adults, mice acquire an increase in abundance of Bacteroidota [72]. In IL13-transgenic adult mice with allergic lung inflammation, there is increased Proteobacteria and decreased Bacteroides in airways [69]. Similarly, in infants, hypopharyngeal Streptococcus pneumoniae, Moraxella catarrhalis, and/or *Haemophilus influenzae* at 1 month of age associated with persistent wheeze, increased blood eosinophil counts and elevated total IgE at age 4, and asthma diagnosis at age 5 [33]. In children, infection of M. catarrhalis or S. pneumoniae and rhinovirus associated with greater severity of respiratory illness, including asthma exacerbations, suggesting that respiratory bacteria may contribute to airway inflammation [86]. In our studies, we demonstrated that the neonatal mice of allergic mothers have an altered BAL microbiome with increased Proteobacteria and decreased Bacteroidota as early as postnatal day 4. Then after allergen challenge, there was also a microbiome dysbiosis with a combination of increased Proteobacteria or Firmicutes and decreased Bacteroidota; this is consistent with the reports of associations of increased Proteobacteria and decreased Bacteroides in infants with wheeze [34]. We also demonstrated that Shannon alpha-diversity was reduced in groups with allergic inflammation and that the groups with allergic inflammation separated in the PCA plot of the Bray-Curtis beta-diversity. These data are consistent with studies demonstrating changes in airway microbiota community diversity in subjects with wheeze or asthma [62, 87].
Reports have demonstrated airway dysbiosis with allergic asthma but have not assessed whether lung microbial community dysbiosis has a functional effect on allergen sensitization. To go beyond associations of lung microbe composition with allergic lung inflammation in neonates, analyses of the function of lung microbiome in neonates are necessary. The function of microbial communities in tissues has mostly been studied in the gut, including mouse models with the transfer of gut microbial communities in disease states or in germ-free conditions. In an adult mouse model, administration of *Escherichia coli* to the lung can skew the type of Th responses and protect the mice from induction of allergic airway inflammation [88]. Also, Herbst and colleagues reported that induction of allergy in germ-free neonatal mice was protected by colonization by co-housing with specific-pathogen-free (SPF) non-allergic mice before allergen sensitization, suggesting a potential protective effect of microbiota of SPF mice, although this was colonization with microbiota composition of non-allergic mice [89]. However, the function of this control microbiota in non-germ-free conditions is not known. In non-germ-free conditions, microbe transfers have been studied in adult mice that were most often pretreated with antibiotics to disrupt the microbe taxa abundance and provide a niche for transferred microbes. In our studies with SPF conditions, antibiotics were not needed as the PND4 pups are rapidly growing neonates, likely providing niches for establishment of the transferred microbial communities. In our studies, transfer of the dysbiotic microbial communities from pups of allergic mothers to pups of non-allergic mothers established a dysbiotic microbial composition in the recipient pups and increased responsiveness to allergen. Thus, the dysbiotic microbial communities of neonates is sufficient for enhanced responsiveness to allergen and a dysbiotic microbial composition is sustained in recipient pups through the allergen challenge. However, the transfer of microbiome from pups not susceptible to allergen hyperresponsiveness (neonates of non-allergic mothers) to pups of allergic mothers did not protect the pups of allergic mothers from hyperresponsiveness to allergen.
Early life development of asthma and allergic disease results from complex interactions of genetic and environmental factors [90], including the dietary lipids, tocopherols [13, 14]. In adults and children, increasing plasma αT concentrations associates with better lung function and γT associates with lower lung function [24, 25, 65, 67, 68]. In mechanistic studies in adult and neonate mouse models of allergic lung inflammation, αT blocks development of lung eosinophilia and this is counteracted by γT (13–15, 21, 46, 68). In mechanistic studies, αT blocks the development of allergic inflammation in adults and neonates, at least, through functioning as an antagonist for protein kinase C during VCAM-1 signaling in endothelial cells, thereby, blocking VCAM-1-dependent eosinophil recruitment into tissues [15, 22, 91]. αT also blocks development of subsets of dendritic cells involved in allergic disease (CD11c + CD11b+ DCs) in vitro and in vivo [13, 18]. In our studies herein, maternal supplementation with αT blocked development of microbiome dysbiosis and allergen-induced lung eosinophilia in offspring of allergic mothers. However, transfer of microbial communities from pups of non-allergic mothers or from allergic mothers with αT-supplemented diet did not block microbial community dysbiosis or the hyperresponsiveness to allergen, as detected by increased BAL eosinophilia when the recipient pups were from allergic mothers with a basal diet. This suggests that the microbiome of offspring of allergic mothers is dominant. Consistent with a dominant microbiome profile, the transfer of BAL microbial communities from pups of allergic mothers to recipient pups of allergic mothers or non-allergic mothers with or without αT yielded a dysbiotic lung microbial composition and responses to allergen challenge with development of BAL eosinophilia.
Inflammation with eosinophilia is regulated by the chemokines CCL11 and the cytokines IL-13, IL-5, and IL-33. These mediators were induced in the no donor → OVA/B group of pups; in these studies only IL-33 expression was blocked by maternal supplementation of allergic mothers with αT (no donor → OVA/αT group). Because CCL11, IL-13 and Muc5ac are produced by airway epithelium, it suggests that αT did not intervene in epithelial-generated mediators of allergic inflammation. In pups with lung microbiome transfers, all pups with either microbiome of OVA/B group as donor or recipients had increases in several of the mediators of allergic inflammation, including CCL11, IL-13, IL-5 or IL-33. This indicates that in the presence of the microbial community dysbiosis, there was elevated responsiveness to allergen by increasing multiple allergen-induced mediators of allergic inflammation. Some variation in effects on CCL11, IL-13, IL-5 and IL-33 in the microbiome transfer studies with donor or recipient pups of the OVA/αT group may result from αT effects on cell signaling because besides αT anti-oxidant functions, αT is also an antagonist of PKCs by binding the regulatory domain of PKCs (15, 18, 20–23). This suggests that maternal αT has some effects on cytokine expression that may, in part, be independent of the transferred dysbiotic microbiota. We also found that recipient pups of allergic mothers on basal diet (OVA/B) had increased serum anti-OVA IgE, except for the pups of the OVA/αT → OVA/B group. However, suggesting although there were low undetectable levels of serum anti-OVA IgE, anti-OVA specific IgE bound to FcεR on leukocytes in tissues may participate in the allergic response to allergen in this group of pups. It is suggested that in humans, serum IgE does not always associate with severity of allergic asthma and allergic diseases [92, 93]. Allergen-specific IgE-mediated responses can occur at low serum levels of allergen-specific IgE when there is allergen-specific IgE bound to FcεR on leukocytes in the tissue [94, 95] Thus, supplementation of allergic mothers with αT not only blocked the development of allergic inflammation in mouse neonates, but also blocked the development of neonate lung dysbiosis.
In the airway microbial composition of infants in the INHANCE cohort [62, 63], nasal microbiome diversity at 3 months of age is lower in the infants with wheeze in the first year of life, with a decrease in Corynebacterium and increase in the proteobacteria Moraxella [62]. Increased *Moraxella is* seen in episodes of wheeze (35, 96–100). Also, lung function in children associates with profiles of tocopherol isoforms [24]. But it was not reported if the tocopherol isoform profiles early in life associate with infant airway microbiome composition. In our analyses of the INHANCE cohort, infants at 3–5 months of age in group Q4 with an anti-inflammatory tocopherol profile of higher αT and lower γT (13–15, 21, 24, 25, 46, 65, 67, 68) had a significant lower abundance of the family *Staphylococcaceae genus* Jeotgalicoccus as compared to the opposing group Q1 that had a proinflammatory profile of tocopherol isoforms with low αT high γT (13–15, 21, 24, 25, 46, 65, 67, 68). Because Staphylococcaceae in the lung has been associated with asthma [32], group Q1 might associate with development of asthma later in life, but this will need further longitudinal study. Additionally, group Q4 had a significantly higher % abundance of the phyla *Firmicute genus* Enterococcus as compared to group Q3 which has a low αT low γT. The *Enterococcus genus* are present as commensals in lung and gut and Enterococcus can limit the growth of pathogenic bacteria, although outgrowth of some Enteroccocus spp. can be pathogens (101–104). The 3–5 months old infants in group Q4 also had a significantly higher % abundance for the phyla *Firmicute genus* Peptoniphilus and a trend for higher % abundance of the genus Dialister as well as the Bacteriodota family Chitinophagaceae as compared to group Q3. Peptoniphilus genus is present in a healthy microbiome community of the nasopharynx [105] and is decreased in the nasopharynx with disease such as chronic rhinosinusitis [106]. Diallister spp. are enriched in healthy human lung compared to lung with infection [107] and is increased in nasopharynx for adult subjects with non-exacerbated asthma as compared to exacerbated asthma [106, 108]. Chitinophagaceae are present in myconium and the lung (109–111). Moreover, Chitinophagaceae are considered beneficial bacteria in the lower respiratory tract because Chitinophagaceae limit colonization by pathogens in animal models (111–113). Thus, the higher abundance in these taxa in group Q4, as compared to group Q3 at 3–5 months of age, is consistent with an anti-inflammatory tocopherol isoform profile and is consistent with a potential for lower airway disease.
In infants at age 12–18 months in the INHANCE cohort, there was a significantly lower % abundance in a Firmicute family Lachnospiraceae in group Q4 compared to group Q3. This was similar to our neonate mouse studies with αT supplementation demonstrating less Lachnospiraceae and less allergic lung inflammation. In the 12–18 months age group, group Q4 had a lower abundance of phyla Firmicutes family Oscillospiraceae and phyla Proteobactera class Alpha-proteobacteria family Paracaedibacteraceae as compared to group Q2, that has a pro-inflammatory tocopherol isoform profile for allergic inflammation. Interestingly, Oscillospiraceae are not detected in lung until inflammation is induced by TLR stimulation in mice and rats [114, 115]. This is interesting because TLR stimulation plays a role in allergic responses [116]. Proteobacteria are elevated in infant lungs with increased prevalence of wheeze [62] and in our neonatal mice with increased allergic lung inflammation. These differences in group Q4 and group Q2 bacterial microbiota suggest that group Q2 has a microbial community that may have an increased potential for elevated airway inflammation as compared to group Q4. A limitation of the INHANCE study is use of the readily accessible infant upper airway microbiome as a surrogate of microbiome in lower airways as these sites have both microbiome similarities and differences, as previously discussed [62, 85]. Nevertheless, the studies of upper airway in infants and adults and the studies of lower airway in children and adults suggest that there is a lower microbiota diversity and increased Proteobacteria with wheeze or asthma (62, 78–81).
In conclusion, unbiased analyses of the lung microbial communities of mouse neonates of allergic mothers indicate that both before and after allergen challenge, there is an altered lung bacterial microbiome composition that is consistent with that found in infants with wheeze and adults with allergic asthma. Most importantly, we have gone beyond associations of lung microbial composition dysbiosis with allergic lung inflammation and have demonstrated that the lung microbiome composition of offspring of allergic mothers confers neonate responsiveness to allergen and development of allergic disease in mouse models. Thus, an early life airway microbiota dysbiosis may have a significant function in development of wheeze and allergic asthma in children. This is a novel regulatory mechanism for development of responses to allergen challenge early in life that may inform design of future studies for approaches in the prevention or intervention in asthma and allergic disease. Further mechanisms for microbiome regulation are currently under investigation by our research group. Moreover, the development of lung microbial community dysbiosis in the neonatal mice was blocked by maternal dietary supplementation with αT during pregnancy and nursing, suggesting a potential target for intervention early in life. In human infants at 3–5 months and at 12–18 months of age, there was an anti-inflammatory tocopherol isoform profile associated with microbiome taxa abundance composition that may have the potential to limit development of airway disease.
## Data availability statement
The raw fastq files of the 16S rRNA analysis from mouse BAL in Figures 2–5 are deposited as NCBI BioProject repository, accession number ID PRJNA925891. The raw fastq files of the 16S rRNA analysis in INHANCE cohort in Figure 8 are deposited as NCBI BioProject repository, accession number ID PRJNA928382.
## Ethics statement
The studies involving human participants were reviewed and approved by Indiana University IRB#l308055098A023 for the INHANCE Cohort at Indiana University that has pre-existing stored data and stored serum and nasal swabs from an existing cohort. Written informed consent to participate in this study was provided by the participants’ legal guardian/next of kin. The animal study was reviewed and approved by Indiana University Institutional Review Committee for animals.
## Author contributions
JCB participated in performing, interpreting and analyzing the mouse experiments, analysis of ASV for human microbiome data and manuscript preparation. AH analyzed the mouse ASV data, as well as participated in performing and interpreting experiments and participated in manuscript preparation. GW performed the human serum tocopherol analysis. RKM and NWS participated in ASV analysis of human microbiome, data interpretations and manuscript preparation. JSD participated in design of microbiome analysis and design of microbiota transfer studies, data interpretations and manuscript preparation. CDM participated in 16S rRNA gene sequencing of mouse microbiota, data interpretations and manuscript preparation. KMK is the PI of the INHANCE cohort with microbiome data, provided the samples for the human serum tocopherol analysis and participated in interpretations and manuscript preparation. JMC-M conceived of the study design and participated in performing experiments, statistical analysis, data interpretations, and manuscript preparation. All authors contributed to the article and approved the submitted version.
## Conflict of interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher's note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## Supplementary material
The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/falgy.2023.1135412/full#supplementary-material.
SUPPLEMENT FIGURE 1 Alpha-diversity of donor pup BAL microbiota and allergen-challenged recipient pup microbiota. Data are from experiments in Figures 2, 3 and 4. Shannon Index for alpha-diversity which incorporates the total number of bacterial species and relative differences in the abundance of taxa in the community. ( A) PND4 donor BAL microbiota communities. ( B) PND16 BAL microbiota communities of allergen-challenged pups. Sal/B, saline-treated mother with basal diet. Sal/αT, saline-treated mother with αT-supplemented diet. OVA/B, OVA allergen-treated mother with a basal diet. OVA/αT, OVA allergen-treated mother with αT-supplemented diet. *. $p \leq 0.045$ for decrease as compared to no donor → Sal/B group. **. $p \leq 0.035$ for decrease as compared to Sal/B → Sal/B group. +. $p \leq 0.04$ for decrease as compared to no donor → Sal/αT group. ++. $p \leq 0.045$ for decrease as compared to OVA/αT → OVA/αT group.
SUPPLEMENT FIGURE 2 Beta-diversity of donor pup BAL microbiota and allergen-challenged recipient pup microbiota. Data are from experiments in Figures 2, 3 and 4. PND4 Donor pup and PND16 recipient pup BAL 16S microbiota were analyzed for beta-diversity with donor and transfer microbiome groups. The donor and transfer groups are displayed in separate three-dimension graphs for clarity but were generated by the same Principal Component Analyses (PCA). ( A) PCA plots of the Bray-*Curtis analysis* that generated a dissimilarity distance matrix for microbiota between groups. Bray-Curtis dissimilarity between groups is a measure that incorporates both overall abundance per sample and abundance of each taxa of the microbiota communities. ( B) PCA plots of the Weighted-Unifrac analyses between groups which incorporates the relative abundance of taxa shared between samples. ( C) PCA plots of Unweighted-Unifrac analyses between groups which only incorporates the presence/absence of taxa. Sal/B, saline-treated mother with basal diet. Sal/αT, saline-treated mother with αT-supplemented diet. OVA/B, OVA allergen-treated mother with a basal diet. OVA/αT, OVA allergen-treated mother with αT-supplemented diet.
## References
1. Friebele E. **The attack of asthma**. *Environ Health Perspect* (1996) **104** 22-5. DOI: 10.1289/ehp.9610422
2. Vollmer WM, Osborne ML, Buist AS. **20-year Trends in the prevalence of asthma and chronic airflow obstruction in an HMO**. *Am J Respir Crit Care Med* (1998) **157** 1079-84. DOI: 10.1164/ajrccm.157.4.9704140
3. van Schayck CP, Smit HA. **The prevalence of asthma in children: a reversing trend**. *Eur Respir J* (2005) **26** 647-50. DOI: 10.1183/09031936.05.00019805
4. Lim RH, Kobzik L. **Maternal transmission of asthma risk**. *Am J Reprod Immunol* (2009) **61** 1-10. DOI: 10.1111/j.1600-0897.2008.00671.x
5. Hamada K, Suzaki Y, Goldman A, Ning YY, Goldsmith C, Palecanda A. **Allergen-independent maternal transmission of asthma susceptibility**. *J Immunol* (2003) **170** 1683-9. DOI: 10.4049/jimmunol.170.4.1683
6. Fedulov AV, Leme AS, Kobzik L. **Duration of allergic susceptibility in maternal transmission of asthma risk**. *Am J Reprod Immunol* (2007) **58** 120-8. DOI: 10.1111/j.1600-0897.2007.00496.x
7. Hubeau C, Apostolou I, Kobzik L. **Targeting of CD25 and glucocorticoid-induced TNF receptor family-related gene-expressing T cells differentially modulates asthma risk in offspring of asthmatic and normal mother mice**. *J Immunol* (2007) **178** 1477-87. DOI: 10.4049/jimmunol.178.3.1477
8. Hubeau C, Apostolou I, Kobzik L. **Adoptively transferred allergen-specific T cells cause maternal transmission of asthma risk**. *Am J Pathol* (2006) **168** 1931-9. DOI: 10.2353/ajpath.2006.051231
9. Herz U, Joachim R, Ahrens B, Scheffold A, Radbruch A, Renz H. **Allergic sensitization and allergen exposure during pregnancy favor the development of atopy in the neonate**. *Int Arch Allergy Immunol* (2001) **124** 193-6. DOI: 10.1159/000053708
10. Herz U, Joachim R, Ahrens B, Scheffold A, Radbruch A, Renz H. **Prenatal sensitization in a mouse model**. *Am J Respir Crit Care Med* (2000) **162** S62-65. DOI: 10.1164/ajrccm.162.supplement_2.ras-1
11. Jarrett E, Hall E. **Selective suppression of IgE antibody responsiveness by maternal influence**. *Nature* (1979) **280** 145-7. DOI: 10.1038/280145a0
12. Fedulov AV, Kobzik L. **Allergy risk is mediated by dendritic cells with congenital epigenetic changes**. *Am J Respir Cell Mol Biol* (2011) **44** 285-92. DOI: 10.1165/rcmb.2009-0400OC
13. Abdala-Valencia H, Berdnikovs S, Soveg F, Cook-Mills JM. **alpha-Tocopherol supplementation of allergic female mice inhibits development of CD11c + CD11b+ dendritic cells in utero and allergic inflammation in neonates**. *Am J Physiol Lung Cell Mol Physiol* (2014) **307** L482-496. DOI: 10.1152/ajplung.00132.2014
14. Abdala-Valencia H, Soveg F, Cook-Mills JM. **γ-Tocopherol supplementation of allergic female mice augments development of CD11c + CD11b+ dendritic cells in utero and allergic inflammation in neonates**. *Am J Physiol Lung Cell Mol Physiol* (2016) **310** L759-771. DOI: 10.1152/ajplung.00301.2015
15. Berdnikovs S, Abdala-Valencia H, McCary C, Somand M, Cole R, Garcia A. **Isoforms of vitamin E have opposing immunoregulatory functions during inflammation by regulating leukocyte recruitment**. *J Immunol* (2009) **182** 4395-405. DOI: 10.4049/jimmunol.0803659
16. Cook-Mills JM, Marchese M, Abdala-Valencia H. **Vascular cell adhesion molecule-1 expression and signaling during disease: regulation by reactive oxygen Species and antioxidants**. *Antioxid Redox Signal* (2011) **15** 1607-38. DOI: 10.1089/ars.2010.3522
17. Cook-Mills J, Gebretsadik T, Abdala-Valencia H, Green J, Larkin EK, Dupont WD. **Interaction of vitamin E isoforms on asthma and allergic airway disease**. *Thorax* (2016) **2** 954-6. DOI: 10.1136/thoraxjnl-2016-208494
18. Lajiness JD, Amarsaikhan N, Tat K, Tsoggerel A, Cook-Mills JM. **β-Glucosylceramides and tocopherols regulate development and function of dendritic cells**. *J Immunol* (2022) **209** 1837-50. DOI: 10.4049/jimmunol.2101188
19. Galli F, Azzi A, Birringer M, Cook-Mills JM, Eggersdorfer M, Frank J. **Vitamin E: emerging aspects and new directions**. *Free Radic Biol Med* (2017) **102** 16-36. DOI: 10.1016/j.freeradbiomed.2016.09.017
20. Abdala-Valencia H, Cook-Mills JM. **VCAM-1 Signals activate endothelial cell protein kinase ca via oxidation**. *J Immunol* (2006) **177** 6379-87. DOI: 10.4049/jimmunol.177.9.6379
21. Abdala-Valencia H, Berdnikovs S, Cook-Mills JM. **Vitamin E isoforms differentially regulate intercellular adhesion molecule-1 activation of PKCalpha in human microvascular endothelial cells**. *PLoS One* (2012) **7** e41054. DOI: 10.1371/journal.pone.0041054
22. McCary CA, Yoon Y, Panagabko C, Cho W, Atkinson J, Cook-Mills JM. **Vitamin E isoforms directly bind PKCalpha and differentially regulate activation of PKCalpha**. *Biochem J* (2012) **441** 189-98. DOI: 10.1042/BJ20111318
23. Cook-Mills JM, McCary CA. **Isoforms of vitamin E differentially regulate inflammation**. *Endocr Metab Immune Disord Drug Targets* (2010) **10** 348-66. DOI: 10.2174/1871530311006040348
24. Kumar R, Ferrie RP, Balmert LC, Kienzl M, Rifas-Shiman SL, Gold DR. **Associations of α- and γ-tocopherol during early life with lung function in childhood**. *J Allergy Clin Immunol* (2020) **146** 1349-57. DOI: 10.1016/j.jaci.2020.04.019
25. Marchese ME, Kumar R, Colangelo LA, Avila PC, Jacobs DR, Gross M. **The vitamin E isoforms alpha-tocopherol and gamma-tocopherol have opposite associations with spirometric parameters: the CARDIA study**. *Respir Res* (2014) **15** 31. DOI: 10.1186/1465-9921-15-31
26. Kalayci O, Besler T, Kilinc K, Sekerel BE, Saraclar Y. **Serum levels of antioxidant vitamins (alpha tocopherol, beta carotene, and ascorbic acid) in children with bronchial asthma**. *Turk J Peds* (2000) **42** 17-21. PMID: 10731863
27. Kelly FJ, Mudway I, Blomberg A, Frew A, Sandstrom T. **Altered lung antioxidant status in patients with mild asthma**. *Lancet* (1999) **354** 482-3. DOI: 10.1016/S0140-6736(99)01812-7
28. Shvedova AA, Kisin ER, Kagan VE, Karol MH. **Increased lipid peroxidation and decreased antioxidants in lungs of Guinea pigs following an allergic pulmonary response**. *Tox Appl Pharm* (1995) **132** 72-81. DOI: 10.1006/taap.1995.1088
29. Schunemann HJ, Grant BJ, Freudenheim JL, Muti P, Browne RW, Drake JA. **The relation of serum levels of antioxidant vitamins C and E, retinol and carotenoids with pulmonary function in the general population**. *Am J Respir Crit Care Med* (2001) **163** 1246-55. DOI: 10.1164/ajrccm.163.5.2007135
30. Bousquet J, Anto J, Auffray C, Akdis M, Cambon-Thomsen A, Keil T. **MeDALL (mechanisms of the development of ALLergy): an integrated approach from phenotypes to systems medicine**. *Allergy* (2011) **66** 596-604. DOI: 10.1111/j.1398-9995.2010.02534.x
31. Natalini JG, Singh S, Segal LN. **The dynamic lung microbiome in health and disease**. *Nat Rev Microbiol* (2022) 1-14. DOI: 10.1038/s41579-022-00821-x
32. Losol P, Park HS, Song WJ, Hwang YK, Kim SH, Holloway JW. **Association of upper airway bacterial microbiota and asthma: systematic review**. *Asia Pac Allergy* (2022) **12** e32. DOI: 10.5415/apallergy.2022.12.e32
33. Bisgaard H, Hermansen MN, Buchvald F, Loland L, Halkjaer LB, Bønnelykke K. **Childhood asthma after bacterial colonization of the airway in neonates**. *N Engl J Med* (2007) **357** 1487-95. DOI: 10.1056/NEJMoa052632
34. McCauley K, Durack J, Valladares R, Fadrosh DW, Lin DL, Calatroni A. **Distinct nasal airway bacterial microbiotas differentially relate to exacerbation in pediatric patients with asthma**. *J Allergy Clin Immunol* (2019) **144** 1187-97. DOI: 10.1016/j.jaci.2019.05.035
35. Teo SM, Mok D, Pham K, Kusel M, Serralha M, Troy N. **The infant nasopharyngeal microbiome impacts severity of lower respiratory infection and risk of asthma development**. *Cell Host Microbe* (2015) **17** 704-15. DOI: 10.1016/j.chom.2015.03.008
36. Walker MT, Ferrie RP, Hoji A, Schroeder-Carter LM, Cohen JD, Schnaar RL. **β-Glucosylceramide from allergic mothers enhances offspring responsiveness to allergen**. *Front Allergy* (2021) **2** 647134. DOI: 10.3389/falgy.2021.647134
37. Muller-Schmehl K, Beninde J, Finckh B, Florian S, Dudenhausen JW, Brigelius-Flohe R. **Localization of alpha-tocopherol transfer protein in trophoblast, fetal capillaries’ endothelium and amnion epithelium of human term placenta**. *Free Radic Res* (2004) **38** 413-20. DOI: 10.1080/10715760310001659611
38. Jishage K, Tachibe T, Ito T, Shibata N, Suzuki S, Mori T. **Vitamin E is essential for mouse placentation but not for embryonic development itself**. *Biol Reprod* (2005) **73** 983-7. DOI: 10.1095/biolreprod.105.043018
39. Cook-Mills JM, Watson RR. *Nutrition and functional foods for healthy aging* (2019) 167-88
40. Villar J, Purwar M, Merialdi M, Zavaleta N, Ngoc NT, Anthony J, De Greeff A, Poston L, Shennan A. **World health organisation multicentre randomised trial of supplementation with vitamins C and E among pregnant women at high risk for pre-eclampsia in populations of low nutritional status from developing countries**. *BJOG* (2009) **116** 780-8. DOI: 10.1111/j.1471-0528.2009.02158.x
41. McCance DR, Holmes VA, Maresh MJ, Patterson CC, Walker JD, Pearson DW. **Vitamins C and E for prevention of pre-eclampsia in women with type 1 diabetes (DAPIT): a randomised placebo-controlled trial**. *Lancet* (2010) **376** 259-66. DOI: 10.1016/S0140-6736(10)60630-7
42. Hauth JC, Clifton RG, Roberts JM, Spong CY, Myatt L, Leveno KJ. **Vitamin C and E supplementation to prevent spontaneous preterm birth: a randomized controlled trial**. *Obstet Gynecol* (2010) **116** 653-8. DOI: 10.1097/AOG.0b013e3181ed721d
43. Greenough A, Shaheen SO, Shennan A, Seed PT, Poston L. **Respiratory outcomes in early childhood following antenatal vitamin C and E supplementation**. *Thorax* (2010) **65** 998-1003. DOI: 10.1136/thx.2010.139915
44. Kalpdev A, Saha SC, Dhawan V. **Vitamin C and E supplementation does not reduce the risk of superimposed PE in pregnancy**. *Hypertens Pregnancy* (2011) **30** 447-56. DOI: 10.3109/10641955.2010.507840
45. Gungorduk K, Asicioglu O, Gungorduk OC, Yildirim G, Besimoglu B, Ark C. **Does vitamin C and vitamin E supplementation prolong the latency period before delivery following the preterm premature rupture of membranes? A randomized controlled study**. *Am J Perinatol* (2014) **31** 195-202. DOI: 10.1055/s-0033-1343774
46. McCary CA, Abdala-Valencia H, Berdnikovs S, Cook-Mills JM. **Supplemental and highly elevated tocopherol doses differentially regulate allergic inflammation: reversibility of alpha-tocopherol and gamma-tocopherol's effects**. *J Immunol* (2011) **186** 3674-85. DOI: 10.4049/jimmunol.1003037
47. Traber MG, Siddens LK, Leonard SW, Schock B, Gohil K, Krueger SK. **Alpha-tocopherol modulates Cyp3a expression, increases gamma-CEHC production, and limits tissue gamma-tocopherol accumulation in mice fed high gamma-tocopherol diets**. *Free Radic Biol Med* (2005) **38** 773-85. DOI: 10.1016/j.freeradbiomed.2004.11.027
48. Meydani SN, Yogeeswaran G, Liu S, Baskar S, Meydani M. **Fish oil and tocopherol-induced changes in natural killer cell-mediated cytotoxicity and PGE2 synthesis in young and old mice**. *J Nutr* (1988) **118** 1245-52. DOI: 10.1093/jn/118.10.1245
49. Meydani SN, Shapiro AC, Meydani M, Macauley JB, Blumberg JB. **Effect of age and dietary fat (fish, corn and coconut oils) on tocopherol status of C57BL/6Nia mice**. *Lipids* (1987) **22** 345-50. DOI: 10.1007/BF02534004
50. Abdala-Valencia H, Earwood J, Bansal S, Jansen M, Babcock G, Garvy B. **Nonhematopoietic NADPH oxidase regulation of lung eosinophilia and airway hyperresponsiveness in experimentally induced asthma**. *Am J Physiol Lung Cell Mol Physiol* (2007) **292** L1111-1125. DOI: 10.1152/ajplung.00208.2006
51. Kreda SM, Seminario-Vidal L, van Heusden CA, O'Neal W, Jones L, Boucher RC. **Receptor-promoted exocytosis of airway epithelial mucin granules containing a spectrum of adenine nucleotides**. *J Physiol* (2010) **588** 2255-67. DOI: 10.1113/jphysiol.2009.186643
52. Mandal RK, Jiang T, Kwon YM. **Genetic determinants in Salmonella enterica serotype typhimurium required for overcoming in vitro stressors in the mimicking host environment**. *Microbiol Spectr* (2021) **9** e0015521. DOI: 10.1128/Spectrum.00155-21
53. Lines JL, Hoskins S, Hollifield M, Cauley LS, Garvy BA. **The migration of T cells in response to influenza virus is altered in neonatal mice**. *J Immunol* (2010) **185** 2980-8. DOI: 10.4049/jimmunol.0903075
54. Karstens L, Asquith M, Davin S, Fair D, Gregory WT, Wolfe AJ. **Controlling for contaminants in low-biomass 16S rRNA gene sequencing experiments**. *mSystems* (2019) **4** e00290-19. DOI: 10.1128/mSystems.00290-19
55. Salter SJ, Cox MJ, Turek EM, Calus ST, Cookson WO, Moffatt MF. **Reagent and laboratory contamination can critically impact sequence-based microbiome analyses**. *BMC Biol* (2014) **12** 87. DOI: 10.1186/s12915-014-0087-z
56. Kumar R, Eipers P, Little RB, Crowley M, Crossman DK, Lefkowitz EJ. **Getting started with microbiome analysis: sample acquisition to bioinformatics**. *Curr Protoc Hum Genet* (2014) **82** 18.8.1-18.8.29. DOI: 10.1002/0471142905.hg1808s82
57. Van Der Pol WJ, Kumar R, Morrow CD, Blanchard EE, Taylor CM, Martin DH. **In silico and experimental evaluation of primer sets for species-level resolution of the vaginal microbiota using 16S ribosomal RNA gene sequencing**. *J Infect Dis* (2019) **219** 305-14. DOI: 10.1093/infdis/jiy508
58. Oren A, Garrity GM. **Valid publication of the names of forty-two phyla of prokaryotes**. *Int J Syst Evol Microbiol* (2021) **71** 005056. DOI: 10.1099/ijsem.0.005056
59. Bolyen E, Rideout JR, Dillon MR, Bokulich NA, Abnet CC, Al-Ghalith GA. **Reproducible, interactive, scalable and extensible microbiome data science using QIIME 2**. *Nat Biotechnol* (2019) **37** 852-7. DOI: 10.1038/s41587-019-0209-9
60. Quast C, Pruesse E, Yilmaz P, Gerken J, Schweer T, Yarza P. **The SILVA ribosomal RNA gene database project: improved data processing and web-based tools**. *Nucleic Acids Res* (2013) **41** D590-596. DOI: 10.1093/nar/gks1219
61. Vázquez-Baeza Y, Pirrung M, Gonzalez A, Knight R. **EMPeror: a tool for visualizing high-throughput microbial community data**. *Gigascience* (2013) **2** 16. DOI: 10.1186/2047-217X-2-16
62. Kloepfer KM, Ross SE, Hemmerich CM, Slaven JE, Rusch DB, Davis SD. **Increased microbiota diversity associated with higher FEV(0.5) in infants**. *Pediatr Pulmonol* (2020) **55** 27-9. DOI: 10.1002/ppul.24559
63. Aguilera AC, Dagher IA, Kloepfer KM. **Role of the microbiome in allergic disease development**. *Curr Allergy Asthma Rep* (2020) **20** 44. DOI: 10.1007/s11882-020-00944-2
64. Wolff G, Slaven J, Colbert J, Andres U, Cook-Mills J, Kloepfer K. **Food deserts and lung function during early life**. *J Allergy Clin Immunol* (2022) **149** AB84. DOI: 10.1016/j.jaci.2021.12.302
65. Stone CA, Cook-Mills J, Gebretsadik T, Rosas-Salazar C, Turi K, Brunwasser SM. **Delineation of the individual effects of vitamin E isoforms on early life incident wheezing**. *J Pediatr* (2019) **206** 156-163.e153. DOI: 10.1016/j.jpeds.2018.10.045
66. Stone CA, McEvoy CT, Aschner JL, Kirk A, Rosas-Salazar C, Cook-Mills JM. **Update on vitamin E and its potential role in preventing or treating bronchopulmonary dysplasia**. *Neonatology* (2018) **113** 366-78. DOI: 10.1159/000487388
67. Cook-Mills JM, Abdala-Valencia H, Hartert T. **Two faces of vitamin E in the lung**. *Am J Respir Crit Care Med* (2013) **188** 279-84. DOI: 10.1164/rccm.201303-200503ED
68. Cook-Mills JM, Averill SH, Lajiness JD. **Asthma, allergy and vitamin E: current and future perspectives**. *Free Radic Biol Med* (2022) **179** 388-402. DOI: 10.1016/j.freeradbiomed.2021.10.037
69. Sohn KH, Baek MG, Choi SM, Bae B, Kim RY, Kim YC. **Alteration of lung and gut Microbiota in IL-13-transgenic mice simulating chronic asthma**. *J Microbiol Biotechnol* (2020) **30** 1819-26. DOI: 10.4014/jmb.2009.09019
70. Durack J, Christian LS, Nariya S, Gonzalez J, Bhakta NR, Ansel KM. **Distinct associations of sputum and oral microbiota with atopic, immunologic, and clinical features in mild asthma**. *J Allergy Clin Immunol* (2020) **146** 1016-26. DOI: 10.1016/j.jaci.2020.03.028
71. Hufnagl K, Pali-Schöll I, Roth-Walter F, Jensen-Jarolim E. **Dysbiosis of the gut and lung microbiome has a role in asthma**. *Semin Immunopathol* (2020) **42** 75-93. DOI: 10.1007/s00281-019-00775-y
72. Gollwitzer ES, Saglani S, Trompette A, Yadava K, Sherburn R, McCoy KD. **Lung microbiota promotes tolerance to allergens in neonates via PD-L1**. *Nat Med* (2014) **20** 642-7. DOI: 10.1038/nm.3568
73. Mika M, Mack I, Korten I, Qi W, Aebi S, Frey U. **Dynamics of the nasal microbiota in infancy: a prospective cohort study**. *J Allergy Clin Immunol* (2015) **135** 905-12. DOI: 10.1016/j.jaci.2014.12.1909
74. Toivonen L, Karppinen S, Schuez-Havupalo L, Waris M, He Q, Hoffman KL. **Longitudinal changes in early nasal Microbiota and the risk of childhood asthma**. *Pediatrics* (2020) **146** e20200421. DOI: 10.1542/peds.2020-0421
75. Peterson SW, Knox NC, Golding GR, Tyler SD, Tyler AD, Mabon P. **A study of the infant nasal microbiome development over the first year of life and in relation to their primary adult caregivers using cpn60 universal target (UT) as a phylogenetic marker**. *PLoS One* (2016) **11** e0152493. DOI: 10.1371/journal.pone.0152493
76. Boutin RCT, Petersen C, Finlay BB. **Microbial insights into asthmatic immunopathology. A forward-looking synthesis and commentary**. *Ann Am Thorac Soc* (2017) **14** S316-s325. DOI: 10.1513/AnnalsATS.201707-534AW
77. Haahtela T, Laatikainen T, Alenius H, Auvinen P, Fyhrquist N, Hanski I. **Hunt for the origin of allergy—comparing the Finnish and Russian Karelia**. *Clin Exp Allergy* (2015) **45** 891-901. DOI: 10.1111/cea.12527
78. Huang YJ, Nariya S, Harris JM, Lynch SV, Choy DF, Arron JR. **The airway microbiome in patients with severe asthma: associations with disease features and severity**. *J Allergy Clin Immunol* (2015) **136** 874-84. DOI: 10.1016/j.jaci.2015.05.044
79. Zhang Q, Cox M, Liang Z, Brinkmann F, Cardenas PA, Duff R. **Airway Microbiota in severe asthma and relationship to asthma severity and phenotypes**. *PLoS One* (2016) **11** e0152724. DOI: 10.1371/journal.pone.0152724
80. Li N, Qiu R, Yang Z, Li J, Chung KF, Zhong N. **Sputum microbiota in severe asthma patients: relationship to eosinophilic inflammation**. *Respir Med* (2017) **131** 192-8. DOI: 10.1016/j.rmed.2017.08.016
81. Marri PR, Stern DA, Wright AL, Billheimer D, Martinez FD. **Asthma-associated differences in microbial composition of induced sputum**. *J Allergy Clin Immunol* (2013) **131** 346-52. DOI: 10.1016/j.jaci.2012.11.013
82. Hilty M, Burke C, Pedro H, Cardenas P, Bush A, Bossley C. **Disordered microbial communities in asthmatic airways**. *PLoS One* (2010) **5** e8578. DOI: 10.1371/journal.pone.0008578
83. Taylor SL, Leong LEX, Choo JM, Wesselingh S, Yang IA, Upham JW. **Inflammatory phenotypes in patients with severe asthma are associated with distinct airway microbiology**. *J Allergy Clin Immunol* (2018) **141** 94-103. DOI: 10.1016/j.jaci.2017.03.044
84. Green BJ, Wiriyachaiporn S, Grainge C, Rogers GB, Kehagia V, Lau L. **Potentially pathogenic airway bacteria and neutrophilic inflammation in treatment resistant severe asthma**. *PLoS One* (2014) **9** e100645. DOI: 10.1371/journal.pone.0100645
85. Kloepfer KM, Deschamp AR, Ross SE, Peterson-Carmichael SL, Hemmerich CM, Rusch DB. **In children, the microbiota of the nasopharynx and bronchoalveolar lavage fluid are both similar and different**. *Pediatr Pulmonol* (2018) **53** 475-82. DOI: 10.1002/ppul.23953
86. Kloepfer KM, Lee WM, Pappas TE, Kang TJ, Vrtis RF, Evans MD. **Detection of pathogenic bacteria during rhinovirus infection is associated with increased respiratory symptoms and asthma exacerbations**. *J Allergy Clin Immunol* (2014) **133** 1301-7. DOI: 10.1016/j.jaci.2014.02.030
87. Al Bataineh MT, Hamoudi RA, Dash NR, Ramakrishnan RK, Almasalmeh MA, Sharif HA. **Altered respiratory microbiota composition and functionality associated with asthma early in life**. *BMC Infect Dis* (2020) **20** 697. DOI: 10.1186/s12879-020-05427-3
88. Nembrini C, Sichelstiel A, Kisielow J, Kurrer M, Kopf M, Marsland BJ. **Bacterial-induced protection against allergic inflammation through a multicomponent immunoregulatory mechanism**. *Thorax* (2011) **66** 755-63. DOI: 10.1136/thx.2010.152512
89. Herbst T, Sichelstiel A, Schär C, Yadava K, Bürki K, Cahenzli J. **Dysregulation of allergic airway inflammation in the absence of microbial colonization**. *Am J Respir Crit Care Med* (2011) **184** 198-205. DOI: 10.1164/rccm.201010-1574OC
90. Martinez FD. **Genes, environments, development and asthma: a reappraisal**. *Eur Respir J* (2007) **29** 179-84. DOI: 10.1183/09031936.00087906
91. Cook-Mills JM. **Isoforms of vitamin E differentially regulate PKC alpha and inflammation: a review**. *J Clin Cell Immunol* (2013) **4** 1000137. DOI: 10.4172/2155-9899.1000137
92. Palmer LJ, Rye PJ, Gibson NA, Burton PR, Landau LI, Lesouef PN. **Airway responsiveness in early infancy predicts asthma, lung function, and respiratory symptoms by school age**. *Am J Respir Crit Care Med* (2001) **163** 37-42. DOI: 10.1164/ajrccm.163.1.2005013
93. Anvari S, Miller J, Yeh CY, Davis CM. **IgE-Mediated food allergy**. *Clin Rev Allergy Immunol* (2019) **57** 244-60. DOI: 10.1007/s12016-018-8710-3
94. Walker M, Green J, Ferrie R, Queener A, Kaplan MH, Cook-Mills JM. **Mechanism for initiation of food allergy: dependence on skin barrier mutations and environmental allergen co-stimulation**. *J Allergy Clin Immunol* (2018) **15** 30233-1. DOI: 10.1016/j.jaci.2018.02.003
95. Kanagaratham C, El Ansari YS, Lewis OL, Oettgen HC. **Ige and IgG antibodies as regulators of mast cell and basophil functions in food allergy**. *Front immunol* (2020) **11** 603050. DOI: 10.3389/fimmu.2020.603050
96. Rosas-Salazar C, Shilts MH, Tovchigrechko A, Schobel S, Chappell JD, Larkin EK. **Differences in the nasopharyngeal microbiome during acute respiratory tract infection with human rhinovirus and respiratory syncytial virus in infancy**. *J Infect Dis* (2016) **214** 1924-8. DOI: 10.1093/infdis/jiw456
97. Chun Y, Do A, Grishina G, Grishin A, Fang G, Rose S. **Integrative study of the upper and lower airway microbiome and transcriptome in asthma**. *JCI insight* (2020) **5** e133707. DOI: 10.1172/jci.insight.133707
98. Rosas-Salazar C, Shilts MH, Tovchigrechko A, Schobel S, Chappell JD, Larkin EK. **Nasopharyngeal Lactobacillus is associated with a reduced risk of childhood wheezing illnesses following acute respiratory syncytial virus infection in infancy**. *J Allergy Clin Immunol* (2018) **142** 1447-1456.e1449. DOI: 10.1016/j.jaci.2017.10.049
99. Biesbroek G, Tsivtsivadze E, Sanders EA, Montijn R, Veenhoven RH, Keijser BJ. **Early respiratory microbiota composition determines bacterial succession patterns and respiratory health in children**. *Am J Respir Crit Care Med* (2014) **190** 1283-92. DOI: 10.1164/rccm.201407-1240OC
100. Calışkan M, Bochkov YA, Kreiner-Møller E, Bønnelykke K, Stein MM, Du G. **Rhinovirus wheezing illness and genetic risk of childhood-onset asthma**. *N Engl J Med* (2013) **368** 1398-407. DOI: 10.1056/NEJMoa1211592
101. Dubin K, Pamer EG. **Enterococci and their interactions with the intestinal microbiome**. *Microbiol Spectrum* (2014) **5**. DOI: 10.1128/microbiolspec.BAD-0014-2016
102. Savini V, Gherardi G, Astolfi D, Polilli E, Dicuonzo G, D'Amario C. **Insights into airway infections by enterococci: a review**. *Recent Pat Antiinfect Drug Discov* (2012) **7** 36-44. DOI: 10.2174/157489112799829774
103. Mourani PM, Harris JK, Sontag MK, Robertson CE, Abman SH. **Molecular identification of bacteria in tracheal aspirate fluid from mechanically ventilated preterm infants**. *PLoS One* (2011) **6** e25959. DOI: 10.1371/journal.pone.0025959
104. Gallacher DJ, Kotecha S. **Respiratory microbiome of new-born infants**. *Front Pediatr* (2016) **4** 10. DOI: 10.3389/fped.2016.00010
105. Kumpitsch C, Koskinen K, Schöpf V, Moissl-Eichinger C. **The microbiome of the upper respiratory tract in health and disease**. *BMC Biol* (2019) **17** 87. DOI: 10.1186/s12915-019-0703-z
106. Elgamal Z, Singh P, Geraghty P. **The upper airway Microbiota, environmental exposures, inflammation, and disease**. *Medicina (B Aires)* (2021) **57** 823. DOI: 10.3390/medicina57080823
107. Hérivaux A, Willis JR, Mercier T, Lagrou K, Gonçalves SM, Gonçales RA. **Lung microbiota predict invasive pulmonary aspergillosis and its outcome in immunocompromised patients**. *Thorax* (2022) **77** 283-91. DOI: 10.1136/thoraxjnl-2020-216179
108. Fazlollahi M, Lee TD, Andrade J, Oguntuyo K, Chun Y, Grishina G. **The nasal microbiome in asthma**. *J Allergy Clin Immunol* (2018) **142** 834-43. DOI: 10.1016/j.jaci.2018.02.020
109. Prevaes SM, de Winter-de Groot KM, Janssens HM, de Steenhuijsen Piters WA, Tramper-Stranders GA, Wyllie AL. **Development of the nasopharyngeal Microbiota in infants with cystic fibrosis**. *Am J Respir Crit Care Med* (2016) **193** 504-15. DOI: 10.1164/rccm.201509-1759OC
110. Stinson LF, Keelan JA, Payne MS. **Comparison of meconium DNA extraction methods for use in microbiome studies**. *Front Microbiol* (2018) **9** 270. DOI: 10.3389/fmicb.2018.00270
111. Megahed A, Zeineldin M, Evans K, Maradiaga N, Blair B, Aldridge B. **Impacts of environmental complexity on respiratory and gut microbiome community structure and diversity in growing pigs**. *Sci Rep* (2019) **9** 13773. DOI: 10.1038/s41598-019-50187-z
112. Ober RA, Thissen JB, Jaing CJ, Cino-Ozuna AG, Rowland RRR, Niederwerder MC. **Increased microbiome diversity at the time of infection is associated with improved growth rates of pigs after co-infection with porcine reproductive and respiratory syndrome virus (PRRSV) and porcine circovirus type 2 (PCV2)**. *Vet Microbiol* (2017) **208** 203-11. DOI: 10.1016/j.vetmic.2017.06.023
113. Siqueira FM, Pérez-Wohlfeil E, Carvalho FM, Trelles O, Schrank IS, Vasconcelos ATR. **Microbiome overview in swine lungs**. *PLoS One* (2017) **12** e0181503. DOI: 10.1371/journal.pone.0181503
114. Goossens E, Li J, Callens C, Van Rysselberghe N, Kettunen H, Vuorenmaa J. **Acute endotoxemia-induced respiratory and intestinal dysbiosis**. *Intl J Molecular Sci* (2022) **23** 11602. DOI: 10.3390/ijms231911602
115. Hosang L, Canals RC, van der Flier FJ, Hollensteiner J, Daniel R, Flügel A. **The lung microbiome regulates brain autoimmunity**. *Nature* (2022) **603** 138-44. DOI: 10.1038/s41586-022-04427-4
116. Zakeri A, Russo M. **Dual role of toll-like receptors in human and experimental asthma models**. *Front immunol* (2018) **9** 1027. DOI: 10.3389/fimmu.2018.01027
|
---
title: 'The ABCG2 rs2231142 polymorphism and the risk of nephrolithiasis: A case–control
study from the Taiwan biobank'
authors:
- Ching-Tsai Lin
- I-Chieh Chen
- Yen-Ju Chen
- Ying-Cheng Lin
- Jui-Chun Chang
- Tsai-Jung Wang
- Wen-Nan Huang
- Yi-Hsing Chen
- Yi-Huei Chen
- Ching-Heng Lin
- Yi-Ming Chen
journal: Frontiers in Endocrinology
year: 2023
pmcid: PMC10036833
doi: 10.3389/fendo.2023.1074012
license: CC BY 4.0
---
# The ABCG2 rs2231142 polymorphism and the risk of nephrolithiasis: A case–control study from the Taiwan biobank
## Abstract
### Background
Hyperuricemia and gout are risk factors of nephrolithiasis. However, it is unclear whether the ABCG2 gene contributes to the development of nephrolithiasis. We aimed to investigate the interaction between the ABCG2 rs2231142 variant and incident nephrolithiasis in the Taiwanese population.
### Methods
A total of 120,267 adults aged 30–70 years were enrolled from the Taiwan Biobank data-base in this retrospective case–control study and genotyped for rs2231142. The primary outcome was the prevalence of self-reported nephrolithiasis. The odds ratio (OR) of incident nephrolithiasis was analyzed by multivariable logistic regression models with adjustment for multifactorial confounding factors. Associations of the ABCG2 rs2231142 variant with serum uric acid levels, and the incident nephrolithiasis were explored.
### Results
The frequency of rs2231142 T allele was $53\%$, and 8,410 participants had nephrolithiasis. The multivariable-adjusted OR ($95\%$ confidence interval) of nephrolithiasis was 1.18 (1.09–1.28) and 1.12 (1.06–1.18) for TT and GT genotypes, respectively, compared with the GG genotype ($p \leq 0.001$), specifically in the male population with hyperuricemia. Higher age, male sex, hyperlipidemia, hypertension, diabetes mellitus, hyperuricemia, smoking and overweight were independent risk factors for nephrolithiasis. In contrast, regular physical exercise is a protective factor against nephrolithiasis.
### Conclusions
ABCG2 genetic variation is a significant risk of nephrolithiasis, independent of serum uric acid levels. For rs2231142 T allele carriers, our result provides evidence for precision healthcare to tackle hyperuricemia, comorbidities, smoking, and overweight, and recommend regular physical exercise for the prevention of nephrolithiasis.
## Introduction
Nephrolithiasis (kidney stones) is a common disease that clinically present as acute colicky pain and often recurs in association with morbidities. Worldwide, the prevalence of nephrolithiasis ranges from $1\%$ to $19.1\%$ in Asia, $5\%$ to $10\%$ in Europe, $4\%$ in South Africa, $7\%$ to $13\%$ in North America, and as $20\%$ to $25\%$ in the Middle East (1–3). In the United States, the $7.4\%$ absolute increase in the prevalence of self-reported nephrolithiasis over four decades – from $3.2\%$ in 1980 to $10.6\%$ in 2018 is noteworthy [4, 5]. In China, the prevalence of nephrolithiasis almost doubled from $5.95\%$ in 1991–2000 to $10.63\%$ in 2011 [6]. The 5–$15\%$ increase in the prevalence of nephrolithiasis in developed and developing countries during the past few decades has increased of health-care cost by approximately $50\%$, which includes both the direct treatment costs and the indirect costs associated with loss of work productivity [3, 7]. Incident and recurrent nephrolithiasis confers a high risk of comorbidity that results in acute and chronic renal failure [8]. Despite the availability of various therapeutic approaches, up to $50\%$ and $75\%$ of patients with nephrolithiasis develop recurrence within 5–10 years and 20 years, respectively [9].
Significant predisposing factors of nephrolithiasis can be classified into five categories: lifestyle, genetics, diet, environment, and systemic comorbidities [3, 10]. The contribution of aforementioned risk factors to nephrolithiasis varies in different populations. Moreover, biochemical abnormalities in the urinary composition have been associated with the risk of nephrolithiasis [10]. Nephrolithiasis characteristically includes four phenotypes – calcium, urate, struvite (magnesium ammonium phosphate), and cystine stones [10]. Approximately $80\%$ of nephrolithiasis, either homogenously or heterogeneously is constituted by the commonest phenotypes of calcium oxalate (CaOx) or calcium phosphate [11]. Comprising less than $10\%$ of cases, urate stones constitute the third most common type of nephrolithiasis and are attributed to predisposing factors such as persistently low urinary pH level, hypovolemia (decreased urinary volume), and hyperuricosuria [12]. Moreover, the dissolved urate in the native urine can salt out CaOx in those with hyperuricosuria and hypercalciuria [13]. Allopurinol may effectively prevent the hyperuricosuric and hypercalciuric patients from recurrence of calcium nephrolithiasis [13, 14]. However, it is unclear whether hyperuricemia contributes to the development of calcium nephrolithiasis.
In genome-wide association studies (GWAS), the strong association of the ABCG2 rs2231142 variant with gout that is mediated by both decreased intestinal urate excretion and renal overload hyperuricemia was confirmed in the Asian population [15, 16]. Additionally, obesity, diet, lifestyle, genetics and underlying comorbidities are important risks factors that predispose to hyperuricemia and contribute to the pathogenesis of gout [17]. The risk of hyperuricemia was markedly increased in the Taiwanese population through the interaction of the rs2231142 variant with obesity [18]. Systematic reviews of the predisposition to kidney stones revealed a significant link between obesity, associated disease, and hyperuricosuria [3, 10, 11]. However, it is unclear whether the rs2231142 variant plays a crucial role in the phathogenesis of kidney stone, and the interaction between genetic factors and other risk factors of nephrolithiasis remains elusive.
In this study, we aimed to explore the association between ABCG2 rs2231142 variant and the risk of nephrolithiasis in a community-based Taiwanese population. The primary objective of the study was to determine the prevalence of self-reported nephrolithiasis in Taiwan and to identify the associations between the ABCG2 rs2231142 variant and incident nephrolithiasis with hyperuricemia.
## Study design, participants and ethics statement
This retrospective case–control study was undertaken using the data collected in the Taiwan Biobank (TWB) database between September 2014 and May 2021 and included 120,267 adult Taiwanese Han Chinese participants aged 30–70 years. The TWB is a prospective population-based research project to recruit volunteers from the general population. Participants with a history of cancer were excluded from enrollment. To study the complex interaction between genomics and comorbidities of public importance, subjects with chronic diseases were not excluded. All participants from 29 recruitment medical centers in Taiwan provided written informed consent before the sample collection and the subsequent analysis. The TWB datasets used and/or analyzed in this study comprised specimens and information that were collected using a completely standardized procedure to fit researchers’ needs in different fields [19, 20]. For this study, we obtained genotyping information, demographics (i.e., sex and age), medical history, lifestyle modalities (i.e., alcohol consumption, smoking and physical activity), physical examination (i.e., body mass index [BMI; kg/m2], and blood pressure [BP; mmHg]), and biochemical reports (i.e., serum uric acid [UA], creatinine, fasting glucose levels, total cholesterol [TC], triglyceride [TG], high-density lipoprotein cholesterol [HDL-C], low-density lipoprotein cholesterol [LDL-C] in mg/dL etc.) from the TWB database in conformance with the Declaration of Helsinki and with ethical approval by the Institutional Review Board of Taichung Veterans General Hospital, Taichung, Taiwan (IRB No. CE16270B-2).
## Genotyping and quality controls
Blood samples for DNA analysis were collected from TWB participants and the genotype was determined by using the Axiom Genome-Wide TWB array (Affymetrix, Santa Clara, CA, USA) at the National Center for Genome Medicine in Academia Sinica, Taiwan [19, 20]. TWB implemented Affymetrix Power Tools (APT) as a standard quality-control procedure to select specific single-nucleotide polymorphisms (SNPs) that are suitable for analyzing the genetic traits of the Taiwanese Han Chinese populations. SNPs on the X and Y chromosomes, as well as those on mitochondrial DNA, were included for data release [19]. In total, 653,291 SNPs were included in the Affymetrix TWB 2.0 SNP chip. Details about the TWB are available from the official website (https://taiwanview.twbiobank.org.tw/index). PLINK was used for analyzing and working with Affymetrix microarray data, as well as for controlling the quality of the procedure with the Hardy–*Weinberg equilibrium* test [21].
## Data collection and outcome identification
Data on medical and family history, personal history, and history of systemic comorbidity were collected in a self-reported questionnaire that was completed during a face-to–face interview. conducted by a trained interviewer. The primary outcome was the prevalence of self-reported nephrolithiasis that was ascertained from the history/-treatment of nephrolithiasis (4–6). Accordingly, we enrolled 120,267 individuals (44,151 men and 76,116 women) for whom information of genotyping and nephrolithiasis were available. Among them, 5,086 men and 3,324 women with self-reported nephrolithiasis identified from TWB database were included in the case group. The remaining 111,857 (39,065 male and 72,792 female) participants without a history or family history of nephrolithiasis were included in the control group.
The relevant biochemical and lifestyle data of the participants were extracted from the TWB database, and the following covariates were evaluated: sex, age, BMI, alcohol consumption, smoking, physical activity, serum UA levels and systemic comorbidity.
BMI ≥24 kg/m2 was defined as overweight for the East Asian population. Habitual alcohol consumption was defined as the intake of more than 150 mL alcohol per week for at least 6 months and smoking was defined as daily use of tobacco continuously for at least 6 months. The extent of physical activity was dichotomized as non-regular and regular exercise (>30 minutes of exercise at least three times a week). Systemic comorbidity included hyperlipidemia, which was defined based on administration of lipid-lowering therapy or a physician’s diagnosis is based on objective parameters (TC ≥200 mg/dL, LDL-C ≥ 130 mg/dL, TG ≥ 150 mg/dL, or HDL-C < 40 or <50 mg/dL in men or women respectively), hypertension (systolic and/or diastolic BP ≥ $\frac{140}{90}$ mmHg), diabetes mellitus (receiving hypoglycemic agents, HbA1c ≥ $6.5\%$, random plasma glucose ≥200 mg/dL, fasting plasma glucose ≥ 126mg/dL, and/or diagnosed by physicians), and hyperuricemia (serum UA level >7.0 mg/dL measured by the uricase method using an Architect i2000SR Analyzer, Abbott Diagnostics, Abbott Park, Chicago, IL, United States).
## Statistical analysis
Quantitative variables are expressed as the mean ± standard deviation (SD). Mann-Whitney U test for continuous variables and Chi-square test for categorical variables were conducted to compare variables between nephrolithiasis cases and non-nephrolithiasis controls. In the sex-stratified analysis, statistical differences between the cases and controls with the rs2231142 genotypes and relationships between categorical variables were analyzed by the chi-square test. The associations between the ABCG2 rs2231142 variant and incident nephrolithiasis with hyperuricemia was analyzed using a multivariable-adjusted logistic regression model for potential confounding factors. Odds ratios (OR) and $95\%$ confidence intervals (CI) were calculated. All statistical analysis was performed in SAS version 9.4 (SAS Institute Inc. Cary NC). Significance was set at $p \leq 0.05$ and $p \leq 0.005$ when appropriate. Post-hoc analysis with Bonferroni correction was utilized to reduce the chance of false-positive results in multiple pairwise tests.
## Demographics at baseline and incident nephrolithiasis
Among the 120,267 participants aged 30–70 years, including 44,151 men and 76,116 women, 8,410 participants were identified with nephrolithiasis (prevalence: $11.52\%$ in men and $4.37\%$ in women). The characteristics of the participants are shown in Table 1. The mean age of the cases was significantly higher than that of the controls (53.6 ± 9.8 vs. 49.7 ± 11.0, $p \leq 0.001$), and men had significantly higher risk of nephrolithiasis than women ($60.5\%$ vs. $39.5\%$, $p \leq 0.001$). Compared with the GG genotype, the ABCG2 rs2231142 TT ($11.1\%$ vs. $9.9\%$ $p \leq 0.001$) and GT ($45.0\%$ vs$.42.9\%$, $p \leq 0.001$) genotypes were more frequently observed in the cases. The prevalence of hyperlipidemia ($14.8\%$ vs. $7.0\%$, $p \leq 0.001$), hypertension ($26.3\%$ vs. $11.4\%$, $p \leq 0.001$), diabetes mellitus ($10.0\%$ vs. $5.0\%$, $p \leq 0.001$), and hyperuricemia ($23.0\%$ vs. $12.7\%$, $p \leq 0.001$) were significantly higher in the nephrolithiasis group than in the controls. Moreover, lifestyle factors including smoking ($41.3\%$ vs. $26.4\%$, $p \leq 0.001$), alcohol consumption ($8.7\%$ vs. $5.8\%$, $p \leq 0.001$), regular physical exercise ($44.8\%$ vs. $40.2\%$, $p \leq 0.001$) and overweight ($61.5\%$ vs. $46.8\%$, $p \leq 0.001$) were significantly higher in the cases compared with that in the controls.
**Table 1**
| Variables | Without nephrolithiasis(n=111,857) | Without nephrolithiasis(n=111,857).1 | With nephrolithiasis(n=8,410) | With nephrolithiasis(n=8,410).1 | p-value |
| --- | --- | --- | --- | --- | --- |
| Variables | n | (%) | n | (%) | p-value |
| Age, years a | 49.7 ± 11.0 | 49.7 ± 11.0 | 53.6 ± 9.8 | 53.6 ± 9.8 | <0.001 |
| Sex b | | | | | <0.001 |
| female | 72792 | (65.1) | 3324 | (39.5) | |
| male | 39065 | (34.9) | 5086 | (60.5) | |
| Overweight (BMI≧24 kg/m2) b | | | | | <0.001 |
| No | 59511 | (53.2) | 3233 | (38.5) | |
| Yes | 52276 | (46.8) | 5166 | (61.5) | |
| ABCG2 rs2231142 b | | | | | <0.001* |
| GG | 45642 | (47.3) | 3206 | (43.9) | |
| GT | 41387 | (42.9) | 3291 | (45.0) | |
| TT | 9536 | (9.9) | 812 | (11.1) | |
| Hyperlipidemia b | | | | | <0.001 |
| No | 104044 | (93.0) | 7162 | (85.2) | |
| Yes | 7813 | (7.0) | 1248 | (14.8) | |
| Hypertension b | | | | | <0.001 |
| No | 99130 | (88.6) | 6202 | (73.7) | |
| Yes | 12727 | (11.4) | 2208 | (26.3) | |
| Diabetes mellitus b | | | | | <0.001 |
| No | 106270 | (95.0) | 7568 | (90.0) | |
| Yes | 5587 | (5.0) | 842 | (10.0) | |
| Uric acid (mg/dL) b | | | | | <0.001* |
| <5 | 46831 | (41.9) | 2244 | (26.7) | |
| 5-7 | 50742 | (45.4) | 4227 | (50.3) | |
| >7 | 14208 | (12.7) | 1933 | (23.0) | |
| Smoking b | | | | | <0.001 |
| No | 82243 | (73.6) | 4931 | (58.7) | |
| Yes | 29570 | (26.4) | 3476 | (41.3) | |
| Alcohol consumption b | | | | | <0.001 |
| No | 105309 | (94.2) | 7669 | (91.3) | |
| Yes | 6453 | (5.8) | 731 | (8.7) | |
| Regular physical exercise b | | | | | <0.001 |
| No | 66907 | (59.8) | 4636 | (55.2) | |
| Yes | 44895 | (40.2) | 3768 | (44.8) | |
## Sex-related differences in nephrolithiasis across the rs2231142 genotype categories
The sex-stratified frequency of the ABCG2 rs2231142 genotype and association with nephrolithiasis is shown in Table 2. Significantly more TT ($10.5\%$ vs. $9.9\%$) and GT ($44.8\%$ vs. $43.0\%$) genotypes were found in the female population with nephrolithiasis compared with female controls ($p \leq 0.05$). A similar distribution of the ABCG2 rs2231142 variants in the male population with nephrolithiasis was observed, with TT and GT genotypes occurring more frequently in participants with nephrolithiasis ($p \leq 0.001$). The serum UA- and sex-stratified association between the rs2231142 genotype and nephrolithiasis are shown in Figure 1. The OR of incident nephrolithiasis in the male study cohort with UA >7mg/dL was 1.30 ($95\%$ CI: 1.15–1.48, $$p \leq 0.001$$) and 1.33 ($95\%$ CI: 1.12–1.58, $p \leq 0.001$) for GT and TT carries, respectively, and with UA 5–7mg/dL was 1.28 for TT carriers ($95\%$ CI: 1.10–1.49, $$p \leq 0.002$$).
## Association of demographics, comorbidities, the ABCG2 rs2231142 variant, and lifestyle factors, with the risk of nephrolithiasis
To evaluate the association of demographics, comorbidities, rs2231142 variant, lifestyle factors with nephrolithiasis, we created a multivariable-adjusted logistic regression model (Table 3). In comparison with the GG genotype, the multivariable-adjusted OR ($95\%$ CI) of the GT and TT genotypes for the risk of nephrolithiasis was 1.12 (1.06–1.18) and 1.18 (1.09–1.28), respectively ($p \leq 0.001$ for both). Moreover, the OR ($95\%$ CI) of incident nephrolithiasis was 1.03 (1.02–1.03), 2.21 (2.08–2.36), 1.40 (1.29–1.51), 1.67 (1.57–1.78), 1.16 (1.06–1.26), 1.25 (1.16–1.36), 1.15 (1.09–1.22), and 1.22 (1.15–1.28) for higher age, male sex, hyperlipidemia, hypertension, diabetes mellitus, hyperuricemia, smoking, and overweight, respectively ($p \leq 0.001$ for all, except $$p \leq 0.002$$ for diabetes mellitus). Interestingly, regular physical exercise is a protective factor against nephrolithiasis (OR=0.95, $95\%$ CI: 0.9–1.0, $$p \leq 0.048$$).
**Table 3**
| Variables | OR | 95% CI | p-value |
| --- | --- | --- | --- |
| Age, years | 1.03 | (1.02-1.03) | <0.001 |
| Sex | Sex | Sex | Sex |
| female | 1.00 | ─ | ─ |
| male | 2.21 | (2.08-2.36) | <0.001 |
| Overweight (BMI ≥24 kg/m2) | Overweight (BMI ≥24 kg/m2) | Overweight (BMI ≥24 kg/m2) | Overweight (BMI ≥24 kg/m2) |
| No | 1.00 | ─ | ─ |
| Yes | 1.22 | (1.15-1.28) | <0.001 |
| ABCG2rs2231142 | ABCG2rs2231142 | ABCG2rs2231142 | ABCG2rs2231142 |
| GG | 1.00 | ─ | ─ |
| GT | 1.12 | (1.06-1.18) | <0.001 |
| TT | 1.18 | (1.09-1.28) | <0.001 |
| Hyperlipidemia | Hyperlipidemia | Hyperlipidemia | Hyperlipidemia |
| No | 1.00 | ─ | ─ |
| Yes | 1.40 | (1.29-1.51) | <0.001 |
| Hypertension | Hypertension | Hypertension | Hypertension |
| No | 1.00 | ─ | ─ |
| Yes | 1.67 | (1.57-1.78) | <0.001 |
| Diabetes mellitus | Diabetes mellitus | Diabetes mellitus | Diabetes mellitus |
| No | 1.00 | ─ | ─ |
| Yes | 1.16 | (1.06-1.26) | 0.002 |
| Uric acid (mg/dL) | Uric acid (mg/dL) | Uric acid (mg/dL) | Uric acid (mg/dL) |
| <5 | 1.00 | ─ | ─ |
| 5-7 | 1.07 | (1.01-1.14) | <0.026 |
| >7 | 1.25 | (1.16-1.36) | <0.001 |
| Smoking | Smoking | Smoking | Smoking |
| No | 1.00 | ─ | ─ |
| Yes | 1.15 | (1.09-1.22) | <0.001 |
| Alcohol consumption | Alcohol consumption | Alcohol consumption | Alcohol consumption |
| No | 1.00 | ─ | ─ |
| Yes | 0.92 | (0.84-1.01) | 0.083 |
| Regular physical exercise | Regular physical exercise | Regular physical exercise | Regular physical exercise |
| No | 1.00 | ─ | ─ |
| Yes | 0.95 | (0.90-1.00) | 0.048 |
## Discussion
In this study, we confirmed the importance of the ABCG2 rs2231142 variant with which the risk of nephrolithiasis is significantly associated. The risk of incident nephrolithiasis increased with hyperlipidemia, hypertension, diabetes mellitus, and hyperuricemia as well as with lifestyle factors (overweight and smoking) rather than active alcohol consumption. Regular physical exercise conferred a protective effect against nephrolithiasis.
The lifetime prevalence of nephrolithiasis increases with age. The highest prevalence of nephrolithiasis in the general population was observed in those over 60 years in the United States and in Asia individuals aged 30–60 years [1, 5]. The prevalence of nephrolithiasis showed a sex difference and was more common among men than in women, with a declined ratio to 1.26 in the United States and 1.3–5.0 in Asia [1, 5]. As the rates of overweight and metabolic syndrome increased significantly, they promoted an increase of $2.9\%$ nephrolithiasis among US women from 2007 through 2018 [5, 22]. In the Taiwanese population, the prevalence of nephrolithiasis that was ascertained using the National Health Insurance Research Database in 2010 was $7.38\%$ in the entire cohort, with a male predominance ($9.01\%$ vs. $5.79\%$ [in women]) [23]. In the present study, the mean age at onset of nephrolithiasis was 53.6 ± 9.8 years, and men were more likely to have nephrolithiasis than women (OR=2.21).
The ABCG2 rs2231142 genotype has a strong association with hyperuricemia and gout through the mechanisms of both decreased intestinal urate excretion and renal overload in the Asian population [15, 16]. The history of gout was evidently linked to a higher prevalence of nephrolithiasis [5]. The ABCG2 gene contributes to inflammation in gouty arthritis that is mediated via the release of Interleukin 8 (IL-8) following MSU crystals-stimulation in an endothelial cell model [24]. Furthermore, the ABCG2 gene played a crucial role in the aberrant generation of pro-inflammatory cytokines such as interleukin-1 beta (IL-1β), tumor necrosis factor-alpha (TNF-α) and IL-8 in gout. The oxidative stress triggered by reactive oxygen species (ROS) and pro-inflammatory cytokines results in gouty arthropathy, and also leads to renal damage in the kidney [25]. Moreover, the rs2231142 T-allele was associated with early-onset gout and tophaceous disease in Western Polynesian individuals with gout [16, 26]. A 10-year observational study showed significantly higher tophaceous gout in the early-onset group [27]. The expression of pro-inflammatory cytokines and receptor activator of nuclear factor κB ligand (RANKL) in T-cell induced by the MSU crystals of chronic gouty arthritis leads to osteoclastogenesis [28]. These mechanisms occur in gout with a 1.2-fold increased risk of osteoporosis [29] which may play an independent risk factor for development of nephrolithiasis [30]. Our study is the first to reveal that the risk of nephrolithiasis was markedly increased in participants with the ABCG2 rs2231142 T genotypes. Our finding concurs with the results in previous studies and suggests that the association between the ABCG2 genetic variants may contribute to the formation of nephrolithiasis through inflammation elicited by hyperuricemia and gout.
Dyslipidemia is an independent risk factor, not only for recurrent multi-stone nephrolithiasis but also for abnormalities in urinary constituents [31, 32]. In addition, a study from NorthShore University Health System showed that incident nephrolithiasis significantly increased with higher TG levels and could be partly prevented by statin use [33]. In the present study, we discovered a significant association between hyperlipidemia and incident nephrolithiasis. Hypertension may contribute to the incidence of nephrolithiasis through increased urinary calcium excretion [34]. In addition, the risk of hypertension increased by $58\%$ after the first symptomatic kidney stone event [35]. Although the biological mechanism underlying the association between hypertension and nephrolithiasis is unknown, there appears to be a bidirectional relationship [34, 35]. Consistent with the results of a previous study, this study demonstrated that hypertension is significantly associated with incident nephrolithiasis [36]. Furthermore, insulin resistance may lead to lower urinary pH values in the renal proximal tubule and hyperuricosuria, which is linked to nephrolithiasis [37]. The urate stones were more frequently observed in patients with diabetes mellitus than in patients without diabetes [38]. In this study, we discovered a significant association between type 2 diabetes and nephrolithiasis, and found that hyperuricemia is an independent risk factor for incident nephrolithiasis. Asymptomatic hyperuricemia was an independent risk factor for nephrolithiasis in a large cohort study [39]. Hyperuricosuria, low urinary pH level, and low urinary volume (dehydration) were crucial drivers that promoted nephrolithiasis [3, 11]. A possible mechanism underlying the aforementioned relationship is that the dissolved sodium urate could salt out CaOx from native urine [13]. Taken together, our data supported an independent role of genetic factors in nephrolithiasis after controlling for potential contributing comorbidities. Further investigations are needed to elucidate the pathogenesis of ABCG2 in nephrolithiasis.
Smoking is associated with oxidative stress and leads to renal damage and chronic kidney disease [40]. Moreover, crystallization modulators, such as osteopontin, bikunin and α-microglobulin, could be generated by smoking and thus contribute to lithogenesis in the kidneys [41]. We discovered an association between incident nephrolithiasis and cigarette smoking, rather than with alcohol consumption. This result was supported by another study which demonstrated that current smoking, but not active alcohol use, was an attributable risk for incident calcium urolithiasis [42].
It is well-known that BMI is associated with biomarker of systemic inflammation of the metabolic syndrome [43]. A previous study demonstrated that the prevalence of overweight and obesity was significantly higher in patients with urolithiasis [44]. In addition, overweight and obesity are liked to insulin resistance; lower urinary pH level; and excretion of more urate, sodium, calcium, oxalate and phosphate in urine [44, 45]. The excess dietary intake of lithogenic substances and a lithogenic urinary profile may predispose obese patients to nephrolithiasis [44]. Therefore, it would be beneficial to implement dietary modifications and control comorbidities to reduce the risk of nephrolithiasis in obese patients.
We discovered a protective association between incident nephrolithiasis and regular physical exercise. During physical exercise, body fluid is lost and the sensation of thirst could induce increased water intake. Moreover, sodium is lost through diaphoresis in exercise and resulted in an approximately $50\%$ reduction in the excretion of urinary sodium and decreased the urinary output [46]. Collectively, these mechanisms may lead to an expansion of 20–$25\%$ of the circulating blood volume and reduce sympathetic tone and cardiovascular diseases after exercise [46]. Furthermore, resistance exercise confers a benefit by decreasing urinary calcium excretion [47]. Our result supports and advocates that regular physical activity could reduce incident nephrolithiasis [48].
The ABCG2 rs2231142 variants not only lead from hyperuricemia to gouty arthropathy, but also increase the risk of incident nephrolithiasis. Thus, the screening of ABCG2 rs2231142 variants is as important as the evaluation of systemic comorbidities and lifestyle factors for patients with nephrolithiasis and hyperuricemia. We infer that the first choice in urate-lowering therapy for those with nephrolithiasis, the ABCG2 rs2231142 T allele, and gout should be xanthine oxidase inhibitors rather than an uricosuric agent.
There are some limitations of this study that to be mentioned. First, this study was a cross-sectional design from TWB database. However, all of the risk factors for nephrolithiasis have been prospectively collected. We believe that ABCG2 rs2231142 T-allele might contribute to the formation of nephrolithiasis through inflammation elicited by hyperuricemia and gout. Second, we did not include all lifestyle risk factors such as detailed dietary information and that could affect the risk of nephrolithiasis. Nevertheless, our data support weight control and smoking cessation to mitigate the risk of nephrolithiasis. Third, the occurrence of nephrolithiasis was collected by self-reported history without confirmation of medical review and the urine chemistry and stone composition were unavailable. Therefore, we cannot exclude the possibility of information bias. Moreover, the impact of concomitant urate-lowering agents was not taken into consideration. However, our result provides robust evidence that ABCG2 rs2231142 genetic variations are associated with both hyperuricemia and incident nephrolithiasis.
## Conclusion
This retrospective case-control study using data from TWB revealed that besides higher age, males sex, smoking, overweight, comorbidities and hyperuricemia, the ABCG2 rs2231142 genotypes was independently associated with nephrolithiasis. To implement precision healthcare, a test of ABCG2 rs2231142 polymorphism could offer additional guidance for lifestyle modification in the prophylactic management of patients with multiple systemic comorbidities and hyperuricemia.
## Data availability statement
The original contributions presented in the study are included in the article/ Supplementary Material. Further inquiries can be directed to the corresponding author.
## Ethics statement
This research project was approved by the ethics committee of Taichung Veterans General Hospital Institutional Review Board (IRB No. CE16270B-2). The patients/participants provided their written informed consent to participate in this study.
## Author contributions
C-TL and Y-MC conceptualized the study. C-TL, Y-HuC, I-CC, C-HL, and Y-MC were responsible for data curation. Y-HuC, and I-CC were responsible for formal analysis. Y-MC and C-HL were responsible for funding acquisition. Y-JC, Y-CL, J-CC, T-JW, W-NH, and Y-HsC were responsible for investigation. C-TL, W-NH, Y-HsC, C-HL, and Y-MC were responsible for methodology. Y-JC, Y-CL, J-CC, and T-JW, C-HL, and Y-MC were responsible for the resources. W-NH, Y-HsC, C-HL, and Y-MC provided supervision. C-TL, Y-HuC, I-CC, C-HL, and Y-MC were responsible for the validation. C-TL, I-CC, and Y-MC were responsible for visualization and wrote the original draft. C-TL, I-CC, C-HL and Y-MC reviewed and edited the manuscript. Each author contributed important intellectual content during manuscript drafting or revision and agrees to be personally accountable for the individual’s own contributions and to ensure that questions pertaining to the accuracy or integrity of any portion of the work, even one in which the author was not directly involved, are appropriately investigated and resolved, including with documentation in the literature if appropriate. All authors contributed to the article and approved the submitted version.
## Conflict of interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher’s note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## Supplementary material
The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fendo.2023.1074012/full#supplementary-material
## References
1. Liu Y, Chen Y, Liao B, Luo D, Wang K, Li H. **Epidemiology of urolithiasis in Asia**. *Asian J Urol* (2018) **5**. DOI: 10.1016/j.ajur.2018.08.007
2. Sorokin I, Mamoulakis C, Miyazawa K, Rodgers A, Talati J, Lotan Y. **Epidemiology of stone disease across the world**. *World J Urol* (2017) **35**. DOI: 10.1007/s00345-017-2008-6
3. Pak CY. **Kidney stones**. *Lancet (London England)* (1998) **351**. DOI: 10.1016/s0140-6736(98)01295-1
4. Stamatelou KK, Francis ME, Jones CA, Nyberg LM, Curhan GC. **Time trends in reported prevalence of kidney stones in the united states: 1976-1994**. *Kidney Int* (2003) **63**. DOI: 10.1046/j.1523-1755.2003.00917.x
5. Abufaraj M, Xu T, Cao C, Waldhoer T, Seitz C, D'Andrea D. **Prevalence and trends in kidney stone among adults in the USA: Analyses of national health and nutrition examination survey 2007-2018 data**. *Eur Urol Focus* (2021) **7**. DOI: 10.1016/j.euf.2020.08.011
6. Wang W, Fan J, Huang G, Li J, Zhu X, Tian Y. **Prevalence of kidney stones in mainland China: A systematic review**. *Sci Rep* (2017) **7**. DOI: 10.1038/srep41630
7. Pearle MS, Calhoun EA, Curhan GC. **Urologic diseases in America project: Urolithiasis**. *J Urol* (2005) **173**. DOI: 10.1097/01.ju.0000152082.14384.d7
8. Shoag J, Halpern J, Goldfarb DS, Eisner BH. **Risk of chronic and end stage kidney disease in patients with nephrolithiasis**. *J Urol* (2014) **192**. DOI: 10.1016/j.juro.2014.05.117
9. Trinchieri A, Ostini F, Nespoli R, Rovera F, Montanari E, Zanetti G. **A prospective study of recurrence rate and risk factors for recurrence after a first renal stone**. *J Urol* (1999) **162** 27-30. DOI: 10.1097/00005392-199907000-00007
10. Khan SR, Pearle MS, Robertson WG, Gambaro G, Canales BK, Doizi S. **Kidney stones**. *Nat Rev Dis Primers* (2016) **2** 16008. DOI: 10.1038/nrdp.2016.8
11. Moe OW. **Kidney stones: Pathophysiology and medical management**. *Lancet (London England)* (2006) **367**. DOI: 10.1016/s0140-6736(06)68071-9
12. Sakhaee K, Adams-Huet B, Moe OW, Pak CY. **Pathophysiologic basis for normouricosuric uric acid nephrolithiasis**. *Kidney Int* (2002) **62**. DOI: 10.1046/j.1523-1755.2002.00508.x
13. Grover PK, Marshall VR, Ryall RL. **Dissolved urate salts out calcium oxalate in undiluted human urine in vitro: Implications for calcium oxalate stone genesis**. *Chem Biol* (2003) **10**. DOI: 10.1016/s1074-5521(03)00057-7
14. Ettinger B, Tang A, Citron JT, Livermore B, Williams T. **Randomized trial of allopurinol in the prevention of calcium oxalate calculi**. *N Engl J Med* (1986) **315**. DOI: 10.1056/nejm198611273152204
15. Yamagishi K, Tanigawa T, Kitamura A, Köttgen A, Folsom AR, Iso H. **The Rs2231142 variant of the Abcg2 gene is associated with uric acid levels and gout among Japanese people**. *Rheumatol (Oxford England)* (2010) **49**. DOI: 10.1093/rheumatology/keq096
16. Matsuo H, Yamamoto K, Nakaoka H, Nakayama A, Sakiyama M, Chiba T. **Genome-wide association study of clinically defined gout identifies multiple risk loci and its association with clinical subtypes**. *Ann Rheum Dis* (2016) **75**. DOI: 10.1136/annrheumdis-2014-206191
17. Dehlin M, Jacobsson L, Roddy E. **Global epidemiology of gout: Prevalence, incidence, treatment patterns and risk factors**. *Nat Rev Rheumatol* (2020) **16**. DOI: 10.1038/s41584-020-0441-1
18. Chen YJ, Chen IC, Lin HJ, Lin YC, Chang JC, Chen YM. **Association of Abcg2 Rs2231142 allele and bmi with hyperuricemia in an East Asian population**. *Front Genet* (2021) **12**. DOI: 10.3389/fgene.2021.709887
19. Chen CH, Yang JH, Chiang CWK, Hsiung CN, Wu PE, Chang LC. **Population structure of han Chinese in the modern Taiwanese population based on 10,000 participants in the Taiwan biobank project**. *Hum Mol Genet* (2016) **25**. DOI: 10.1093/hmg/ddw346
20. Pan WH, Fann CS, Wu JY, Hung YT, Ho MS, Tai TH. **Han Chinese cell and genome bank in Taiwan: Purpose, design and ethical considerations**. *Hum Heredity* (2006) **61** 27-30. DOI: 10.1159/000091834
21. Purcell S, Neale B, Todd-Brown K, Thomas L, Ferreira MA, Bender D. **Plink: A tool set for whole-genome association and population-based linkage analyses**. *Am J Hum Genet* (2007) **81**. DOI: 10.1086/519795
22. Wang Y, Beydoun MA, Min J, Xue H, Kaminsky LA, Cheskin LJ. **Has the prevalence of overweight, obesity and central obesity levelled off in the united states? trends, patterns, disparities, and future projections for the obesity epidemic**. *Int J Epidemiol* (2020) **49**. DOI: 10.1093/ije/dyz273
23. Huang WY, Chen YF, Carter S, Chang HC, Lan CF, Huang KH. **Epidemiology of upper urinary tract stone disease in a Taiwanese population: A nationwide, population based study**. *J Urol* (2013) **189**. DOI: 10.1016/j.juro.2012.12.105
24. Chen CJ, Tseng CC, Yen JH, Chang JG, Chou WC, Chu HW. **Abcg2 contributes to the development of gout and hyperuricemia in a genome-wide association study**. *Sci Rep* (2018) **8** 3137. DOI: 10.1038/s41598-018-21425-7
25. Zamudio-Cuevas Y, Hernández-Díaz C, Pineda C, Reginato AM, Cerna-Cortés JF, Ventura-Ríos L. **Molecular basis of oxidative stress in gouty arthropathy**. *Clin Rheumatol* (2015) **34**. DOI: 10.1007/s10067-015-2933-y
26. Zaidi F, Narang RK, Phipps-Green A, Gamble GG, Tausche AK, So A. **Systematic genetic analysis of early-onset gout: Abcg2 is the only associated locus**. *Rheumatol (Oxford England)* (2020) **59**. DOI: 10.1093/rheumatology/kez685
27. Gao Q, Cheng X, Merriman TR, Wang C, Cui L, Zhang H. **Trends in the manifestations of 9754 gout patients in a Chinese clinical center: A 10-year observational study**. *Joint Bone Spine* (2021) **88** 105078. DOI: 10.1016/j.jbspin.2020.09.010
28. Lee SJ, Nam KI, Jin HM, Cho YN, Lee SE, Kim TJ. **Bone destruction by receptor activator of nuclear factor κb ligand-expressing T cells in chronic gouty arthritis**. *Arthritis Res Ther* (2011) **13** R164. DOI: 10.1186/ar3483
29. Kok VC, Horng JT, Wang MN, Chen ZY, Kuo JT, Hung GD. **Gout as a risk factor for osteoporosis: Epidemiologic evidence from a population-based longitudinal study involving 108,060 individuals**. *Osteoporos Int* (2018) **29**. DOI: 10.1007/s00198-018-4375-2
30. Rendina D, D'Elia L, Evangelista M, De Filippo G, Giaquinto A, Barone B. **Osteoporosis is a predictive factor for nephrolithiasis in an adult free-living Caucasian population from southern Italy: A longitudinal retrospective study based on a general practice database**. *Calcified Tissue Int* (2020) **107**. DOI: 10.1007/s00223-020-00737-9
31. Kohjimoto Y, Sasaki Y, Iguchi M, Matsumura N, Inagaki T, Hara I. **Association of metabolic syndrome traits and severity of kidney stones: Results from a nationwide survey on urolithiasis in Japan**. *Am J Kidney Dis* (2013) **61**. DOI: 10.1053/j.ajkd.2012.12.028
32. Masterson JH, Woo JR, Chang DC, Chi T, L'Esperance JO, Stoller ML. **Dyslipidemia is associated with an increased risk of nephrolithiasis**. *Urolithiasis* (2015) **43** 49-53. DOI: 10.1007/s00240-014-0719-3
33. Cohen AJ, Adamsky MA, Nottingham CU, Pruitt J, Lapin B, Wang CH. **Impact of statin intake on kidney stone formation**. *Urology* (2019) **124** 57-61. DOI: 10.1016/j.urology.2018.01.029
34. Cappuccio FP, Strazzullo P, Mancini M. **Kidney stones and hypertension: Population based study of an independent clinical association**. *BMJ (Clinical Res ed)* (1990) **300**. DOI: 10.1136/bmj.300.6734.1234
35. Kittanamongkolchai W, Mara KC, Mehta RA, Vaughan LE, Denic A, Knoedler JJ. **Risk of hypertension among first-time symptomatic kidney stone formers**. *Clin J Am Soc Nephrol* (2017) **12**. DOI: 10.2215/cjn.06600616
36. Borghi L, Meschi T, Guerra A, Briganti A, Schianchi T, Allegri F. **Essential arterial hypertension and stone disease**. *Kidney Int* (1999) **55**. DOI: 10.1046/j.1523-1755.1999.00483.x
37. Abate N, Chandalia M, Cabo-Chan AV, Moe OW, Sakhaee K. **The metabolic syndrome and uric acid nephrolithiasis: Novel features of renal manifestation of insulin resistance**. *Kidney Int* (2004) **65**. DOI: 10.1111/j.1523-1755.2004.00386.x
38. Daudon M, Traxer O, Conort P, Lacour B, Jungers P. **Type 2 diabetes increases the risk for uric acid stones**. *J Am Soc Nephrol* (2006) **17**. DOI: 10.1681/asn.2006030262
39. Kim S, Chang Y, Yun KE, Jung HS, Lee SJ, Shin H. **Development of nephrolithiasis in asymptomatic hyperuricemia: A cohort study**. *Am J Kidney Dis* (2017) **70**. DOI: 10.1053/j.ajkd.2017.01.053
40. Jaimes EA, Tian RX, Raij L. **Nicotine: The link between cigarette smoking and the progression of renal injury**. *Am J Physiol Heart Circulatory Physiol* (2007) **292**. DOI: 10.1152/ajpheart.00693.2006
41. Khan SR. **Renal tubular Damage/Dysfunction: Key to the formation of kidney stones**. *Urol Res* (2006) **34** 86-91. DOI: 10.1007/s00240-005-0016-2
42. Liu CC, Huang SP, Wu WJ, Chou YH, Juo SH, Tsai LY. **The impact of cigarette smoking, alcohol drinking and betel quid chewing on the risk of calcium urolithiasis**. *Ann Epidemiol* (2009) **19**. DOI: 10.1016/j.annepidem.2009.02.006
43. West B, Luke A, Durazo-Arvizu RA, Cao G, Shoham D, Kramer H. **Metabolic syndrome and self-reported history of kidney stones: The national health and nutrition examination survey (Nhanes iii) 1988-1994**. *Am J Kidney Dis* (2008) **51**. DOI: 10.1053/j.ajkd.2007.12.030
44. Trinchieri A, Croppi E, Montanari E. **Obesity and urolithiasis: Evidence of regional influences**. *Urolithiasis* (2017) **45**. DOI: 10.1007/s00240-016-0908-3
45. Kahn BB, Flier JS. **Obesity and insulin resistance**. *J Clin Invest* (2000) **106**. DOI: 10.1172/jci10842
46. Convertino VA. **Blood volume: Its adaptation to endurance training**. *Med Sci sports Exercise* (1991) **23**. DOI: 10.1249/00005768-199112000-00004
47. Vezzoli G, Soldati L, Arcidiacono T, Terranegra A, Biasion R, Russo CR. **Urinary calcium is a determinant of bone mineral density in elderly men participating in the inchianti study**. *Kidney Int* (2005) **67**. DOI: 10.1111/j.1523-1755.2005.00302.x
48. Sorensen MD, Chi T, Shara NM, Wang H, Hsi RS, Orchard T. **Activity, energy intake, obesity, and the risk of incident kidney stones in postmenopausal women: A report from the women's health initiative**. *J Am Soc Nephrol* (2014) **25**. DOI: 10.1681/asn.2013050548
|
---
title: Fatty acid transport protein inhibition sensitizes breast and ovarian cancers
to oncolytic virus therapy via lipid modulation of the tumor microenvironment
authors:
- Abera Surendran
- Monire Jamalkhah
- Joanna Poutou
- Rayanna Birtch
- Christine Lawson
- Jaahnavi Dave
- Mathieu J. F. Crupi
- Justin Mayer
- Victoria Taylor
- Julia Petryk
- Christiano Tanese de Souza
- Neil Moodie
- Jacob Lecompte Billingsley
- Bradley Austin
- Nicole Cormack
- Natalie Blamey
- Reza Rezaei
- Curtis W. McCloskey
- Emily E. F. Fekete
- Harsimrat K. Birdi
- Serge Neault
- Taylor R. Jamieson
- Brenna Wylie
- Sarah Tucker
- Taha Azad
- Barbara Vanderhyden
- Lee-Hwa Tai
- John C. Bell
- Carolina S. Ilkow
journal: Frontiers in Immunology
year: 2023
pmcid: PMC10036842
doi: 10.3389/fimmu.2023.1099459
license: CC BY 4.0
---
# Fatty acid transport protein inhibition sensitizes breast and ovarian cancers to oncolytic virus therapy via lipid modulation of the tumor microenvironment
## Abstract
### Introduction
Adipocytes in the tumour microenvironment are highly dynamic cells that have an established role in tumour progression, but their impact on anti-cancer therapy resistance is becoming increasingly difficult to overlook.
### Methods
We investigated the role of adipose tissue and adipocytes in response to oncolytic virus (OV) therapy in adipose-rich tumours such as breast and ovarian neoplasms.
### Results
We show that secreted products in adipocyte-conditioned medium significantly impairs productive virus infection and OV-driven cell death. This effect was not due to the direct neutralization of virions or inhibition of OV entry into host cells. Instead, further investigation of adipocyte secreted factors demonstrated that adipocyte-mediated OV resistance is primarily a lipid-driven phenomenon. When lipid moieties are depleted from the adipocyte-conditioned medium, cancer cells are re-sensitized to OV-mediated destruction. We further demonstrated that blocking fatty acid uptake by cancer cells, in a combinatorial strategy with virotherapy, has clinical translational potential to overcome adipocyte-mediated OV resistance.
### Discussion
Our findings indicate that while adipocyte secreted factors can impede OV infection, the impairment of OV treatment efficacy can be overcome by modulating lipid flux in the tumour milieu.
## Introduction
The complex tumor microenvironment (TME) consists of a collection of malignant cell types as well as infiltrating and resident cells (e.g., fibroblasts, adipocytes, vascular endothelial cells, and immune cells), secreted factors, and extracellular matrix [1]. Due to recent advancements in our understanding of tumor biology, it is well known that the TME plays a critical role in cancer progression; however, there is increasing evidence to suggest that many stromal elements can also significantly modulate the therapeutic effects of cancer drugs [2]. Among the TME players, cancer-associated adipocytes are an underappreciated driver of both tumor progression and anti-cancer therapy resistance [3, 4].
The composition of the TME varies significantly between tumor types [1]. A hallmark feature of breast tumors and metastatic ovarian cancers is the abundant presence of surrounding and infiltrating adipocytes (3, 5–9). Adipocytes in the fatty breast TME play critical roles in promoting tumor progression by stimulating cancer cell motility and invasion [10, 11]. Over $80\%$ of women diagnosed with ovarian cancer exhibit metastasis in a large abdominal pad of fat cells, called the omentum [7]. Nieman and colleagues eloquently showed that adipocytes promote both homing of ovarian cancer cells to the omentum and tumor growth through adipocyte-secreted interleukins and other factors [8]. In addition, adipocytes have a demonstrated role in actively promoting anti-breast and ovarian cancer therapy resistance across a wide range of treatment modalities, including chemotherapy, radiation, targeted therapy, monoclonal antibody therapy, and even immunotherapies (4, 6, 12–15). Recent studies revealed that the transfer of bioactive molecules from the omental microenvironment decreased chemotherapy-induced apoptosis of ovarian tumor cells [16, 17]. In the context of breast cancer, surrounding adipocytes promote neoplastic cell resistance to the anti-human epidermal growth factor receptor 2 protein antibody Trastuzumab [3], and expression of programmed cell death-ligand 1 (PD-L1) in mammary adipocytes attenuates anti-tumor immunity [6]. The mechanisms of adipocyte-driven resistance are diverse, and this is strongly reflective of the complex ways in which adipocytes can communicate with cancer cells and condition the TME.
Cancer-lysing or oncolytic viruses (OVs) are a unique class of cancer immunotherapeutic drugs [18]. Infected cancer cells eventually succumb to oncolysis, which facilitates the spread of the virus in the TME to further infection, tumor debulking, and even abscopal effects due to a systemic response despite a localized treatment [19]. Moreover, tumor lysis and the release of tumor-associated antigens act as in situ vaccination in the TME and contribute to generating a robust host-mediated anti-tumor immune response to wake up immunologically inert or ‘cold’ tumors to become pro-inflammatory or ‘hot’ [19]. Several decades of pre-clinical optimization and clinical testing preceded the regulatory approval of different virotherapeutics for cancer treatment. An oncolytic picornavirus, named Rigvir®, was the first platform that achieved approval in 2004 for treating people with advance melanoma in Latvia [20]. Shortly after, China approved Oncorine (H101), an attenuated adenoviral vector for the treatment of head and neck cancer in combination with chemotherapy [21]. Nearly a decade later, the FDA and the EMA approved T-VEC (Imlygic®), a genetically modified herpes simplex virus-1 (HSV-1) for treating unresectable melanoma [22]. Since then, many other viruses, including oncolytic vesicular stomatitis virus (VSVΔ51) (23–25), vaccinia virus (VACV) [26, 27], and measles virus (MeV) [28] have been exploited as cancer-killing agents and evaluated in pre-clinical and clinical studies globally.
The hallmarks of cancer also act as cellular properties that confer selectivity for OV therapy [29]; however, successful tumor colonization by OVs is partly governed by the tumor ecosystem and crosstalk between its cellular compartments. As obligate intracellular parasites, viruses, including OVs, seek a niche favorable for their growth and spread. Given the enrichment of adipocyte-derived fatty acid species in the adipose-rich TME, we speculated about the role of adipocyte-derived lipids in the context of OV infection. Numerous studies evaluating the effect of fatty acids on virus infection have demonstrated pleiotropic effects (30–33). To determine if a lipid-rich niche impacts the activity of clinically staged OVs, we evaluated tumors in adipose-rich TMEs in vivo or in cancer cells cultured with adipocyte-secreted factors in vitro. We found that tumors in a fatty niche were more resistant to OV infection compared to tumors grown in less fatty tissues. When lipid-derived constituents were depleted from the adipocyte-conditioned medium (ACM), OV infection and OV-mediated cell death were re-instated. These findings led us to investigate the effects of depleting or inhibiting specific fatty acid transport proteins (FATPs) on OV infection and OV-mediated killing in the presence of adipocyte-secreted factors. The combination of virotherapy with a FATP inhibitor improved OV infection in vitro and enhanced survival in models of syngeneic fat-pad localized breast cancer or intraperitoneal ovarian cancer. Our findings show, for the first time, that fatty acid blockade in lipid-rich TMEs can sensitize resistant tumors to oncolytic virotherapy.
## A fat-rich microenvironment impedes productive OV infection
Tumor cells can interact with neighboring adipocytes, and this crosstalk appears to reduce the efficacy of certain anti-cancer drugs (34–37). To explore the effect of tumor localization on responses to OV infection in vivo, we implanted mouse breast tumors in the mouse hind flank (HF) or mammary fat pad (FP) and then treated them intratumorally with VSVΔ51. In all three syngeneic breast tumor models evaluated (EO771, 4T1, EMT6), tumors seeded near the adipocyte-rich mammary fat pad were significantly less infected than their HF counterparts (Figures 1A–C). We observed similar OV infection patterns in a spontaneously transformed ovarian surface epithelial (STOSE) model [38, 39]. Tumors in the FP or intrabursal (IB) to the ovary were significantly more resistant to OV infection than subcutaneous HF tumors (Figure 1D). C57BL/6 mice bearing mammary FP breast tumors were fed either a high-fat diet (HFD) or a regular chow diet (RD) to impact their body weight (Supplementary Figure 1A) and adipose tissue invasion into the tumor (Supplementary Figure 1B). Mice fed with HFD showed decreased intratumoral OV titer in comparison to mice on RD (Figure 1E), suggesting that the increased presence of adipocytes in the fatty TME impairs virus replication in the tumor.
**Figure 1:** *A Fatty TME correlates with OV resistance. (A–C) Syngeneic EO771 (A), 4T1 (B), and EMT6 (C) breast tumors were seeded in the fat-pad (FP) or subcutaneously in the hind flank (HF) of Balb/c or C57BL/6 mice. Tumors were intratumorally treated with VSVΔ51 [1.5E8 plaque forming units (PFU)/tumor] and collected after 48 hours for quantification of infectious particles. ND, non-detected. Two-tailed, unpaired t-test. (D) STOSE cells were seeded in the HF, FP or intrabursally (IB) in FVB/N mice and treated with Maraba MG1 (1E7 PFU/tumor). Tumors were harvested 72 hours post-infection (hpi) for virus quantification by plaque assay. ND, non-detected. Lines indicates the median. One-way ANOVA, Tukey’s multiple comparison test. (E) Following a period of high-fat or regular diet feeding of C57BL/6 mice, EO771 cells were seeded in the FP. Tumors were intratumorally treated with Maraba MG1 (5E8 PFU/tumor) and harvested for quantification by plaque assay 48 hpi. Two-tailed, unpaired t-test. (F) Relative percentage (% compared to uninfected cells) of VSVΔ51-induced cytotoxicity in OVCAR8, SKOV3, and a primary ovarian cancer cell culture (AF2068) cultured in a regular growth medium (CTL) or a human breast adipocyte-conditioned medium (ACM) for 16 hours prior to infection with indicated OVs (48 hours, MOI 0.1). Results are displayed as mean ± SEM of four biological replicates. Two-tailed, unpaired t-test for each CTL and ACM pair. (G) Oncolytic virus titers from an OV-infected panel of breast and ovarian cancer cell lines [OVCAR8, MDA MB 231 and SKOV3 (MOI 0.1, 48 hpi), and OVCAR433 and BT549 (MOI 1, 48 hpi)] cultured in a regular growth medium or ACM were quantified by plaque assay. Data indicate the mean ± SEM of 3 to 5 biological replicates. Two-tailed, unpaired t-test for each CTL vs ACM pair. *p <0.05, **p<0.01, *p<0.05, **p<0.01, ****p<0.0001. (H) Representative immunofluorescence microscopy images of OV-infected (VSVΔ51 MOI 0.1 24 hpi, VACV MOI 0.1 48 hpi, HSV-1 MOI 0.1, 48 hpi, MeV MOI 1, 48 hpi) OVCAR8 cells are shown (VSVΔ51 and VACV, scale bar 20μm; HSV-1 and MeV scale bar represents 50μm). ns, non-significant.*
To explore potential effects of adipocytes on OV infection, we assessed the role of adipocyte-secreted factors in promoting or inhibiting viral replication. A representative panel of breast and ovarian cancer cell lines, including an ascites-derived ovarian cancer cell culture (AF2068) were exposed to adipocyte-conditioned medium [(ACM), Supplementary Figure 1C] and infected with clinically relevant OVs [19], including oncolytic VSVΔ51 [23, 24], Maraba MG1 [40], VACV [26, 27], MeV [28] and HSV-1 [22]. While ACM treatment did not significantly impact cell proliferation in our experimental settings (Supplementary Figure 1D), we observed decreased OV-induced cytotoxicity (Figure 1F) and virus replication in the presence of ACM compared to control medium (CTL) (Figure 1G) or pre-adipocyte conditioned medium-receiving cells (Supplementary Figure 1E). The stark nature of this resistant effect is dose-dependent (Supplementary Figure 1F) and apparent in the immunofluorescence microscopy images of OV-infected ovarian and breast cancer cells, as shown in Figure 1H and Supplementary Figure 1G. Furthermore, ACM from an alternative adipose depot, such as human visceral adipocytes (Supplementary Figure 2A–C), or different species [(i.e., mouse adipocytes) Supplementary Figure 2D, F] also inhibited OV infection as observed with human breast adipocytes. These data suggest that one or more secreted products of human or murine adipocyte lineage contribute to OV resistance.
## ACM-mediated impairment to OV replication occurs post-virus entry
To better understand the phenomenon of adipocyte-driven OV resistance, we sought to examine various steps of the virus replication cycle. A virus must first attach to and penetrate a host cell to make copies of itself to generate new virus particles that can then spread to neighboring cells [19]. Thus, we first evaluated whether the pre-treatment of virions in ACM was sufficient to neutralize the infectious particles and thus impair their attachment and internalization in cancer cells. Our data revealed that a direct virus-neutralizing agent is unlikely to be the driver of ACM-mediated OV inhibition. Pre-incubation of virus particles with ACM did not compromise OV infection or virus-induce cell death (Figure 2A). Furthermore, we conducted a modified plaque formation assay in which cancer cells were infected in the presence of ACM for varying lengths of time. Next, the virus inoculum-containing medium was replaced with a semi-solid medium that permitted the formation of viral plaques. While OV infection in ACM conditions leads to a significant decrease in viral transcripts (Figure 2B), the plaque formation experiment revealed that exposure to ACM solely at the time of infection or early-on in the infectious cycle (0-4 hrs) does not significantly impair OV infection (Figure 2C). To further investigate the role of ACM on virus entry, we employed virus-like particles (VLPs). VLPs, also known as viral “empty shells”, can mimic the structural properties of native viruses without the capacity for replication since they do not carry a viral genome. We employed a retrovirus-derived VLP platform [41] containing Gag proteins fused to the green fluorescent protein (Gag-GFP) and pseudotyped with VSV glycoprotein (VSV-G) (Figure 2D). In ACM treated cells, the percentage of GFP positive cells following VSV infection was diminished considerably, whereas the percentage of GFP positive VLP-transduced cells was not significantly ($$p \leq 0.35$$) altered (Figures 2E, F, Supplementary Figure 3). Thus, VSV-G-mediated cell entry appeared unhindered in ACM despite a severe impairment to virus replication. A comprehensive evaluation of the effect of the timing of ACM exposure on OV titers corroborated the findings from the VLP studies, suggesting that ACM exposure at later stages of the infection cycle has the most significant influence on mounting OV resistance (Figure 2G). Taken together, these findings suggest that adipocyte-secreted factors reduce OV infection at a step after virus entry.
**Figure 2:** *ACM-mediated OV resistance occurs post-virus entry and is a Type I IFN-independent phenomenon. (A) Cytotoxicity induced by VSVΔ51 inoculums (MOI 1, 48hpi) that were pre-incubated in CTL medium or ACM at room temperature (RT) or 37°C for the indicated lengths of times and then used to infect OVCAR8 cells in CTL medium. (B) OVCAR8 cells in CTL medium were infected with MG1ΔG (MOI 3). After 1 hour, the virus inoculum was removed and changed to CTL medium or ACM. The RNA was extracted at 24 hpi, and the expression of VSV gene L was evaluated by qPCR relative to cellular Rplp0 loading control. Data indicate the mean ± SEM of three biological replicates. Two-tailed, unpaired t-test. (C) Cells were incubated in CTL medium or ACM for the indicated amount of time prior to infection with VSVΔ51 (150 infectious particles). Following 1 hour of infection, the virus inoculum was removed and replaced with a semi-solid overlay. After 48, viral titers (in PFU) were quantified and plotted. Lines connect means. Two-way ANOVA, Sidak’s multiple comparison test. (D) Cartoon comparing a VSV-eGFP virion and a GFP-loaded VLP pseudotyped with VSV-G. (E, F) CTL medium or ACM-receiving cells were infected with GFP-expressing VSVΔ51 (MOI 0.1) or transduced with GFP-loaded VLPs pseudotyped with VSV-G for 24 hours and percentage of GFP positive cells was assessed by flow cytometry. Data indicate the mean ± SEM (n=7). Two-tailed, unpaired t-test for VLP or VSVΔ51-infected samples. (G) As summarized in the displayed experimental timeline, OVCAR8 cells were cultured overnight in CTL medium or ACM, and the medium was changed to CTL medium at the time of VSVΔ51 infection. Alternatively, the medium was changed to a CTL medium or ACM at the time of infection. The virus inoculum-containing medium was removed following an hour of infection and replaced with the same medium at the time of infection. Viral titers were quantified by plaque assay. Data indicate the mean ± SEM of 3 biological replicates. Two-tailed, unpaired t-test for each CTL vs ACM pair. (H, I) Cell lines or patient ascites cells were cultured in CTL medium or ACM overnight and treated with JAK I inhibitor (1μM), IFN-α (200 U/mL), or both prior to infection with VSVΔ51 (MOI 1) for 48 hours. Cytotoxicity assay was performed to evaluate virus-mediated cell death (H), and the supernatant was collected for quantification of infectious particles by plaque assay (I). Data indicate the mean ± SEM of 3 biological replicates. Two-way ANOVA, Tukey’s multiple comparison test on log 10 transformed data.*
## ACM-mediated OV resistance is not driven by Type I interferon
Adipocytes are endocrine cells that secrete a variety of bioactive molecules, including cytokines and adipokines [7, 8, 42]. Type I interferons (IFN-I) are often the first line of defense in the innate anti-viral response [29, 43]. Given the robust resistance of otherwise OV-sensitive cancer cells in the presence of adipocyte-secreted factors, we first sought to determine the potential role of IFN-I signaling on ACM-mediated OV resistance. IFN-α engages with the cellular interferon-α/β receptor (IFNAR). Topical treatment with IFN-α served as a positive control for engaging anti-viral signaling associated with an IFNAR-mediated interferon signaling cascade. When IFN-α treated cells were infected with the highly IFN-sensitive oncolytic VSVΔ51, little to no infection was observed (Supplementary Figure 4A). When cells were treated with both IFN-α and an IFNAR blocking antibody, the cells became vulnerable to VSVΔ51 infection, demonstrating that the concentration range of IFNAR blocking antibody was sufficient to block the binding and engagement of IFNAR ligands. Yet, in cancer cells receiving ACM, the addition of IFNAR blocking antibody did not alleviate resistance to VSVΔ51 infection, suggesting an IFNAR-independent mechanism of OV resistance (Supplementary Figure 4A). Like most proteins, IFNs are thermolabile and sensitive to proteolytic treatment [44]. We complemented the above findings by evaluating the effects of proteinase K treatment, heat-inactivation, or boiling of ACM, which led to no notable effects on mitigating OV resistance (Supplementary Figure 4B–G). Interferons work through activation of the Janus kinase-signal transducer and activator of transcription (JAK-STAT) pathway to activate a plethora of genes that are collectively known as interferon-stimulated genes [29]. To further validate that the ability of ACM to induce an anti-viral state is IFN-I independent, we evaluated the effect of blocking downstream IFN signaling using JAK inhibitors. JAK inhibitor-treated ACM-receiving cancer cell lines or patient ascites-derived cell cultures (AF2068 and AF2149) remained firmly resistant to infection despite a sensitizing effect in controls (Figures 2H, I). Collectively, these findings suggest that the ACM-driven OV resistance is a type-I IFN-independent phenomenon.
## Adipocyte-derived lipid moieties are the primary driver of OV resistance
Adipocytes are highly metabolic cells that store lipids and release them as free fatty acids [12]. Consistent with previous findings [7, 8], we found that the level of fatty acid in ACM compared to control medium was significantly higher ($$p \leq 0.004$$) (Supplementary Figure 5A), and cancer cells grown in the presence of ACM accumulated intracellular lipids over time (Figures 3A, B). Moreover, we observed an inverse correlation between intracellular lipid accumulation and OV infection (Figure 3B). Close examination of the transcriptome of ACM-treated cancer cells also revealed an increase in the expression of critical genes involved in fatty acid metabolism, including carnitine o-acyl transferase (CRAT), carnitine acylcarnitine translocase (SLC25A2O) and carnitine palmitoyl transferase I (CPT1), a critical mitochondrial enzyme involved in the β-oxidation of fatty acids [45] (Figure 3C and Supplemental Figure 5B–D). We therefore tested whether the accumulation of lipids and changes in the expression of genes involved in fatty acid oxidation contribute to altering the metabolism of cancer cells by assessing mitochondrial metabolic parameters using Agilent Seahorse technology. When etomoxir, a non-reversible inhibitor of CPT1 [45, 46], was added to OVCAR8 ovarian cancer cells, CTL medium-receiving cells showed no notable change in their oxygen consumption rate. In contrast, ACM-receiving cells demonstrated a drastic reduction in the window assessing spare respiratory capacity (Figure 3D), suggesting a reliance on CPT1-mediated respiration. These results suggest a metabolic dependency on fatty acids as a fuel source in ACM-receiving cancer cells.
**Figure 3:** *Adipocyte-derived lipid moieties are the key drivers of OV resistance. (A) OVCAR8 cells cultured in ACM for the indicated lengths of time and stained with Bodipy to label neutral lipid content. DAPI denotes nuclei staining. (B) SKOV3 cells were infected with an RFP-expressing VSVΔ51 (MOI 3, 18 hours) and stained with Bodipy for flow cytometry analysis. Data indicate the mean ± SEM of 3 biological replicates. Two-tailed, unpaired t-test. (C) OVCAR8 cells were cultured in ACM or CTL-medium for 18 hours prior to harvesting total RNA for RNA sequencing analysis. The data was analyzed relative to ACM to visualize the relative changes of gene expression in ACM from two biological replicates. Y-axis dotted line denotes a p-value cut-off of 0.05. X-axis dotted lines denote a 2-fold change cut off. (D) Oxygen consumption rate (OCR) in OVCAR8 cells cultured in a control medium or ACM was assessed using Agilent Seahorse technology. Oligomycin, trifluoromethoxy carbonylcyanide phenylhydrazone (FCCP), and antimycin and rotenone treatment at designated intervals. Acute treatment with Etomoxir (Eto) was assessed in parallel. (E–J) Lipid-depleted control or ACM medium was added to OVCAR8 cells for overnight culturing and infected with VSVΔ51 (MOI 0.1) or VACV (MOI 1) or HSV-1 (MOI 0.1). Virus-induced cytotoxicity is shown by crystal violet staining of attached cells (E–G). The number of infectious particles released by cancer cells cultured in the indicated conditions after 48 hours of infection was assessed by plaque assay (H–J). Data indicate the mean ± SEM of 3 biological replicates. Two-way ANOVA, Tukey’s multiple comparison test on log 10 transformed data.*
Adipocytes can release large amounts of soluble mediators that can directly or indirectly promote lipid accumulation in cancer cells, thereby contributing to their resistance to oncolytic virotherapy. Accordingly, we next explored whether we could restore OV infection and killing by blocking lipid biosynthesis in cancer cells cultured in ACM. TOFA (5-tetradecyloxy-2-furoic acid) is an allosteric inhibitor of acetyl-CoA carboxylase (ACC), an enzyme critical for long chain fatty acid synthesis [47]. Interestingly, we found that when cancer cells were treated with increasing concentrations of TOFA, there was an increase in susceptibility to OV infection and cell death; however, this sensitizing effect was absent in cells that also received adipocyte secreted factors (Supplementary Figure 6A, B). These findings may indicate that despite the beneficial effects of combination treatment with TOFA, OV resistance is reinstated when an exogenous source of lipids counteracts the absence of intracellular lipid synthesis. On the contrary, intracellular lipid accumulation decreased when cancer cells received lipid-depleted ACM (Supplementary Figure 6C), and this phenomenon correlated with an increase in OV titer and recovery in OV-mediated cytotoxicity (Figures 3E–J). Moreover, the inhibitory effect of ACM was similar across two different OV platforms when regular culture medium was supplemented with the fatty acid palmitate or with a chemically defined lipid mixture containing seven fatty acids and cholesterol (Supplementary Figure 6D–I). These studies provide functional evidence for the uptake of lipid moieties into receiving cancer cells and strongly suggest that ACM-mediated OV resistance is primarily a lipid-driven phenomenon.
## Blocking FA uptake as a combinatorial strategy to sensitize tumors in a fatty niche to virotherapy
To develop a strategy that can improve oncolytic virotherapy in adipocyte-rich TMEs, we sought to investigate the effect of blocking the uptake of lipids by cancer cells. Numerous proteins have been implicated in transporting fatty acids, including fatty acid translocase/CD36, caveolin-1, and fatty acid transport proteins (FATP1-6) [48, 49]. FATPs are one of the most heavily studied families of proteins involved in the transport of fatty acids. However, of the FATP family members, only FATP1, 2, and 4 have been shown to directly participate in fatty acid transport [50, 51]. Thus, we evaluated the effect of FATP1, 2, and 4 downregulation (Supplementary Figure 7A) on OV infection of cancer cells. We found that while FATP1 or FATP4 knockdown had no apparent effect on bolstering OV infection (Supplementary Figure 7B); FATP2 downregulation noticeably increased OV infection in OVCAR8 cells cultured in ACM (Figure 4A, Supplementary Figure 7B). Alternatively, when we overexpressed FATP2 in OVCAR8 cells using a tetracycline-inducible system (Supplementary Figure 7C), OV infection was reduced in both conditions receiving control medium or ACM (Figure 4B). To determine whether adipocyte-secreted lipid molecules and their uptake in cancer cells via FATP2 are responsible for the observed OV resistance, we treated FATP2-mediated fatty acid accumulation with a small molecule inhibitor. Lipofermata is a widely studied FATP2 inhibitor (FATP2i) that was identified in a yeast system expressing human FATP2 and has been demonstrated to inhibit FATP2 in cell lines and animal models [52, 53]. We found that Lipofermata recapitulated the phenomenon observed in the FATP2 knockdown experiments. While Lipofermata did not impact the viability of OVCAR8 cancer cells (Supplementary Figure 7D), we observed that Lipofermanta-treated cells were more susceptible to OV infection and killing in the presence of ACM than the vehicle control-treated counterparts (Figure 4C and Supplementary Figure 7E).
**Figure 4:** *Blocking fatty acid uptake can resensitize cancer cells to OV infection and improve response to OV therapy. (A) OVCAR8 cells were transfected with siRNAs targeting FATP2 (SLC27A2) for 16 hours and then infected with VSVΔ51 (MOI 0.05) for an hour. The virus inoculum was removed and replaced with CTL medium or ACM. The supernatants were titrated by plaque assay 48 hpi. Data indicate the mean ± SEM (n=3). Two-way ANOVA, Bonferroni’s multiple comparisons test on log 10 transformed data. (B) Representative images of doxycycline (Dox) inducible FATP2 expressing OVCAR8 cells cultured in ACM or CTL medium and infected with VSVΔ51 (MOI 0.01, 48 hpi) or oncolytic HSV-1 (MOI 0.01, 72 hpi) are shown. (C) OVCAR8 cells cultured in CTL medium or ACM and treated with vehicle control or a FATP2i (Lipofermata 0.24 μM) were infected with VSVΔ51 (MOI 0.1) and stained with crystal violet (96 hpi). (D, E) A single dose of VSVΔ51 was delivered IT into FP E0771 (n=14-20 per group) or OVCAR8 (n=22 per group) tumors. After 48 hours post-delivery, virus titers were quantified. Data represent mean values ± SEM. Unpaired two-tailed t-test. (F) Timeline for VSVΔ51 and FATP2i treatment of EO771 bearing mice and Kaplan Meier survival analysis of EO771 fat-pad tumor bearing C57bl/6 mice received vehicle (n=3), FATP2i (3mg/kg) (n=4), VSVΔ51 (1E8 PFU) (n=5) or both FATP2i and VSVΔ51 (n=5) by intratumoral delivery as indicated in the experimental timeline. (G) ID8-F3 p53 -/- intraperitoneal (IP) tumor bearing C57bl/6 mice received vehicle (n=4), FATP2i (3mg/kg) (n=4), VSVΔ51 (3E8 PFU) (n=5) or both FATP2i and VSVΔ51 (n=5) by IP delivery and assessed for survival as shown in the timeline. Log-rank (Mantel-Cox) test, *p <0.05, **p <0.01. ns, non-significant.*
Next, we examined whether FATP2 inhibition can improve the therapeutic efficacy of an OV in pre-clinical breast and ovarian cancer models. Similar to our observations in the in vitro studies, evaluation of the intratumoral oncolytic VSVΔ51 titers in syngeneic breast EO771 mammary fat pad-localized tumors revealed a Lipofermata-driven increase in OV replication at the site of the tumor (Figure 4D). We found similar results when we used a human ovarian OVCAR8 tumor model engrafted in immunodeficient animals (Figure 4E). Moreover, mice bearing breast EO771 orthotopic tumors and receiving a combination treatment of Lipofermata and oncolytic VSVΔ51 showed the best survival advantage compared to either treatment alone (Figure 4F). Similarly, in an immunocompetent model of intraperitoneal ovarian cancer (ID8 p53-/-), combination treatment with Lipofermata provided the best survival advantage, while neither treatment alone provided a survival benefit (Figure 4G). The aforementioned pre-clinical studies demonstrate that combination treatment with a FATP2i can enhance intratumoral virus growth and the therapeutic benefits of OV therapy in fatty tumor models.
## Discussion
The cellular composition of the TME varies among tumor types creating permissive or hostile niches for oncolytic virus activity. For example, the defining feature of the pancreatic TME is its fibrotic stroma consisting primarily of cancer-associated fibroblasts. We have shown previously that these cells sensitize neoplastic pancreatic cells to OV killing via the secretion of FGF-2 [54]. Conversely, our in vitro and in vivo studies presented herein suggest adipose-rich microenvironments as negative modulators of OV therapeutic activity in the breast and ovarian cancer settings. For the first time and as summarized in Figure 5, we showed that adipocyte-secreted molecules could severely impair OV replication and impede OV-mediated tumor cell destruction (Figure 1, and Supplementary Figures 1 and 2). Of note, the inhibitory effect of ACM was observed across viruses from four distinct families (i.e., Rhabdoviridae, Poxviridae, Herpesviridae, and Paramyxoviridae). These oncolytic viruses are genetically diverse and unique in the host receptors they require for cell entry and their modes and sites of replication. The concordant inhibitory effect of ACM on multiple viruses with such wide diversity in their methods of infection and replication suggests that ACM-mediated impediment to OV infection is more likely an indirect effect, for example, through alterations to the host cell, rather than a direct effect on the virions themselves or the virus-receptor interactions. Our data suggest a post-virus entry (Figures 2A–G and Supplementary Figure 3) and Type I interferon-independent (Figures 2H, I and Supplementary Figure 4) mechanism of resistance. Viruses are obligate intracellular parasites, and their activity and successful growth depends entirely on the host cell nutrients, energy, and metabolites [55]. Future work should focus on deepen our understanding of the potential cellular changes induced by adipocyte-secreted factors, including reprogramming of metabolic pathways, which can impact OV infection and replication in cancer cells.
**Figure 5:** *Working model of the effects of adipocyte-derived lipids on OV resistance and the OV sensitizing effects of combination therapy with FATP2 inhibition. (A) Tumors in an adipocyte-rich microenvironment accumulate intracellular lipids that lead to a decreased sensitivity to oncolytic virus infection and cancer cell killing. (B) Intracellular lipid accumulation can be mitigated by blocking the uptake of exogenous adipocyte-derived lipids by depletion of the lipid-derived constituents in the ACM or by using specific inhibitors to block the activity of fatty acid transporter proteins, such as FATP2. For example, Lipofermanta, a specific FATP2 inhibitor, can re-sensitize cancer cells to OV infection and virotherapy-mediated cell death.*
One of the earliest experiments demonstrating fatty acid-mediated anti-viral effects was observed in breast milk [56]. Since then, several studies revealed a potential role of lipids in perturbing infection and replication of various viruses (30–33). The identification of lipid species, including fatty acids, as the primary driver of ACM-mediated OV resistance was a landmark finding in this study (Figure 3 and Supplementary Figures 5, 6). Our investigations showed that cancer cells treated with ACM accumulated neutral lipids and displayed a resistant state against OVs. We hypothesized that limiting lipid uptake by cancer cells in a fat-rich microenvironment, could override the inhibitory effects of adipocyte-derived metabolites on OV platforms. There are numerous proteins involved in fatty acid transport. Among them, FATPs are one of the most heavily studied. In our study, FATP2 downregulation by siRNAs or FATP2 inhibition with the small molecule Lipofermata improved OV-mediated cell death and OV infection in vitro and in vivo (Figure 4 and Supplementary Figure 7). Of note, *Lipofermata is* one of the best-characterized inhibitors of FATP2 with promise for use in the clinic. To date, only a handful of studies have evaluated Lipofermata in the context of cancer [53, 57]. These studies showed that the in vivo anti-tumor effects of Lipofermata are driven by modification of immune cells in the TME, specifically neutrophils, MDSCs, and indirectly, T-cells [57]. While Lipofermata has previously been shown to slow tumor growth, in our studies, Lipofermata monotherapy did not provide significant tumor control or survival benefit. Given that Lipofermata has been demonstrated to have direct anti-cancer activity on cancer cells in a concentration-specific manner in vitro, we speculate that the non-toxic concentrations used in our in vivo study provided a sufficient fatty acid limiting effect without inducing cellular toxicity that may impair productive OV replication, ultimately supporting the therapeutic benefit of OV combination therapy. Of note, Lipofermata has been shown to be a non-competitive inhibitor for long and very long fatty acids that prevent cellular dysfunction and cell death induced by excessive exposure to fatty acids [52]. Here, we have demonstrated that FATP2i improves the therapeutic potential of OV therapy in both immunocompetent and immunodeficient tumor models (Figure 4). Our findings build on a growing body of evidence for FATP2 blockade as a therapeutic strategy to overcome anti-cancer therapy resistance [58]. However, future work should explore the therapeutic benefits of blocking other modes of fatty acid transport. While downregulation of other fatty acid transporters, such as FATP1 and FATP4, did not improve the sensitivity of ACM-treated cells to OV infection in our models (Supplementary Figures 7A, B), there are many other membrane-associated putative fatty acid carriers that display increased expression in cancer cells, including FATPs other than FATP2, the fatty acid translocase CD36 and fatty acid binding proteins (FABPs) (59–61). Notably, CD36 has demonstrated roles in potentiating tumor growth and anti-cancer treatment resistance by facilitating the transport FA substrates in the tumor microenvironment (62–64). Given that CD36 has been implicated in driving both ovarian and breast cancers, it could be an attractive target in CD36-expressing tumors [62, 64]. Future studies may seek to determine if tailored FA blockade or simultaneously targeting of FA transporters can alter exogenous lipid uptake to enhance the sensitivity of tumor cells in lipid-rich microenvironments to virotherapy. Collectively, our work shows for the first time that fatty acid blockade can improve the therapeutic impact of OV therapy in lipid-rich TMEs.
The therapeutic advantage of FATP2 targeting during virotherapy treatment could be due to several reasons. FATP2 transports long and very long chain fatty acids but has a demonstrated preference for polyunsaturated fatty acids (PUFAs) [65]. Several PUFAs, including docosahexaenoic acid and arachidonic acid, have previously been shown to have anti-viral activity by diverse mechanisms, including by interfering with binding to host entry receptors or a post-entry mechanism such as inhibiting genome replication [66, 67]. Characterization of PUFA species and their relative quantity may provide clues about the lipid environment contributing to FATP2-mediated OV resistance. Further characterization of FATP2-mediated fatty acid transport and the intracellular metabolic fate of translocated fatty acids may shed more light on how FATP2 guided uptake of fatty acid species drives the OV resistance observed in our studies.
While the mechanisms of lipid-mediated virus impairment are not yet well understood, it is plausible to speculate a link between the endoplasmic reticulum (ER) stress, lipid metabolism and the antiviral state observed in cancer cells exposed to adipocyte-secreted molecules. ER is a key organelle involved synthesizing, folding, and modifying proteins. When the protein folding capacity of the ER is disrupted, increasing quantities of unfolded or misfolded proteins are detected by ER resident sensors. This leads to the activation of the unfolded protein response as several mechanisms intended to reinstate ER homeostasis. Lipid metabolism and ER stress are closely linked and bi-directional. For example, some lipid species, including fatty acids, have been shown to induce ER stress and the expression of the thioredoxin-interacting protein (TXNIP) [68]. TXNIP inhibits the antioxidative effects of thioredoxin, an important regulator in redox signaling, and it is also implicated in regulating cellular metabolism and ER stress [69, 70]. Our RNA sequencing data revealed TXNIP as an upregulated gene in ACM conditions (Supplementary Figure 5B). Given that viruses exploit the ER for replication, assembly, and egress, an impairment to ER function due to a lipotoxic load can form a significant barrier to successful infection. In fact, some evidence suggests that downregulating TXNIP may have a sensitizing effect on virus infection. In a study by Tiwarekar et al., silencing of this gene increased *Measles virus* replication [71]. Albeit speculative, it is reasonable to hypothesize that the benefit of combinatorial treatment with the FATP2 inhibitor Lipofermata may at least partially derive from Lipofermata-mediated ER stress relief. Previous studies showed that Lipofermata-attenuated palmitate transport corresponded with a decrease in the expression of lipotoxicity mediated cell stress markers [52]. Future studies may seek to determine the role of TXNIP and ER stress in lipotoxic TMEs, in the context of OV therapy. Testing combinatorial treatments that include drugs that restore ER homeostasis and virotherapy in tumors homed in adipose-rich microenvironments could provide opportunities to characterize new OV resistance mechanisms and develop new therapeutic approaches. Overall, a deeper understanding of how the adipocyte-cancer cell crosstalk influences virus-based therapies would allow the enhancement of OV therapy and unlock other potential new therapeutic avenues for tumors surrounded by environments where fat cells are abundant.
## Methods
The research presented herein complies with all ethical regulations at OHRI and the University of Ottawa (biohazardous material use certificate GC317-125-12). All animal studies were approved by the institutional animal care committee of the University of Ottawa (Protocol ID: OHRI2870) and carried out following the guidelines of the National Institutes of Health and the Canadian Council on animal care.
## Cell lines and cell culture conditions
Cell lines (OVCAR8, SKOV3, OVCAR433, MDA MB 231, BT549, MCF-7, 4T1, EMT6, EO771, B16-F10, 3T3-L1) were purchased from American Type Culture Collection (ATCC, Manassas, VA), except ID8 p53-/- which was a gift from Dr. Iain McNeish (Barts Cancer Institute, Queen Mary University of London, UK), and STOSE cells which were a gift from Dr. Barbara Vanderhyden’s lab (Ottawa Hospital Research Institute, Ottawa, ON). Cells were cultured in RPMI-1640 with $10\%$ FBS and buffered with 25mM HEPES, except for MDA MB 231 and MCF-7, which were maintained in DMEM with $10\%$ FBS. Vero cells were cultured in a medium containing %10 NCS/FBS mix (3:1 NCS: FBS). STOSE cells were cultured in DMEM containing $5\%$ FBS and ITS (5μg/mL insulin, 5μg/mL transferrin and 5ng/mL sodium selenite; Gibco). Human breast and visceral preadipocytes (Zenbio) were differentiated with the appropriate proprietary adipocyte differentiation medium (Zenbio). Briefly, the preadipocytes were cultured in RPMI containing $10\%$ FBS in T75 flasks until at least $80\%$ confluency. The preadipocytes were cultured in the appropriate differentiation medium for ten days, with a medium change every 3-4 days. Following differentiation, the cells were acclimatized to a regular culture medium (RPMI-1640 with $10\%$ FBS and containing HEPES) for at least five days prior to the collection of adipocyte-conditioned medium (ACM). ACM was harvested every 48-72 hours. Each ACM batch was quality controlled by conducting a test with cancer cells that are highly susceptible to oncolytic virus infection. The cells received control media or ACM and were infected with a GFP-expressing oncolytic virus. After a 24-hour period, microscopy studies were conducted to assess GFP expression, a proxy for infection, and oncolytic virus mediated cell death. For each ACM batch, we looked for resistance that was >60-$70\%$ (vs CTL media). The inhibitory effect did not vary greatly between batches. The medium was centrifuged (1500rpm, 5mins), and the supernatant was stored at -20°C until further use. Murine 3T3-L1 cells were differentiated in the recommended differentiation medium (Zenbio). All cell lines were incubated at 37°C in a $5\%$ CO2 humidified incubator. All cells were tested with the e-Myco VALiD Myco PCR detection kit (FroggaBio) or Hoescht’s staining to ensure they were devoid of mycoplasma contamination prior to the experiments described.
## Generation of OVCAR8 cells over-expressing FATP2
Tetracycline inducible over-expressing cell lines were generated using lentiviral vectors. Briefly, Lenti-X™ 293T cells (Takara Bio) were transfected with packaging plasmids pMD2G and pPAX2 and a doxycycline inducible pTRIPZ lentivirus vector encoding human FATP2-FLAG tag (synthesized by GenScript). OVCAR8 cells were transduced with cell-free lentiviral vector-containing supernatant in the presence of 1 µg/mL of polybrene (Sigma, St Louis, MO, USA) and were selected with 1 μg/mL of puromycin. Human FATP2 expression was induced with doxycycline (1 μg/mL), and 48 hours post-induction, the FATP2 overexpression was demonstrated by western blot analysis using FATP2 polyclonal antibody (Proteintech, 14048-1-AP) and monoclonal Anti-FLAG® M2 antibody (Millipore Sigma, F1804-200UG).
## Propagation and purification
The oncolytic VSVΔ51, Maraba MG1, oncolytic HSV-1, oncolytic MeV, and VV TK- VGF- virus backbones and propagation and purification protocols have previously been described [24].
## Titration of VSV and HSV infectious particles by plaque assay
The day prior, 8E5 or 2E5 Vero cells were seeded on 6-well or 12-well, respectively. Infectious viral particles from the supernatant of VSVΔ51-infected cells or virus particles from the supernatant and cell-associated viral particles from HSV-infected cells were pooled (cell-associated viruses were liberated by bursting the cells with 3 freeze-thaw cycles by freezing the sample at -80°C and thawing in a 37°C water bath until thawed but before warm). The sample was clarified of cell debris by centrifugation (1500rpm, 10mins, 4°C). The virus samples were then diluted in a 10-fold serial dilution in cold serum-free DMEM medium in deep-well plates. The medium in the plates containing the Vero cells was aspirated and replaced with 500-600μL of the serially diluted virus. The inoculated plates were incubated in the cell incubator (37°C) with gentle rocking every 10-15mins. After 45 minutes, the inoculum was aspirated and replaced with an overlay medium [1:1 mixture of $6\%$ carboxymethyl cellulose (CMC; Sigma): DMEM (2X concentrated with $20\%$ NCS/FBS mix)]. After a 48-hour incubation, the overlay medium was aspirated, and the cells were gently washed with PBS twice and stained with crystal violet ($0.1\%$ crystal violet in $80\%$ MeOH in water) for 20-30 minutes. The titers were reported as plaque-forming units (PFU) per mL of sample.
## Titration of VACV by plaque assay
Infected cells undergo three freeze-thaw cycles to release intracellular particles. Following centrifugal clarification of cell debris, the sample is tittered similarly to the plaque assay of VSV described above, with a few modifications. The samples were tittered on U2OS cells, and the overlay medium is made with $3\%$ Carboxymethyl cellulose (CMC). The titers were reported as PFU per mL of sample.
## MeV titration
The supernatant of MeV infected cells was serially diluted in a 10-fold serial dilution. 10μL of the diluted viral stock was combined with 90μL of Vero cells prepared in a 150 000 cells/mL solution in DMEM ($10\%$FBS) in the wells of a 96-well plate. The plate was incubated in a 37°C incubator for three days. The titer was determined based on the $50\%$ tissue culture infective dose endpoint method described by Spearman–Karber. The titers were reported as PFU per mL of sample.
## Titration of tumor samples
Tumors were harvested, weighed, and transferred to 2mL Qiagen tubes containing 500uL of PBS-containing Complete™ Mini EDTA-free Protease inhibitor cocktail (1 tablet per 10mL PBS). The tumors were homogenized using a tissue lyser (TissueLyser II, Qiagen). The cell debris was spun down with a benchtop centrifuge (14000 rpm, 10 mins). The remaining cell debris was removed by passing the sample through a 70μm cell strainer (Fisherband). The sample was titered by plaque assay as described above; however, penicillin streptomycin (100 U/mL; Thermo Fisher Scientific) was added to the overlay medium. The titers were reported as PFU per milligram of tumor.
## OV pre-incubation with ACM
VSVΔ51-GFP or VACV were incubated in Eppendorf tubes containing RPMI-1640 ($10\%$ FBS, HEPES) or ACM in a 37°C incubator or at room temperature for various lengths of time. These pre-treated viral stocks were then used to infect cancer cells.
## Modified plaque assay to assess virus entry
The modified plaque assay to assess virus entry was adapted from Xu et al. Briefly, OVCAR8 cells were incubated in ACM at the indicated lengths of time before infection with 150 PFU of VSVΔ51-GFP. After an hour period of infection, the virus inoculum was removed and replaced with an overlay medium (1:1 mixture of $6\%$ CMC: 2X DMEM with $20\%$ NCS/FBS mix). After 48 hours, the overlay was removed, the cell monolayer was washed with PBS, and the wells were stained with crystal violet solution ($0.1\%$ crystal violet with $80\%$ MeOH). The plates were left to air dry overnight, and the PFUs were counted the next day.
## Using VLPs to assess viral entry
Gag-GFP loaded VLPs pseudotyped with VSV-G glycoprotein were generated using a previously described protocol [72]. Briefly, following transfection of plasmids encoding VLP components in HEK293T cells (ATCC), VLPs were harvested and concentrated as per manufacturer’s instructions (Lenti-X concentrator; Takara) to a 1 mL suspension in PBS. OVCAR8 cells were seeded at $80\%$ confluency in 6-well plates and were cultured in ACM for at least 4 hours before they were transduced with 30μL of VLPs [in the presence of 0.8μg/mL polybrene (Sigma)] or infected with VSVΔ51-GFP (MOI 0.1). After 24 hours, the cells were harvested, stained with propidium iodide, and assessed by flow cytometry.
## Quantitative real-time PCR
Cell monolayers were washed twice with PBS and the total RNA was extracted using the NucleoZOL kit (Macherey-Nagel). The RNA pellets were resuspended in 20 uL of nuclease-free water. RNA quantification was achieved using the Nanodrop™ One Microvolume UV-Vis Spectrophotometer (Thermo Fisher Scientific). cDNA was generated using the iScriptTM cDNA Synthesis Kit (Bio-Rad) using 1μg of RNA. Amplification was achieved using the QuantiTect SYBR Green PCR Kit (Qiagen) in conjunction with Applied Biosystems™ 7500 Fast Real-Time PCR System analysis. Gene expression was determined relative to rPlp0 using the delta delta Ct method [2^(ΔΔCt)]. Primers were acquired from IDT (Integrated DNA Technologies), and primer optimization was determined using the standard curve method. Primer sequences are shown in Supplemental Table 1.
## Flow cytometry
Visualization of neutral lipids was achieved with Bodipy $\frac{493}{503}$ (Molecular Probes) using a previously described protocol. Briefly, cells in 6-well plates were washed with PBS to remove medium and serum. They were then incubated with Bodipy staining solution (1:1000 diluted in PBS) for 15 minutes in the dark at 37°C. The cells were rinsed with PBS to remove the staining solution and trypsinized (Trypsin-EDTA $0.25\%$) to dislodge the cells from the well. The cells were resuspended in 5mL PBS, transferred to a 15mL conical tube, and pelleted by centrifugation (1500 rpm, 5 minutes). Next, the cells were resuspended in 200 μL of PBS and transferred to a 96-well V-bottom plate (Corning® 96-well Clear V-Bottom TC-treated Microplate). After centrifugation (1500 rpm, 5 minutes) to pellet the cells, they were stained with fixable viability stain 510 (BD Horizon™; 0.5μL of stain was added per well in a total volume of 25 μL PBS) and incubated for 15 minutes in the dark. The cells were diluted with 150 μL PBS and centrifuged (1500 rpm, 5 minutes). Then, the cells were washed with flow buffer once before resuspending in 200 μL of flow buffer and transferred to 1.1 mL tall microtubes (Thermo Scientific™ Microtubes for 1.1mL Microtube System). The samples were analyzed on the BD LSR Fortessa or BD Celesta flow cytometers at the University of Ottawa Flow Cytometry and Virometry core facility (Roger Guindon Hall, University of Ottawa). Data analysis was conducted on FlowJo software (FlowJo, LLC, Ashland, OR).
## Crystal violet staining
The medium was aspirated, and the cells were stained with crystal violet solution ($0.5\%$ crystal violet in $80\%$ MeOH in water) for 40 minutes on a benchtop rocker. The crystal violet was removed, and the monolayer was gently washed with tap water twice and left to air dry overnight. Qualitative analysis of the stained plates was achieved by scanning the plates. Cytotoxicity was quantified by lifting the crystal violet staining from the stained plates. The plates were incubated with $100\%$ methanol on a shaker for 40 minutes. The methanol containing crystal violet was transferred to a 96-well plate, and the OD570 was read using the Multiskan Ascent plate reader.
## LDH assay
A CyQuantTM cytotoxicity assay (ThermoFisher) was used as per the manufacture’s protocol to measure virus-induced cell death.
## Immunofluorescence microscopy
Brightfield and GFP filter images of cells seeded on plates were acquired with an AMG EVOS fluorescence microscope (Advanced Microscopy Group, Washington, USA). Immunofluorescence images were also obtained of cells that were grown on glass coverslips. As per a previously described protocol, lipid staining was achieved with Bodipy $\frac{493}{503}$ (Molecular Probes). Briefly, cells were washed with PBS and incubated with a Bodipy staining solution (1:1000 in PBS; 15 minutes, 37°C) in the dark. Next, the cells were washed with PBS and fixed with $4\%$ PFA (Thermo Fisher) for 20 minutes. The PFA was aspirated, and the cells were washed twice with PBS before the coverslips were mounted on slides with a mounting medium containing DAPI (ProLong™ Gold Antifade Mounting solution, Thermo Fisher Scientific). The slides were visualized with Zeiss Fluorescent Microscopes (ZEISS, Germany).
## IFNAR blocking
ACM-receiving cells incurred a medium change to ACM the evening before the day of the experiment. Anti-human IFNAR2 neutralizing monoclonal antibody (2ug; PBL Assay Science) was added to wells, and four hours later, human IFN-α 2b (200U; Sigma Aldrich) was added to the appropriate wells. Four hours post-IFN-α addition, the cells were infected with VSVΔ51-GFP (MOI 1).
## JAK inhibition
Cell lines or ovarian patient ascites cells were cultured in ACM overnight. The next day the cells were treated with 1μM Jaki (EMD Millipore) for 3 hours before receiving 200 U/mL human IFN-α 2b (Sigma Aldrich). After 3 hours of IFN-α treatment, the cells were infected with VSVΔ51.
## Proteinase-K treatments, heat-inactivation, and boiling studies
Each sample was treated with 1U of proteinase K-linked to agarose beads (Sigma-Aldrich) for 2-4 hours at 37°C. The proteinase-K linked agarose beads were removed by pelleting via centrifugation (500 x g, 5 minutes). Untreated ACM or proteinase-K treated ACM were run on an SDS-PAGE gel and stained with Coomassie blue. The efficacy of protein digestion was evaluated by monitoring the density of the band corresponding to BSA. The effect of heat inactivation or boiling of ACM was evaluated by incubation of ACM in a 56°C water bath or boiling (95-100°C) on a heat block, respectively, for 30 minutes before cooling to room temperature and transferring to cells. The cells were then infected with VSVΔ51.
## LRA-mediated lipid depletion
Lipid Removal Agent (MilliporeSigma™) was mixed with PBS to achieve a working solution of 100 mg/mL. The LRA solution was added to CTL or ACM medium to reach a final LRA concentration of 4.3 mg/mL. This concentration was determined based on optimization experiments that sought to minimize cellular toxicity. CTL medium or ACM treated with LRA solution were incubated on a benchtop shaker for 30mins. Next, the samples were spun down (1500rpm, 10 minutes), and the supernatant was transferred to a new tube. The supernatant was gently drawn up while leaving a 1-2 mL buffer between the pellet and the tip of the serological pipette. Centrifugation of the supernatant was repeated twice to eliminate any LRA contaminants. Finally, the lipid-depleted medium was transferred to cells.
## Fatty acid quantification
Quantification of fatty acids in the ACM was achieved using the Free Fatty Acid Quantification Colorimetric/Fluorometric Kit (BioVision) as per the manufacture’s protocol.
## Fatty acid supplementation
Sodium palmitate (Sigma-Aldrich) was combined with heated $100\%$ ethanol and vortexed for 15 seconds. The fatty acid solution was then heated for 5-10 minutes at 65°C with periodic vortexing. The suspension was combined with an equal volume of heated sterile water and vortexed immediately. The fatty acid solution was stored at -20°C until use. A $2\%$ BSA-containing medium solution was prepared by adding BSA (fatty acid-free BSA; Sigma-Aldrich) to a serum-free RPMI medium. The sample was briefly mixed by swirling and warmed in a 37°C water bath to facilitate the dissolving of the BSA. The BSA-containing medium was filtered with a 0.22 μM syringe filter. The stock of sodium-palmitate in $50\%$ ethanol was warmed on a heat block at 65°C until the fatty acid was in solution. While still warm, the sodium-fatty acid solution was added to the $2\%$ BSA- containing medium in a dropwise fashion, immediately capped and vortexed. The palmitate supplemented medium was incubated in a 37°C water bath for three hours. The prepared fatty acid-BSA medium was supplemented with $10\%$ FBS right before being added to cells. Cytotoxicity quantification of lifted crystal violet or OV titers were normalized to samples containing equivalent volume of $2\%$ BSA-containing medium.
## Lipid mixture supplementation
A chemically defined lipid mixture (Sigma-Aldrich) containing non-animal derived fatty acids (2 μg/ml arachidonic and 10 μg/ml each linoleic, linolenic, myristic, oleic, palmitic, and stearic) and 0.22 mg/ml cholesterol, was diluted in RPMI medium containing $10\%$ FBS to final concentrations of 1, 5, 10, 25, 50, 75 and 100 mL/L and added to OVCAR8 cells. Cells were pre-incubated with the lipid mixture for 1 hour before infecting with VSVΔ51-EGFP at MOI 0.1. After 48 hours, infectivity was assessed by quantifying the mean fluorescent intensity (MFI) of EGFP (excitation 488 nm, emission 510 nm) using the BioTek Synergy™ Mx Microplate Reader. Infectivity quantification was plotted as percent MFI-EGFP and normalized to samples receiving no lipid mixture. Cell viability was also assessed with the REDOX indicator resazurin (Sigma Aldrich) according to the manufacturer’s protocol. Fluorescence was measured (excitation 530 nm, emission 590 nm) using the BioTek Synergy™ Mx Microplate Reader. Cell viability (metabolism) was plotted as percent viability and normalized to samples receiving a $50\%$ lipid mixture. Cytotoxicity was assessed using a crystal violet assay. The medium was removed from wells, and cells were washed once with PBS before $0.5\%$ crystal violet ($80\%$ MeOH in water) was added to each well. Plates were incubated at room temperature on a shaker for 20 minutes, then crystal violet solution was removed, and fixed cells were washed three times with water. Plates were left to air dry overnight; once dry, plates were scanned.
## Assessment of mitochondrial respiration using seahorse technology
Oligomycin (1.5 μM), trifluoromethoxy carbonylcyanide phenylhydrazone (FCCP) (0.5 μM), antimycin, and rotenone (0.5 μM) treatment were injected at designated intervals as per the manufacturer’s instructions for Seahorse XF Cell Mito Stress Test Kit (Agilent). The effect of acute etomoxir (Cayman Chemical Company) treatment was evaluated by injection of the drug after three baseline readings.
## siRNA knockdown of genes involved in FA transport
siRNAs [SMARTpool ON-TARGETplus Non-targeting Control Pool, SMARTpool ON-TARGETplus Human SLC27A1, SMARTpool ON-TARGETplus Human SLC27A2, SMARTpool ON-TARGETplus Human SLC27A4)] were purchased from Dharmacon and resuspended in siRNA Buffer (5X siRNA Buffer, Dharmacon) to generate 20 μM stocks. For transfection, 5 μL of Lipofectamine™ RNAiMAX Transfection Reagent (Thermo Fisher Scientific) was combined with 250 μL Opti-MEM and incubated for 5 minutes. The RNAiMAX complex was then combined with siRNAs diluted in Opti-MEM (2 μL of siRNA with 250 μL of Opti-MEM) in a gentle dropwise fashion. The siRNA-RNAiMAX complex was then incubated for 20 minutes at room temperature and added to wells with 5E5 cells suspended in 1.5 mL RMPI-1640 (containing $10\%$ FBS). 18-24 hours post-transfection, designated wells were infected with VSVΔ51-GFP for 40 minutes. The virus inoculum-containing medium was removed and replaced with a CTL medium or ACM ($50\%$ v/v).
## TOFA
Cells were treated with 0.1, 0.5, 1, 2, 5, 8 or 10μM of 5-tetradecyloxy-2-furoic (TOFA, Millipore Sigma) for 3hours. The medium was then changed to CTL medium or ACM medium containing half the previous concentration of TOFA in the well. Two hours later, the cells were infected with VSVΔ51 (MOI 0.1).
## Lipofermata
Cells were infected with VSVΔ51 for 40 minutes. The virus inoculum-containing medium was removed and replaced with CTL medium containing Lipofermata (Cayman Chemical Company) at the indicated concentration for 2 hours before an equivalent volume of CTL medium or ACM was added to the wells. For in vivo studies, Lipofermata was prepared from a 25mg/mL stock (in DMSO) to its working concentrating with dilution with DMSO alone or DMSO and $30\%$ v/v of Kolliphor®EL (Sigma Aldrich). The target injection volume was 25 uL to minimize DMSO toxicity to receiving mice.
## Hematoxylin and eosin staining
Tumors were fixed in formalin as per standard protocols [24] and were stained with hematoxylin and eosin by the Pathology Department at the Ottawa Hospital.
## RNA sequencing analysis
OVCAR8 cells were cultured in CTL medium or ACM for 18 hours or cultured in CTL medium or ACM. The wells were washed with sterile PBS and the RNA was extracted (NucleoZOL kit; Macherey-Nagel). Tidyverse, EdgeR, and Heatmaply were used for bioinformatics analysis and heatmap generation.
## Animal studies and tumor models
All animal studies complied with ethical regulations and were approved by the Institutional Animal Care Committee of the University of Ottawa (animal protocol # OHRI2870) and carried out in accordance with guidelines of the National Institutes of Health and the Canadian Council on Animal Care. Balb/c or C57BL/6 or nude CD-1 (6 to 8 weeks old) female mice were acquired from Charles River Laboratories. Mice with palpable tumors were monitored daily. Mice were euthanized at the indicated experimental time point. Otherwise, animals were euthanized at the pre-established human endpoint criteria. Animals displaying signs of pain, lethargy, labored breathing, lack of responsiveness, significant abdominal distension due to ascites build up, or when tumor volume reached 1500 mm3, were sacrificed. Animals were blindly randomized to treatment groups upon tumor implantation and before treatments. Animals that did not develop palpable tumors were excluded from the study. All animal manipulations were conducted with the operator blinded to the experimental condition and allocation group. Tumor volume was calculated using the following formula: tumor volume = $\frac{1}{2}$(length × width2). HFD (TD.06414) or a RD control (T. 08806) fed C57BL/6 mice received a specialized diet from Harlan Laboratories (Teklad Custom Diet; Envigo) for 8-10 weeks until a substantial change in the average mouse weight was observed (at least 20-$25\%$). The mice remained on the corresponding diet until the end of the experiment. Cells for implantation were cultured until 60-$70\%$ confluency. The cells were washed twice with PBS and passed through a 70 μM cell strainer. 2E5 cells (EMT6, 4T1), 5E5 cells (EO771) or 5E6 cells (ID8-F3 p53-/-, OVCAR8) resuspended in sterile PBS were implanted SC/FP or IP, respectively. Tumor models in mice were generated by implanting syngeneic tumor cells in the FP, IP, or subcutaneously. Intrabursal STOSE models were generated by surgical implantation as previously described [38, 39].
## Statistical analysis
Statistical significance was determined using an unpaired t-test, one-way or two-way analysis of variance (ANOVA) as indicated in the figure legend. Log-rank Mantel-Cox test was used to assess survival. The number of biological replicates and the statistical test used are indicated in the figure legends. Error bars represent the standard error of mean or standard deviation as indicated. P-values less than 0.05 were deemed significant. If no indication is shown, the results are non-significant. Statistical analyses were performed using GraphPad Prism 9 software. The statistical significance of all p-values is: ns $p \leq 0.05$, *$p \leq 0.05$, **$p \leq 0.01$, ***$p \leq 0.001$ and ****$p \leq 0.0001.$ Whenever possible exact p-values are provided in the text, figure legends or figures.
## Data availability statement
The data presented in this study are deposited in the NCBI (BioProject) repository, accession number PRJNA937284 (https://www.ncbi.nlm.nih.gov/bioproject/PRJNA937284).
## Ethics statement
The animal study was reviewed and approved by Animal Care Committee of the University of Ottawa.
## Author contributions
AS, MJ, JP, RB, MC, CL, JD JM, VT, RR, EF, HB, CT, SN, BW, ST, TA, and CI conducted in vitro experiments. AS, MJ, JPo, JPe, CT, NM, JB, BA, NC, NB, and CI. performed mouse experiments. AS, MJ, JP, RB, MC, and CI wrote the manuscript. AS, BV, L-HT, JB, and CI contributed to design of studies. All authors contributed to the article and approved the submitted version.
## Conflict of interest
We declare that JB has an interest in Turnstone Biologics, which developed the oncolytic Maraba MG1 virus as an OV platform.
The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher’s note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## Supplementary material
The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fimmu.2023.1099459/full#supplementary-material
## References
1. Anderson NM, Simon MC. **The tumor microenvironment**. *Curr Biol* (2020) **30** R921-r925. DOI: 10.1016/j.cub.2020.06.081
2. Achard C, Surendran A, Wedge ME, Ungerechts G, Bell J, Ilkow CS. **Lighting a fire in the tumor microenvironment using oncolytic immunotherapy**. *EBioMedicine* (2018) **31**. DOI: 10.1016/j.ebiom.2018.04.020
3. Duong MN, Cleret A, Matera EL, Chettab K, Mathe D, Valsesia-Wittmann S. **Adipose cells promote resistance of breast cancer cells to trastuzumab-mediated antibody-dependent cellular cytotoxicity**. *Breast Cancer Res* (2015) **17** 57. DOI: 10.1186/s13058-015-0569-0
4. Cao Y. **Adipocyte and lipid metabolism in cancer drug resistance**. *J Clin Invest Ed* (2019) **129**. DOI: 10.1172/JCI127201
5. Dai L, Song K, Di W. **Adipocytes: active facilitators in epithelial ovarian cancer progression**. *J Ovarian Res* (2020) **13** 115. DOI: 10.1186/s13048-020-00718-4
6. Wu B, Sun X, Gupta HB, Yuan B, Li J, Ge F. **Adipose PD-L1 modulates PD-1/PD-L1 checkpoint blockade immunotherapy efficacy in breast cancer**. *Oncoimmunology* (2018) **7**. DOI: 10.1080/2162402X.2018.1500107
7. Nieman KM, Romero IL, Van Houten B, Lengyel E. **Adipose tissue and adipocytes support tumorigenesis and metastasis**. *Biochim Biophys Acta Ed* (2013) **1831**. DOI: 10.1016/j.bbalip.2013.02.010
8. Nieman KM, Kenny HA, Penicka CV, Ladanyi A, Buell-Gutbrod R, Zillhardt MR. **Adipocytes promote ovarian cancer metastasis and provide energy for rapid tumor growth**. *Nat Med* (2011) **17**. DOI: 10.1038/nm.2492
9. Massa M, Gasparini S, Baldelli I, Scarabelli L, Santi P, Quarto R. **Interaction between breast cancer cells and adipose tissue cells derived from fat grafting**. *Aesthet Surg J Ed* (2016) **36**. DOI: 10.1093/asj/sjv194
10. Carter JC, Church FC. **Mature breast adipocytes promote breast cancer cell motility**. *Exp Mol Pathol Ed* (2012) **92**. DOI: 10.1016/j.yexmp.2012.03.005
11. Dirat B, Bochet L, Dabek M, Daviaud D, Dauvillier S, Majed B. **Cancer-associated adipocytes exhibit an activated phenotype and contribute to breast cancer invasion**. *Cancer Res* (2011) **71**. DOI: 10.1158/0008-5472.CAN-10-3323
12. Wang H, Airola MV, Reue K. **How lipid droplets “TAG” along: Glycerolipid synthetic enzymes and lipid storage**. *Biochim Biophys Acta Mol Cell Biol Lipids* (2017) **1862**. DOI: 10.1016/j.bbalip.2017.06.010
13. Incio J, Ligibel JA, McManus DT, Suboj P, Jung K, Kawaguchi K. **Obesity promotes resistance to anti-VEGF therapy in breast cancer by up-regulating IL-6 and potentially FGF-2**. *Sci Transl Med* (2018) **10**. DOI: 10.1126/scitranslmed.aag0945
14. Wu Q, Li B, Li Z, Li J, Sun S, Sun S. **Cancer-associated adipocytes: key players in breast cancer progression**. *J Hematol Oncol* (2019) **12** 1-15. DOI: 10.1186/s13045-019-0778-6
15. Yang J, Zaman MM, Vlasakov I, Roy R, Huang L, Martin CR. **Adipocytes promote ovarian cancer chemoresistance**. *Sci Rep* (2019) **9** 1-12. DOI: 10.1038/s41598-019-49649-1
16. Nowicka A, Marini FC, Solley TN, Elizondo PB, Zhang Y, Sharp HJ. **Human omental-derived adipose stem cells increase ovarian cancer proliferation, migration, and chemoresistance**. *PloS One* (2013) **8**. DOI: 10.1371/journal.pone.0081859
17. Chkourko Gusky H, Diedrich J, MacDougald OA, Podgorski I. **Omentum and bone marrow: how adipocyte-rich organs create tumour microenvironments conducive for metastatic progression**. *Obes Rev* (2016) **17**. DOI: 10.1111/obr.12450
18. Jamieson TR, Poutou J, Ilkow CS. **Redirecting oncolytic viruses: Engineering opportunists to take control of the tumour microenvironment**. *Cytokine Growth Factor Rev* (2020) **56**. DOI: 10.1016/j.cytogfr.2020.07.004
19. Twumasi-Boateng K, Pettigrew JL, Kwok YYE, Bell JC, Nelson BH. **Oncolytic viruses as engineering platforms for combination immunotherapy**. *Nat Rev Cancer Nat Publishing Group* (2018) **18**. DOI: 10.1038/s41568-018-0009-4
20. Alberts P, Tilgase A, Rasa A, Bandere K, Venskus D. **The advent of oncolytic virotherapy in oncology: The rigvir® story**. *Eur J Pharmacol* (2018) **837**. DOI: 10.1016/j.ejphar.2018.08.042
21. Xia ZJ, Chang JH, Zhang L, Jiang WQ, Guan ZZ, Liu JW. **[Phase III randomized clinical trial of intratumoral injection of E1B gene-deleted adenovirus (H101) combined with cisplatin-based chemotherapy in treating squamous cell cancer of head and neck or esophagus]**. *Ai Zheng* (2004) **23**
22. Rehman H, Silk AW, Kane MP, Kaufman HL. **Into the clinic: Talimogene laherparepvec (T-VEC), a first-in-class intratumoral oncolytic viral therapy**. *J Immunother Cancer* (2016) **4** 53. DOI: 10.1186/s40425-016-0158-5
23. Stojdl DF, Lichty BD, tenOever BR, Paterson JM, Power AT, Knowles S. **VSV strains with defects in their ability to shutdown innate immunity are potent systemic anti-cancer agents**. *Cancer Cell* (2003) **4**. DOI: 10.1016/S1535-6108(03)00241-1
24. Wedge M-E, Jennings V, Crupi MJF, Poutou J, Jamieson T, Pelin A. **Virally programmed extracellular vesicles sensitize cancer cells to oncolytic virus and small molecule therapy**. *Nat Commun* (2022) **13** 1898. PMID: 35393414
25. Stojdl DF, Lichty B, Knowles S, Marius R, Atkins H, Sonenberg N. **Exploiting tumor-specific defects in the interferon pathway with a previously unknown oncolytic virus**. *Nat Med* (2000) **6**. DOI: 10.1038/77558
26. Guo ZS, Lu B, Guo Z, Giehl E, Feist M, Dai E. **Vaccinia virus-mediated cancer immunotherapy: Cancer vaccines and oncolytics**. *J Immunother Cancer* (2019) **7** 6. DOI: 10.1186/s40425-018-0495-7
27. Pelin A, Foloppe J, Petryk J, Singaravelu R, Hussein M, Gossart F. **Deletion of apoptosis inhibitor F1L in vaccinia virus increases safety and oncolysis for cancer therapy**. *Mol Ther Oncol* (2019). DOI: 10.1016/j.omto.2019.06.004
28. Pidelaserra-Martí G, Engeland CE. **Mechanisms of measles virus oncolytic immunotherapy**. *Cytokine Growth Factor Rev* (2020) **56** 28-38. DOI: 10.1016/j.cytogfr.2020.07.009
29. Ilkow CS, Swift SL, Bell JC, Diallo JS. **From scourge to cure: Tumour-selective viral pathogenesis as a new strategy against cancer**. *PloS Pathog* (2014) **10**. DOI: 10.1371/journal.ppat.1003836
30. Vázquez-Calvo Á, Saiz JC, Sobrino F, Martín-Acebes MA. **Inhibition of enveloped virus infection of cultured cells by valproic acid**. *J Virol* (2011) **85**. DOI: 10.1128/JVI.01717-10
31. Bartolotta S, García CC, Candurra NA, Damonte EB. **Effect of fatty acids on arenavirus replication: Inhibition of virus production by lauric acid**. *Arch Virol* (2001) **146**. DOI: 10.1007/s007050170146
32. Hornung B, Amtmann E, Sauer G. **Lauric acid inhibits the maturation of vesicular stomatitis virus**. *J Gen Virol* (1994) **75**. DOI: 10.1099/0022-1317-75-2-353
33. Kapadia SB, Chisari FV. **Hepatitis c virus RNA replication is regulated by host geranylgeranylation and fatty acids**. *Proc Natl Acad Sci U.S.A.* (2005) **102**. DOI: 10.1073/pnas.0409834102
34. Pramanik R, Sheng X, Ichihara B, Heisterkamp N, Mittelman SD. **Adipose tissue attracts and protects acute lymphoblastic leukemia cells from chemotherapy**. *Leuk Res* (2013) **37** 503. DOI: 10.1016/j.leukres.2012.12.013
35. Coelho P, Silva L, Faria I, Vieria M, Monteiro A, Pinto G. **Adipocyte secretome increases radioresistance of malignant melanocytes by improving cell survival and decreasing oxidative status**. *Radiat Res* (2017) **187**. DOI: 10.1667/RR14551.1
36. Chi M, Chen J, Ye Y, Tseng HY, Lai F, Tay KH. **Adipocytes contribute to resistance of human melanoma cells to chemotherapy and targeted therapy**. *Curr Med Chem* (2014) **21**. DOI: 10.2174/0929867321666131129114742
37. Yang Y, Yang Y, Yang J, Zhao X, Wei X. **Tumor microenvironment in ovarian cancer: Function and therapeutic strategy [Internet]**. *Front Cell Dev Biol* (2020) **8**. DOI: 10.3389/fcell.2020.00758
38. McCloskey CW, Goldberg RL, Carter LE, Gamwell LF, Al-Hujaily EM, Collins O. **A new spontaneously transformed syngeneic model of high-grade serous ovarian cancer with a tumor-initiating cell population**. *Front Oncol* (2014) **4**. DOI: 10.3389/fonc.2014.00053
39. McCloskey CW, Rodriguez GM, Galpin KJC, Vanderhyden BC. **Ovarian cancer immunotherapy: Preclinical models and emerging therapeutics [Internet]**. *Cancers* (2018) **10** 244. DOI: 10.3390/cancers10080244
40. Bourgeois-Daigneault MC, St-Germain LE, Roy DG, Pelin A, Aitken AS, Arulanandam R. **Combination of paclitaxel and MG1 oncolytic virus as a successful strategy for breast cancer treatment**. *Breast Cancer Res* (2016) **18** 83. DOI: 10.1186/s13058-016-0744-y
41. Mangeot PE, Risson V, Fusil F, Marnef A, Laurent E, Blin J. **Genome editing in primary cells and in vivo using viral-derived nanoblades loaded with Cas9-sgRNA ribonucleoproteins**. *Nat Commun* (2019) **10** 1-15. DOI: 10.1038/s41467-018-07845-z
42. Grant RW, Stephens JM. **Fat in flames: influence of cytokines and pattern recognition receptors on adipocyte lipolysis**. *Am J Physiol-Endocrinol Metab* (2015) **309**. DOI: 10.1152/ajpendo.00053.2015
43. Crupi MJF, Bell JC, Singaravelu R. **Concise review: Targeting cancer stem cells and their supporting niche using oncolytic viruses**. *Stem Cells* (2019) **37**. DOI: 10.1002/stem.3004
44. Jariwalla R, Grossberg SE, Sedmak JJ. **The influence of physicochemical factors on the thermal inactivation of murine interferon**. *Arch Virol* (1975) **49**. DOI: 10.1007/BF01317544
45. Schlaepfer IR, Joshi M. **CPT1A-mediated fat oxidation, mechanisms, and therapeutic potential**. *Endocrinology* (2020) **161**. DOI: 10.1210/endocr/bqz046
46. Wang W, Ishibashi J, Trefely S, Shao M, Cowan AJ, Sakers A. **A PRDM16-driven metabolic signal from adipocytes regulates precursor cell fate**. *Cell Metab* (2019) **30** 174-189.e5. DOI: 10.1016/j.cmet.2019.05.005
47. Guseva NV, Rokhlin OW, Glover RA, Cohen MB. **TOFA (5-tetradecyl-oxy-2-furoic acid) reduces fatty acid synthesis, inhibits expression of AR, neuropilin-1 and mcl-1 and kills prostate cancer cells independent of p53 status**. *Cancer Biol Ther* (2011) **12**. DOI: 10.4161/cbt.12.1.15721
48. Schwenk RW, Holloway GP, Luiken JJFP, Bonen A, Glatz JFC. **Fatty acid transport across the cell membrane: Regulation by fatty acid transporters**. *Prostaglandins Leukotrienes Essential Fatty Acids (PLEFA)* (2010) **82**. DOI: 10.1016/j.plefa.2010.02.029
49. Pilch PF, Meshulam T, Ding S, Liu L. **Caveolae and lipid trafficking in adipocytes**. *Clin Lipidol* (2011) **6** 49-58. DOI: 10.2217/clp.10.80
50. DiRusso CC, Li H, Darwis D, Watkins PA, Berger J, Black PN. **Comparative biochemical studies of the murine fatty acid transport proteins (FATP) expressed in yeast**. *J Biol Chem* (2005) **280**. DOI: 10.1074/jbc.M409598200
51. Black PN, Ahowesso C, Montefusco D, Saini N, Dirusso CC. **Fatty acid transport proteins: Targeting FATP2 as a gatekeeper involved in the transport of exogenous fatty acids**. *Medchemcomm* (2016) **7**. DOI: 10.1039/C6MD00043F
52. Ahowesso C, Black PN, Saini N, Montefusco D, Chekal J, Malosh C. **Chemical inhibition of fatty acid absorption and cellular uptake limits lipotoxic cell death**. *Biochem Pharmacol* (2015) **98**. DOI: 10.1016/j.bcp.2015.09.004
53. Adeshakin AO, Liu W, Adeshakin FO, Afolabi LO, Zhang M, Zhang G. **Regulation of ROS in myeloid-derived suppressor cells through targeting fatty acid transport protein 2 enhanced anti-PD-L1 tumor immunotherapy**. *Cell Immunol* (2021) **362** 104286. DOI: 10.1016/j.cellimm.2021.104286
54. Ilkow CS, Marguerie M, Batenchuk C, Mayer J, Ben Neriah D, Cousineau S. **Reciprocal cellular cross-talk within the tumor microenvironment promotes oncolytic virus activity**. *Nat Med* (2015) **21**. DOI: 10.1038/nm.3848
55. Eisenreich W, Rudel T, Heesemann J, Goebel W. **How viral and intracellular bacterial pathogens reprogram the metabolism of host cells to allow their intracellular replication**. *Front Cell Infect Microbiol* (2019) **9** 42. PMID: 30886834
56. Thormar H, Isaacs CE, Brown HR, Barshatzky MR, Pessolano1 T. **Inactivation of enveloped viruses and killing of cells by fatty acids and monoglycerides**. *Antimicrob Agents Chemother* (1987) **31**. DOI: 10.1128/AAC.31.1.27
57. Veglia F, Tyurin VA, Blasi M, de Leo A, Kossenkov AV, Donthireddy L. **Fatty acid transport protein 2 reprograms neutrophils in cancer**. *Nat* (2019) **569**. DOI: 10.1038/s41586-019-1118-2
58. Alicea GM, Rebecca VW, Goldman AR, Fane ME, Douglass SM, Behera R. **Changes in aged fibroblast lipid metabolism induce age-dependent melanoma cell resistance to targeted therapy**. *Cancer Discovery* (2020) **10**. DOI: 10.1158/2159-8290.CD-20-0329
59. Koundouros N, Poulogiannis G. **Reprogramming of fatty acid metabolism in cancer**. *Br J Cancer* (2020) **122** 4-22. DOI: 10.1038/s41416-019-0650-z
60. Zhang M, di Martino JS, Bowman RL, Campbell NR, Baksh SC, Simon-Vermot T. **Adipocyte-derived lipids mediate melanoma progression**. *Cancer Discovery* (2018) **8**. DOI: 10.1158/2159-8290.CD-17-1371
61. Geng QS, Yang MJ, Li LF, Shen ZB, Wang LH, Zheng YY. **Over-expression and prognostic significance of FATP5, as a new biomarker, in colorectal carcinoma**. *Front Mol Biosci* (2022) **8**. DOI: 10.3389/fmolb.2021.770624
62. Ladanyi A, Mukherjee A, Kenny HA, Johnson A, Mitra AK, Sundaresan S. **Adipocyte-induced CD36 expression drives ovarian cancer progression and metastasis**. *Oncogene* (2018) **37**. DOI: 10.1038/s41388-017-0093-z
63. Watt M, Clark A, Selth L, Haynes V, Lister N, Rebello R. **Suppressing fatty acid uptake has therapeutic effects in preclinical models of prostate cancer**. *Sci Transl Med* (2019) **11**. DOI: 10.1126/scitranslmed.aau5758
64. Zhao J, Zhi Z, Wang C, Xing H, Song G, Yu X. **Exogenous lipids promote the growth of breast cancer cells**. *Oncol Rep* (2017) **38**. DOI: 10.3892/or.2017.5864
65. Melton EM, Cerny RL, Dirusso CC, Black PN. **Overexpression of human fatty acid transport protein 2/very long chain acyl-CoA synthetase 1 (FATP2/Acsvl1) reveals distinct patterns of trafficking of exogenous fatty acids**. *Biochem Biophys Res Commun* (2013) **440**. DOI: 10.1016/j.bbrc.2013.09.137
66. de Toledo-Piza AR, de Oliveira MI, Negri G, Mendonça RZ, Figueiredo CA. **Polyunsaturated fatty acids from phyllocaulis boraceiensis mucus block the replication of influenza virus**. *Arch Microbiol* (2018) **200**. DOI: 10.1007/s00203-018-1507-1
67. Goc A, Niedzwiecki A, Rath M. **Polyunsaturated ω-3 fatty acids inhibit ACE2-controlled SARS-CoV-2 binding and cellular entry**. *Sci Rep* (2021) **11** 5207. DOI: 10.1038/s41598-021-84850-1
68. Szpigel A, Hainault I, Carlier A, Venteclef N, Batto AF, Hajduch E. **Lipid environment induces ER stress, TXNIP expression and inflammation in immune cells of individuals with type 2 diabetes**. *Diabetologia* (2018) **61** 399-412. DOI: 10.1007/s00125-017-4462-5
69. Oslowski CM, Hara T, O’Sullivan-Murphy B, Kanekura K, Lu S, Hara M. **Thioredoxin-interacting protein mediates ER stress-induced β cell death through initiation of the inflammasome**. *Cell Metab* (2012) **16**. DOI: 10.1016/j.cmet.2012.07.005
70. Lerner AG, Upton JP, Praveen PVK, Ghosh R, Nakagawa Y, Igbaria A. **IRE1α induces thioredoxin-interacting protein to activate the NLRP3 inflammasome and promote programmed cell death under irremediable ER stress**. *Cell Metab* (2012) **16**. DOI: 10.1016/j.cmet.2012.07.007
71. Tiwarekar V, Wohlfahrt J, Fehrholz M, Scholz CJ, Kneitz S, Schneider-Schaulies J. **APOBEC3G-regulated host factors interfere with measles virus replication: Role of REDD1 and mammalian TORC1 inhibition**. *J Virol* (2018) **92**. DOI: 10.1128/JVI.00835-18
72. Kaczmarczyk SJ, Sitaraman K, Young HA, Hughes SH, Chatterjee DK. **Protein delivery using engineered virus-like particles**. *Proc Natl Acad Sci U S A* (2011) **108**. DOI: 10.1073/pnas.1101874108
|
---
title: Relationship between disorders of the intestinal microbiota and heart failure
in infants with congenital heart disease
authors:
- Qi-Liang Zhang
- Xiu-Hua Chen
- Si-Jia Zhou
- Yu-Qing Lei
- Jiang-Shan Huang
- Qiang Chen
- Hua Cao
journal: Frontiers in Cellular and Infection Microbiology
year: 2023
pmcid: PMC10036851
doi: 10.3389/fcimb.2023.1152349
license: CC BY 4.0
---
# Relationship between disorders of the intestinal microbiota and heart failure in infants with congenital heart disease
## Abstract
### Purpose
There is a close relationship between the intestinal microbiota and heart failure, but no study has assessed this relationship in infants with congenital heart disease. This study aimed to explore the relationship between heart failure and intestinal microbiota in infants with congenital heart disease.
### Methods
Twenty-eight infants with congenital heart disease with heart failure admitted to a provincial children’s hospital from September 2021 to December 2021 were enrolled in this study. A total of 22 infants without heart disease and matched for age, sex, and weight were selected as controls. Faecal samples were collected from every participant and subjected to 16S rDNA gene sequencing.
### Results
The composition of the intestinal microbiota was significantly disordered in infants with heart failure caused by congenital heart disease compared with that in infants without heart disease. At the phylum level, the most abundant bacteria in the heart failure group were Firmicutes, Actinobacteria, Proteobacteria, and Bacteroidetes, and the most abundant bacteria in the control group were Firmicutes, Proteobacteria, Actinobacteria, and Bacteroidetes. At the genus level, the most abundant bacteria in the heart failure group were Enterococcus, Bifidobacterium, Subdoligranulum, Shigella, and Streptococcus, and the most abundant bacteria in the control group were Bifidobacterium, Blautia, Bacteroides, Streptococcus, and Ruminococcus. The alpha and beta diversities of the gut bacterial community in the heart failure group were significantly lower than those in the control group ($p \leq 0.05$). Compared with the control group, retinol metabolism was significantly downregulated in the heart failure group.
### Conclusion
Heart failure in infants with congenital heart disease caused intestinal microbiota disorder, which was characterised by an increase in pathogenic bacteria, a decrease in beneficial bacteria, and decreases in diversity and richness. The significant downregulation of retinol metabolism in the intestinal microbiota of infants with heart failure may be related to the progression of heart failure, and further study of the underlying mechanism is needed.
## Introduction
Heart failure is the terminal stage of all types of cardiovascular disease and is a common cause of death (Mosterd and Hoes, 2007; Targher et al., 2017). The pathophysiological mechanisms of heart failure are complex and include haemodynamic abnormalities, neuroendocrine system activation, cardiac remodelling, and inflammatory responses (Mudd and Kass, 2008). To reduce the disease and economic burden associated with heart failure, it is important to elucidate the mechanisms of heart failure development and explore new potential therapeutic targets (Zhang et al., 2021). Accumulating evidence indicates that there is a close relationship between the intestinal microbiota and heart failure. Heart failure can cause a disturbance in the intestinal microbiota, and the intestinal microbiota has potential significance in mediating or regulating the pathophysiology of heart failure (Sandek et al., 2012; Brown and Hazen, 2015; Miele et al., 2015; Marques et al., 2017; Tang et al., 2017; Tang et al., 2019).
Congenital heart disease is a deformity caused by disorders of foetal heart and large blood vessel development (Hoffman and Kaplan, 2002). It is one of the most common congenital malformations and the most common cause of heart failure in children (Hinton and Ware, 2017). Left-to-right shunt congenital heart disease, including patent ductus arteriosus, ventricular septal defect, atrial septal defect, and endocardial pad defect, accounts for approximately $50\%$–$70\%$ of congenital heart disease. Patients with severe disease often develop heart failure in infancy due to a large left-to-right shunt and are seriously ill, which places a heavy burden on families and society. Studies of the relationship between heart failure and the intestinal microbiota have mainly focused on heart disease in adults, and few studies have explored this relationship in infants with congenital heart disease (Ellis et al., 2013). Therefore, we conducted a cohort study of infants with heart failure caused by left-to-right shunt congenital heart disease to explore the relationship between heart failure and the intestinal microbiota. We hypothesised that heart failure in infants with congenital heart disease can cause intestinal microbiota disorder and that this disorder can aggravate the progression of heart failure.
## Research design and study cohort
The present study was approved by the ethics committee of our hospital and adhered to the tenets of the Declaration of Helsinki. Additionally, all parents of the patients signed the consent form before participating in the study.
This cohort study aimed to explore the relationship between heart failure and the intestinal microbiota in infants with congenital heart disease. A total of 28 infants with heart failure caused by left-to-right shunt congenital heart disease who were admitted to the cardiac surgery department of a provincial children’s hospital in southeast China from September 2021 to December 2021 were enrolled in this study. A total of twenty-two infants without heart disease and matched for age, sex, and weight were selected as controls.
Inclusion criteria were infants with heart failure caused by left-to-right shunt congenital heart disease. Exclusion criteria the following: [1] other major organ diseases, such as digestive tract malformation, liver failure, or kidney failure; [2] digestive tract diseases, such as diarrhoea, constipation, or jaundice; [3] infection or receiving antibiotics; and [4] parental refusal to participate in the study.
## Faecal sample collection
Faecal samples (1 ml) were collected from each patient, immediately frozen in liquid nitrogen, and stored at −80°C.
## 16S rDNA sequencing
Total genomic DNA samples were extracted using the OMEGA Soil DNA Kit (M5635-02) (Omega Bio-Tek, Norcross, GA, USA). DNA extracted from the sample was used as a template. PCR amplification of the bacterial 16S rRNA genes V3–V4 region was performed using the forward primer 338F (5′-ACTCCTACGGGAGGCAGCA-3′) and the reverse primer 806R (5′-GGACTACHVGGGTWTCTAAT-3′). The amplified product was purified and recycled by using clean beads. The purified and recycled products were subjected to fluorescence quantitation with a Quant-iT PicoGreen dsDNA Assay Kit in a microplate reader (BioTek, FLx800). According to the fluorescence quantitation results, each sample was mixed in proportion according to the sequencing volume requirements of each sample. The sequencing library was prepared with a TruSeq Nano DNA LT Library Prep Kit (Illumina). Finally, paired-end sequencing was carried out on a NovaSeq sequencer with a NovaSeq 6000 SP Reagent Kit (500 cycles).
## Bioinformatics analysis
Microbiome bioinformatics were performed with QIIME2 2019.4 with slight modification according to the official tutorials. Briefly, raw sequence data were demultiplexed using the demux plugin followed by primers cutting with cutadapt plugin. Sequences were then quality filtered, denoised, and merged, and chimera removed using the DADA2 plugin. Venn diagram was generated to visualise the shared and unique ASVs among samples or groups using R package “VennDiagram,” based on the occurrence of ASVs across samples/groups regardless of their relative abundance. Alpha-diversity metrics, such as Chao1 richness estimator, observed species, Shannon diversity index, Simpson index, Faith’s PD, Pielou’s evenness, and Good’s coverage were calculated using the ASV table in QIIME2, and visualised as box plots. Beta diversity analysis was performed to investigate the structural variation of microbial communities across samples using Jaccard metrics, Bray–Curtis metrics, and UniFrac distance metrics, and visualised via principal coordinate analysis, non-metric multidimensional scaling, and unweighted pair-group method with arithmetic means hierarchical clustering. Microbial functions were predicted by PICRUSt2 upon MetaCyc and KEGG databases.
## Statistical analysis
SPSS 25.0 software was used for statistical analysis. Continuous variables with a normal distribution were expressed as the mean ± standard deviation and were compared via the t-test. Continuous variables without a normal distribution were compared via the non-parametric test. Categorical variables were described as integers and percentages, and comparisons between groups were performed using Fisher’s exact test. A p-value <0.05 was considered statistically significant.
## Result
A total of 28 infants with heart failure caused by left-to-right shunt congenital heart disease were enrolled as the heart failure group. There were 21 infants with ventricular septal defects, 5 infants with ductus arteriosus defects, and 2 infants with complete endocardial pad defects. The heart failure group comprised 15 male and 13 female infants with a mean age of 2.8 ± 2.9 months, weight of 4.6 ± 1.8 kg, pulmonary artery pressure of 59.8 ± 10.1 mmHg, and NT-proBNP of 4,218.7 ± 3,757.1 pg/ml. Nine of the infants in the heart failure group were exclusively breastfed, 5 were formula fed, and 14 were both breastfed and formula fed. A total of 22 age- and sex-matched children without heart disease served as controls. The control group comprised 12 male and 10 female infants with a mean age of 2.6 ± 1.7 months and weight of 5.1 ± 1.6 kg. Eight of the infants in the control group were exclusively breastfed, 3 were formula fed, and 11 were both breastfed and formula fed. There were no differences in sex, age, weight, or feeding style between the two groups (Table 1).
**Table 1**
| Unnamed: 0 | Intervention group | Control group | P value |
| --- | --- | --- | --- |
| Age (month) | 2.8±2.9 | 2.6±1.7 | 0.745 |
| Weight (kg) | 4.6±1.8 | 5.1±1.6 | 0.343 |
| Boys/girls | 15/13 | 12/10 | 0.945 |
| pulmonary artery pressure (mmHg) | 59.8±10.1 | – | – |
| NT-proBNP (pg/ml) | 4218.7±3757.1 | – | – |
| Feeding methods | | | |
| Breastfeeding | 9 | 8 | 0.904 |
| Formula feeding | 14 | 11 | 0.904 |
| Mixed Feeding | 5 | 3 | 0.904 |
To analyse the differences in intestinal microbial species between the heart failure group and control group, a Venn diagram was constructed using the ASV/OTU abundance table, and the number of members in each set was counted according to presence or absence in the ASV/OTU abundance table. The results showed that there were 1,744 ($10.9\%$) ASVs/OTUs that were shared between the heart failure group and the control group, 7,758 ($48.48\%$) ASVs/OTUs unique to the heart failure group, and 6,502 ($40.63\%$) ASVs/OTUs unique to the control group (Figure 1).
**Figure 1:** *There were 1,744 identical ASVs/OTUs in the heart failure group and the control group. There were 7,758 unique ASVs/OTUs in the heart failure group and 6,502 unique ASVs/OTUs in the control group.*
We further analyse the species composition and abundance of intestinal microorganisms in the two groups at the phylum and genus levels. At the phylum level, the most abundant bacteria in the heart failure group were Firmicutes ($58.8\%$), Actinobacteria ($20.6\%$), Proteobacteria ($10.3\%$), and Bacteroidetes ($3.7\%$), and the most abundant bacteria in the control group were Firmicutes ($47.8\%$), Proteobacteria ($22.1\%$), Actinobacteria ($21.8\%$), and Bacteroidetes ($3.1\%$). At the genus level, the most abundant bacteria in the heart failure group were Enterococcus ($30.3\%$), Bifidobacterium ($13.1\%$), Subdoligranulum ($5.7\%$), Shigella ($3.6\%$), and Streptococcus ($3.1\%$), and the most abundant bacteria in the control group were Bifidobacterium ($24.3\%$), Blautia ($6.0\%$), Bacteroides ($3.3\%$), Streptococcus ($3.0\%$), and Ruminococcus ($2.5\%$) (Figure 2).
**Figure 2:** *At the phylum level, the most abundant bacteria in the heart failure group were Firmicutes, Actinobacteria, Proteobacteria, and Bacteroidetes, and the most abundant bacteria in the control group were Firmicutes, Proteobacteria, Actinobacteria, and Bacteroidetes. At the genus level, the most abundant bacteria in the heart failure group were Enterococcus, Bifidobacterium, Subdoligranulum, Shigella, and Streptococcus, and the most abundant bacteria in the control group were Bifidobacterium, Blautia, Bacteroides, Streptococcus, and Ruminococcus.*
The comparison of the alpha diversity of the intestinal microbiota between the two groups showed that the Chao1 and observed species indices were significantly lower in the heart failure group than in the control group (Figure 3).
**Figure 3:** *The Chao1 and observed species indices of abundance were significantly lower in the heart failure group than in the control group. The meaning of each symbol in the boxplot was as follows. The upper and lower end line of the box were the upper and lower Interquartile range. Median line of the box was the median. The upper and lower margins were the maximum and minimum internal circumference (1.5 times of interquartile range). Numbers under the diversity index label are p-values.*
The beta diversity index focuses on the comparison of diversity between different environments, which are represented by infant groups in the present study. We visually analyse the data using principal coordinate analysis, and differences were further examined using Adonis. The PCoA of beta diversity based on the Bray−Curtis distance matrix revealed that the differences in community composition between the two groups were significant (Figure 4).
**Figure 4:** *The PCoA of beta diversity based on the Bray−Curtis distance matrix revealed that the differences in community composition between the two groups were significant.*
The functional properties of the intestinal microbiota were predicted using PICRUSt2. The analysis of Kyoto Encyclopedia of Genes and Genomes pathway annotations at level 2 as determined by PICRUSt2 revealed significant differences in retinol metabolism between the two groups. Compared with the control group, retinol metabolism was significantly downregulated in the heart failure group (Figure 5).
**Figure 5:** *Compared with the control group, retinol metabolism was significantly downregulated in the heart failure group. The negative value of the horizontal axis logFC represents the downregulation of metabolism in the heart failure group compared with the control group. The ordinate represents the different KEGG metabolic pathway labels. The degree of saliency is shown in different colors.*
## Discussion
Heart failure is a formidable global public health challenge and is responsible for more than 1 million hospitalisations each year (Jackevicius et al., 2019). Heart failure describes a variety of cardiac structural or functional diseases that result in impaired ventricular filling or ejection capacity, insufficient blood perfusion in organs and tissues, and pulmonary or systemic circulation stasis (Jia et al., 2019). The gut is an endocrine organ that is rich in blood, accounting for approximately $40\%$ of the body’’s total blood, and it is significantly affected by reduced blood supply (Takala, 1996). During the heart failure process, the gut is the first organ to undergo ischaemia and the last organ to recover. Intestinal ischaemia or hyperaemia reduces the intestinal oxygen supply, which can lead to changes in intestinal microbial composition and, in turn, metabolic disorders and intestinal microbiota dysfunction. Furthermore, the intestinal microbiota affects the body’s physiology, including the progression of heart failure (Polsinelli et al., 2019).
The gut contains trillions of microbes—as many cells as the total number of human cells (Sender et al., 2016a). The microbes that colonise the intestinal tract play an important role in the physiological and pathological processes of the body. They participate in nutrient metabolism and absorption, regulate intestinal epithelial barrier function, and affect local or systemic immune inflammatory responses (Nicholson et al., 2012). The composition of the intestinal microbiome is dynamic and may differ in the same individual under different physiological conditions and at different times. These changes may play an important role in human health and disease. Many studies have shown that there is a close relationship between the intestinal microbiota and heart failure (Sandek et al., 2012; Brown and Hazen, 2015; Miele et al., 2015; Marques et al., 2017; Tang et al., 2017; Tang et al., 2019). However, these studies have focused on adults, and few studies have explored the relationship between heart failure and the intestinal microbiota in infants with congenital heart disease. We conducted a cohort study to analyse the relationship between heart failure and intestinal microbiota in infants with congenital heart disease.
Heart failure causes intestinal ischemia or hyperemia, which can result in a reduction in the intestinal oxygen supply. As a result, the intestinal environment changes, which leads to changes in the composition of the intestinal microbiota, mainly a reduction in beneficial intestinal bacteria and an increase in pathogenic bacteria. Chen et al. showed that intestinal microbiota disorder in patients with heart failure manifested as decreases in Bacteroides and Bifidobacteria and increases in Firmicutes and Proteus (Chen et al., 2017). Pasini et al. found increased abundances of intestinal pathogens such as Candida, Salmonella, Shigella, and *Campylobacter in* faecal samples from patients with heart failure (Pasini et al., 2016). Sandek et al. also observed excessive growth and adhesion of pathogenic bacteria in the intestinal mucosa of patients with heart failure (Sandek et al., 2007). A similar phenomenon was observed in the present study. Compared with infants without heart disease, the proportion of pathogenic bacteria such as Enterococcus, Shigella, and Subdoligranulum in the intestinal tract was significantly higher in infants with congenital heart disease with heart failure, while the proportion of beneficial bacteria such as Bifidobacterium, Blautia, and Bacteroides was significantly lower. Nagatomo et al. and Hooper et al. demonstrated that an increase in pathogenic bacteria and a decrease in beneficial bacteria lead to an increase in inflammatory factors in the intestine and an increase in the body’s inflammatory response, which aggravate the progression of heart failure (Hooper et al., 2012; Nagatomo and Tang, 2015).
The intestinal microbiota coexists with the host. The intestine provides a good colonisation environment for microbes, and the intestinal microbiota plays an important role in maintaining nutrient metabolism, the stability of the intestinal environment, and the health of the body. The intestinal microbiota is the most complex ecosystem in the human body and is composed of a variety of microorganisms residing in the human intestine. These microorganisms coexist mutually in the intestinal tract to jointly maintain the stability of the intestinal environment. Once homeostasis is disrupted, the disorder of the intestinal microbiota will aggravate the pathological state (Sender et al., 2016b). Kummen showed that intestinal hypoxia caused by heart failure reduces the diversity and richness of the intestinal microbiota (Kummen et al., 2018). In severe left-to-right shunt congenital heart disease, heart failure often occurs in infancy due to a large left-to-right shunt, which results in gastrointestinal congestion and disruption of the ecology of the intestinal microbiota. In this study, compared with infants without heart disease, the alpha and beta diversities of the intestinal microbiota were significantly lower in infants with heart failure. These findings indicate that heart failure reduced the diversity of the intestinal microbiota, with a decrease in beneficial bacteria and an increase in pathogenic bacteria, and ultimately aggravated the progression of heart failure.
Many studies have suggested that retinol metabolism is also involved in the progression of heart failure in infants with congenital heart disease. Yang et al. found that cardiac retinoic acid levels decline in heart failure patients (Yang et al., 2021). Osorio found that the protein expression of retinoid X receptor was reduced in pacing-induced heart failure (Osorio et al., 2002). Choudhary showed that retinoic acid can prevent angiotensin II- and mechanical stretch-induced reactive oxygen species generation and cardiomyocyte apoptosis (Choudhary et al., 2008). Subramanian observed that all-trans retinoic acid supplementation can prevent cardiac fibrosis and cytokines induced by methylglyoxal (Subramanian and Nagarajan, 2017). Huang et al. reported that lower levels of serum retinoic acid were associated with more serious cardiovascular disease and higher mortality (Huang et al., 2021). Hence, retinol metabolism was studied in this study. Compared with the control group, retinol metabolism was significantly downregulated in the heart failure group. Disorders of the intestinal microbiota will inevitably lead to changes in the metabolic function of the microbiota, which may cause downregulation of retinol metabolism. Adjusting the intestinal microbiota to upregulate retinol metabolism may be a new way to treat heart failure in infants with congenital heart disease, but further studies are needed.
There are some limitations to this study. First, the main food of infants is milk. In this study, infants were selected as the participants. Although the influence of some foods on the microbiota has been ruled out, interference from differences in formula and the diets of breastfeeding mothers cannot be ruled out. Second, this study only analyse the intestinal microbiota of infants with left-to-right shunt congestive heart failure and did not analyse other types of heart failure. Third, this study was a single-centre study with a small sample size, and a multicentre study with a large sample size should be conducted next.
## Conclusion
There is a close relationship between heart failure and the intestinal microbiota in infants with congenital heart disease. Heart failure in infants with congenital heart disease caused intestinal microbiota disorder, which was characterized by an increase in pathogenic bacteria, a decrease in beneficial bacteria, and decreases in diversity and richness. We also found that retinol metabolism of the intestinal microbiota was significantly downregulated in infants with heart failure, which may be related to the progression of heart failure. The underlying mechanism needs to be studied further.
## Data availability statement
The datasets presented in this study are deposited in online repositories. The names of the repositories and accession numbers can be found below: https://www.ncbi.nlm.nih.gov/bioproject/?term=PRJNA862324.
## Ethics statement
The studies involving human participants were reviewed and approved by Fujian Children’s Hospital. Written informed consent to participate in this study was provided by the participants’ legal guardian/next of kin.
## Author contributions
Q-LZ designed the study, acquired and interpreted the data, and drafted the manuscript. X-HC analysed and interpreted the data. S-JZ,acquired and analysed the data. J-SH and Y-QL collected and analyse faeces specimens. QC has made substantial contributions to conception and design. HC was involved in analysis and interpretation of data, revision of the manuscript, and given final approval of the version to be published.
## Conflict of interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher’s note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## References
1. Brown J. M., Hazen S. L.. **The gut microbial endocrine organ: bacterially derived signals driving cardiometabolic diseases**. *Annu. Rev. Med.* (2015) **66** 343-359. DOI: 10.1146/annurev-med-060513-093205
2. Chen K., Zheng X., Feng M., Li D., Zhang H.. **Gut microbiota-dependent metabolite trimethylamine n-oxide contributes to cardiac dysfunction in Western diet-induced obese mice**. *Front. Physiol.* (2017) **8**. DOI: 10.3389/fphys.2017.00139
3. Choudhary R., Baker K. M., Pan J.. **All-trans retinoic acid prevents angiotensin II- and mechanical stretch-induced reactive oxygen species generation and cardiomyocyte apoptosis**. *J. Cell Physiol.* (2008) **215** 172-181. DOI: 10.1002/jcp.21297
4. Ellis C. L., Bokulich N. A., Kalanetra K. M., Mirmiran M., Elumalai J., Haapanen L.. **Probiotic administration in congenital heart disease: a pilot study**. *J. Perinatol.* (2013) **33** 691-697. DOI: 10.1038/jp.2013.41
5. Hinton R. B., Ware S. M.. **Heart failure in pediatric patients with congenital heart disease**. *Circ. Res.* (2017) **120** 978-994. DOI: 10.1161/CIRCRESAHA.116.308996
6. Hoffman J. I., Kaplan S.. **The incidence of congenital heart disease**. *J. Am. Coll. Cardiol.* (2002) **39** 1890-1900. DOI: 10.1016/S0735-1097(02)01886-7
7. Hooper L. V., Littman D. R., Macpherson A. J.. **Interactions between the microbiota and the immune system**. *Science* (2012) **336** 1268-1273. DOI: 10.1126/science.1223490
8. Huang J., Weinstein S. J., Yu K., Männistö S., Albanes D.. **Association between serum retinol and overall and cause-specific mortality in a 30-year prospective cohort study**. *Nat. Commun.* (2021) **12** 6418. DOI: 10.1038/s41467-021-26639-4
9. Jackevicius C. A., Page R. L., Buckley L. F., Jennings D. L., Nappi J. M., Smith A. J.. **Key articles and guidelines in the management of heart failure: 2018 update**. *J. Pharm. Pract.* (2019) **32** 77-92. DOI: 10.1177/0897190018819413
10. Jia Q., Xie Y., Lu C., Zhang A., Lu Y., Lv S.. **Endocrine organs of cardiovascular diseases: Gut microbiota**. *J. Cell Mol. Med.* (2019) **23** 2314-2323. DOI: 10.1111/jcmm.14164
11. Kummen M., Mayerhofer C. C. K., Vestad B., Broch K., Awoyemi A., Storm-Larsen C.. **Gut microbiota signature in heart failure defined from profiling of 2 independent cohorts**. *J. Am. Coll. Cardiol.* (2018) **71** 1184-1186. DOI: 10.1016/j.jacc.2017.12.057
12. Marques F. Z., Nelson E., Chu P. Y., Horlock D., Fiedler A., Ziemann M.. **High-fiber diet and acetate supplementation change the gut microbiota and prevent the development of hypertension and heart failure in hypertensive mice**. *Circulation* (2017) **135** 964-977. DOI: 10.1161/CIRCULATIONAHA.116.024545
13. Miele L., Giorgio V., Alberelli M. A., De Candia E., Gasbarrini A., Grieco A.. **Impact of gut microbiota on obesity, diabetes, and cardiovascular disease risk**. *Curr. Cardiol. Rep.* (2015) **17** 120. DOI: 10.1007/s11886-015-0671-z
14. Mosterd A., Hoes A. W.. **Clinical epidemiology of heart failure**. *Heart* (2007) **93** 1137-1146. DOI: 10.1136/hrt.2003.025270
15. Mudd J. O., Kass D. A.. **Tackling heart failure in the twenty-first century**. *Nature* (2008) **451** 919-928. DOI: 10.1038/nature06798
16. Nagatomo Y., Tang W. H.. **Intersections between microbiome and heart failure: Revisiting the gut hypothesis**. *J. Card Fail.* (2015) **21** 973-980. DOI: 10.1016/j.cardfail.2015.09.017
17. Nicholson J. K., Holmes E., Kinross J., Burcelin R., Gibson G., Jia W.. **Host-gut microbiota metabolic interactions**. *Science* (2012) **336** 1262-1267. DOI: 10.1126/science.1223813
18. Osorio J. C., Stanley W. C., Linke A., Castellari M., Diep Q. N., Panchal A. R.. **Impaired myocardial fatty acid oxidation and reduced protein expression of retinoid X receptor-alpha in pacing-induced heart failure**. *Circulation* (2002) **106** 606-612. DOI: 10.1161/01.CIR.0000023531.22727.C1
19. Pasini E., Aquilani R., Testa C., Baiardi P., Angioletti S., Boschi F.. **Pathogenic gut flora in patients with chronic heart failure**. *JACC Heart Fail.* (2016) **4** 220-227. DOI: 10.1016/j.jchf.2015.10.009
20. Polsinelli V. B., Marteau L., Shah S. J.. **The role of splanchnic congestion and the intestinal microenvironment in the pathogenesis of advanced heart failure**. *Curr. Opin. Support Palliat. Care* (2019) **13** 24-30. DOI: 10.1097/SPC.0000000000000414
21. Sandek A., Bauditz J., Swidsinski A., Buhner S., Weber-Eibel J., von Haehling S.. **Altered intestinal function in patients with chronic heart failure**. *J. Am. Coll. Cardiol.* (2007) **50** 1561-1569. DOI: 10.1016/j.jacc.2007.07.016
22. Sandek A., Bjarnason I., Volk H. D., Crane R., Meddings J. B., Niebauer J.. **Studies on bacterial endotoxin and intestinal absorption function in patients with chronic heart failure**. *Int. J. Cardiol.* (2012) **157** 80-85. DOI: 10.1016/j.ijcard.2010.12.016
23. Sender R., Fuchs S., Milo R.. **Are we really vastly outnumbered? revisiting the ratio of bacterial to host cells in humans**. *Cell* (2016) **164** 337-340. DOI: 10.1016/j.cell.2016.01.013
24. Sender R., Fuchs S., Milo R.. **Revised estimates for the number of human and bacteria cells in the body**. *PloS Biol.* (2016) **14**. DOI: 10.1371/journal.pbio.1002533
25. Subramanian U., Nagarajan D.. **All-trans retinoic acid supplementation prevents cardiac fibrosis and cytokines induced by methylglyoxal**. *Glycoconj. J.* (2017) **34** 255-265. DOI: 10.1007/s10719-016-9760-5
26. Takala J.. **Determinants of splanchnic blood flow**. *Br. J. Anaesth.* (1996) **77** 50-58. DOI: 10.1093/bja/77.1.50
27. Tang W. H. W., Li D. Y., Hazen S. L.. **Dietary metabolism, the gut microbiome, and heart failure**. *Nat. Rev. Cardiol.* (2019) **16** 137-154. DOI: 10.1038/s41569-018-0108-7
28. Tang W. H., Kitai T., Hazen S. L.. **Gut microbiota in cardiovascular health and disease**. *Circ. Res.* (2017) **120** 1183-1196. DOI: 10.1161/CIRCRESAHA.117.309715
29. Targher G., Dauriz M., Laroche C., Temporelli P. L., Hassanein M., Seferovic P. M.. **In-hospital and 1-year mortality associated with diabetes in patients with acute heart failure: results from the ESC-HFA heart failure long-term registry**. *Eur. J. Heart Fail.* (2017) **19** 54-65. DOI: 10.1002/ejhf.679
30. Yang N., Parker L. E., Yu J., Jones J. W., Liu T., Papanicolaou K. N.. **Cardiac retinoic acid levels decline in heart failure**. *JCI Insight* (2021) **6**. DOI: 10.1172/jci.insight.137593
31. Zhang Y., Wang Y., Ke B., Du J.. **TMAO: how gut microbiota contributes to heart failure**. *Transl. Res.* (2021) **228** 109-125. DOI: 10.1016/j.trsl.2020.08.007
|
---
title: Mayo Adhesive Probability Score Does Not Have Prognostic Ability in Locally
Advanced Renal Cell Carcinoma
authors:
- Benjamin N. Schmeusser
- Tad A. Manalo
- Yuan Liu
- Yash B. Shah
- Adil Ali
- Manuel Armas-Phan
- Dattatraya H. Patil
- Reza Nabavizadeh
- Kenneth Ogan
- Viraj A. Master
journal: Journal of Kidney Cancer and VHL
year: 2023
pmcid: PMC10036918
doi: 10.15586/jkcvhl.v10i1.269
license: CC BY 4.0
---
# Mayo Adhesive Probability Score Does Not Have Prognostic Ability in Locally Advanced Renal Cell Carcinoma
## Abstract
Nephrectomy remains standard treatment for renal cell carcinoma (RCC). The Mayo Adhesive Probability (MAP) score is predictive of adherent perinephric fat and associated surgical complexity, and is determined by assessing perinephric fat and stranding. MAP has additionally predicted progression-free survival (PFS), though primarily reported in stage T1-T2 RCC. Here, we examine MAP’s ability to predict overall survival (OS) and PFS in T3-T4 RCC. From our prospectively maintained RCC database, patients that underwent radical nephrectomy [2009-2016] with available abdominal imaging (<90 days preop) and T3/T4 RCC underwent MAP scoring. Survival analyses were conducted with MAP scores as individual [0-5] and dichotomized (0-3 vs 4-5) using Kaplan-Meier method. Multivariable Cox proportional hazard regression models for PFS and OS were built with backward elimination. 141 patients were included. 134 ($95\%$) and 7 ($5\%$) had pT3 and pT4 disease, respectively. $46.1\%$ of patients had an inferior vena cava thrombus. Mean MAP score was 3.22±1.52, with 75 ($53\%$) patients having a score between 0-3 and 66 ($47\%$) having a score of 4-5. Both male gender ($$p \leq 0.006$$) and clear cell histology ($$p \leq 0.012$$) were associated with increased MAP scores. On Kaplan-Meier and multivariable analysis, no significant associations were identified between MAP and PFS (HR=1.01, $95\%$ CI 0.85-1.20, $$p \leq 0.93$$) or OS (HR=1.01, $95\%$ CI 0.84-1.21, $$p \leq 0.917$$). In this cohort of patients with locally advanced RCC, high MAP scores were not predictive of worse PFS or OS.
## Introduction
In 2021, renal cell carcinoma (RCC) was responsible for nearly 14,000 deaths in the United States [1]. Diagnosis of RCC has rapidly risen in recent decades, with a doubling in incidence since 1975 [1]. Nephrectomy with curative intent remains the gold standard in RCC management; however, image-guided procedures, conservative treatment approaches, and active surveillance have gained popularity. Great interest persists in patient-specific preoperative risk stratification to inform management, rather than relying on postoperative information such as tumor pathology. Specifically, measurements on preoperative imaging may be informative and assist in preoperative prognostication to further guide clinical decision-making.
One radiographic feature that has demonstrated the ability to predict surgical risks and outcomes in RCC is the Mayo Adhesive Probability (MAP) score. MAP estimates the probability of encountering adherent perinephric fat (APF) [2] and has been associated with increased surgical complexity, operative time, and blood loss during partial nephrectomy (PN) [3]. Moreover, Thiel et al. explored the association between MAP scores and progression-free survival (PFS). In their analysis, patients with high MAP scores (4–5) experienced inferior PFS (HR = 2.16, $95\%$ CI 1.15–4.06, $$P \leq 0.017$$) following surgery for clinically localized RCC [4].
Accordingly, MAP score appears useful in clinically localized disease and is appealing given its quick and convenient measurement on routine preoperative imaging. However, little is known about its utility in locally advanced RCC. In the study by Thiel et al., $82\%$ of patients had T1–T2 disease. As novel preoperative prognostic factors continue to emerge, understanding their value in all patient populations is necessary. To further elucidate the prognostic utility of MAP, we retrospectively analyzed the associations between preoperative MAP and both PFS and overall survival (OS) in patients with locally advanced nonmetastatic RCC.
## Patient selection and data acquisition
Patients that underwent radical nephrectomy (RN) for RCC from 2009 to 2016 were identified in our institutional database. MAP scores were calculated for patients with available computerized tomography (CT) or magnetic resonance imaging (MRI) within 90 days before surgery, as previously described [2]. MAP scores were acquired by two Medical Doctorate (MD) candidates pursuing urology residency training and familiar with renal imaging under the direct supervision of an attending urologic oncologist. Patients with T1–T2 disease were excluded. Patient characteristics included race, gender, age of surgery, Eastern Cooperative Oncology Group (ECOG) score, and BMI (<25 or ≥25). Clinical factors including presence of inferior vena cava (IVC) thrombus; laterality of kidney tumor; Fuhrman nuclear grade; presence of necrosis; pathologic N and T stage; stage, size, grade, and necrosis (SSIGN) score; University of California Los Angeles Integrated Staging System (UISS) score; systemic therapy history; corrected calcium; modified Glasgow prognostic score (mGPS); and histology (clear cell [ccRCC] or nonclear cell) were also obtained. All patients provided their informed consent in this study approved by the Institutional Review Board.
## Data analysis
The primary objective of this study was to analyze the prognostic ability of MAP in patients with locally advanced, nonmetastatic RCC. The primary endpoints were PFS and OS.
For survival analyses, MAP scores were analyzed as individual scores (0–5) and dichotomized groups (0–3 vs. 4–5) using the Kaplan–Meier method. In addition, multivariable Cox proportional hazard regression models were built with backward elimination using an alpha level of removal of 0.1. All patient clinicopathologic and demographic features were included in the model. For both PFS and OS, two separate multivariable models were generated to include and exclude SSIGN score, which is only validated in patients with clear cell RCC (ccRCC). Additional subanalyses were conducted in patients with and without presence of IVC tumor thrombus. All statistical tests were two-sided with type I error set at 0.05. Statistical analysis was conducted using SAS Version 9.4 (Cary, NC, USA) and SAS macros developed by the Biostatistics and Bioinformatics Shared Resource at Winship Cancer Institute.
## Results
A summary of patient demographics and clinicopathologic data is represented in Table 1. In total, 141 patients were included, of whom 134 ($95\%$) had pT3 and 7 ($5\%$) had pT4 disease. One hundred and seven ($75.9\%$) patients had clear-cell histology and 65 ($46.1\%$) patients had the presence of an IVC tumor thrombus. The final cohort was primarily male ($$n = 100$$, $71\%$) and white ($$n = 104$$; $74\%$). The median age was 63 years (IQR: 54–72) and median BMI was 28.5 kg/m2 (IQR: 24.6–32.6). In total, 47 ($33.3\%$) patients received some form of systemic therapy, all of which were administered postoperatively. Mean MAP score was 3.22 ± 1.52, with 75 ($53\%$) patients having a score between 0–3 and 66 ($47\%$) having a score of 4–5. Both male gender ($$P \leq 0.006$$) and ccRCC histology ($$P \leq 0.012$$) were significantly associated with increased MAP scores, though pathologic staging, ECOG status, and various clinical scoring systems were not. Interestingly, low BMI patients appeared to have lower MAP scores, although the association between these two measures was not significant ($$P \leq 0.059$$).
**Table 1:**
| Unnamed: 0 | MAP score, n (%) | MAP score, n (%).1 | MAP score, n (%).2 | MAP score, n (%).3 | MAP score, n (%).4 | MAP score, n (%).5 | Unnamed: 7 | Unnamed: 8 |
| --- | --- | --- | --- | --- | --- | --- | --- | --- |
| Covariate | 0 (n = 14) | 1 (n = 3) | 2 (n = 22) | 3 (n = 36) | 4 (n = 31) | 5 (n = 35) | Total (n = 141) | P |
| Gender | Gender | Gender | Gender | Gender | Gender | Gender | Gender | Gender |
| Male | 9 (64.3) | 1 (33.3) | 14 (63.6) | 19 (52.8) | 26 (83.9) | 31 (88.6) | 100 (70.9) | 0.006 |
| Female | 5 (35.7) | 2 (66.7) | 8 (36.4) | 17 (47.2) | 5 (16.1) | 4 (11.4) | 41 (29.1) | 0.006 |
| Race | Race | Race | Race | Race | Race | Race | Race | Race |
| White | 7 (50) | 1 (33.3) | 15 (68.2) | 28 (77.8) | 25 (80.6) | 28 (80) | 104 (73.8) | 0.12 |
| Non-white | 7 (50) | 2 (66.7) | 7 (31.8) | 8 (22.2) | 6 (19.4) | 7 (20) | 37 (26.2) | 0.12 |
| IVC thrombus | 5 (35.7) | 0 (0) | 11 (50) | 17 (47.2) | 18 (58.1) | 14 (40) | 65 (46.1) | 0.343 |
| ECOG | ECOG | ECOG | ECOG | ECOG | ECOG | ECOG | ECOG | ECOG |
| ≥1 | 0 (0) | 1 (33.3) | 5 (22.7) | 14 (38.9) | 8 (25.8) | 5 (14.3) | 33 (23.4) | 0.052 |
| BMI* | 23.9 (20.1–29.0) | 30.0 (28.6–31.0) | 29.0 (24–35) | 26.0 (23.1–30.3) | 28.8 (26.2–32.9) | 29.9 (27–34.3) | 28.50 (24.6–32.6) | 0.059 |
| Age* | 54.2 (44.2–68.1) | 62.7 (51.2–77.7) | 59.4 (45.6–71.6) | 61.6 (51.5–71.3) | 66.9 (55.4–73.4) | 63.3 (54.5–73.4) | 62.7 (53.8–71.5) | 0.212 |
| Nephrectomy side | Nephrectomy side | Nephrectomy side | Nephrectomy side | Nephrectomy side | Nephrectomy side | Nephrectomy side | Nephrectomy side | Nephrectomy side |
| Right | 8 (57.1) | 0 (0) | 12 (54.5) | 16 (44.4) | 17 (54.8) | 19 (54.3) | 72 (51.1) | 0.494 |
| Histology | Histology | Histology | Histology | Histology | Histology | Histology | Histology | Histology |
| ccRCC | 6 (42.9) | 2 (66.7) | 14 (63.6) | 32 (88.9) | 24 (77.4) | 29 (82.9) | 107 (75.9) | 0.012 |
| non-ccRCC | 8 (57.1) | 1 (33.3) | 8 (36.4) | 4 (11.1) | 7 (22.6) | 6 (17.1) | 34 (24.1) | 0.012 |
| pT stage | pT stage | pT stage | pT stage | pT stage | pT stage | pT stage | pT stage | pT stage |
| T3 | 11 (78.6) | 3 (100) | 22 (100) | 35 (97.2) | 30 (96.8) | 33 (94.3) | 134 (95.0) | 0.077 |
| T4 | 3 (21.4) | 0 (0) | 0 (0) | 1 (2.8) | 1 (3.2) | 2 (5.7) | 7 (5.0) | 0.077 |
| pN stage | pN stage | pN stage | pN stage | pN stage | pN stage | pN stage | pN stage | pN stage |
| N1 | 2 (14.3) | 0 (0) | 2 (9.1) | 4 (11.1) | 5 (16.1) | 2 (6.3) | 15 (10.9) | 0.821 |
| Fuhrman nuclear grade | Fuhrman nuclear grade | Fuhrman nuclear grade | Fuhrman nuclear grade | Fuhrman nuclear grade | Fuhrman nuclear grade | Fuhrman nuclear grade | Fuhrman nuclear grade | Fuhrman nuclear grade |
| 2 | 0 (0) | 2 (66.7) | 4 (18.2) | 3 (8.3) | 6 (19.4) | 5 (14.3) | 20 (14.2) | 0.246 |
| 3 | 8 (57.1) | 1 (33.3) | 11 (50) | 22 (61.1) | 15 (48.4) | 16 (45.7) | 73 (51.8) | 0.246 |
| 4 | 6 (42.9) | 0 (0) | 7 (31.8) | 11 (30.6) | 10 (32.3) | 14 (40) | 48 (34.0) | 0.246 |
| Necrosis | | | | | | | | |
| Yes | 10 (71.4) | 2 (66.7) | 17 (77.3) | 24 (66.7) | 17 (54.8) | 24 (68.6) | 94 (66.7) | 0.659 |
| SSIGN score*** | SSIGN score*** | SSIGN score*** | SSIGN score*** | SSIGN score*** | SSIGN score*** | SSIGN score*** | SSIGN score*** | SSIGN score*** |
| n (%) | 6 (5.6) | 2 (1.8) | 14 (13.08) | 32 (29.9) | 24 (22.4) | 29 (27.1) | 107 (100) | |
| Mean (std) | 7.3 (±1.8) | 3.5 (±2.1) | 5.1 (±1.6) | 6.1 (±1.7) | 6.3 (±1.8) | 6 (±1.6) | 6 (±1.76) | |
| UISS score** | 2.5 (±0.9) | 2 (±0.6) | 4 (±1.0) | 3 (±0.9) | 3.5 (±0.9) | 2 (± 1.0) | 2.94 (±0.93) | 0.57 |
| mGPS | mGPS | mGPS | mGPS | mGPS | mGPS | mGPS | mGPS | mGPS |
| Low | 8 (66.7) | 2 (100) | 8 (38.1) | 13 (38.2) | 13 (44.8) | 12 (35.3) | 56 (42.4) | 0.211 |
| Intermediate | 2 (16.7) | 0 (0) | 4 (19) | 7 (20.6) | 9 (31) | 14 (41.2) | 36 (27.3) | 0.211 |
| High | 2 (16.7) | 0 (0) | 9 (42.9) | 14 (41.2) | 7 (24.1) | 8 (23.5) | 40 (30.3) | 0.211 |
| Missing | – | – | – | – | – | – | 9 (6.4%) | |
| Corrected calcium** | 9.8 (±0.6) | 9.8 (±0.6) | 9.9 (±0.7) | 9.8 (±0.8) | 9.7 (±0.6) | 9.6 (±0.4) | 9.74 (±0.65) | 0.447 |
| Received systemic therapy | 4 (28.6) | 0 (0) | 6 (27.3) | 12 (33.3) | 14 (45.2) | 11 (31.4) | 47 (33.3) | 0.547 |
| MAP score | MAP score | MAP score | MAP score | MAP score | MAP score | MAP score | MAP score | MAP score |
| Mean** | – | – | – | – | – | – | 3.22 (±1.52) | |
| 0–3 | – | – | – | – | – | – | 75 (53.2) | |
| 4–5 | – | – | – | – | – | – | 66 (46.8) | |
On Kaplan–Meier analysis, there were no significant associations between continuous or dichotomized MAP scores and PFS or OS (Figure 1). Similarly, multivariable Cox proportional hazard models (Table 2) demonstrated no significant associations both statistically and clinically between MAP and PFS (HR = 1.01, $95\%$ CI 0.85–1.20, $$P \leq 0.93$$) or OS (HR = 1.01, $95\%$ CI 0.84–1.21, $$P \leq 0.917$$). However, no receipt of systemic therapy was associated with better PFS (HR = 0.28, $95\%$ CI 0.18–0.46, $P \leq 0.001$). The presence of IVC tumor thrombus was predictive of significantly worse OS (HR = 2.05, $95\%$ CI 1.23–3.39, $$P \leq 0.006$$). These results were similar with and without the inclusion of SSIGN score.
**Figure 1::** *Kaplan–Meier curves for stage T3/T4 renal cell carcinoma patients ($$n = 141$$) displaying median progression-free survival (PFS) or overall survival (OS) with either individualized (0–5) or dichotomized (0–3 vs 4–5) Mayo Adhesive Probability (MAP) scores. (A) Median PFS with individualized MAP Scores. (B) Median PFS with dichotomized MAP scores. (C) Median PFS with individualized MAP Scores. (D) Median OS with dichotomized MAP scores.* TABLE_PLACEHOLDER:Table 2: *On subanalysis* of patients with the presence of IVC tumor thrombus (Table 3), MAP score continued to have no significant predictive value. Non-white race was associated with worse PFS and no receipt of systemic therapy was associated with improved PFS. For OS, non-white race was additionally associated with worse survival. Notably, only 14 of the patients in the thrombus cases were non-white.
**Table 3:**
| Covariate | N | Hazard ratio (95% CI) | P |
| --- | --- | --- | --- |
| Progression-free survival* | Progression-free survival* | Progression-free survival* | Progression-free survival* |
| MAP score | 64 | 0.92 (0.67–1.25) | 0.589 |
| Non-white race | 14 | 2.38 (1.07–5.33) | 0.034 |
| Received systemic therapy | Received systemic therapy | Received systemic therapy | Received systemic therapy |
| Yes | 26 | – | – |
| No | 38 | 0.29 (0.14–0.57) | <0.001 |
| Age | 64 | 1.02 (1.00–1.05) | 0.082 |
| Corrected calcium | 64 | 1.56 (0.98–2.47) | 0.06 |
| Overall survival** | Overall survival** | Overall survival** | Overall survival** |
| MAP score | 65 | 1.09 (0.83–1.44) | 0.539 |
| Non-white race | 14 | 2.31 (1.10–4.83) | 0.027 |
## Discussion
Ultimately, a significant association of MAP with PFS and OS in patients with nonmetastatic T3–T4 RCC was not identified. These findings are important as our understandings of diagnostics, patient-specific prognostication, and the role of body composition in RCC continue to evolve. It is believed that the association between MAP and survival outcomes in RCC exists because perinephric fat thickness and stranding may serve as a proxy for visceral obesity and inflammation [4]. Visceral adiposity and inflammation are interconnected, and each has additionally independently been identified as a risk factor for poorer RCC survival outcomes and more aggressive disease [5, 6].
It is unclear why our population of locally advanced RCC patients does not corroborate previous literature demonstrating the prognostic value of MAP scores. It is likely that, for locally advanced disease, extent of disease extension, Tumor, Node, Metastasis (TNM) stage, and other tumor-specific factors play a stronger role in determining survival, thus overpowering the effects of factors such as visceral obesity [7, 8]. Moreover, patients with higher visceral obesity may be more likely to present with confounding factors harming survival, including advanced disease or cardiovascular comorbidities.
An important consideration in this cohort of patients with locally advanced disease is the utility of effective systemic therapy (i.e., immune-checkpoint inhibitors [ICI], tyrosine kinase inhibitors) in the neoadjuvant or adjuvant setting. While neoadjuvant therapy prior to nephrectomy has been reported as feasible (9–11), our patient cohort only received adjuvant systemic therapy since neoadjuvant systemic therapy outside of clinical trials is currently not routinely used for nonmetastatic RCC. Patients not receiving any systemic therapy actually experienced better survival, likely as a result of patient selection. Furthermore, in our cohort, there was no difference in receipt of adjuvant systemic therapy by MAP score. Though unable to be captured in this cohort, ICI-induced inflammatory response [12] could potentially be captured by radiographic features measured by MAP scoring given it may partially serve as a proxy for perinephric inflammation [4]. Therefore, future studies are warranted in this patient population, specifically.
## Conclusions
This study is the first to examine MAP score in a locally advanced RCC cohort. No significant associations with survival outcomes were identified. Limitations of this study include its retrospective nature and relatively limited sample size. The data enhances our best use of MAP scores to the T1/T2 population. As we move forward in an age of precision medicine and patient-specific risk stratification, a comprehensive understanding of potential prognostic tools is crucial for decision-making and patient counseling.
## Funding
We gratefully acknowledge support of the John Robinson Family Foundation, Christopher Churchill Foundation, and Cox Immunology Fund.
## Declaration of Interest
All of the authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
## References
1. Siegel RL, Miller KD, Fuchs HE, Jemal A. **Cancer statistics, 2021**. *CA Cancer J Clin* (2021) **71** 7-33. DOI: 10.3322/caac.21654
2. Davidiuk AJ, Parker AS, Thomas CS, Leibovich BC, Castle EP, Heckman MG. **Mayo adhesive probability score: an accurate image-based scoring system to predict adherent perinephric fat in partial nephrectomy**. *Eur Urol* (2014) **66** 1165-71. DOI: 10.1016/j.eururo.2014.08.054
3. Yao Y, Xu Y, Gu L, Liu K, Li P, Xuan Y. **The Mayo adhesive probability score predicts longer dissection time during laparoscopic partial nephrectomy**. *J Endourol* (2020) **34** 594-9. DOI: 10.1089/end.2019.0687
4. Thiel DD, Davidiuk AJ, Meschia C, Serie D, Custer K, Petrou SP. **Mayo adhesive probability score is associated with localized renal cell carcinoma progression-free survival**. *Urology* (2016) **89** 54-60. DOI: 10.1016/j.urology.2015.10.034
5. Park YH, Lee JK, Kim KM, Kook HR, Lee H, Kim KB. **Visceral obesity in predicting oncologic outcomes of localized renal cell carcinoma**. *J Urol* (2014) **192** 1043-9. DOI: 10.1016/j.juro.2014.03.107
6. Hu X, Wang Y, Yang WX, Dou WC, Shao YX, Li X. **Modified Glasgow prognostic score as a prognostic factor for renal cell carcinomas: a systematic review and meta-analysis**. *Cancer Manag Res* (2019) **11** 6163-73. DOI: 10.2147/CMAR.S208839
7. Klatte T, Rossi SH, Stewart GD. **Prognostic factors and prognostic models for renal cell carcinoma: a literature review**. *World J Urol* (2018) **36** 1943-52. DOI: 10.1007/s00345-018-2309-4
8. Volpe A, Patard JJ. **Prognostic factors in renal cell carcinoma**. *World J Urol* (2010) **28** 319-27. DOI: 10.1007/s00345-010-0540-8
9. Bilen MA, Jiang JF, Jansen CS, Brown JT, Harik LR, Sekhar A. **Neoadjuvant cabozantinib in an unresectable locally advanced renal cell carcinoma patient leads to downsizing of tumor enabling surgical resection: a case report**. *Front Oncol* (2020) **10** 622134. DOI: 10.3389/fonc.2020.622134
10. Master VA, Schmeusser BN, Osunkoya AO, Palacios AR, Midenberg E, Yantorni L. **Neoadjuvant nivolumab and ipilimumab for nonmetastatic renal cell carcinoma with tumor thrombus**. *J Immunother Precis Oncol* (2023) **6** 50-5. DOI: 10.36401/JIPO-22-16
11. Gorin MA, Patel HD, Rowe SP, Hahn NM, Hammers HJ, Pons A. **Neoadjuvant nivolumab in patients with high-risk nonmetastatic renal cell carcinoma**. *Eur Urol Oncol* (2022) **5** 113-17. DOI: 10.1016/j.euo.2021.04.002
12. Pignot G, Thiery-Vuillemin A, Walz J, Lang H, Bigot P, Werle P. **Nephrectomy after complete response to immune checkpoint inhibitors for metastatic renal cell carcinoma: a new surgical challenge?**. *Eur Urol* (2020) **77** 761-3. DOI: 10.1016/j.eururo.2019.12.018
|
---
title: Association between diet history and symptoms of individuals having recovered
from COVID-19
authors:
- Ola T. Sahloul
- Talaat M. Sahloul
journal: Journal of Health, Population, and Nutrition
year: 2023
pmcid: PMC10036967
doi: 10.1186/s41043-023-00365-7
license: CC BY 4.0
---
# Association between diet history and symptoms of individuals having recovered from COVID-19
## Abstract
### Background
Many studies show that people who eat a balanced diet have stronger immunity. The present work aimed to identify the effects of the diet history of COVID-19 patients having recovered from the disease on the occurrence and severity of symptoms.
### Methods
The study sample consisted of 346 individuals aged 20–65 years. The participants’ data and answers to an electronic questionnaire regarding their diet history and symptoms were collected. The study focused on four hard symptoms (fever, body pain, cough, and dyspnoea) to investigate the relationship between these symptoms and the consumption of specific immunity foods.
### Results
Symptoms were reported by $88.1\%$ of the participants eating none of the foods investigated, whereas $85.54\%$ and $85.55\%$ of the individuals with little or intensive food intake, respectively, experienced symptoms.
### Conclusions
Intake of specific functional foods might slightly reduce the occurrence of some symptoms.
## Introduction
The coronavirus disease 2019 or COVID-19 is characterized by serious acute respiratory symptoms [1]. This disease originated at the end of 2019 in Wuhan, China [2, 3]. In March 2020, the World Health Organization (WHO) considered the COVID-19 outbreak a global pandemic [4]. Since 14 February 2020, when the first cases of COVID-19 infections were reported in Egypt, the Egyptian people have lived in a different reality [5, 6]. The first symptom usually experienced by COVID-19 patients is pain, particularly headaches, myalgia, or arthralgia [7, 8]. The pain appears 1.6 days after the onset of illness. The second set of symptoms is fever, followed by cough and diarrhea. Then, patients suffer from anosmia, which occurs days after the onset of infection. All symptoms persist for 10 ± 4.9 (mean ± standard deviation) days (range of 3–27 days), except for the fever, cough, and anosmia that last 5.5 ± 4.4 (range of 1–19 days), 7.7 ± 4.3 (range of 1–18 days), and 7.3 ± 5 (range of 1–19 days), respectively [9].
In addition to the outbreak, governments have imposed "lock-downs" to mitigate its virus's spread [10, 11]. Therefore, the prevalence of psychological disorders' symptoms has increased widely (e.g., depression, somatic symptoms, anxiety, negative feelings, emotional exhaustion, and panic disorder) [12–14]. Consequently, these symptoms could potentially disrupt sleep patterns [11, 15]. Plus, a reduction in daily levels of physical activity could also negatively affect sleep [16] and thus contribute to an impaired immune system [17].
The nutritional status, which results from the diet history, can significantly impact the overall health and reduce the risk of developing infections [18]. Healthy nutritional habits help prevent non-communicable diseases, which are risk factors for developing COVID-19 [19]. Additionally, nutrition has been linked to systemic infectious diseases through its effects on the immune system [20]. Thus, malnutrition increases the host’s susceptibility to infectious diseases. These infections negatively affect the metabolism, worsening the nutritional state [21]. Moreover, getting used to a healthy lifestyle is important for reducing cholesterol levels and increasing antioxidant levels from fruits, vegetables, and monounsaturated fatty acids present in fish, nuts, and olive oil [22]. Moreover, the frequent consumption of healthy foods such as vegetables, fruits, and fish contributes to supplying the body with sufficient amounts of essential nutrients and antioxidants [23, 24]. In Damietta (Egypt), most COVID-19 patients have poor nutritional habits and present severe symptoms of fatigue. Additionally, most COVID-19 patients are overweight or obese, and these patients have more severe symptoms of fatigue [25]. Interestingly, the Mediterranean diet is one of the healthiest diets in the world and is associated with lower rates of mortality, obesity, type 2 diabetes mellitus, low-grade inflammation, cancer, Alzheimer’s disease, depression, and COVID-19 [22, 26].
Therefore, the current study aims to identify the effects of the nutritional history of patients having recovered from COVID-19 on the occurrence and intensity of some of their symptoms.
## Study design and participants
This study took place in Egypt (Damietta Governorate). A random sample of 346 individuals who had recovered from COVID-19 was selected. The sample consisted of 212 women and 134 men aged 20 to 65 years, 131 from rural areas, and 215 from urban areas. The illness lasted 14 to 21 days for most people ($52.3\%$ subjects).
All participants consented to share their data by sending back the electronic questionnaire. Several specialized faculty members from the Faculty of Specific Education at Damietta University verified the test’s validity. The two tests’ stability coefficient was calculated before using the data. The average time spent filling out the survey was 10 min.
## The study collected
[1] Personal information (gender, location, age, and illness duration).[2] Answers to a structured questionnaire on diet history.
Then, the relationship between the diet history, particularly consuming certain foods, and the degree of some symptoms was determined.
An electronic questionnaire (in the Arabic language) was built using the Google Form application [27, 28] and can be viewed at the following URL: https://docs.google.com/forms/d/e/1FAIpQLSfNudaJ3Hpn1XonEFU-rJkq5zZnywYhDzoV9Y28rQ9NkmQ8LA/viewform?usp=sf_link.
Translated version of the survey.https://docs.google.com/forms/d/e/1FAIpQLSd9gyg9-FqB1fpTzTeXqv6DCwJqROjj6kuAAbZCnjhvWxip1Q/viewform?usp=sf_link.
## COVID-19 cases
COVID-19 cases have been defined as symptomatic (with fever, cough, nasal congestion and runny nose, sore throat, dyspnoea, loss of smell or taste, body pain, and diarrhea) or asymptomatic (defined as a positive PCR or antibody test without typical COVID-19 symptoms) [29].
## Severity and duration of COVID-19 illness
Participants rated their COVID-19 symptoms from three options: asymptomatic, moderate symptoms, and severe symptoms. In addition, they had to indicate the number of days spent presenting COVID-19 symptoms [29].
## Statistical analysis
SPSS statistical software (version 11.5.1) was used to analyse the data collected using Pearson’s correlation coefficient (R) [30] (Fig. 1).Fig. 1Flowchart to disclosing the relationship between some functional food and COVID-19 symptoms
## Results
Figure 2 recapitulates the COVID-19 symptoms that were investigated, namely fever, body pain, cough, nasal congestion, runny nose, sore throat, diarrhea, dyspnoea, and loss of smell or taste. Most participants reported having had a moderate fever, cough, nasal congestion, runny nose, sore throat, diarrhea, and dyspnoea ($52\%$, $63.6\%$, $59\%$, $53.8\%$, $49.1\%$, and $53.2\%$, respectively). Then, the rest of participants had severe body pain and smell or taste loss ($67.6\%$ and $48.6\%$, respectively).Fig. 2COVID-19 symptoms investigated Thus, most participants had moderate symptoms, some had severe symptoms, and a minority was asymptomatic.
Tables 1 to Fig. 3 present the relationship between COVID–19 symptoms (fever, body pain, cough, and dyspnea in Tables 1, 2, 3, and 4, respectively and intake of some foods (fruits or vegetables, yogurt, onions, and garlic). Each table was divided into three levels of food intake: [1] no intake, [2] little intake (weekly or monthly), and [3] intensive intake (more than once a week to daily).Table 1Relationship between the consumption of specific foods and the occurrence of feverLevel of intakeFoodsFruits or vegetablesYogurtOnionsGarlicTotal symptomsSymptom rateN%N%N%N%N%No intakeAsymptomatic42027.1610.7812.92012SymptomsModerate84014502646.43251.68048.2Severe8401242.92442.92235.56639.8Total16802692.95089.35487.114688Total20100281005610062100166100Little intake (monthly or weekly)Asymptomatic812.5141368.11010.63811.2SymptomsModerate3453.15853.73851.44851.117852.3Severe2234.43633.33040.53638.312436.5Total5687.594876891.98489.430288.8Total641001081007410094100340100Intensive intake (daily or more than once a week)Asymptomatic3613.73215.33616.73015.813415.3SymptomsModerate13852.710851.411653.710052.646252.6Severe8833.67033.36429.66031.628232.1Total22686.317884.718083.316084.274484.7Total262100210100216100190100878100Total3463463463461384R0.0630.0750.141** − 0.0680.135**N: Number of participantsR: Pearson’s correlation coefficient**: P-value of correlation was significant ($p \leq 0.01$), 2-tailed testFig. 3Relationship between the consumption of all investigated foods and the proportion of paticipants having symptomsTable 2Relationship between the consumption of specific foods and the occurrence of body painLevel of intakeFoodsFruits or vegetablesYogurtOnionsGarlicTotal symptomsSymptom rateN%N%N%N%N%No intakeAsymptomatic0000000000SymptomsModerate840828.6142518294828.9Severe12602071.44275447111871.1Total20100281005610062100166100Total20100281005610062100166100Little intake (monthly or weekly)Asymptomatic0021.90022.141.2SymptomsModerate16254642.63243.23234.112637Severe48756055.54256.86063.821061.8Total6410010698.1741009297.933698.8Total641001081007410094100340100Intensive intake (daily or more than time weekly)Asymptomatic41.52141.921.1121.4SymptomsModerate8432.15425.76228.75830.525829.4Severe17466.415473.315069.413068.460869.2Total25898.52089921298.118898.986698.6Total262100210100216100190100878100Total3463463463461384R − 0.0570.0240.069 − 0.0180.037N Number of participantsR Pearson’s correlation coefficientTable 3Relationship between the consumption of specific foods and the occurrence of coughLevel of intakeFoodsFruits or vegetablesYogurtOnionsGarlicTotal symptomsSymptom rateN%N%N%N%N%No intakeAsymptomatic210621.4610.769.72012.1SymptomsModerate15751864.33460.73454.810160.8Severe315414.31628.62235.54527.1Total18902278.65089.35690.314687.9Total20100281005610062100166100Little intake (monthly or weekly)Asymptomatic2031.22624.11418.92223.48224.1SymptomsModerate3859.46257.43445.95255.318654.7Severe69.42018.52635.22021.37221.2Total4468.88275.96081.17276.625875.9Total641001081007410094100340100Intensive intake (daily or more than time weekly)Asymptomatic5219.8422054254624.219422.1SymptomsModerate16763.79846.711050.99248.446753.2Severe4316.47033.35224.15227.421724.7Total21080.2168801627514475.868477.9Total262100210100216100190100878100Total3463463463461384R0.022 − 0.0860.014 − 0.116* − 0.002N Number of participantsR Pearson’s correlation coefficient*: P-value of correlation was significant ($p \leq 0.05$), 2-tailed testTable 4Relationship between the consumption of specific foods and the occurrence of dyspnoeaLevel of intakeFoodsFruits or vegetablesYogurtOnionsGarlicTotal symptomsSymptoms rateN%N%N%N%N%No intakeAsymptomatic630621.41017.81422.63621.7SymptomsModerate8401864.32646.53454.88651.8Severe630414.32035.71422.64426.5Total14702278.64682.24877.413078.3Total20100281005610062100166100Little intake (monthly or weekly)Asymptomatic1421.92825.91418.91819.17421.8SymptomsModerate2843.85248.23648.64851.116448.2Severe2234.32825.92432.52829.810230Total5078.18074.16081.17680.926678.2Total641001081007410094100340100Intensive intake (daily or more than time weekly)Asymptomatic5019.13617.14621.3382017019.4SymptomsModerate14856.511454.312256.510253.748655.4Severe6424.46028.64822.25026.322225.2Total21280.917482.917078.71528070880.6Total262100210100216100190100878100Total3463463463461384R0.063 − 0.0830.0910.0180.062*N Number of participantsR Pearson’s correlation coefficient*: P-value of correlation was significant ($p \leq 0.05$), 2-tailed test Table 1 shows that participants experienced fever independently of their food intake. Data shows that among 166 recovered individuals who did not eat any of the investigated foods, 20 ($12.5\%$) had been asymptomatic, whereas 146 ($87.5\%$) had suffered from either moderate or severe fever. Among the 338 participants with little food intake (weekly or monthly), only 38 cases ($11.24\%$) had no fever, whereas 300 ($88.76\%$) had experienced moderate or severe fever. Finally, among the 877 subjects with intensive food intake (daily or more than once a week), 134 ($15.26\%$) had had no fever, and 744 ($84.74\%$) had suffered from moderate or severe fever.
These data indicate comparable fever occurrence rates ($88\%$ and $88.8\%$) in participants with no or little food uptake, whereas fever occurred less in subjects with intensive intake ($84.7\%$). Additionally, the foods investigated here had different effects on the occurrence of fever. For example, fever was reported less ($80\%$) by participants consuming no vegetables or fruits (no intake), whereas it occurred the most ($92.9\%$) in subjects who did not eat yogurt. Participants with little intake (weekly or monthly) of yogurt, vegetables or fruits ($87\%$ and $87.5\%$, respectively) experienced fever less often. Fevers were more intense in subjects consuming only small amounts of onions and garlic ($91.9\%$ and $89.4\%$, respectively). Finally, $86.3\%$, $84.7\%$, $84.2\%$, and $83.3\%$) of participants with an intensive intake (daily or more than once a week) of fruits or vegetables, yogurt, garlic, and onion, respectively, had a fever.
There is a significant correlation between fever occurrence, onion, and total food consumption (Pearson’s correlation coefficient $R = 0.141$ and 0.135, respectively, $p \leq 0.01$).
Data in Table 2 showed that $100\%$ of participants not eating any of the investigated foods (no intake) had moderate or severe body pain symptoms. Moderate or severe body pain was experienced by $98.8\%$ of the participants with little food intake (weekly or monthly) and $98.6\%$ of those with an intensive intake (daily or more than once a week) of the foods. In view of the fact that the proportion of patients suffering from body pain decreased, these foods and their bioactive compounds may have affected the body’s resistance to pain.
Additionally, data in Table 2 reveals that $100\%$ of participants who had experienced body pain were eating few fruits or vegetables and onions (weekly or monthly). In contrast, a slight decrease in the proportion of participants who had suffered from body pain occurred with the intensive consumption of yogurt, garlic, fruits or vegetables, and onions ($99\%$, $98.9\%$, $98.5\%$, and $98.1\%$, respectively).
Data in Table 3 shows Cough symptoms occurred in participants who did or did not eat the investigated foods. According to the findings, 166 recovered individuals who did not consume any food had experienced either a moderate or severe cough. Among the 340 participants having little food intake (weekly or monthly), 258 ($75.9\%$) had suffered from moderate or severe cough symptoms. Finally, among the 878 individuals with intensive food intake (daily or more than once a week), 684 ($77.9\%$) had moderate or severe cough symptoms.
These data revealed a similar cough occurrence in individuals with little or intensive dietary intake ($75.9\%$ and $77.9\%$, respectively). This rate increased in participants eating none of the investigated foods ($87.9\%$). Additionally, the foods tested in this study had different effects on cough occurrence. Participants eating no yogurt reported fewer cough symptoms ($78.6\%$) than those who consumed no garlic, onions, or fruits or vegetables ($90.3\%$, $90\%$, or $89.3\%$, respectively). Few individuals appeared to have cough symptoms ($68.8\%$) with little intake (weekly or monthly) of fruits or vegetables, compared to those rarely eating yogurt or garlic ($75.9\%$ or $76.6\%$, respectively). Participants with little onion intake reported experiencing cough symptoms ($81.1\%$). Finally, fewer subjects who often (daily or more than once a week) ate onions or garlic developed cough symptoms ($75\%$ and $75.8\%$, respectively). In contrast, the cough was most experienced by individuals who often had yogurt, fruits or vegetables ($80\%$ or $80.2\%$, respectively). There was a significant correlation between garlic consumption and cough severity ($R = 0.116$, p 0.05).
Data in Table 4 presents the link between the consumption of some foods and the development of dyspnea. Data revealed that 36 out of 166 recovering participants who did not eat the foods did not develop symptoms ($21.7\%$), whereas 130 ($78.3\%$) had suffered from either moderate or severe dyspnea. Among the 340 who had a small intake (weekly or monthly) of foods, only 74 ($21.8\%$) had not had dyspnea, whereas 266 ($78.2\%$) had suffered from either moderate or severe dyspnea. Finally, among the 878 individuals consuming the foods regularly (daily or more than once a week), 170 ($19.4\%$) had not experienced dyspnea, whereas 708 cases ($80.6\%$) developed moderate or severe dyspnea. The data also showed that many individuals had developed dyspnea regardless of their general food consumption ($78.3\%$, $78.2\%$, and $80.6\%$, for no, little, and intensive uptake, respectively). Among the participants having no intake of the foods, dyspnea symptoms were less observed in those who did not eat fruits or vegetables ($70\%$), followed by those who did not eat garlic ($77.4\%$), yogurt ($78.6\%$), or onions ($82.2\%$).
Additionally, fewer individuals with a low intake (weekly or monthly) of yogurt reported dyspnea symptoms ($74.1\%$) than those eating little fruit or vegetables, garlic, or onions ($78.1\%$, $80.9\%$, or $81.1\%$, respectively). Finally, participants eating regularly (daily or more than once a week) yogurt experienced dyspnea symptoms more often ($82.9\%$) than those eating high levels of onions, garlic, or fruits or vegetables ($78.7\%$, $80.9\%$, or $80.9\%$, respectively).
There was a significant correlation between total food consumption and the development of dyspnea ($R = 0.062$, $p \leq 0.05$).
Figure 3 summarizes the data from Tables 1, 2, 3, and 4 regarding the link between the food consumption history of individuals having recovered from COVID-19 and the rate of specific symptoms. The proportions of individuals who consumed a specific food, never (no intake), sometimes (little intake), or regularly (intensive intake) and developed a specific symptom were calculated. Moreover, the average proportion of participants developing symptoms according to their average intake of specific foods and the average proportion of individuals developing a specific symptom according to their global food consumption level were calculated. The data showed that symptoms were present in $88.1\%$ of participants consuming none of the listed foods, $85.54\%$ of those consuming these foods sometimes, and $85.55\%$ of those eating these foods regularly.
This data draws attention to the fact that the food consumption history of recovered people might have increased their resistance to some symptoms, as fewer participants with little or intensive food intake presented symptoms than those who did not consume any of the foods.
## Discussion
The authors think that this study is the first to link COVID-19 symptoms with consumption levels of some foods by recovered individuals from the disease. Then choose three food intake levels: no intake, little intake (weekly or monthly), and intensive intake (daily or more than once a week) and Four food types that positively affect the immune system were chosen. Indeed, fruits or vegetables provide high amounts of vitamins and minerals that are important for the immune system [31]. Additionally, a lower COVID-19 infection rate in individuals drinking yogurt daily than that of people not drinking yogurt has been reported [32]. Yogurt, a fermented dairy product, exhibits interesting properties related to the presence of bioactive peptides and probiotics that might play a beneficial role in COVID-19 presentation and outcome [33]. Kumar et al. [ 2015] indicated that onions possess immune-stimulatory activities toward murine lymphocytes [34]. Hirayama et al. [ 2019] also suggested that the intake of low or high doses of onion green leaf extract might positively regulate immune competence [35]. Garlic essential oil is also a valuable natural antiviral agent that contributes to preventing the invasion of the human body by a coronavirus [36]. Finally, the immune system is highly affected by malnutrition, which leads to decreased immune responses and a consequent augmented risk of infection and disease severity [37].
On the other hand, participants reported suffering from different symptoms, namely fever, body pain, cough, nasal congestion, runny nose, sore throat, diarrhea, dyspnea, and loss of smell or taste. However, the study found that four symptoms, i.e., fever, body pain, cough, and dyspnea, were more often experienced and were linked with the participants’ history of immune food consumption. These findings were similar to those reported by (Carfì et al. 2020) [38], who found that, after the onset of the first COVID-19 symptom, only 18 ($12.6\%$) participants were completely free of any COVID-19 symptom, whereas $32\%$ had 1 or 2 symptoms and $55\%$ had 3 or more. A worse quality of life was observed among $44.1\%$ of the patients. The data also shows that a high proportion of individuals reported myalgia ($53.1\%$), dyspnea ($43.4\%$), and joint pain ($27.3\%$). In this respect, (Çalıca et al., 2020) reported that out of 297 patients, 143 had positive symptoms, and 154 had negative symptoms [39]. The most common symptoms in the positive group were cough ($56.6\%$), weakness ($56.6\%$), taste disorder ($35.7\%$), myalgia ($34.3\%$), and fever ($33.6\%$), whereas in the negative group they were cough ($63\%$), weakness ($45.5\%$), dyspnea ($29.9\%$), headache ($27.3\%$), and fever ($24.7\%$).
The present study compared the occurrence of symptoms according to the general and specific consumption of foods. Although there was no impact of food consumption on symptoms in general, the consumption of specific foods in given amounts affected the occurrence of some symptoms. The most notable difference was body pain, experienced by $100\%$ of the participants who had not eaten any of the foods, whereas it was less present in individuals with low or high food intake. Moreover, as shown in Fig. 3, fewer symptoms were experienced by participants consuming the investigated foods sometimes (little intake) or regularly (intensive intake) compared with the group having no intake of these foods. These data contradict those of a previous study (Kim et al., 2021), indicating that no association was observed between diets and the odds of COVID-19-like illness or duration of symptoms [29].
## Conclusions
The present study suggests that although there is no impact of food consumption on symptoms in general, the consumption of some foods i.e., (Fruits or vegetables, Yogurt, Onions and Garlic) affected the occurrence of some symptoms. The most notable difference is body pain.
## Recommendation
The authors suggested to conduct more research by adding more foods related to COVID-19 symptoms especially for children and adolescent. Also, governments should provide more information related immunity foods content in information, education and communication on pandemic targeted to the communities. Furthermore, civil society organizations should pay special attention to assist social including immunity foods vulnerable groups in the family.
## The strengths and limitations of the study are
The strength of the study’s findings lies in the fact that the relationship between different consumption levels of specific foods by people having recovered from COVID-19 and the symptoms experienced by these individuals was investigated.
The limitation of the study was insufficient participants.
## References
1. 1.Centers for Disease Control and Prevention. "Centers for Disease Control and Prevention Coronavirus disease 2019 (COVID-19) " United States: centers for disease control and prevention, Available from: https://www.cdc.gov/coronavirus/2019-ncov/index.html Accessed 25 July 2020.
2. Cucinotta D, Vanelli M. **WHO declares COVID-19 a pandemic**. *Acta bio medica: Atenei parmensis.* (2020.0) **91** 157. PMID: 32191675
3. Jiang S, Xia S, Ying T, Lu L. **A novel coronavirus (2019-nCoV) causing pneumonia-associated respiratory syndrome**. *Cell Mol Immunol* (2020.0) **17** 554-554. DOI: 10.1038/s41423-020-0372-4
4. Sohrabi C, Alsafi Z, O’Neill N, Khan M, Kerwan A, Al-Jabir A, Iosifidis C, Agha R. **World health organization declares global emergency: a review of the 2019 novel coronavirus (COVID-19)**. *Int J Surg* (2020.0) **12** 682-692
5. Morais De, Medeiros C. **Nutritional therapy in COVID-19 management**. *Kompass Nutrition & Dietetics* (2021.0) **1** 10-12
6. 6.The Arab Republic of Egypt the prime minister [homepage on the Internet] (2020). Available from: https://www.care.gov.eg/EgyptCare/News/Preview.aspx?id=15059.Accessed Accessed 13 Mar, 2020.
7. Dergaa I, Abubakera M, Souissi A, Mohammed AR, Varmaa A, Musa S, Al Naama A, Mkaouer B, Saad HB. **Age and clinical signs as predictors of COVID-19 symptoms and cycle threshold value**. *Libyan J Med* (2022.0) **17** 10. DOI: 10.1080/19932820.2021.2010337
8. Varma A, Dergaa I, Mohammed AR, Abubaker M, Al Naama A, Mohammed S, Rafique MA, Manu L, Vedasalam S, Parveze P, Hammouda O. **Covid-19 and diabetes in primary care–How do hematological parameters present in this cohort?**. *Expert Rev Endocrinol Metabol.* (2021.0) **16** 147-153. DOI: 10.1080/17446651.2021.1909472
9. 9.Zayet, S., Gendrin, V., & Klopfenstein, T. "Natural history of COVID-19: back to basics" New Microbes and New Infections, 38, PP100815, 2020.
10. Rubin GJ, Wessely S. **The psychologicaleffects of quarantining a city**. *BMJ* (2020.0) **368** m313. DOI: 10.1136/bmj.m313
11. Trabelsi K, Ammar A, Masmoudi L, Boukhris O, Chtourou H, Bouaziz B, Brach M, Bentlage E, How D, Ahmed M, Mueller P. **Globally altered sleep patterns and physical activity levels by confinement in 5056 individuals: ECLB COVID-19 international online survey**. *Biol Sport* (2021.0) **38** 495-506. DOI: 10.5114/biolsport.2021.101605
12. Ammar A, Mueller P, Trabelsi K, Chtourou H, Boukhris O, Masmoudi L, Bouaziz B, Brach M, Schmicker M, Bentlage E, How D. **Psychological consequences of COVID-19 home confinement: The ECLB-COVID19 multicenter study**. *PLoS ONE* (2020.0) **15** e0240204. DOI: 10.1371/journal.pone.0240204
13. Jiang HJ, Nan J, Lv ZY, Yang J. **Psychological impacts of the COVID-19 epidemic on Chinese people: exposure, post-traumatic stress symptom, and emotion regulation**. *Asian Pacific J Tropical Med* (2020.0) **13** 252. DOI: 10.4103/1995-7645.281614
14. Trabelsi K, Ammar A, Masmoudi L, Boukhris O, Chtourou H, Bouaziz B, Brach M, Bentlage E, How D, Ahmed M, Mueller P. **Sleep quality and physical activity as predictors of mental wellbeing variance in older adults during COVID-19 lockdown: ECLB COVID-19 international online survey**. *Int J Environ Res Publ Health* (2021.0) **18** 4329. DOI: 10.3390/ijerph18084329
15. Xiao H, Zhang Y, Kong D, Li S, Yang N. **The effects of social support on sleepquality of medical staff treating patientswith coronavirus disease (COVID-19) in January and February 2020 in China**. *Med Sci Monitor: Int Med J Experiment Clinic Res* (2019.0) **26** 2020
16. Akbari HA, Pourabbas M, Yoosefi M, Briki W, Attaran S, Mansoor H, Ghram A. **"How physical activity behavior affected well-being, anxiety and sleep quality during COVID-19 restrictions in Iran**. *Eur Rev Med Pharmacol Sci* (2021.0) **25** 7847-7857. PMID: 34982447
17. 17.National Heart, Lung, and BloodInstitute. COVID and sleep: betterslumber during the pandemic may helpprotect your health, as of September 1, 2020.
18. 18.World Health Organization. Off-label use of medicines for COVID-19 (2020). Available from: https://www.who.int/news-room/commentaries/detail/off-label-use-of-medicines-for-covid-19.
19. Zabetakis I, Lordan R, Norton C, Tsoupras A. **COVID-19: the inflammation link and the role of nutrition in potential mitigation**. *Nutrients* (2020.0) **12** 1466. DOI: 10.3390/nu12051466
20. Farhadi S, Ovchinnikov RS. **The relationship between nutrition and infectious diseases: a review**. *Biomed Biotechnol Res J* (2018.0) **2** 168-172. DOI: 10.4103/bbrj.bbrj_69_18
21. Guerrant RL, Reinaldo BO, Sean SM, Mônica OBO, Aldo AML. **Malnutrition as an enteric infectious disease with long-term effects on child development**. *Nutrition Reviews* (2008.0) **66** 487-505. DOI: 10.1111/j.1753-4887.2008.00082.x
22. Soldati L, Di Renzo L, Jirillo E, Ascierto PA, Marincola FM, De Lorenzo A. **The influence of diet on anti-cancer immune responsiveness**. *J Transl Med* (2018.0) **16** 1-18. DOI: 10.1186/s12967-018-1448-0
23. Wallace TC, Bailey RL, Blumberg JB, Burton-Freeman B, Chen CO, Crowe-White KM, Drewnowski A, Hooshmand S, Johnson E, Lewis R, Murray R, Shapses SA, Wang DD. **Fruits, vegetables, and health: a comprehensive narrative, umbrella review of the science and recommendations for enhanced public policy to improve intake**. *Crit Rev Food Sci Nutr* (2020.0) **60** 2174-2211. DOI: 10.1080/10408398.2019.1632258
24. Hosseini B, Berthon BS, Wark P, Wood LG. **Effects of fruit and vegetable consumption on risk of asthma, wheezing and immune responses: a systematic review and meta-analysis**. *Nutrients* (2017.0) **9** 341. DOI: 10.3390/nu9040341
25. Sahloul OT, El-Kholey HM. **Impact of nutritional habits and status on the symptoms degree with COVID-19**. *J. Res Fields Specific Edu* (2022.0) **8** 525-544
26. Cani PD, Van Hul M. **Mediterranean diet, gut micro biota and health: when age and calories do not add up**. *Gut* (2020.0) **69** 1167-1168. DOI: 10.1136/gutjnl-2020-320781
27. Di Renzo L, Gualtieri P, Pivari F, Soldati L, Attinà A, Cinelli G, Leggeri C, Caparello G, Barrea L, Scerbo F, Esposito E, De Lorenzo A. **Eating habits and lifestyle changes during COVID-19 lockdown: an Italian survey**. *J Transl Med* (2021.0) **18** 1-15
28. Heuer T, Krems C, Moon K, Brombach C, Hoffmann I. **Food consumption of adults in Germany: results of the German National nutrition survey II based on diet history interviews**. *Br J Nutr* (2015.0) **113** 1603-1614. DOI: 10.1017/S0007114515000744
29. Kim H, Rebholz CM, Hegde S, LaFiura C, Raghavan M, Lloyd JF, Cheng SB, Seidelmann SB. **Plant-based diets, pescatarian diets and COVID-19 severity: a population-based case–control study in six countries**. *BMJ Nutrition, Prevent Health* (2021.0) **4** 257-266. DOI: 10.1136/bmjnph-2021-000272
30. Artimage GY, Berry WG. *Statistical methods* (1987.0) 39-63
31. Elsayh KI, El-Badawy O, Saad K, Zahran AM. **Wheat bran as a natural remedy for COVID-19**. *Current Trends Immunol* (2020.0) **21** 123-128. DOI: 10.31300/CTIMM.21.2020.123-128
32. Silverio R, Gonçalves DC, Andrade MF, Seelaender M. **Coronavirus disease 2019 (COVID-19) and nutritional status: the missing link?**. *Adv Nutr* (2021.0) **12** 682-692. DOI: 10.1093/advances/nmaa125
33. 33.Shahbazi S, Hajimohammadebrahim-Ketabforoush M, Shariatpanahi MV, Shahbazi E, Shariatpanahi ZV. The validity of the global leadership initiative on malnutrition criteria for diagnosing malnutrition in critically ill patients with COVID-19: a prospective cohort study. Clinical nutrition. ESPEN. 2021;43:377-82.
34. Kumar VP, Prashanth KH, Venkatesh YP. **Structural analyses and immunomodulatory properties of fructo-oligosaccharides from onion (**. *Carbohydrate Polym* (2015.0) **6** 115-122. DOI: 10.1016/j.carbpol.2014.09.039
35. Hirayama Y, Takanari J, Goto K, Ueda H, Tanaka A, Nishihira J. **Effect of Welsh onion (**. *Funct Foods Health Disease* (2019.0) **9** 123-133. DOI: 10.31989/ffhd.v9i2.569
36. 36.Thuy BT, My TT, Hai NT, Hieu LT, Hoa TT, Thi Phuong Loan H, Triet NT, Anh TT, Quy PT, Tat PV, Hue NV. Investigation into SARS-CoV-2 resistance of compounds in garlic essential oil. ACS omega. 2020;5(14):8312-20.
37. 37.Mösbauer, K., Fritsch, V. N., Adrian, L., Bernhardt, J., Gruhlke, M. C. H., Slusarenko, A. J., ... & Antelmann, H. (2021), Allicin inhibits SARS-CoV-2 replication and abrogates the antiviral host response in the Calu-3 proteome. BioRxiv 14–49, 2021.
38. Carfì A, Bernabei R, Landi F. **Persistent symptoms in patients after acute COVID-19**. *JAMA* (2020.0) **324** 603-605. DOI: 10.1001/jama.2020.12603
39. Çalıca Utku A, Budak G, Karabay O, Güçlü E, Okan HD, Vatan A. **Main symptoms in patients presenting in the COVID-19 period**. *Scott Med J* (2020.0) **65** 127-132. DOI: 10.1177/0036933020949253
40. 40.Sahlou O T and Sahlou T M " Associations between diet history and symptoms in Covid-19 recovered people." Research square, "preprint", 2022.
|
---
title: 'Association between decreased grip strength in preschool children and the
COVID-19 pandemic: an observational study from 2015 to 2021'
authors:
- Atsumu Yuki
- Yumi Tamase
- Mika Nakayama
journal: Journal of Physiological Anthropology
year: 2023
pmcid: PMC10036968
doi: 10.1186/s40101-023-00321-8
license: CC BY 4.0
---
# Association between decreased grip strength in preschool children and the COVID-19 pandemic: an observational study from 2015 to 2021
## Abstract
### Background
Coronavirus disease 2019 (COVID-19) has reduced people’s physical activity. It is essential to accumulate knowledge regarding the influence of COVID-19 on the stimulation of physical fitness and physical functions. Several studies have reported the effects of COVID-19 on physical fitness; however, there are very few reports regarding preschoolers. This study aimed to compare the physical fitness of preschoolers before and during the COVID-19 pandemic to clarify the effects of curtailment of outings implemented to control the pandemic on physical fitness among preschoolers.
### Methods
The subjects were 593 Japanese preschool children enrolled at a kindergarten during 2015–2019 and in 2021 who received a physical fitness test. Children enrolled in 2020 who did not receive a physical fitness test because of the COVID-19 pandemic were excluded. The physical fitness test included grip strength, standing long jump, and a 25-m run. The relationship between physical fitness level and survey year was analyzed using a general linear model, with grip strength and standing long jump as dependent variables, year of study as the independent variable, and sex and age in months as adjusted variables. Kruskal–Wallis test was used to analyze data for the 25-m run. Multiple comparisons were used to compare fitness levels between 2021 (during the COVID-19 pandemic) with levels in previous years.
### Results
Significant relationships were found between survey year and each of grip strength ($p \leq 0.001$), standing long jump ($p \leq 0.05$), and 25-m run ($p \leq 0.001$) among the overall subjects. Grip strength was significantly lower in 2021 compared with the 2016–2019 period. Similarly, sub-stratification analysis by sex showed that grip strength was lower in 2021 than in previous survey years, in both sexes. However, there was no difference in standing long jump or 25-m run times between before and during the pandemic among the overall subjects or according to sex.
### Conclusions
These findings indicate that the COVID-19 pandemic has had a negative effect on the development of muscle strength in preschoolers, and suggest the need to develop strategies that could promote the development of muscle strength of preschool children when limitations are placed on activity during prolonged infectious disease pandemics.
### Supplementary Information
The online version contains supplementary material available at 10.1186/s40101-023-00321-8.
## Background
Coronavirus disease (COVID-19) caused by severe acute respiratory syndrome-associated coronavirus-2 (SARS-CoV-2), of which the first outbreak was in 2019 in Wuhan, China, severely impacted people's wellbeing. Its impacts on physical fitness include a decrease in physical activity resulting from lockdowns and the refraining of the general population from non-essential outings [1], and an associated increase in obesity and decline in physical fitness [2, 3]. Impairments of physical function and fitness following SARS-CoV-2 infection have also been reported [4]. In view of the after-effects of COVID-19, it is essential to accumulate knowledge concerning the stimulation of physical fitness and physical functions.
Several reports have indicated a decline in physical fitness and physical function in school-aged children following the COVID-19 pandemic [5–9]. According to the National Survey on Physical Fitness, Exercise Ability and Exercise Habits in Japan in 2021 [10], the total physical fitness scores of both elementary and junior high school students were lower in 2021 compared to the results of the 2019 survey. Effects of existing risk factors such as decreased exercise time and increased screen time, as well as those of activity limitation due to COVID-19, have been suggested as reasons for the decline in physical fitness. Knowledge of the effects of COVID-19 behavioral restrictions on children’s fitness is accumulating; however, there are very few reports regarding preschoolers [11]. The aim of this study was to compare the physical fitness of preschool children before and during the COVID-19 pandemic to clarify the effects of curtailment of outings implemented to control the pandemic on physical fitness among preschoolers.
## Subjects
The subjects were 708 preschool children enrolled in kindergartens affiliated with the Faculty of Education of the National University in Kochi city between 2015 and 2021. Of these, all children enrolled in 2020 ($$n = 93$$), who did not receive a physical fitness test due to the COVID-19 pandemic, and those with missing variables for the analysis ($$n = 22$$) were excluded from the study. The final number of subjects included in the analysis was 593. The number of participants in each year from 2015 to 2021 (excluding 2020) was 107, 103, 104, 106, 90, and 83 respectively; 301 boys and 292 girls. The median and interquartile range of age in months and distribution of subjects by grade and sex are shown in Table 1.Table 1Distribution of subjects by year of survey2015($$n = 107$$)2016($$n = 103$$)2017($$n = 104$$)2018($$n = 106$$)2019($$n = 90$$)2021($$n = 83$$)χ2pAge, months63, 53–7059, 53–7162, 57–6862, 53–7161, 56–7061, 51–681.110.953Grade, nFirst (age 3 years)252720272026Second (age 4 years)4138463337258.050.624Third (age 5 years)413838463332Sex, nBoys5658524644454.220.518Girls514552604638Age data are presented as the median and interquartile rangep values were obtained using Kruskal–Wallis test or χ-square test
## Physical fitness test
The physical fitness tests included grip strength, standing long jump, and a 25-m run. Grip strength was measured twice each, alternating left and right, using an infant grip dynamometer (T.K.K.5825; Takei Scientific Instruments, Niigata, Japan) and the average value was calculated from the maximum values for each of the left and right sides. The standing long jump was performed using a sheet with lines drawn at 1 cm intervals (KH-164, Kaneya Industry, Osaka, Japan). The distance of the jump was measured from the take-off line to the nearest point of contact during landing, and was performed twice, with the higher recorded distance used for analysis. The 25-m run was measured using a runway set at 30 m and the children were instructed to run through the 30 m line at the highest speed possible. The test was performed once and recorded to the nearest 0.1 s (SVAJ001; Seiko Watch Corporation, Tokyo, Japan).
## Statistical analysis
Statistical analysis was performed using EZR ver. 1.60 (Saitama Medical Center, Jichi Medical University, Saitama, Japan), which is a graphical user interface for R 4.2.1 (The R Foundation for Statistical Computing, Vienna, Austria), and SAS ver. 9.4 (SAS Institute, Inc., Cary, NC, USA). Normality of the data was tested using the Shapiro–Wilk test. Analysis of the relationship between physical fitness and survey year was performed using a general linear model with grip strength and standing long jump as dependent variables, year of survey as an independent variable, and sex and age in months as adjusted variables. Multiple comparisons were made using the Dunnett–Hsu method with the year 2021 as the control group. For analysis of the 25-m run data, which were not normally distributed, the Kruskal–Wallis test was used with the year of the survey as the independent variable. Multiple comparisons were compared between 2021 and earlier years using the Dwass–Steele–Critchlow–Fligner test. Partial η2 was calculated for the effect size of the main effect with general linear model and r for the effect size of multiple comparisons.
## Results
Table 2 shows the relationship between the results of the physical fitness test and survey year. In the overall subjects, there was a significant association between grip strength, standing long jump, 25-m run, and survey year. Multiple comparisons revealed that grip strength was lower in 2021 than in 2016–2019. However, there was no difference in standing long jump and 25-m run in 2021 compared to earlier years. The sex-stratified analysis showed a significant association between grip strength and 25-m run and survey year among the boys. Multiple comparisons revealed that grip strength was lower in 2021 than in 2016–2018. There was no difference in 25-m run in 2021 compared to earlier years. Among the girls, there was a significant association between grip strength, standing long jump, 25-m run, and survey year. Multiple comparisons revealed that grip strength was lower in 2021 than in 2015–2019. The standing long jump was lower in 2021 than in 2016.Table 2Relationship between physical fitness and the COVID-19 pandemicYear of survey201520162017201820192021F or χ2ppartial η2Overall ($$n = 593$$)$$n = 107$$$n = 103$n = 104n = 106n = 90n = 83 Grip strength, kg7.7 ± 0.2(0.13)8.9 ± 0.2***(0.40)8.9 ± 0.2***(0.41)8.6 ± 0.2***(0.34)8.3 ± 0.2***(0.29)7.2 ± 0.212.14 < 0.0010.09 Standing long jump, cm90.8 ± 1.4(0.04)94.0 ± 1.5(0.15)89.5 ± 1.5(< 0.00)87.3 ± 1.5(0.08)92.9 ± 1.6(0.11)89.6 ± 1.62.720.0190.02 25-m run, s6.6, 5.9–7.4(– 0.11)6.8, 6.0–7.5(– 0.03)6.6, 6.1–7.3(– 0.07)6.6, 6.2–7.5(– 0.04)7.2, 6.7–8.1(0.19)6.6, 6.3–7.622.23 < 0.001–Boys ($$n = 301$$)$$n = 56$$$n = 58$n = 52n = 46n = 44n = 45 Grip strength, kg7.8 ± 0.3(0.01)9.1 ± 0.3**(0.32)9.4 ± 0.3***(0.36)9.1 ± 0.3**(0.32)8.6 ± 0.3(0.20)7.8 ± 0.36.15 < 0.0010.09 Standing long jump, cm94.5 ± 2.1(0.01)94.2 ± 2.1(0.01)95.4 ± 2.2(0.03)90.4 ± 2.3(0.13)96.8 ± 2.4(0.08)94.5 ± 2.30.850.5170.01 25-m run, s6.6, 5.8–7.4(– 0.11)6.8, 6.3–7.5(0.04)6.3, 5.9–6.9(– 0.19)6.6, 6.2–7.2(– 0.04)7.0, 6.3–7.9(0.14)6.6, 6.2–7.614.150.015–Girls ($$n = 292$$)$$n = 51$$$n = 45$n = 52n = 60n = 46n = 38 Grip strength, kg7.6 ± 0.3*(0.27)8.6 ± 0.3***(0.50)8.4 ± 0.3***(0.46)8.0 ± 0.2***(0.37)8.0 ± 0.3***(0.39)6.5 ± 0.36.91 < 0.0010.11 Standing long jump, cm87.1 ± 2.0(0.08)94.9 ± 2.1**(0.34)83.3 ± 1.9(0.05)84.3 ± 1.8(0.02)88.9 ± 2.1(0.15)84.8 ± 2.34.40 < 0.0010.07 25-m run, s6.5, 6.0–7.5(– 0.11)6.9, 6.0–7.5(– 0.09)6.9, 6.3–7.7(0.03)6.7, 6.1–7.5(–0.06)7.5, 6.7–8.1(0.23)6.8, 6.3–7.815.390.009–Values are mean ± standard error or median and interquartile rangeValues in parentheses indicate effect size (r)*$p \leq 0.05$; **$p \leq 0.01$; ***$p \leq 0.001$ vs. survey in 2021
## Discussion
The present study investigated whether the COVID-19 pandemic caused a decline in physical fitness among preschool children in Japan by comparing physical fitness tests of preschoolers before and during the COVID-19 pandemic. The results showed that grip strength was lower during the COVID-19 pandemic than before the pandemic.
This study appears to be the first to report a reduction in grip strength in preschool children due to the COVID-19 pandemic. Several studies have evaluated the association between the COVID-19 pandemic and grip strength in children and adolescents. A study of adolescents reported an association between the COVID-19 pandemic and decreased grip strength in male students [3]. However, studies of school-aged children close to the age of the present subjects have reported that the COVID-19 pandemic and related lockdown were not associated with decreased grip strength [5, 8, 9]. The inconsistency between the present results and the previous results of school-aged children may be due to differences in the year in which the fitness tests were performed during the COVID-19 pandemic, rather than to differences in school grades between the studies. The World Health Organization (WHO) declared the COVID-19 outbreak as a global pandemic on 11 March 2020 [12]. These previous studies began to assess grip strength within one year at the latest after the pandemic declaration [5, 8, 9]. In contrast, the present study used fitness test data collected approximately 18 months after the declaration of the pandemic. Consistent with studies that reported a decrease in physical activity in preschool children that was associated with the COVID-19 pandemic [13–16], the present subjects might have been exposed for a relatively longer time to low levels of physical activity. Although the present study did not assess physical activity, some studies have reported that COVID-19 behavioral restriction has increased sedentary behavior in preschool children [15, 17]. Although there are no consistent results on whether sedentary behavior in preschool children inhibits the development of grip strength [18–20], if sedentary behavior caused by COVID-19 behavioral restrictions had increased in the present subjects, this could possibly be the reason for the inhibition of grip strength development. Furthermore, Japan experienced repeated waves of the COVID-19 pandemic during this 18-month period [21] during which restrictions on activities were ordered. In addition, attendance at kindergarten and leaving the home were prohibited for children infected with SARS-CoV-2 or identified as close contacts of a SARS-CoV-2-infected person. These environments might have been risk factors for increased sedentary behavior in preschool children.
Decreased grip strength during the COVID-19 pandemic might have been caused by insufficiency of physical development resulting from measures taken against COVID-19, including behavioral restrictions. However, as height and weight data were available only for third-grade children (aged 5 years), it was difficult to provide sufficient results. Therefore, to examine whether there was any association of stunted growth due to the COVID-19 pandemic with reduced grip strength, we performed additional analysis using the complete data for third-grade children to examine differences in body mass index between the survey years. The results showed no differences in body mass index between the survey years (Supplementary Table S1). This finding indicates that in the specific case of third-grade children, the decline in grip strength in 2021 was not due to impairment of physical development.
Although the trend of lower grip strength during the COVID-19 pandemic than before the pandemic was generally consistent for both sexes, the effect size was greater in girls than in boys. Therefore, the impact of the COVID-19 pandemic on grip strength might have been stronger in girls than in boys. No previous studies have assessed the effects of the COVID-19 pandemic on grip strength in preschool children. A study of US children aged 5–13 years reported that girls were exposed to longer sitting times than boys in the early post-pandemic period [22]. A study of Canadian children reported that fewer girls aged 5–11 years were achieving sufficient levels of physical activity after the pandemic than boys of the same age group [23]. These reports suggest the interesting possibility that undesirable lifestyle habits may have appeared in girls specifically. However, as studies assessing muscle strength and physical activity in preschool children are limited, further research is essential to clarify this issue.
The present study found no association of the COVID-19 pandemic with reduced jump and sprint performance. A previous study that compared fundamental motor skills such as standing long jump and 25-m run before and after the COVID-19 pandemic in 608 Japanese preschool children reported no change in the standing long jump before and after the pandemic, in all age groups. A decline in the 25-m running record after the COVID-19 pandemic was reported only in the 5-year-old group [11]. The present results appear to generally support the findings of this previous report [11]. It is well known that jumping and sprinting ability are affected by coordinated motions as well as by lower limb muscular strength. Coordination of these motions occurs through play in early childhood. In addition, moderate-to-vigorous physical activity (MVPA) is positively associated with the development of jumping and sprinting ability [18, 19]. A study of preschool children aged 3–5 years in 14 countries found no significant change in MVPA in preschool children between before and during the COVID-19 pandemic [17]. In addition, a study of Japanese preschool children reported no change in MVPA on weekends before and during the COVID-19 pandemic [15]. We did not assess physical activity in the present study; however, it is possible that COVID-19 behavioral restrictions might not have reduced MVPA in preschool children. Nevertheless, a minor discrepancy remains, as the results of the 25-m run in the 5-year-old population differ between studies [11]. Among the present female subjects, there were differences between the standing long jump in 2016 and in 2021. Furthermore, several studies of children aged 6 years and older have noted that standing long jump ability was reduced following the COVID-19 pandemic [7, 8]. Further research is necessary to address these conflicting findings.
This study has several limitations. First, the study design did not follow individuals. Therefore, the effect of the COVID-19 pandemic on the development of the physical fitness of preschool children is unknown. Second, as the study was based on one kindergarten in a rural city, the results cannot be generalized. Third, data on physical fitness for 2020 and on physical activity were not obtained. These data might be necessary to further elucidate the association between physical fitness of preschoolers and the COVID-19 pandemic. Furthermore, other background factors affecting physical fitness were not adequately considered in the analysis. Play with physical movement is considered particularly important for the physical development of preschool children. There might have been changes in play behavior, such as a shift from outdoor play to indoor play and fewer opportunities to play with friends.
## Conclusion
This study examined the impact of the COVID-19 pandemic on physical fitness of preschool children by comparing physical fitness data obtained in 2015–2019 with those in 2021. Grip strength was significantly lower in 2021 than in 2016–2019. These findings indicate that the COVID-19 pandemic may have had a negative effect on the development of muscle strength in preschool children, and suggest the need to develop strategies that could promote the development of muscle strength in preschool children when prolonged infectious disease pandemics occur.
## Supplementary Information
Additional file 1: SupplementalTable S1. Relationship between body mass index and the COVID-19pandemic in five-year-old children.
## References
1. Castaneda-Babarro A, Arbillaga-Etxarri A, Gutierrez-Santamaria B, Coca A. **Physical activity change during COVID-19 confinement**. *Int J Environ Res Public Health* (2020.0) **17** 6878. DOI: 10.3390/ijerph17186878
2. Robinson E, Boyland E, Chisholm A, Harrold J, Maloney NG, Marty L. **Obesity, eating behavior and physical activity during COVID-19 lockdown: a study of UK adults**. *Appetite* (2021.0) **156** 104853. DOI: 10.1016/j.appet.2020.104853
3. Tsoukos A, Bogdanis GC. **The effects of a five-month lockdown due to COVID-19 on physical fitness parameters in adolescent students: a comparison between cohorts**. *Int J Environ Res Public Health* (2021.0) **19** 326. DOI: 10.3390/ijerph19010326
4. Rooney S, Webster A, Paul L. **Systematic review of changes and recovery in physical function and fitness after severe acute respiratory syndrome-related coronavirus infection: implications for COVID-19 rehabilitation**. *Phys Ther* (2020.0) **100** 1717-1729. DOI: 10.1093/ptj/pzaa12910.1093/ptj/pzaa129
5. Ito T, Sugiura H, Ito Y, Noritake K, Ochi N. **Effect of the COVID-19 emergency on physical function among school-aged children**. *Int J Environ Res Public Health* (2021.0) **18** 9620. DOI: 10.3390/ijerph18189620
6. Pajek SV. **Impact of the COVID-19 pandemic on the motor development of schoolchildren in rural and urban environments**. *Biomed Res Int* (2022.0) **2022** 8937693. DOI: 10.1155/2022/8937693
7. Pombo A, Luz C, de Sa C, Rodrigues LP, Cordovil R. **Effects of the COVID-19 lockdown on portuguese children's motor competence**. *Children (Basel)* (2021.0) **8** 199. DOI: 10.3390/children8030199
8. Ramos Alvarez O, Arufe Giraldez V, Cantarero Prieto D, Ibanez Garcia A. **Changes in physical fitness, dietary habits and family habits for spanish children during SARS-CoV-2 lockdown**. *Int J Environ Res Public Health* (2021.0) **18** 13293. DOI: 10.3390/ijerph182413293
9. Rua-Alonso M, Rial-Vazquez J, Nine I, Lete-Lasa JR, Clavel I, Giraldez-Garcia MA. **Comparison of physical fitness profiles obtained before and during COVID-19 pandemic in two independent large samples of children and adolescents: DAFIS project**. *Int J Environ Res Public Health* (2022.0) **19** 3963. DOI: 10.3390/ijerph19073963
10. 10.Japan Sports Agency. Results of the 2021 National Physical Fitness and Exercise Ability and Exercise Habits Survey. https://www.mext.go.jp/sports/b_menu/toukei/kodomo/zencyo/1411922_00003.html. Accessed 10 Nov.
11. Abe T, Kitayuguchi J, Fukushima N, Kamada M, Okada S, Ueta K. **Fundamental movement skills in preschoolers before and during the COVID-19 pandemic in Japan: a serial cross-sectional study**. *Environ Health Prev Med* (2022.0) **27** 26. DOI: 10.1265/ehpm.22-00049
12. 12.World Health Organization. WHO Director-General's opening remarks at the media briefing on COVID-19 - 11 March 2020. https://www.who.int/director-general/speeches/detail/who-director-general-s-opening-remarks-at-the-media-briefing-on-covid-19---11-march-2020. Accessed 10 Nov.
13. Clarke J, Kipping R, Chambers S, Willis K, Taylor H, Brophy R. **Impact of COVID-19 restrictions on preschool children's eating, activity and sleep behaviours: a qualitative study**. *BMJ Open* (2021.0) **11** e051497. DOI: 10.1136/bmjopen-2021-051497
14. Hossain MS, Deeba IM, Hasan M, Kariippanon KE, Chong KH, Cross PL. **International study of 24-h movement behaviors of early years (SUNRISE): a pilot study from Bangladesh**. *Pilot Feasibility Stud* (2021.0) **7** 176. DOI: 10.1186/s40814-021-00912-1
15. Hyunshik K, Jiameng M, Sunkyoung L, Ying G. **Change in Japanese children's 24-hour movement guidelines and mental health during the COVID-19 pandemic**. *Sci Rep* (2021.0) **11** 22972. DOI: 10.1038/s41598-021-01803-4
16. Sum KK, Cai S, Law E, Cheon B, Tan G, Loo E. **COVID-19-related life experiences, outdoor play, and long-term adiposity changes among preschool- and school-aged children in singapore 1 year after lockdown**. *JAMA Pediatr* (2022.0) **176** 280-289. DOI: 10.1001/jamapediatrics.2021.5585
17. Okely AD, Kariippanon KE, Guan H, Taylor EK, Suesse T, Cross PL. **Global effect of COVID-19 pandemic on physical activity, sedentary behaviour and sleep among 3- to 5-year-old children: a longitudinal study of 14 countries**. *BMC Public Health* (2021.0) **21** 940. DOI: 10.1186/s12889-021-10852-3
18. Leppanen MH, Nystrom CD, Henriksson P, Pomeroy J, Ruiz JR, Ortega FB. **Physical activity intensity, sedentary behavior, body composition and physical fitness in 4-year-old children: results from the ministop trial**. *Int J Obes (Lond)* (2016.0) **40** 1126-1133. DOI: 10.1038/ijo.2016.54
19. Leppanen MH, Henriksson P, Delisle Nystrom C, Henriksson H, Ortega FB, Pomeroy J. **Longitudinal physical activity, body composition, and physical fitness in preschoolers**. *Med Sci Sports Exerc* (2017.0) **49** 2078-2085. DOI: 10.1249/MSS.0000000000001313
20. Potter M, Spence JC, Boule N, Stearns JA, Carson V. **Behavior tracking and 3-year longitudinal associations between physical activity, screen time, and fitness among young children**. *Pediatr Exerc Sci* (2018.0) **30** 132-141. DOI: 10.1123/pes.2016-0239
21. 21.Ministry of Health, Labour and Welfare, Japan. Visualizing the data: information on COVID-19 infections. https://covid19.mhlw.go.jp/en/. Accessed 3 Mar.
22. Dunton GF, Do B, Wang SD. **Early effects of the COVID-19 pandemic on physical activity and sedentary behavior in children living in the US**. *BMC Public Health* (2020.0) **20** 1351. DOI: 10.1186/s12889-020-09429-3
23. Moore SA, Faulkner G, Rhodes RE, Brussoni M, Chulak-Bozzer T, Ferguson LJ. **Impact of the COVID-19 virus outbreak on movement and play behaviours of Canadian children and youth: a national survey**. *Int J Behav Nutr Phys Act* (2020.0) **17** 85. DOI: 10.1186/s12966-020-00987-8
|
---
title: A mixed-methods study exploring women’s perceptions and recommendations for
a pregnancy app with monitoring tools
authors:
- Natasa Lazarevic
- Carol Pizzuti
- Gillian Rosic
- Céline Bœhm
- Kathryn Williams
- Corinne Caillaud
journal: NPJ Digital Medicine
year: 2023
pmcid: PMC10036977
doi: 10.1038/s41746-023-00792-0
license: CC BY 4.0
---
# A mixed-methods study exploring women’s perceptions and recommendations for a pregnancy app with monitoring tools
## Abstract
Digital health tools such as apps are being increasingly used by women to access pregnancy-related information. Conducted during the COVID-19 pandemic, this study investigated: (i) pregnant women’s current usage of digital health tools to self-monitor and (ii) their interest in theoretical pregnancy app features (a direct patient-to-healthcare-professional communication tool and a body measurement tool). Using a mixed methods approach, 108 pregnant women were surveyed and 15 currently or recently pregnant women were interviewed online. We found that pregnant women used digital health tools to mainly access pregnancy related information and less so to self-monitor. Most participants were interested and enthusiastic about a patient-to-healthcare-professional communication tool. About half of the survey participants ($49\%$) felt comfortable using a body measurement tool to monitor their body parts and $80\%$ of interview participants were interested in using the body measurement to track leg/ankle swelling. Participants also shared additional pregnancy app features that they thought would be beneficial such as a “Digital Wallet” and a desire for a holistic pregnancy app that allowed for more continuous and personalised care. This study highlights the gaps and needs of pregnant women and should inform all stakeholders designing pregnancy digital healthcare. This study offers a unique insight into the needs of pregnant women during a very particular and unique period in human history.
## Introduction
Maternal and neonatal disorders were among the top ten causes of global burden of disease in 20191. Better access to perinatal healthcare would help to reduce preventable morbidity associated with pregnancy2. The increase in access to and use of smartphones presents a unique opportunity to transform and improve how women monitor their own health during pregnancy, especially for those living in remote regions. Furthermore, the COVID-19 pandemic has emphasized the need for more effective digital health interventions, better data collection, and continuity of care3–5.
Research demonstrates that pregnant women frequently search the internet and use digital health tools such as apps to access pregnancy-related information6–8. The usage of pregnancy apps either persisted or increased during the COVID-19 pandemic, partly due to the decreased access to in-person healthcare services9. However, several recent scoping reviews and cross-sectional studies have highlighted the need for better quality apps with greater content credibility8,10–14. Most pregnancy apps are primarily focused on providing educational information, and do not include self-monitoring features - even though there is evidence that these features can assist with behaviour changes that lead to improved healthcare outcomes12,15. For instance, Willcox et al. [ 2017]16 demonstrated that their mHealth digital health intervention, which required users to set goals and self-monitor, had regular user engagement that led to lower gestational weight gain and increased physical activity. Moreover, Dahl et al. [ 2018]17 showed that their Healthy Motivations for Moms-To-Be app that used behaviour change techniques such as goal setting was able to promote healthy eating behaviours.
Cross-sectional studies have suggested that pregnant women are eager for app features that connect them more directly to their chosen healthcare professionals9,18. As has been previously suggested, an app feature that allows healthcare professionals to remotely monitor their patients’ progress has the potential to make healthcare more continuous and accessible at a reduced cost19,20. A recent study tested the feasibility of remote monitoring and asked pregnant participants to monitor their weight, heart rate, blood pressure, physical activity and sleep patterns daily using a smartwatch, a blood pressure machine and a scale alongside attending 4 in-person clinic consults. They found that home monitoring with devices alongside in-person consults was feasible21. Thus, there is an opportunity to develop a digital health intervention that integrates tools that allow self-monitoring of multiple health parameters with remote monitoring by and communication with healthcare professionals.
When combined with other clinical parameters, monitoring body shape changes using anthropometry (e.g., measuring weight) during pregnancy can indicate the risk of developing certain health conditions such as gestational diabetes (GDM), obesity, pre-eclampsia, and the need for a caesarean delivery22–25. These potential risks are not isolated to the expectant mother, as they can result in neonatal premature birth, neonatal mortality, and early childhood obesity26–28. Historically, the most common anthropometric measurement used for perinatal health assessment is pre-pregnancy Body Mass Index (BMI). However, this measure does not capture the distribution and the percentage of a person’s body fat nor their genetic risk of disease. As such, two people with completely different body shapes and sizes can fall into the same BMI category yet have drastically different risk profiles for developing adverse health outcomes29–33.
Monitoring BMI as well as gestational weight gain (GWG) is now the standard approach to assess risk and provide recommendations to pregnant patients. However, other anthropometric and clinical parameters in addition to BMI and GWG are needed to help stratify disease risk as not all individuals are at equal risk of developing certain diseases such as GDM and preeclampsia33,34. For instance, a prediction tool for the early identification of GDM in pregnant women with obesity combined anthropometric measures with other measures such as blood pressure35.
In addition to this, studies reporting the reliability and accuracy by which doctors take anthropometric measurements revealed that while BMI is reliably measured, measures of hip circumference, waist circumference, and waist-to-hip ratio are measured less reliably especially in patients with obesity36,37. Considering this, novel digital approaches to taking body measurements should be explored [1] to allow for more consistent and convenient measurements for patients and [2] to support patients self-monitor these anthropometric measures that are generally difficult and inconvenient to take themselves using a measuring tape.
The overarching aim of this study was to evaluate pregnant women’s interest and willingness to use digital health monitoring tools. To help gauge their attitudes more tangibly, we asked them about a hypothetical app, with a body measurement self-monitoring tool that extracts digital measurements of the body from photos taken by users on their smartphones, and a patient-to-healthcare professional communication tool that allows for direct communication with user’s healthcare professionals. Thus, using a combination of online surveys and interviews, this study investigated: (i) pregnant women’s current usage of digital health tools to self-monitor, and (ii) their interest in the hypothetical pregnancy app.
## Demographic data
The demographic characteristics of the 108 survey participants (101 completed and 7 partial responses) are presented in Table 1. Three quarters of the participants ($76\%$) were in Australia. The majority of the participants were between the ages of 18 and 35 ($71\%$), were in their second or third trimesters ($84\%$), were experiencing their first pregnancy ($55\%$), were in the normal and overweight BMI ranges ($71\%$), completed tertiary education ($71\%$), and reported no health conditions ($65\%$) (See Supplementary Table 1 for more demographic details). The majority of the interview participants were recently pregnant and in their postnatal period ($\frac{10}{15}$, $67\%$), while the rest were pregnant at the time of the interview. Fifty-three percent ($\frac{8}{15}$) reported that it was their first pregnancy. Majority of participants were in Australia ($\frac{10}{15}$) (Supplementary Table 2).Table 1Demographics of the total of 108 survey respondents.n%Continent Australia8275.9 North America87.4 Africa76.5 Europe76.5 Asia43.7Age 18–347671.0 35–503130.0Trimesters First trimester (1–13 weeks)1715.7 Second trimester (14–27 weeks)4844.4 Third trimester (28–42 weeks)4339.8Gravidity None5954.6 One3633.3 More than one1312.0BMI Underweight76.5 Normal/Healthy4743.5 Overweight2926.9 Obesity2523.1Education Tertiary education7771.3 Secondary education2119.4 Vocational qualification98.3 No formal education10.9Reported health condition Yes3835.9 No6965.1The health conditions respondents selected from include: Type 1 and 2 diabetes, hypertension, depression, Polycystic Ovarian Syndrome, non-alcoholic fatty liver disease, obstructive sleep apnoea and other. Ancestry was entered by respondents as free text and grouped into regions.
## Outcomes
Two main themes arose from the interviews: (i) Self-monitoring behaviours using digital health tools, and (ii) interest and recommendations regarding the hypothetical app. The survey and interview results are described within these themes. It must be noted that this study was conducted during the COVID-19 pandemic, and this context is relevant to the study findings (Fig. 1). In fact, the pandemic overall influenced how willing participants were to use digital health. Of the interview participants, $73\%$ ($\frac{11}{15}$) agreed that the pandemic influenced how willing they were to use digital health tools or access information online: “But yes, definitely having COVID as a restriction. Yeah, far more inclined to use a digital tool” [IP 1].Fig. 1Themes and subthemes arising from the interviews. Themes: orange boxes. Subthemes: grey boxes. The relationship between the themes is depicted using a grey arrow. The dashed orange box outlines the context, COVID-19 pandemic.
## Self-monitoring behaviours using digital health tools
Most survey participants ($72\%$) reported that they used pregnancy apps during their pregnancy. As shown in Table 2, the apps were primarily used as a source of information ($65\%$), for education ($42\%$), for self-monitoring ($45\%$), and for reassurance that the pregnancy is going well ($36\%$). Similarly, $67\%$ ($\frac{10}{15}$) of interview participants stated that they used pregnancy apps (See Supplementary Table 3 for a breakdown of which apps were used). While one of the most common features of pregnancy apps is baby development information and baby size-to-fruit-size comparison, only 4 people used it for that purpose. Table 2Survey participant self-monitoring behaviours.n%I feel that I can monitor my pregnancy easily from home Strongly Agree54.6 Agree4743.5 Do not agree or disagree2321.3 Disagree2825.9 Strongly disagree54.6Do you use any activity trackers such as fitbits or smart watches during pregnancy? I use activity trackers4138.0 I do not used activity trackers6762.0For what purposes do you use pregnancy apps? *As a* source of information7064.8 For education4541.7 For self-monitoring4945.4 For reassurance that the pregnancy is going well3936.1 Other87.4 Of those interview participants that didn’t use pregnancy apps, three explained that they felt they did not need to use it because it wasn’t their first pregnancy, and they knew what to expect. All the interview participants that used pregnancy apps emphasized that the app user experience, interface, and customisability were important factors when deciding which apps to use: “I’m very impatient when it comes to user interfaces, so if something doesn’t work easily for me. I’m gone” [Interview Participant (IP) 7]. In this regard, participants mentioned that there was a lack of personalization in apps, especially for symptom tracking. For instance: “…I mean, sometimes it made me less reassured…For example, when I felt so awful, and that carried on into my third trimester and all the apps were telling that I was going to start to feel better…but because it sort of contradicted what the health professionals were saying” [IP 4].
Regarding the utilisation of pregnancy apps to monitor health, we found that while most participants used digital health resources to access pregnancy-related information, only $45\%$ ($\frac{49}{108}$) of survey participants and $30\%$ ($\frac{5}{15}$) of the interview participants reported using an app to monitor their health during pregnancy. Also, when survey participants were asked whether they agreed or disagreed with the statement “I feel that I can monitor my pregnancy easily from home”, only $44\%$ agreed. One of the more common reasons why interview participants reported not using an app to self-monitor their health during pregnancy was because they had been pregnant before and knew what to expect. However, interview participants who tracked a health condition during pregnancy tended to monitor their health more frequently, “…I have chronic hypertension. And I actually just wound up using a spreadsheet to track that” [IP 3]. Among the interview participants who used an app to monitor their health during pregnancy, four stated that the app chosen was not designed for pregnancy because they could not find pregnancy specific versions. These included Kegel exercise, diet, and calorie counting apps, “…All of the like health tracking…none of it has support for pregnancy, which is very annoying” [IP 3].
However, apart from using digital health tools, some women had reported that they have their own approaches to self-monitoring, such as note taking, their personal memory, photos, and excel spreadsheets. Participants used such self-monitoring methods to track health data such as, their blood pressure, blood glucose insulin, calorie intake, medications, physical activity, and symptoms. Another method of recording health data included activity trackers, which only $38\%$ of survey participants reported using during pregnancy. Three quarters of those participants reported that they achieved their exercise goals more than once weekly using their activity trackers (Table 2).
## Self-monitoring behaviours using digital health tools – monitoring and data security and privacy concerns
The interview participants that used apps to self-monitor weight or diet said that they tracked their weight only when prompted by a notification to record their weight. One remarked that when weight was tracked, no feedback was provided, “And I did track my weight in it. But I mean, that was really just a place to put it, it didn’t really provide any feedback or anything” [IP 7]. Notably, only $4\%$ ($\frac{4}{108}$) of survey participants reported using any app to monitor their diet during pregnancy. Another factor affecting willingness to use digital health tools was not related to feeling that they could monitor themselves from home, but concerns about data privacy and security. $63\%$ of the survey participants reported having concerns about data privacy and security in relation to using the pregnancy apps. Moreover, the multiple logistic regression of survey responses found that, holding all other predictor variables constant, having “no concerns about data privacy and security issues in pregnancy apps” ($p \leq 0.0001$) was a significant predictor for the likelihood that survey participants used pregnancy apps ($95\%$ CI 2.81,7.58: odds ratio = 127.84). BMI, age, gravidity, “feel that they can monitor themselves from home”, “use digital health more now that they are pregnant” and “reported a health condition” were non-significant predictors (Table 3). Some interview participants mentioned that they had surpassed their privacy concerns during the pandemic because of the unorthodox situations they had to face and because of their need for care and remote monitoring by healthcare professionals: “I worked with a lactation consultant when I got home from the hospital. And we had to do virtual appointments because of the pandemic. And that meant that I literally texted her on WhatsApp, like pictures of naked pictures of my boobs, which, you know, is not something that I would have opted to do if there hadn’t been a pandemic and I could have seen her in person…But because of the pandemic, I was like, “Great, I hope this doesn’t get hacked and leaked somewhere”” [IP15].Table 3Results of statistical analyses using logistic regression to determine likelihood of pregnancy app usage. Predictor variablesP valueEstimateOdds ratio$95\%$ Confidence IntervalLikelihood of pregnancy app usage (response variable) Pseudo R2 (0.482) and β (−3.631) Concerns about data privacy and security issues [No]<0.00014.85127.842.81, 7.58 Concerns about data privacy and security issues [Yes]0.1431.584.86−0.40, 3.97 Use digital health more now that they are pregnant [Yes]0.1301.243.46−0.40, 2.89 Health condition [yes]0.1391.343.80−0.32, 3.30 Feel that they can monitor themselves from home [Strongly Agree]0.370−1.550.21−5.00, 2.06 Feel that they can monitor themselves from home [Agree]0.870−0.160.85−2.16, 1.76 Feel that they can monitor themselves from home [Disagree]0.4830.8662.37−1.52, 3.40 Feel that they can monitor themselves from home [Strongly Disagree]0.4901.223.38−2.00, 5.29 BMI [Obesity]0.7790.261.30−1.54, 2.20 BMI [Overweight]0.3181.052.86−0.89, 3.32 BMI [Underweight]0.369−1.280.28−4.29, 1.42 Age [35–50]0.2510.972.64−0.61, 2.79 Gravidity0.549−0.300.74−1.21, 0.80Odds ratios and Confidence Intervals are reported for each test.
## Interest and recommendations regarding the hypothetical app - patient-to-healthcare professional communication tool
All survey and interview participants were asked how comfortable they would be to use a patient-to-healthcare professional communication tool (Fig. 2). The vast majority of survey participants ($83\%$) expressed that they would be comfortable sharing the health results generated from the theoretical app with their healthcare professional (e.g., their clinician, midwife or obstetrician) through a secure network. Complementing the survey findings, all the interview participants ($100\%$, $\frac{15}{15}$) shared that they would be happy to communicate via an app with their chosen healthcare professional/s: “Oh, of course, I’d love that…It would make it a lot more accessible. But as long as I know that she [their healthcare professional] is the one whom I’m going to be seeing even in my next appointment…*That is* valuable, you know, that sort of established care, continuous care with one person like that one point of contact.” [ IP 11]. In addition to this, participants wished not only to communicate with their healthcare professionals, but also share relevant health data. Participants mentioned that they would communicate with their healthcare professional via an app because communication via the phone or email was not possible or feasible, “…Because sometimes you just have the simple question, and, you know, having to wait for an appointment and go through everything is just harder.” [ IP 2]. The form of communication within an app that participants said they would prefer were communicating via chat, voice message, or email. For example, a participant shared their experiences communicating with their strength coach via voice message: “Just being able to send her a voice message or a thing on the other communication app we used and then she’d get back to me when it was convenient for her without, you know, having for me to wake up at a certain time…I guess asynchronous communication. And I found that quite helpful” [IP 9]. In line with the participants’ willingness for asynchronous communication, a common app feature participants mentioned they would find helpful was sending the questions they plan to ask their healthcare professional before their upcoming appointment. Fig. 2Participant thoughts about 2 theoretical app features. The figure summarises both the survey and interview participant responses when asked about a tool to communicate with their healthcare professionals and the body measurement digital health tool.
Another point participants raised was how digital health could bridge the communication not only between patients and health professionals, but also between specialists themselves. Several participants reported the difficulties they encountered during pregnancy when they needed to communicate with multiple healthcare professionals or specialists. One participant explained, for instance, how difficult it was to remember what their obstetrician wanted them to ask their endocrinologist and suggested how the communication between specialists could be bridged by digital health, “…My obstetrician is a bit old school, but he would often use a voice recorder and record his notes…Maybe something that could just capture a couple of points and then I could play that back to the endocrinologist or vice versa…I think that would be really useful.” [ IP 8].
## Interest and recommendations regarding the hypothetical app - body measurement tool
All participants were also asked how comfortable they would be to use the body measurement tool described in the study design section of the methods (Fig. 2).
About half of the survey participants were willing to take photos of their entire body ($43\%$) and/or body parts ($49\%$) so the app could automatically measure their body changes over time. The majority of people who were willing to take photos of their body parts were also willing to take photos of their entire body ($\frac{43}{52}$). In line with the survey results, $33\%$ ($\frac{5}{15}$) of interview participants immediately agreed that they would be comfortable taking photos of themselves and using a body measurement tool, as it would help them monitor their health, quantify changes, and store a collection of pregnancy photos. Most of the interview participants ($\frac{12}{15}$) were interested in using the body measurement tool to track leg/ankle swelling: “My feet, they started swelling. So, I was taking pictures of them to compare like, okay, is it swollen from last week or this week, but that was towards the end. So, I would use that feature, like to track the swelling on my feet” [IP 2].
Notably, a larger proportion of interview participants ($47\%$, $\frac{7}{15}$) expressed that they would use a body measurement tool to measure specific body parts only. These participants were hesitant about taking photos that would include their faces due to privacy and body image concerns: “Because you mentioned ankles, I think I would be totally okay, taking pictures of my ankles like that…But in my mind, I was thinking about bumps or my face or something that feels a bit different somehow…faces are obviously identifiable, and then bumps are not identifiable in the same way, but it feels a bit more personal than taking a photo of my ankles or fingers or something else. So yeah, I think that would definitely play into it. What I was taking the photo of for sure” [IP 10]. The privacy concerns were not isolated to the inclusion of their faces in photos, but to concerns about, “Who keeps that information?” [ IP 1]. When the interviewer explained that only the digital measurements extracted from their photos would be stored and used to train a machine learning system, and that they would choose who their data was shared with (family/friends/healthcare professional), participants reported that their concerns were alleviated. Similarly, $68\%$ of survey participants responded “yes” to being comfortable with the use of their anonymised data [digital measurements extracted from photos] to train and develop the app technology while using the app ($21\%$ = I am not sure and $11\%$ = No). In addition, survey participants were asked how interested they were to use an app that learns from their anonymous data (i.e., machine learning) to assist in the identification, prediction, and prevention of adverse health outcomes during pregnancy and there was moderate-high interest ($76\%$) in using such an app. The interview participants that were not comfortable using a digital body measurement tool ($20\%$, $\frac{3}{15}$) had both privacy and body image concerns, “I think that would make me very uncomfortable…And I’d be worried about my privacy too” [IP 4]. In fact, when survey participants who responded “no” ($\frac{28}{106}$) to the question of being comfortable taking photos of their body parts regularly to extract digital measurements were asked to elaborate why, their explanations included discomfort/uneasiness, privacy/security concerns, body image/mental health issues, and credibility. In line with participants’ concerns around body image issues, one participant with a nutritionist background voiced their concerns about the lack of diversity of current anthropometric standards/guidelines, “*What is* typical? How do you come up with references or standards, when there’s this level of heterogeneity in body shapes and body types? And just in general, who stores fat where, you know, so for all you know its fluid accumulation, it’s fat, or they have goiter?” [ IP 11] and, consequently, on the potential unindented yet negative consequences of the use of monitoring tools on pregnant women’s mental health.
## Interest and recommendations regarding the hypothetical app - suggested educational information
Interview participants were given the opportunity to make additional comments regarding the use of digital health tools to monitor pregnancy. In response, participants shared which educational information (Table 4) and additional features (Table 5) they would include in an app or digital health tool. The following are a selection of the issues they experienced during pregnancy and some of their ideas for potential solutions. All the 15 interview participants mentioned that the educational information provided in apps or digital health tools could be improved. Some participants noted that more support and information should be provided about the initial stages of pregnancy, before their first pregnancy confirmation appointment at 8–9 weeks, and before their 20-week morphology scan. One participant explained the anxiety that they felt before their first pregnancy confirmation appointment, and how other users in a pregnancy app they used (Peanut app) felt the same need of support and reassurance. Four participants added how challenging it was to access guidelines that were pregnancy specific and that there is a need for credible information to help interpret health results. Aside from access to more pregnancy specific guidelines, there was a desire to learn more about how their bodies change during pregnancy outside of how their baby is developing. Table 4Educational information for a pregnancy app or other digital health solution. What types of educational information do interview participants want?Participant quotesSupport and information about initial pregnancy stages before the first doctor’s appointment (at 8-9 weeks) and before the 20-week morphology scan“A lot of us on the app [Peanut], were very worried, like, you know, who had confirmed pregnancies only based on the home kit, but who had to wait till the eighth or the ninth week… And it was just such a real experience for me because, you know, you don’t know what it is like till the time you actually go for your first ultrasound…So I think some sort of support in the very initial phases that will be useful to incorporate in a pregnancy app. ( nervous tone throughout)” [IP 11].Not only information about fetal development (and fruit comparison), but information about how the female body changes during pregnancy“…And I think what would have been great was to understand how my body’s changing, how my hips are spreading apart, the strains putting on my muscles, and then things I can do to help with that, to help manage that pain…Even then having like a digital tool to help you understand like, you know, there’s loads of things that help you visualize how your baby’s developing, but there’s less about how your body’s changing, and then what you can do to help you understand and help you manage that would be super helpful” [IP 13].Pregnancy specific guidelines within the app to help interpret tests such as blood tests and risk factors (related to pre-pregnancy BMI)“I think there’s still a space for a really good pregnancy tracking app that shows, you know, all of the common things that they’re looking for you. You take millions of blood tests, and they give you back like all of these, like, random numbers that you spend a bajillion years trying to figure out what they mean…My chart [the app] is the thing that most people use. They’re all calibrated for non-pregnant people. And so, the app will come back and tell you that you have, you know, you have an elevated white cell blood count. And it’s like, who cares all pregnant women have an elevated white blood cell count, which is only after a bunch of googling…And so there’s this situation where the medical system is giving you this data, but there’s nothing useful to interpret it at all (frustrated tone). And the tools that they give you for interpretation often aren’t geared towards pregnancy. And so having an app that would show me how things were progressing on all of these things that they were tracking…It would have been really, really nice” [IP 3].Breast feeding - information about diverse experiences and the top 10 common complaints or problems postpartum“…honestly, for me, the biggest information gap came with breastfeeding. Um, and I definitely, I felt like a complete absence of support, especially from health care providers in that regard. And so, if there had been resources related to breastfeeding, that were a little bit beyond just sort of, like, you know, your baby should latch. I don’t know, it was just terrible” [Participant IP 15].The table summarises the points interview participants raised regarding the educational information that they would like to be included in a pregnancy app or other digital health solution. The Table includes selected interview participant quotes related to some of the points raised. Table 5Additional features for a pregnancy app or other digital health solutions. What types of features did participants want an app or digital tool to include?Participant quotesExport/share tool of app data and educational information“Yes, but the user interface would have to be very sensitive. So, the app that I have doesn’t allow screenshots at all for privacy reasons. And that was frustrating because sometimes I was talking to my mom or my friend who had preeclampsia. And I was like, here are the numbers they got for me today on my protein test. And it would be really weird to have a button under my urine protein test that says share with a friend. And yet, that’s exactly the feature I wanted. So, you know, something that said, like export might have been a little more appropriate” [IP 3].Connect users to research publication database“I found the app that I used, like they had a bit of information, but it was super basic and general, which of course, that’s what you would expect in the app. But if they actually had links to solid research, and a database where you could go and look at that stuff. I think that would be really good” [IP7].Shared user interface with healthcare professionals“So particularly with the preeclampsia, I had to report the data back to my doctor. And so, it was a little kludgie, the medical apps, they just have like this text box for you to fill stuff in…And I would come in, and I would have my own graphs…And so, I would have my own graphs. And I would, like, bring them printed out so that we could talk about them. And it would have been so nice if there was some sort of like shared interface that we could go over the data together with and be like, here’s, you know, here’s what we’re seeing” [IP 3].'Digital Wallet’ that includes digital copies of resources provided by hospitals, pregnancy antenatal cards, receipts, scripts etc. “I would much prefer to have gotten everything digitally…Even scripts…I just never really understood why I had to keep it with me, why couldn’t they just email it to each other, it’s just like, on my file. So that kind of thing I felt was really weird. And, you know, obviously very old systems” [IP 13].“I do think that having a digital wallet would have been literally life changing. That would be amazing, something secure, that you could store all of that documentation in” [IP 7].Symptom tracker - could work like a contraction/ kick counter“I just really would have loved to have something to monitor how much I was being sick… I don’t know even if you just like you know, push a button every time you were sick, and then it could calculate over a period of time how often it would be happening…Well, the app, I had actually had a kick count. So, like, you could press it every time you felt kicks and then it would track that…So yeah, like something similar to that for the vomiting would have probably really helped me because I’d go into the hospital, and they would be like”, “How many times have you been sick in the last 24 h?” I’d be like, “I don’t know. 1000? Like, I fill buckets? I don’t know” [IP 5].Gestational diabetes monitoring“Maybe something about the gestational diabetes, because I think that’s quite common. You know, a lot of women, you know, particularly older women sort of have that issue…I think would have been really useful” [IP 8].A holistic app (from antenatal to postpartum)“…There’s just so many different apps for so many different things, it would be good to have something that kind of combines everything into one, especially now that I’m looking towards, like downloading apps for sleep for the baby to track their sleep, there’s breastfeeding apps to track how much milk they’re drinking…So, like I’m just finding that my phone is getting filled with all these different apps. And you have to keep inputting your information into every app, you know, if it’s about me, it’s my age, my height, my weight, how far along I’m in my pregnancy and things like that. And if it’s about the baby, you have to input all that information. So, if there was some sort of app that could follow you from pre pregnancy right through to baby being born, I think that would be quite beneficial” [IP 12].The table summarises the points interview participants raised regarding the app features that they would like to be included in a pregnancy app or other digital health solution. The Table includes selected interview participant quotes related to some of the features raised.
Despite these limitations, many participants felt that the support and information they received during pregnancy was still satisfactory. They reported though that postpartum support was inadequate - such as access to more diverse breast-feeding resources - and that better communication about potential postpartum complications should be encouraged.
## Interest and recommendations regarding the hypothetical app - additional app features
With regard to possible app features, many participants stated that they would find a “Digital Wallet” that stores copies of their medical data (such as resources provided by hospitals, pregnancy antenatal cards, receipts, scripts, and referrals) extremely helpful. One of the major factors that annoyed participants was that they had to carry physical scripts or referrals to appointments. Another additional feature many participants mentioned was a tool to track their symptoms during pregnancy. A participant who experienced severe morning sickness or hyperemesis gravidarum during pregnancy mentioned how difficult it was to track the number of times they vomited. In particular, two participants expressed that there is a current lack of support for pregnant women who are diagnosed with gestational diabetes and suggested that a gestational diabetes tracking tool should be developed.
Several participants suggested that they would find a personalised list of appointments and appointment reminders useful because their busy schedule (due to their work and childcare responsibilities) makes it difficult to track their health appointments, as well as information about what to expect during those appointments.
In terms of data sharing, participants indicated that they would not only like to share their data with their healthcare professionals, but with their family and friends using an export or share tool.
To summarize, all interview participants besides one felt that there was room for an app that provided personalised support, tracked pregnancy at all stages from preconception to postpartum, and provided integrated and continuous care, in other words a more ‘holistic’ pregnancy app.
## Discussion
This study sought to investigate (i) pregnant women’s current usage of digital health tools to self-monitor, and (ii) their interest regarding two theoretical pregnancy app features. Using our mixed methods study design, we found that the majority of participants already used pregnancy apps and other digital health tools, but most did not use them to self-monitor. We found that participants primarily used pregnancy apps to access pregnancy-related information and receive updates about their baby’s development – which is consistent with previous cross-sectional studies18,38 – but many also expressed their desire for apps to also provide information about the changes occurring to their own bodies.
The majority of participants were interested and enthusiastic about a patient-to-healthcare-professional communication tool. While less than half of survey participants ($43\%$) were comfortable taking photos of their bodies for the app, most of the interview participants ($80\%$) were interested in using the body measurement tool to track leg/ankle swelling. Additionally, participants consistently raised the need for additional educational information and app/digital tool features that allow for more personalised and holistic care. These findings contribute to the growing literature on the needs and preferences of pregnant women during the COVID-19 pandemic9,39.
However, what are the barriers to using digital health to self-monitor? Based on both the survey responses and interviews, the largest barrier for using digital health to self-monitor was that the available tools did not meet the consumer demands. This is illustrated by our finding that only $51\%$ of survey participants felt that they could monitor themselves from home and that interview participants often relied on other approaches to self-monitor such as using their memory or an excel spreadsheet.
In contrast to previous findings9,18,40, a significant barrier to using digital health found in this study was security and privacy concerns as well as concerns about information credibility and quality. This shift in concerns is likely due to: [1] the increasing spread of misinformation online, which participants mentioned as being the driver for their preference for their healthcare professional to be their main point of contact, and [2] poor data privacy, sharing, and security standards in pregnancy apps being widespread41. Thus, the quality of information and data privacy of pregnancy apps should be more formally assessed. Perhaps apps could be reviewed by a panel of experts and scored for information quality and privacy before they are uploaded to app stores, or a central app rating platform could be created42.
Based on our study findings, the features pregnant women wanted to include in an ideal app can be outlined. When we asked about the theoretical app features of a patient-healthcare provider communication tool and a digital body measurement tool, interview participants were enthusiastic to share their views and even shared their thoughts about what an ideal pregnancy app should include. As a cohort, participants outlined that an ideal pregnancy app should: [1] provide holistic care for preconception, prenatal and postnatal support, [2] include credible information developed by experts/clinicians, [3] ensure data privacy and security, [4] include a patient-healthcare provider communication tool, [5] include a “Digital Wallet” with their patient data, [6] include a body measurement tool that measures body regions such as the ankle/foot region, [7] include monitoring tools for other health parameters such as diet, physical activity and mental health, and [8] include behaviour change techniques, such as reminders, goal setting, and providing personalised feedback on progress towards goals. A user-interface mock-up of these app requirements is depicted in Fig. 3.Fig. 3User-interface mock-up of pregnancy app features. The figure outlines which app features (grey boxes) pregnant women outlined that they would like an ideal pregnancy app to include.
Several pregnancy app reviews have shown that most commercial pregnancy apps include low numbers of behaviour change techniques especially for providing personalised feedback, goal setting and planning11,43,44. Additionally, pregnant women mentioned that a pregnancy app should have a well-designed user-experience and user-interface, though as demonstrated by numerous pregnancy app reviews the majority of currently available commercial pregnancy apps perform highly on usability10,45–48.
However, there are several benefits and barriers that would need to be considered when implementing the ideal app. The first benefit is that the body measurement tool within this ideal app could allow for consistent and convenient monitoring of body shape changes over time and allow for the collection of more diverse and complete data about body changes during pregnancy. With sufficient data, machine learning could help mitigate and stratify risk (i.e., your ankle swelling has not decreased for x days, please contact your healthcare provider). It is also promising that most participants were open to their anonymised data being used to train the machine learning system of a theoretical app. And that patient image-based assessment is becoming more widely accepted, especially for wound-care management49. However, an associated barrier of the body measurement tool and the other monitoring tools mentioned (such as for diet and physical activity tracking) is that with the introduction of any new health assessment measures or digital screening tools, there are concerns regarding overinterpretation of results and inappropriate linkage to disease risk. Results and feedback presented to users could also induce anxiety or lead to unnecessary clinical consults. As described by Capurro et al.50, digital screening tools can lead to overdiagnosis and potential harm to patients, and for this reason should be validated. Another factor to consider is the privacy of machine learning models and how that can be minimised by using anonymisation and data minimization tools51. However, the need to tailor body measures for ethnicity and cultural appropriateness is a challenge, which is evident as there are still no concrete and accepted BMI guidelines for different ethnicities. Thus, the clinical utility of health monitoring tools such as the described body measurement tool should be assessed by healthcare professionals and its accuracy should be validated in an iterative process.
The second benefit is that the patient-to-healthcare-professional communication tool embedded in the ideal app could be integrated within current healthcare systems. Integration of such tools has the potential to allow for asynchronous communication, remote monitoring, and continuous care. For instance, a mobile-based telehealth service in rural Bangladesh that allowed for remote consultation for maternal, neonatal, and infant health or emergencies was able to provide users with advice, preliminary diagnosis, reassurance, referrals, and scripts and promote healthy behaviours, such as regular healthcare consults19. A barrier to the implementation of such digital support apps is the required integration within current healthcare systems and workflows, which is notoriously challenging. However, studies have suggested that embedding technologies within existing systems is the preference20.
Lastly, all the participants emphasized their preference for a tool that enables communication with their chosen healthcare professionals. This is consistent with other studies that have demonstrated patients desire for more continuous care from their healthcare professionals during pregnancy9,18. However, for such interventions to be successful, the involvement of healthcare professionals is a must. A mixed methods study of healthcare providers during COVID-19 revealed that when telehealth consults replaced in-person care, the quality of care was impacted52. Other challenges mentioned in this study were the lack of digital literacy, patient monitoring, and disruption of patient-healthcare professional bonding. Thus, digital communication should not disrupt any in-person care and should be used in conjunction39. The type of communication, feedback, and time commitment provided to patients by healthcare professionals should also be carefully considered and accounted for in staffing profiles.
Thus, digital health interventions such as the described hypothetical pregnancy app in this study could enhance care in rural settings and/or support the flexibility of healthcare delivery especially during pandemics, where face-to-face contact needs to be minimised. Such a digital health service could employ a range of features that drive user engagement, encourage self-monitoring, communication with healthcare professionals, and ultimately lead to behaviour change.
This study has several limitations. Firstly, though the survey was tested for face and construct validity between authors and several external researchers, it was not pilot tested with pregnant women before it was launched. As outlined by several studies, it is recommended to test surveys for their validity and relevance with a pool of intended respondents53,54. However, as raised by Goodyear-Smith et al. [ 2015]55 and experienced by the authors, there are some challenges with pre-defining participatory and co-design approaches to involve research participants for human research ethics committee approval as the process can be iterative and unpredictable.
Secondly, we found that interview participants had several follow-up questions when deciding how willing they were to use the theoretical app features. Perhaps the format of using surveys to assess willingness to use theoretical features may not have been as appropriate of a method as the interviews. Participants may not have fully been able to understand the theoretical features from the short descriptions about them provided in the survey, and this may have impacted their responses. For instance, a relatively large proportion of survey participants responded, “I don’t know” when asked about their willingness to use the body measurement tool to take photos of their entire body ($22\%$) and body parts ($26\%$). Additionally, interviews can be used to not only assess the needs of users but using human-centred design approaches, prototypes can be developed with users56. After such a design process, future studies may consider displaying digital prototypes of theoretical app features created with users alongside questions related to it to provide survey responders more context.
Thirdly, a larger proportion of both survey and interview participants were highly educated (tertiary education: $71\%$ and $53\%$), which could explain their hyperawareness about app privacy issues. This elicits the question of how transferrable the study findings are to more diverse groups. In addition, due to the random sampling method used for recruitment, representation across all demographics could not be achieved (such as for country location, ethnicity, BMI category, and socioeconomic status). This highlights the importance of designing accessible and culturally appropriate digital health solutions, which are reliant on the involvement of more diverse groups or several co-design iterations with different subgroups. Innovators should consider abiding by frameworks that promote digital health equity during development to prevent disparities in access to these tools57,58.
Overall, our findings demonstrate that pregnant women feel that there is a gap for a better pregnancy app or tool that allows for more holistic care from the prenatal to postnatal period and that could be integrated within healthcare systems. Their enthusiasm for a patient-to-healthcare-professional communication tool and interest in using the body measurement tool to track leg/ankle swelling illustrates that pregnant women are willing to use self-monitoring tools as long as they are accompanied by remote monitoring or connection with their healthcare professionals. Based on the findings of this study, researchers, innovators, and developers seeking to improve digital health services for pregnant women should consider incorporating the app features raised by pregnant women. We examined self-monitoring behaviours and recommendations for a hypothetical pregnancy apps during a very particular and unique period in human history. The question remains whether attitudes expressed here will continue in post-pandemic times. However, it is clear that such digital strategies are likely to improve the flexibility of healthcare systems to respond to such unpredictable events into the future.
## Study design
A convergent mixed method approach was used. Quantitative and qualitative data were collected concurrently and analysed separately through surveys and interviews. The participants could complete both the survey and interview, but their data were not linked. The study findings and interpretations were triangulated from the combined data. At the end of both the surveys and interviews, participants were asked to share their thoughts about two theoretical app features: [1] a digital tool that would allow them to communicate with their chosen healthcare professionals, and [2] a body measurement tool, that extracts digital measurements of their body from photos taken on their smartphone. Specifically, participants were asked if they would be comfortable taking photos of their entire body or body parts regularly in an app that monitors how their body is changing and would give feedback about those changes. If participants wanted further elaboration, examples were provided of how the measurement tool could be used (such as, using the tool to monitor whether swelling in the ankle region was increasing or decreasing). Participants were then asked if they would feel comfortable communicating with and sharing health information with their chosen healthcare professionals.
## Ethics approval
Ethics approval was obtained from Nepean Blue Mountains Local Health District Human Research Ethics Committee, Australia (Ethics approval number: ETH00580). Online consent was obtained from all participants.
## Recruitment
Participants were eligible to complete the survey only if they were [1] currently pregnant, [2] able to provide consent, and [3] English literate. Participants were eligible for the interviews only if they met the criteria described above and if they were recently pregnant in the last 12 months. Eligibility was assessed for interview and survey participants upon completion of an online form. Participants were recruited both in person and online. In person recruitment was conducted at the Nepean Hospital Antenatal Clinic, Australia. Online recruitment was conducted via social media posts and advertisements as well as via email newsletter advertisements. All participants had the option to provide their contact details at the end of the survey if they wished to express their interest in participating in the interviews. Survey recruitment occurred during November 2020–May 2022 and the interviews were completed between July 2021–March 2022.
## Survey
The online survey questions were modified from existing validated questionnaires or were newly created for this study by the authors. Questions were selected to address the study aims. Questions were modified from questionnaires designed to: [1] assess at risk pregnant women on a national level (PRIMS – Pregnancy Risk Assessment Monitoring System)59, [2] investigate the use of pregnancy apps18, and [3] assess body perception attitudes60,61. The authors assessed and tested the survey for face and content validity to ensure the questions captured the study aims. Several external researchers were also consulted. There were several rounds of feedback before the survey was finalised. The self-administered survey contained 49 questions and took approximately 20–30 min to complete. The majority of the survey questions were closed-ended ($84\%$, $\frac{41}{49}$) including multiple choice, dropdown, binary, ranking, and Likert questions. Questions were related to respondents’ health and health experiences during pregnancy, their health monitoring behaviour, their attitudes and usage of digital health tools, and thoughts about features included in the proposed app. At the end of the survey, respondents were asked two open-ended questions related to how the COVID-19 pandemic impacted their pregnancy and their use of digital health. Refer to Supplementary Method 1 for the survey in its entirety.
## Interviews
Participants who expressed their interest to participate in an interview at the end of the survey or who signed up for an interview via the advertising link, were asked to complete an online form assessing their eligibility and to provide their consent online. They were then contacted and instructed to schedule an interview via Calendly at a suitable time for them. Seventy percent ($\frac{15}{22}$) of the participants who provided their consent, scheduled an interview via Calendly. All interviews took place online, via Zoom video communications software, and were conducted by the same interviewer (NL) to ensure consistency. The interviewer was a female PhD student trained to conduct the interviews by a qualitative methodologist. Interviews lasted between 30–70 min (mean = 45 min, range = 33–70 min) and were semi-structured to allow for open discussion and elaboration of particular responses.
To ensure the quality of the online interviews, the following measures were taken: (i) all online interviews were conducted with video and audio, to capture both verbal and non-verbal cues; (ii) to build initial rapport, the interviewer would introduce themselves, the study, why it is being conducted, outline what to expect and address any questions/concerns; (iii) the interviewer reminded the participant that there are no right or wrong answers and that they may also refuse to answer any questions that they did not wish to during the interview; (iv) the interviewer also started the interview with a general question such as, “*Is this* a good time for you to talk?” or “Could you tell me a little about your pregnancy?”; ( v) follow-up questions were asked when contextual information would add value to the conversation but was missing, unclear, not detailed. The interviewer also took extra care in reading non-verbal cues and included them in the interview transcripts in brackets when assessing them for accuracy against the video and audio recording. Additionally, the interviewer was aware of the circumstances and the subject of the interviews, and for this reason referred to the principles of trauma-informed research when interacting with the participants62.
Interview questions were created by the authors or modified from questions designed to assess weight related attitudes22. Qualitative interview topics were identified by reviewing the overall aims of the study and the survey questions. The interview was designed to address the study aims in a more in-depth manner: to investigate how digital health usage, self-monitoring behaviours and body image/weight may influence participants willingness to use digital health and their thoughts about the theoretical app features. The interview guide was organised in the following manner: [1] health monitoring and digital health usage; [2] weight and body image during pregnancy and weighing and photo taking behaviours; [3] theoretical app features; and [4] digital health usage during pregnancy in general. No theoretical framework was used to develop the interview guide. Refer to Supplementary Method 2 for the interview guide in its entirety.
## Data analyses
Survey respondents who completed $75\%$ or more of the survey questions were included in the analysis. Multiple logistic regression was used to assess which variable predicted the use of pregnancy apps by survey participants. More specifically, the relationship between (i) BMI, (ii) age, (iii) gravidity, (iv) their usage of digital health more during pregnancy, (v) their belief that they can monitor themselves from home, (vi) any reported health condition and (vii) if they had concerns about data privacy and security issues in pregnancy apps were the predictor variables, and the ‘use of pregnancy apps’ was the response variable. Predictor variables were selected based on their theoretical and conceptual relevance. Only full survey responses were included in the logistic regression analysis. These results are reported as $95\%$ confidence intervals and adjusted odds ratios. Open-source code published by “StatQuest with Josh Starmer”63 was used to run the logistic regressions. All statistical analyses were completed in R Studio version 4.2.0 (see Supplementary Method 3).
Participants reported relevant demographic information during interviews (whether they were currently pregnant, their location and gravidity). And participants education level and health status were recorded if participants chose to mention them during interviews. Interviews were recorded on Zoom and a digital voice recorder as a backup and the audio was transcribed automatically using Otter.ai software. The transcripts were proofread and imported into NVivo 12 for coding and analysis. The coding structure was first defined via an iterative process. Three authors (CP, CC, and NL) coded one initial interview and discussed the coding structure. Two authors (CP and NL) then coded two different interviews and finalised the coding structure. Coding between authors of these interviews was found to be highly consistent. One author coded all 15 interviews (NL) while another author coded 9 interviews (CP). Fig. 1 illustrates the major themes and subthemes that arose from the interviews. Once the interviews were coded, a coding comparison query was run in NVivo to numerically assess percentage agreement between coders (percentage of interview transcripts that should be coded to a specific node or case) and was found to be ≥$93\%$. This confirmed that the coding was consistent. Common themes and subthemes from qualitative data in both the surveys and interviews were identified using a thematic analysis as described by Braun and Clarke64. The minimum sample size for interviews was determined to be 15 based on previous studies65,66. After interviewing, coding, and analysing 15 interviews, the research team (CP, NL, and CC) determined that data saturation was achieved. No new themes could be attained, therefore achieving inductive thematic saturation and data saturation as outlined by Saunders et al. [ 2018]67. All authors then discussed and confirmed the themes and subthemes based on the coding structure and study aims. All Figures were created using a Canva Pro Premium account.
## Reporting summary
Further information on research design is available in the Nature Research Reporting Summary linked to this article.
## Supplementary information
Supplement information REPORTING SUMMARY The online version contains supplementary material available at 10.1038/s41746-023-00792-0.
## References
1. Diseases GBD, Injuries C. **Global burden of 369 diseases and injuries in 204 countries and territories, 1990–2019: a systematic analysis for the Global Burden of Disease Study 2019**. *Lancet* (2020.0) **396** 1204-1222. DOI: 10.1016/S0140-6736(20)30925-9
2. Gon G. **The frequency of maternal morbidity: a systematic review of systematic reviews**. *Int J. Gynaecol. Obstet.* (2018.0) **141** 20-38. DOI: 10.1002/ijgo.12468
3. Kotlar B, Gerson E, Petrillo S, Langer A, Tiemeier H. **The impact of the COVID-19 pandemic on maternal and perinatal health: a scoping review**. *Reprod. Health* (2021.0) **18** 10. DOI: 10.1186/s12978-021-01070-6
4. Noddin K, Bradley D, Wolfberg A. **Delivery outcomes during the COVID-19 pandemic as reported in a pregnancy mobile app: retrospective cohort study**. *JMIR Pediatr. Parent* (2021.0) **4** e27769. DOI: 10.2196/27769
5. Tilahun B, Gashu KD, Mekonnen ZA, Endehabtu BF, Angaw DA. **Mapping the role of digital health technologies in prevention and control of COVID-19 pandemic: review of the literature**. *Yearb. Med Inf.* (2021.0) **30** 26-37. DOI: 10.1055/s-0041-1726505
6. Lima-Pereira P, Bermudez-Tamayo C, Jasienska G. **Use of the Internet as a source of health information amongst participants of antenatal classes**. *J. Clin. Nurs.* (2012.0) **21** 322-330. DOI: 10.1111/j.1365-2702.2011.03910.x
7. Sayakhot P, Carolan-Olah M. **Internet use by pregnant women seeking pregnancy-related information: a systematic review**. *BMC Pregnancy Childbirth* (2016.0) **16** 65. DOI: 10.1186/s12884-016-0856-5
8. Wallwiener S. **Pregnancy eHealth and mHealth: user proportions and characteristics of pregnant women using Web-based information sources—a cross-sectional study**. *Arch. Gynecol. Obstet.* (2016.0) **294** 937-944. DOI: 10.1007/s00404-016-4093-y
9. Ozkan Sat S, Yaman Sozbir S. **Use of mobile applications by pregnant women and levels of pregnancy distress during the COVID-19 (Coronavirus) Pandemic**. *Matern Child Health J.* (2021.0) **25** 1057-1068. DOI: 10.1007/s10995-021-03162-y
10. 10.Musgrave, L. M., Kizirian, N. V., Homer, C. S. E. & Gordon, A. Mobile phone apps in Australia for improving pregnancy outcomes: systematic search on app stores. JMIR mHealth and uHealth8, 10.2196/22340 (2020).
11. Munoz-Mancisidor A, Martin-Payo R, Gonzalez-Mendez X, Fernandez-Alvarez MDM. **Content, behavior change techniques, and quality of pregnancy apps in spain: systematic search on app stores**. *JMIR Mhealth Uhealth* (2021.0) **9** e27995. DOI: 10.2196/27995
12. 12.Brown, H. M., Bucher, T., Collins, C. E. & Rollo, M. E. A review of pregnancy apps freely available in the Google Play Store. Health Promot. J. Aust.: Off. J. Aust. Assoc. Health Promot. Prof.10.1002/hpja.270 (2019).
13. 13.Kraschnewski, J. L. et al. Paging “Dr. Google”: Does technology fill the gap created by the prenatal care visit structure? Qualitative focus group study with pregnant women. J. Med. Internet Res.16, 10.2196/jmir.3385 (2014).
14. Yu H. **A comparison of functional features of Chinese and US mobile apps for pregnancy and postnatal care: a systematic app store search and content analysis**. *Front Public Health* (2022.0) **10** 826896. DOI: 10.3389/fpubh.2022.826896
15. 15.Hughson, J.-A. P., Daly, J. O., Woodward-Kron, R., Hajek, J. & Story, D. The rise of pregnancy apps and the implications for culturally and linguistically diverse women: narrative review. JMIR mHealth and uHealth6, 10.2196/mhealth.9119 (2018).
16. Willcox JC. **A mobile health intervention promoting healthy gestational weight gain for women entering pregnancy at a high body mass index: the txt4two pilot randomised controlled trial**. *BJOG* (2017.0) **124** 1718-1728. DOI: 10.1111/1471-0528.14552
17. 17.Dahl, A. A. Healthy Motivations for Moms-To-Be (Healthy MoM2B) Study: A mobile health intervention targeting gestational weight gain among U.S. women. Doctoral dissertation thesis, University of South Carolina, (2018).
18. Lupton D, Pedersen S. **An Australian survey of women’s use of pregnancy and parenting apps**. *Women Birth* (2016.0) **29** 368-375. DOI: 10.1016/j.wombi.2016.01.008
19. Alam M, Banwell C, Olsen A, Lokuge K. **Patients’ and doctors’ perceptions of a mobile phone-based consultation service for maternal, neonatal, and infant health care in Bangladesh: a mixed-methods study**. *JMIR Mhealth Uhealth* (2019.0) **7** e11842. DOI: 10.2196/11842
20. Marwaha JS, Landman AB, Brat GA, Dunn T, Gordon WJ. **Deploying digital health tools within large, complex health systems: key considerations for adoption and implementation**. *NPJ Digit. Med.* (2022.0) **5** 13. DOI: 10.1038/s41746-022-00557-1
21. Zielinska AP, Mullins E, Lees C. **The feasibility of multimodality remote monitoring of maternal physiology during pregnancy**. *Med. (Baltim.)* (2022.0) **101** e29566. DOI: 10.1097/MD.0000000000029566
22. Padmanabhan U, Summerbell CD, Heslehurst N. **A qualitative study exploring pregnant women’s weight-related attitudes and beliefs in UK: the BLOOM study**. *BMC Pregnancy Childbirth* (2015.0) **15** 99. DOI: 10.1186/s12884-015-0522-3
23. Goldstein RF. **Association of gestational weight gain with maternal and infant outcomes: a systematic review and meta-analysis**. *JAMA* (2017.0) **317** 2207-2225. DOI: 10.1001/jama.2017.3635
24. LifeCycle Project-Maternal O. **Association of gestational weight gain with adverse maternal and infant outcomes**. *JAMA* (2019.0) **321** 1702-1715. DOI: 10.1001/jama.2019.3820
25. Uzan J, Carbonnel M, Piconne O, Asmar R, Ayoubi JM. **Pre-eclampsia: pathophysiology, diagnosis, and management**. *Vasc. Health Risk Manag.* (2011.0) **7** 467-474. PMID: 21822394
26. Rooney BL, Schauberger CW. **Excess pregnancy weight gain and long-term obesity: one decade later**. *Obstet. Gynecol.* (2002.0) **100** 245-252. PMID: 12151145
27. Amorim AR, Rossner S, Neovius M, Lourenco PM, Linne Y. **Does excess pregnancy weight gain constitute a major risk for increasing long-term BMI?**. *Obes. (Silver Spring)* (2007.0) **15** 1278-1286. DOI: 10.1038/oby.2007.149
28. Korkmaz A, Teksam O, Yurdakok M, Yigit S, Tekinalp G. **Fetal malnutrition and its impacts on neonatal outcome in preterm infants**. *Turk. J. Pediatr.* (2011.0) **53** 261-268. PMID: 21980806
29. Rahman SA, Adjeroh D. **Surface-based body shape index and its relationship with all-cause mortality**. *PLoS One* (2015.0) **10** e0144639. DOI: 10.1371/journal.pone.0144639
30. Gomez-Peralta F. **Relationship between “a body shape index (ABSI)” and body composition in obese patients with type 2 diabetes**. *Diabetol. Metab. Syndr.* (2018.0) **10** 21. DOI: 10.1186/s13098-018-0323-8
31. Nuttall FQ. **Body mass index: obesity, bmi, and health: a critical review**. *Nutr. Today* (2015.0) **50** 117-128. DOI: 10.1097/NT.0000000000000092
32. Nevill AM, Stewart AD, Olds T, Holder R. **Relationship between adiposity and body size reveals limitations of BMI**. *Am. J. Phys. Anthropol.* (2006.0) **129** 151-156. DOI: 10.1002/ajpa.20262
33. Relph S, Ong M, Vieira MC, Pasupathy D, Sandall J. **Perceptions of risk and influences of choice in pregnant women with obesity. An evidence synthesis of qualitative research**. *PLoS One* (2020.0) **15** e0227325. DOI: 10.1371/journal.pone.0227325
34. Vieira MC. **Clinical and biochemical factors associated with preeclampsia in women with obesity**. *Obes. (Silver Spring)* (2017.0) **25** 460-467. DOI: 10.1002/oby.21715
35. White SL. **Early antenatal prediction of gestational diabetes in obese women: development of prediction tools for targeted intervention**. *PLoS One* (2016.0) **11** e0167846. DOI: 10.1371/journal.pone.0167846
36. Sebo P, Beer-Borst S, Haller DM, Bovier PA. **Reliability of doctors’ anthropometric measurements to detect obesity**. *Prev. Med.* (2008.0) **47** 389-393. DOI: 10.1016/j.ypmed.2008.06.012
37. Sebo P, Herrmann FR, Haller DM. **Accuracy of anthropometric measurements by general practitioners in overweight and obese patients**. *BMC Obes.* (2017.0) **4** 23. DOI: 10.1186/s40608-017-0158-0
38. 38.Wang, N., Deng, Z., Wen, L. M., Ding, Y. & He, G. Understanding the use of smartphone apps for health information among pregnant Chinese women: Mixed Methods Study. JMIR mHealth and uHealth7, 10.2196/12631 (2019).
39. Wang W, Sun L, Liu T, Lai T. **The use of E-health during the COVID-19 pandemic: a case study in China’s Hubei province**. *Health Sociol Rev.* (2021.0) **31** 1-17
40. 40.Lupton, D. ‘It Just Gives Me a Bit of Peace of Mind’: Australian women’s use of digital media for pregnancy and early motherhood. 10.3390/SOC7030025 (2017).
41. Alfawzan N, Christen M, Spitale G, Biller-Andorno N. **Privacy, data sharing, and data security policies of women’s mhealth apps: scoping review and content analysis**. *JMIR Mhealth Uhealth* (2022.0) **10** e33735. DOI: 10.2196/33735
42. Lee Y, Moon M. **Utilization and content evaluation of mobile applications for pregnancy, birth, and child care**. *Healthc. Inform. Res.* (2016.0) **22** 73-80. DOI: 10.4258/hir.2016.22.2.73
43. Hayman MJ. **Evaluating evidence-based content, features of exercise instruction, and expert involvement in physical activity apps for pregnant women: systematic search and content analysis**. *JMIR Mhealth Uhealth* (2022.0) **10** e31607. DOI: 10.2196/31607
44. Faessen JPM. **Eating for 2: a systematic review of dutch app stores for apps promoting a healthy diet during pregnancy**. *Curr. Dev. Nutr.* (2022.0) **6** nzac087. DOI: 10.1093/cdn/nzac087
45. Frid G, Bogaert K, Chen KT. **Mobile health apps for pregnant women: systematic search, evaluation, and analysis of features**. *J. Med. Internet Res.* (2021.0) **23** e25667. DOI: 10.2196/25667
46. Brunelli L. **Gaps and future challenges of Italian apps for pregnancy and postnatal care: systematic search on app stores**. *J. Med. Internet Res.* (2021.0) **23** e29151. DOI: 10.2196/29151
47. Dahl AA, Dunn CG, Boutt‚ AK, Crimarco A, Turner-McGrievy G. **Mobilizing mHealth for moms: a review of mobile apps for tracking gestational weight gain**. *J. Technol. Behav. Sci.* (2018.0) **3** 32-40. DOI: 10.1007/s41347-017-0030-6
48. Hayman M. **Quality, features, and presence of behavior change techniques in mobile apps designed to improve physical activity in pregnant women: systematic search and content analysis**. *JMIR Mhealth Uhealth* (2021.0) **9** e23649. DOI: 10.2196/23649
49. Gunter R. **Evaluating patient usability of an image-based mobile health platform for postoperative wound monitoring**. *JMIR Mhealth Uhealth* (2016.0) **4** e113. DOI: 10.2196/mhealth.6023
50. Capurro D, Coghlan S, Pires DEV. **Preventing digital overdiagnosis**. *JAMA* (2022.0) **327** 525-526. DOI: 10.1001/jama.2021.22969
51. Goldsteen A, Farkash A, Moffie M, Shmelkin R. **Applying artificial intelligence privacy technology in the healthcare domain**. *Stud. Health Technol. Inf.* (2022.0) **294** 121-122
52. 52.Galle, A. et al. A double-edged sword-telemedicine for maternal care during COVID-19: findings from a global mixed-methods study of healthcare providers. BMJ Glob. Health6, 10.1136/bmjgh-2020-004575 (2021).
53. Tsang S, Royse CF, Terkawi AS. **Guidelines for developing, translating, and validating a questionnaire in perioperative and pain medicine**. *Saudi J. Anaesth.* (2017.0) **11** S80-S89. DOI: 10.4103/sja.SJA_203_17
54. Barthold D, Brah AT, Graham SM, Simoni JM, Hauber B. **Improvements to survey design from pilot testing a discrete-choice experiment of the preferences of persons living with hiv for long-acting antiretroviral therapies**. *Patient* (2022.0) **15** 513-520. DOI: 10.1007/s40271-022-00581-z
55. Goodyear-Smith F, Jackson C, Greenhalgh T. **Co-design and implementation research: challenges and solutions for ethics committees**. *BMC Med. Ethics* (2015.0) **16** 78. DOI: 10.1186/s12910-015-0072-2
56. Bhattacharyya O, Mossman K, Gustafsson L, Schneider EC. **Using human-centered design to build a digital health advisor for patients with complex needs: persona and prototype development**. *J. Med. Internet Res.* (2019.0) **21** e10318. DOI: 10.2196/10318
57. Richardson S, Lawrence K, Schoenthaler AM, Mann D. **A framework for digital health equity**. *NPJ Digit. Med.* (2022.0) **5** 119. DOI: 10.1038/s41746-022-00663-0
58. Buchanan L. **Sources of information and the use of mobile applications for health and parenting information during pregnancy: implications for health promotion**. *Health Inform. J.* (2021.0) **27** 14604582211043146. DOI: 10.1177/14604582211043146
59. Shulman HB, D’Angelo DV, Harrison L, Smith RA, Warner L. **The pregnancy risk assessment monitoring system (prams): overview of design and methodology**. *Am. J. Public Health* (2018.0) **108** 1305-1313. DOI: 10.2105/AJPH.2018.304563
60. Tylka TL, Wood-Barcalow NL. **The body appreciation scale-2: item refinement and psychometric evaluation**. *Body Image* (2015.0) **12** 53-67. DOI: 10.1016/j.bodyim.2014.09.006
61. Watson B, Fuller-Tyszkiewicz M, Broadbent J, Skouteris H. **Development and validation of a tailored measure of body image for pregnant women**. *Psychol. Assess.* (2017.0) **29** 1363-1375. DOI: 10.1037/pas0000441
62. 62.Day, M. On Trauma and Safety: Toward Trauma-Informed Research Methods. (2018).
63. 63.StatQuest. Logistic Regression in R, Clearly Explained. (26 Jul 2018).
64. Braun V, Clarke V. **What can “thematic analysis” offer health and wellbeing researchers?**. *Int J. Qual. Stud. Health Well-Being* (2014.0) **9** 26152. DOI: 10.3402/qhw.v9.26152
65. Hagaman AK, Wutich A. **How many interviews are enough to identify metathemes in multisited and cross-cultural research? another perspective on Guest, Bunce, and Johnson’s (2006) Landmark Study**. *Field Methods* (2017.0) **29** 23-41. DOI: 10.1177/1525822X16640447
66. Vasileiou K, Barnett J, Thorpe S, Young T. **Characterising and justifying sample size sufficiency in interview-based studies: systematic analysis of qualitative health research over a 15-year period**. *BMC Med. Res. Methodol.* (2018.0) **18** 148. DOI: 10.1186/s12874-018-0594-7
67. Saunders B. **Saturation in qualitative research: exploring its conceptualization and operationalization**. *Qual. Quant.* (2018.0) **52** 1893-1907. DOI: 10.1007/s11135-017-0574-8
|
---
title: 'Efficacy and safety of saroglitazar in real‐world patients of non‐alcoholic
fatty liver disease with or without diabetes including compensated cirrhosis: A
tertiary care center experience'
authors:
- Sujit Chaudhuri
- Agnibha Dutta
- Sunil Baran Das Chakraborty
journal: 'JGH Open: An Open Access Journal of Gastroenterology and Hepatology'
year: 2023
pmcid: PMC10037031
doi: 10.1002/jgh3.12878
license: CC BY 4.0
---
# Efficacy and safety of saroglitazar in real‐world patients of non‐alcoholic fatty liver disease with or without diabetes including compensated cirrhosis: A tertiary care center experience
## Abstract
One hundred and twelve NAFLD patients were screened and 76 patients, of whom 11 had compensated cirrhosis (LSM ≥ 14 kPa), were started on saroglitazar 4 mg. There was significant improvement of LSM from baseline (11.03 ± 7.19 kPa) to 24‐week (9.29 ± 6.39 kPa) and 52‐week (8.59 ± 6.35 kPa) values respectively. The favorable response was seen across all the fibrosis stages as categorized based on LSM values (F0–F1: LSM <7 kPa; F2: LSM 7–10 kPa; F3: LSM ≥10 to <14 KPa and F4: LSM ≥14 kPa). Significant reduction was also seen in CAP, ALT, AST, HbA1c, LDL, total cholesterol, and triglyceride values. There was no significant weight change along the study interval. Pruritus occurred in one patient who required treatment discontinuation.
### Background and Aim
Saroglitazar, a dual PPAR α/γ agonist, is useful in management of NAFLD and diabetic dyslipidemia. Here, we report the safety and efficacy of saroglitazar in NAFLD patients with or without diabetes including compensated cirrhosis.
### Methods
Patients, started on saroglitazar 4 mg were prospectively evaluated for 52 weeks in a tertiary care center in Eastern India. Effectiveness was measured in terms of anthropometric measurements, fasting blood glucose, LFT, lipid profile, HbA1c, and elastography parameters (LSM and CAP) measured at baseline, 24, and 52 weeks. Adverse drug reactions were monitored.
### Results
A total of 112 patients were enrolled in the study, of whom 63 patients were taken up for per‐protocol analysis. Mean age was 49.11 ± 11.09 years and 46($73\%$) were male. Thirty‐four ($54\%$) were nondiabetic. Eleven patients had compensated cirrhosis. There was significant improvement of LSM from baseline (11.03 ± 7.19 kPa) to 24‐week (9.29 ± 6.39 kPa) and 52‐week (8.59 ± 6.35 kPa) values respectively ($P \leq 0.001$). Significant reduction was also seen in CAP values, ALT, AST, HbA1c, LDL, total cholesterol, and triglyceride values. There was no significant weight change along the study interval. Pruritus occurred in one patient who required treatment discontinuation and another patient had mild symptomatic loose motion.
### Conclusions
Saroglitazar is effective and safe in improving biochemical parameters and LSM and CAP values irrespective of weight reduction. It may be used in compensated cirrhotics with close monitoring for side effects.
## Background and aim
Non‐alcoholic fatty liver disease (NAFLD) is a modern‐day epidemic. Its prevalence is almost $30\%$ in Asian countries. 1 *It is* a spectrum of illness from non‐alcoholic fatty liver to steatohepatitis (NASH) with a chance of progression to fibrosis ultimately leading to cirrhosis. This entity is being increasingly a cause of cirrhosis and hepatocellular carcinoma (HCC) globally.
Pathogenesis of NAFLD is multifactorial. NASH often develops in the context of excess adiposity and systemic insulin resistance. 2 Pathogenesis involves simple accumulation of fat followed by a variable contribution from different pathogenic drivers, such as lipotoxicity, oxidative stress, inflammation, and immune system activation. Insulin resistance leads to lipid accumulation in hepatocyte as the first hit and is followed by a second hit driven by lipotoxic metabolite‐induced mitochondrial dysfunction and oxidative stress leading to hepatocyte death and inflammation. 3 Lifestyle modification with exercise and achieving weight loss remains the cornerstone of therapy for NAFLD, but it is very difficult to sustain for a prolonged period. Despite better understanding of the pathogenesis and progression of NAFLD‐/NASH‐related liver disease, drug targeting key areas of metabolic pathways is lacking. In the last decade or so, multiple drugs have been tested for this condition, but till now a reliable medicine with a definite therapeutic end point is lacking. Also, in advanced stages of the disease, lifestyle modification alone is unlikely to be adequate management and may not be justifiable as monotherapy.
Saroglitazar is a dual potent PPAR‐α/γ agonist. Synergistic effect of improved lipid oxidation and improved insulin resistance by PPAR‐α and PPAR‐γ, respectively, makes rational its use in NASH/NAFLD. Its efficacy in management of patients of NAFLD and diabetic dyslipidemia is being reported throughout the world. 4, 5, 6, 7 Furthermore, data regarding efficacy and safety of saroglitazar in non‐diabetics and in patients with compensated cirrhosis are lacking.
Diagnosis of baseline severity and assessment of therapeutic response of NAFLD have predominantly shifted to noninvasive methods. Liver biopsy, although considered gold standard, has its drawbacks, such as being invasive and less patient acceptance, particularly in repeated testing.
We aimed to evaluate the real‐world safety and efficacy of saroglitazar 4 mg in patients with NAFLD primarily by seeing the improvement in liver steatosis and stiffness.
## Method
This is an investigator‐initiated, single‐center, prospective, observational, open‐label, single‐arm study to evaluate the safety and efficacy of saroglitazar 4 mg in patients with NAFLD/NASH in real‐life setting, conducted in a tertiary care research institute in Kolkata.
Ethical clearance was obtained from institutional ethical committee.
## Inclusion criteria
The included patients were of age ≥18 years, and diagnosed for NAFLD fulfilling the AASLD guidelines, 8 hepatic steatosis by imaging or histology, absence of significant alcohol consumption, competing etiologies for hepatic steatosis, and co‐existing causes for chronic liver disease, having elevated ALT levels along with liver stiffness value ≥6 kPa and/or liver steatosis CAP >290 dB/m, measured through FibroScan/TE, has been included in the study.
## Exclusion criteria
The patients with any evidence of alcoholic liver disease, significant alcohol use (210 gm/week in male, 140 g/week in females), concomitant use of any steatogenic drugs, and clinical or lab evidence of other liver disease have also been excluded from the study. Patients with decompensated cirrhosis have been excluded. Other liver illnesses like chronic hepatitis B or C infection, Wilson's disease, and drugs causing liver fibrosis like amiodarone, methotrexate, etc., have been excluded. Patients taking thiazolidine diones or saroglitazar in the last 6 months were also excluded.
The patients found to have any confounding factors, which may overestimate the FibroScan values like liver congestion, ascites, liver inflammation due to any recent liver illness or drinking alcohol, benign or cancerous tumors in the liver, biliary obstructions, etc., were also excluded from the study.
The person who has known allergy, sensitivity or intolerance to saroglitazar or formulation ingredients, women with pregnancy or lactation or of childbearing potential and not using appropriate contraceptive measures, has history or other evidence of severe illness or any other conditions that would make the patient, in the opinion of the investigator, unsuitable for the study and those who used vitamin E ≥ 800 IU/day or multivitamins containing vitamin E ≥ 800 IU/day in the 1 month preceding screening visit, were also excluded from the study.
Saroglitazar 4 mg once daily was prescribed in routine clinical practice to eligible patients, who visited the outpatient department of AMRI Hospital, at Salt Lake, Kolkata in April 2019, and fulfilled the inclusion/exclusion criteria, along with routine care of diet and lifestyle modification. Informed consent was taken from every participant. The patients were followed up at an interval of 3 months and up to a total of 1‐year duration and the safety and efficacy data were collected in an excel sheet at every follow‐up visit. The data were collected by interviewing every participant using a set of questionnaires related to demographic features like age, gender, anthropometric measurement, family history for liver disease or other metabolic disorder, ongoing medications, detailed medical history, including personal history of alcohol intake, and any comorbidities like dyslipidemia, diabetes, and hypertension. A trained technical assistant did all the physical examination. Height was measured by a standard stadiometer, and weight was measured using a standard bathroom scale. Anthropometric measurements were cross‐checked to ensure inter‐observer reliability. Body mass index (BMI) was calculated using height in meter and weight in kg.
Baseline patients demographics like age, gender, weight, height, waist circumference, co‐morbid conditions, and ongoing medications were recorded.
Detailed medical history, including personal history of alcohol intake, concomitant medication, and presence of metabolic comorbidities like obesity, hypertension, diabetes mellitus, coronary artery, and cerebrovascular disease, has been recorded. Family history of any metabolic or liver‐related disorders has also been captured in the excel sheet at the baseline visit.
The EchosensFibroScan® 530 Compact machine has been used to measure both fibrosis (liver stiffness) and steatosis (ultrasound attenuation rate/ CAP), together by an expert and Echosens certified technician. M and XL probes were used for subjects with BMI less than 30 kg/m2 and ≥ 30 kg/m2 respectively.
At baseline, patients were categorized into four categories based on liver stiffness (LSM) value: (i) F0–F1: LSM <7 kPa; (ii) F2: LSM 7–10 kPa; (iii) F3: LSM >10 to <14 KPa; and (iv) F4: LSM >14 kPa.
Patients in the last group also underwent upper GI endoscopy for variceal screening.
The patients on continued saroglitazar 4 mg once daily therapy have undergone follow‐up elastography at 24 and 52 weeks. Anthropometric measurements were taken at each clinical visit. Patients underwent testing for lipid profile, LFT, HbA1c, and CBC at baseline, 24 weeks, and 52 weeks. The data were analyzed using paired t‐test on SPSS Version 22 statistical package.
At every follow‐up visit, safety parameters were assessed in the form of history regarding possible side effects as well as clinical examination.
## Statistical analysis
Continuous data with normal distribution were presented as mean and SD and without normal distribution as median and IQR. Normally distributed continuous data were analyzed using paired t‐test. Wilcoxon signed‐rank test was used for all skewed non‐normal data. All categorical data were presented as proportion and analyzed using the χ2 test. P values less than 0.05 were considered significant. All the statistical analysis was done on SPSS ver 22 statistical package.
## Results
Between April and October 2019, 112 patients were screened for the study. Thirty‐six had one or more exclusion criteria, treatment had to be stopped in one due to allergic complication, 12 patients did not complete 12‐month therapy and were lost to follow‐up. Sixty‐three patients were taken up for per‐protocol analysis.
Basic demographic profile of patients is shown in Table 1. The mean age of the population was 49.11 ± 11.09 years and 46($73\%$) were male. Thirty‐four ($54\%$) were nondiabetic.
**Table 1**
| Age (Years) † | 49.11 (11.09) |
| --- | --- |
| Weight (Kg) † | 73.56 (12.47) |
| BMI (Kg/m2) † | 27.25(4.10) |
| Waist circumference (meter) mean (SD) | 1.02(0.15) |
| Male, n (%) | 46 (73.02) |
| Nondiabetic, n (%) | 34 (53.97) |
| Non‐dyslipidemic, n (%) | 46 (73.02) |
| Non‐hypertensive, n (%) | 34 (53.97) |
| Family history of liver disease | 21 (33.33) |
| LSM (KPa) ‡ | 8.5 (3.9) |
| CAP (dB/m) ‡ | 328 (46) |
| FBS (mg/dl) ‡ | 109 (31.5) |
| HbA1c (%) ‡ | 6.45 (1.5) |
| ALT (IU/L) ‡ | 45 (34.5) |
| AST (IU/L) ‡ | 39 (23) |
| LDL (mg/dl) ‡ | 125 (38.4) |
| TG (mg/dl) ‡ | 160 (43.4) |
Baseline investigations are shown in Table 1. Median ALT and AST values were 45 (18–188) and 39 (25–170) respectively. ALT and AST values were greater than 40 IU/L in $86\%$ ($$n = 55$$) and $81\%$ ($$n = 52$$) respectively. Median LSM at baseline was 8.5 (4.2–35.6) kPa. Eleven patients had LSM >14 kPa, of them three were found to have esophageal varices. The median CAP value was 328 (233–400) dB/m.
## Post‐treatment investigation
Compared with baseline, after 24 and 52 weeks of saroglitazar therapy, there was reduction in ALT levels by $37\%$ and $52\%$ respectively, which was statistically significant. Similar trends were seen in AST also. There was statistically significant reduction in cholesterol and triglyceride levels at 52 by $24.1\%$ and $40.6\%$ (Table 2).
**Table 2**
| Secondary parameters | Baseline median (IQR) | % Change* (at 24 weeks) | * P value (z score) | % Change# (at 52 weeks) | † P value (z score) |
| --- | --- | --- | --- | --- | --- |
| Weight (kg) | 72 (16) | 1.05 | 0.139 (−1.479) | 1.42% | 0.055 (−1.922) |
| WC (cm) | 101 (14.5) | −2.25 | 0.038 (−2.075) | −2.97% | 0.049 (−1.971) |
| BMI (kg/m2) | 26.3 (4.6) | 1.03 | 0.163 (−1.393) | 1.31% | 0.13 (−1.515) |
| HDL‐C (mg/dl) | 39 (6) | 8.42 | <0.001 (−4.295) | 15.59% | <0.001 (−5.124) |
| LDL‐C (mg/dl) | 125 (38.4) | −15.9 | <0.001 (−5.388) | −25.54% | <0.001 (−6.069) |
| Tot Chol (mg/dl) | 178 (47) | −16.5 | <0.001 (−5.895) | −24.05% | <0.001 (−6.258) |
| TG (mg/dl) | 160 (43.4) | −29.6 | <0.001 (−6.401) | −40.61% | <0.001 (−6.644) |
| AST (IU/L) | 39 (23) | −25.1 | <0.001 (−3.982) | −40.78% | <0.001 (−5.752) |
| ALT (IU/L) | 45 (34.5) | −36.7 | <0.001 (−5.487) | −52.21% | <0.001 (−6.542) |
Eleven patients had LSM ≥14 kPa corresponding with F4 fibrosis. 9 There was statistically significant improvement in LSM value in all the four groups at 52 weeks with decrease of around 22–$25\%$ in patients with baseline LSM ≥7 kPa (Table 3 and Fig. 1). There was $14\%$ reduction in CAP values from baseline across all the patient population, which was also statistically significant.
There was a significant improvement in median CAP at 24 weeks [281 [70] dB/m, $P \leq 0.001$] and 52 weeks [287 [67] dB/m, $P \leq 0.001$] as compared with the baseline [328 [46] dB/m] (Table 4).
**Table 4**
| Parameters | At baseline | 24 weeks | 52 weeks | % Change ‡ | * P value (z score) |
| --- | --- | --- | --- | --- | --- |
| LSM (kPa) † | 8.5 (3.9) | 6.5 (2.8) | 6.5 (2.9) | −22.11% | <0.001 (−5.467) |
| CAP (dB/m) † | 328 (46) | 281 (70) | 287 (67) | −14.18% | <0.001 (−6.278) |
The median LSM at week 24 [6.5 (2.8) kPa, $P \leq 0.001$] and at week 52 [6.5 (2.9) kPa, $P \leq 0.001$] was also statistically significantly improved than the baseline median LSM of 8.5 (3.9) kPa (Table 4).
There was no significant weight reduction in the patient population, and BMI distribution was similar at baseline and 52 weeks (Table 5); rather overall there was $1.42\%$ weight gain. However, reduction in HbA1c, LDL and increase in HDL was seen after 52 weeks of therapy (Table 2).
**Table 5**
| Obesity categorization † (BMI range in Kg/m2) | At baseline, n (%) | After 52 weeks, n (%) |
| --- | --- | --- |
| Underweight | 0 (0) | 0 (0) |
| Normal weight | 18 (28.6) | 19 (30.2) |
| Overweight | 30 (47.6) | 29 (46.03) |
| Obese (Class I) | 8 (12.7) | 8 (12.7) |
| Obese (Class II) | 4 (6.3) | 4 (6.3) |
| Obese (Class III) | 1 (1.6) | 1 (1.6) |
Saroglitazar 4 mg once daily was found to be safe and well tolerated, as there was no severe reported drug‐related major adverse event, which results in discontinuation of the therapy, except one case of minor itching, which subsided within 2 weeks of treatment discontinuation and one other case who has reported mild symptomatic loose motion.
## Discussion
This study aimed to test the efficacy and tolerability of saroglitazar in an extended population including nondiabetics and ones having significant fibrosis on elastography. It is shown that saroglitazar shows significant reduction in LSM and CAP values as well as improvement in metabolic parameters. One novel finding is that saroglitazar is tolerable, safe, and efficacious in compensated F4 fibrosis also. There has been limited data regarding the efficacy of saroglitazar in the Indian population. Padole et al. 10 have shown that saroglitazar is effective in reducing transaminase value, but LSM and CAP reduction was seen only in patients who achieved at least $5\%$ weight reduction. This study refutes the same showing improvement in LSM irrespective of weight change. Further studies with higher sample size are required to throw further light into it.
This study adds to real‐world experience of saroglitazar in the NASH population. It is predominantly recommended for patients with diabetes, but Sarin et al. conducted a study including nondiabetics as well. 11 This study also included nondiabetic patients and its efficacy is shown in this population as well. There was a significant improvement in hepatic steatosis as measured by improvement in CAP. MR‐PDDF is known to be superior to CAP assessment, but the use of the former is limited by its cost and availability and for real‐world scenario, CAP is a validated parameter for assessment of steatosis. For the assessment of fibrosis and inflammation, biopsy remains the gold standard. But apart from research setting, biopsy for all patients of NASH remains impractical considering the risk–benefit ratio and patient acceptability. Further limitation of biopsy remains sampling bias. Accuracy of LSM to assess fibrosis is an area of ongoing research and there are different studies supporting the same. 12, 13, 14, 15 For real‐life patients, it remains a feasible and repeatable option. Therefore, it has been used in this study. This study showed improvement in LSM and CAP values mostly in the first six months that persisted after 52 weeks of therapy.
As α effect of saroglitazar is more than γ agonism, lipid profile improvement is expected to be more than the improvement of glycemic status. It was corroborated in the result as well with patients having around $25\%$ improvement in total cholesterol and LDL level and $40\%$ reduction in triglyceride level. As expected from γ agonism, there was weight gain in the patient population despite improvement in LSM and CAP, which is shown in other studies 16 as well. This area needs further assessment. Weight loss with lifestyle management has been shown to improve NASH. So, the study of effect of medication adjusted for weight gain or loss would require different study design. This may fill the knowledge gap in this domain. Other γ agonism effect in the form of improvement of blood glucose and HbA1c has been shown in the patient population.
Liver enzymes have been used traditionally to assess the extent of liver injury. It has inherent limitations being nonspecific and not correlative with fibrosis. However, it is very useful in practical setting due to its availability and low cost. Study population showed a significant reduction of ALT and AST both at 24 and 52 weeks. It corroborates with recent real‐life data 17 on saroglitazar in NASH population.
Perhaps this study was the first of its kind to test the efficacy and safety of saroglitazar in compensated cirrhotics. Cirrhosis was diagnosed based on either LSM values or presence of varices on upper GI endoscopy with other causes being excluded. Saroglitazar was well tolerated with no new onset decompensation in this group as well as efficacious with approximately $23\%$ reduction in LSM values at 52 weeks as compared with baseline. This is definitely a novel finding. Although only 11 patients were cirrhotics, further studies with larger patient population are warranted to possibly expand the use of this molecule.
The first strength of the study is its patient population. Inclusion of nondiabetics and cirrhotics allowed to assess efficacy and safety of saroglitazar in this less studied population. The second strength is longer follow‐up duration of 52 weeks where most studies have shorter follow‐up data. The third strength is the use of noninvasive markers at frequent time interval that mimics real‐life experience, which most of the physicians follow.
Limitation of this study remains that it cannot predict superiority over other molecules presently being researched for NASH. Absence of biopsy data would indicate that idea of baseline necro‐inflammatory activity was unknown. Third, follow‐up data were not there on whether discontinuation of drug would lead to reversal of biochemical and stiffness parameter improvement or persistence of benefit of drug. This very important question was not addressed in the study design. Fourth, cardiovascular outcome was not assessed. The cause of mortality and morbidity in NASH population is predominantly cardiovascular. So, whether improvement of metabolic parameter ultimately translates to improvement in cardiovascular outcome or not is a question that would require long‐term follow‐up, for which this study was not designed. Fifth, there was no placebo arm to compare the contribution of lifestyle modification alone to addition of saroglitazar.
## Conclusion
Saroglitazar is effective and safe in improving biochemical parameters and LSM and CAP values irrespective of weight reduction. It may be used in compensated cirrhotics with close monitoring for side effects.
## Ethics statement
The study design was approved by institutional ethical committee of AMRI Hospitals, Salt Lake.
## Patient consent statement
Written consent was obtained from each participant after explaining in their own vernacular, consent forms preserved by the author.
## Data availability statement
The patient related data are accessible from principal author on request.
## References
1. De Roza MA, Goh GB. **The increasing clinical burden of NAFLD in Asia**. *Lancet Gastroenterol. Hepatol.* (2019) **4** 333-4. PMID: 30902671
2. Dowman JK, Tomlinson JW, Newsome PN. **Pathogenesis of non‐alcoholic fatty liver disease**. *QJM* (2010) **103** 71-83. PMID: 19914930
3. Parthasarathy G, Revelo X, Malhi H. **Pathogenesis of Nonalcoholic Steatohepatitis: An Overview**. *Hepatol Commun.* (2020) **4** 478-92. PMID: 32258944
4. Joshi SR. **Saroglitazar for the treatment of dyslipidemia in diabetic patients**. *Expert Opin Pharmacother.* (2015) **16** 597-606. PMID: 25674933
5. Kaul U, Parmar D, Manjunath K. **New dual peroxisome proliferator activated receptor agonist‐Saroglitazar in diabetic dyslipidemia and non‐alcoholic fatty liver disease: integrated analysis of the real world evidence**. *Cardiovasc. Diabetol.* (2019) **18** 80. PMID: 31208414
6. Sosale A, Saboo B, Sosale B. **Saroglitazar for the treatment of hypertrig‐lyceridemia in patients with type 2 diabetes: current evidence**. *Diabetes Metab. Syndr. Obes.* (2015) **15** 189-96
7. Krishnappa M, Patil K, Parmar K. **Effect of saroglitazar 2 mg and 4 mg on glycemic control, lipid profile and cardiovascular disease risk in patients with type 2 diabetes mellitus: a 56‐week, randomized, double blind, phase 3 study (PRESS XII study)**. *Cardiovasc. Diabetol.* (2020) **19** 93. PMID: 32560724
8. Chalasani N, Younossi Z, Lavine JE. **The diagnosis and management of nonalcoholic fatty liver disease: Practice guidance from the American Association for the Study of Liver Diseases**. *Hepatology* (2018) **67** 328-57. PMID: 28714183
9. Wong GL, Wong VW, Choi PC. **Assessment of fibrosis by transient elastography compared with liver biopsy and morphometry in chronic liver diseases**. *Clin. Gastroenterol. Hepatol.* (2008) **6** 1027-35. PMID: 18456573
10. Padole P, Arora A, Sharma P, Chand P, Verma N, Kumar A. **Saroglitazar for Nonalcoholic Fatty Liver Disease: A Single Centre Experience in 91 Patients**. *J. Clin. Exp. Hepatol.* (2022) **12** 435-9. PMID: 35535066
11. Sarin SK. **A prospective, multi‐centre, double‐blind, randomized trial of Saroglitazar 4 mg compared to placebo in patients with nonalcoholic steatohepatitis**. *Hepatol. Int.* (2020) **14** S326
12. Ducancelle A, Leroy V, Vergniol J. **A Single Test Combining Blood Markers and Elastography is More Accurate Than Other Fibrosis Tests in the Main Causes of Chronic Liver Diseases**. *J. Clin. Gastroenterol.* (2017) **51** 639-49. PMID: 28692443
13. Das K, Sarkar R, Ahmed SM. **"Normal" liver stiffness measure (LSM) values are higher in both lean and obese individuals: a population‐based study from a developing country**. *Hepatology* (2012) **55** 584-93. PMID: 21952989
14. Yoshioka K, Hashimoto S, Kawabe N. **Measurement of liver stiffness as a non‐invasive method for diagnosis of non‐alcoholic fatty liver disease**. *Hepatol. Res.* (2015) **45** 142-51. PMID: 25040931
15. Petta S, Vanni E, Bugianesi E. **The combination of liver stiffness measurement and NAFLD fibrosis score improves the noninvasive diagnostic accuracy for severe liver fibrosis in patients with nonalcoholic fatty liver disease**. *Liver Int.* (2015) **35** 1566-73. PMID: 24798049
16. Gawrieh S, Noureddin M, Loo N. **Saroglitazar, a PPAR‐α/γ Agonist, for Treatment of NAFLD: A Randomized Controlled Double‐Blind Phase 2 Trial**. *Hepatology* (2021) **74** 1809-24. PMID: 33811367
17. Goyal O, Nohria S, Goyal P. **Saroglitazar in patients with non‐alcoholic fatty liver disease and diabetic dyslipidemia: a prospective, observational, real world study**. *Sci. Rep.* (2020) **10** 21117. PMID: 33273703
|
---
title: Examining the influence of inflammatory bowel disease medications on sleep
quality
authors:
- Alex Barnes
- Paul Spizzo
- Peter Bampton
- Jane M Andrews
- Robert J Fraser
- Sutapa Mukherjee
- Réme Mountifield
journal: 'JGH Open: An Open Access Journal of Gastroenterology and Hepatology'
year: 2023
pmcid: PMC10037038
doi: 10.1002/jgh3.12871
license: CC BY 4.0
---
# Examining the influence of inflammatory bowel disease medications on sleep quality
## Abstract
In a large population of people with Inflammatory bowel disease, poor sleep was more common in those on opioids, infliximab, corticosteroids, vitamin D, and methotrexate. Opioids remained associated with poor sleep when adjusting for other influences of sleep quality including abdominal pain.
### Background and Aim
Inflammatory bowel disease (IBD) can disrupt sleep, leading to poor sleep quality. This may in part be due to the symptoms of IBD and the influence of pro‐inflammatory cytokines on sleep. This study aimed to investigate the potential influence of IBD medications on sleep quality.
### Methods
An online survey of adults with IBD was conducted, which included measures of sleep quality, IBD activity, anxiety, depression, and physical activity. Logistic regression was used to investigate possible associations between IBD medications (corticosteroids, immunomodulators, biologics, aminosalicyate) and outcome of poor sleep. A generalized linear model was built for outcome of sleep quality score.
### Results
There were 544 participants included in the final analysis, median age of 42, and $61\%$ with Crohn's disease. Increased odds of poor sleep were seen in those taking opioids, medications for anxiety or depression, corticosteroids, vitamin D, methotrexate, and infliximab. A multivariate model was built incorporating demographic and IBD variables with opioids present in the final model and associated with increased odds of poor sleep. This was in addition to medications for sleep, depression, anxiety, IBD activity, and body weight. In a multivariate generalized linear model, opioids and methotrexate were associated with worse sleep quality scores.
### Conclusions
Opioids were associated with increased odds of poor sleep independent of other factors. This provides further support for avoiding these medications in people with IBD. Infliximab was associated with increased body weight and consequently increased odds of poor sleep.
## Introduction
Sleep is an essential biologic function with an important role in overall health. Abnormal sleep has been linked to poor health outcomes including cardiovascular disease, 1 metabolic syndrome, 2 and increased all‐cause mortality in some studies, 3 in addition to significant economic cost in the form of decreased productivity and increased health care utilization. 4 Sleep has been shown to regulate a number of gastrointestinal functions including gastrointestinal motility and secretion. 5 Sleep disruption has been associated with increased levels of inflammatory cytokines, such as IL‐6, and TNF‐α, that have been implicated in the pathogenesis of inflammatory bowel disease. 6, 7, 8 Inflammatory bowel disease (IBD) is a relapsing–remitting autoimmune disorder that results from a complex interaction between genetics and the environment. 9 Poor sleep is prevalent in people with IBD with a recent meta‐analysis suggesting a pooled prevalence of $56\%$. 10 IBD may impair sleep through its myriad of disabling symptoms, including abdominal pain and nocturnal diarrhea. 11 Poor sleep is more common in those with IBD than controls, 11 more common in those with active IBD than inactive IBD, 11, 12 and remains more common in those with inactive IBD than controls. 13 Endoscopically or histologically active IBD in the absence of any IBD symptoms may be sufficient to disrupt sleep. 14, 15 There have been several association studies of sleep and IBD with comorbid depression 16, 17, 18, 19, 20, 21, 22, 23 frequently associated with poor sleep, and low physical activity associated with poor sleep. 19, 24 The effect of IBD medications on sleep has been investigated with a prospective study following the introduction of a biologic medication with subsequent measurement of subjective sleep quality improving, 20 likely accompanied by an improvement in IBD activity. Other cross‐sectional studies have been unable to demonstrate a relationship between biologics, immunomodulators, and sleep quality, 14, 16, 25 although these studies may have been underpowered. Current use of corticosteroids was associated with worse sleep quality although confounded by IBD activity, 15, 23 however, this was not replicated in other studies. 16, 26, 27 Sleep, being an immunologically active state, may be influenced by medications that alter the immune system such as TNF‐a inhibitors. 28 In people with rheumatoid arthritis, 29 infliximab, a TNF‐a inhibitor also commonly used in IBD, 30 was observed to improve some aspects of sleep quality and reduce daytime sleepiness. 31 Adalimumab, another TNF‐a inhibitor commonly used to treat IBD, 32 was associated with improved sleep quality in people with psoriasis, 33 and ankylosing spondylitis. 34
This study aims to explore the relationship between medications used by people with IBD and sleep quality. It will also consider other influences of sleep quality such as IBD activity, physical activity, and mental health.
## Methods
An online questionnaire was made available to people with IBD via tertiary hospital patient email lists, private gastroenterology practice email lists, and social media. This study received ethics approval from the Southern Adelaide Human Research Ethics Committee (203.20). Individuals with a self‐reported diagnosis of IBD over 18 years of age were invited to participate. Demographic data such as age and sex were recorded, along with data on IBD, which included disease duration and previous surgery. Current medications were recorded including those specifically for IBD, sleep, mental health, and pain control. Medications for sleep were subcategorized as melatonin or, benzodiazepines and zolpidem.
## Sleep quality
The Pittsburgh Sleep Quality Index (PSQI) is a validated tool that assesses perceived sleep quality. 35 The index consists of subscales on sleep duration, sleep disturbance, sleep latency, daytime dysfunction, sleep efficiency, overall sleep quality, and medications for sleep. The score ranges from 0 to 21, with a PSQI >5 considered to represent poor sleep quality.
The mean (SD) PSQI for the cohort was 8.80 (4.56). In reference to different IBD medications (Table 2), the mean PSQI was higher in those on opioids, medication for anxiety or depression, benzodiazepines or zolpidem, melatonin, and corticosteroids ($P \leq 0.001$ for all). PSQI subscales for medications with a higher PSQI are detailed in Table S1, Supporting information. Corticosteroids were associated with worse sleep efficiency, increased sleep duration, and worse sleep disturbance. Opioids impacted all PSQI subscales apart from need for medications for sleep.
**Table 2**
| Medication | PSQI (mean, 95% CI) | P value |
| --- | --- | --- |
| Opioids | Yes: 11.92 (11.09–12.75) | <0.0001 |
| Opioids | No: 9.05 (8.71–9.40) | |
| Anti‐anxiety or anti‐depressant | Yes: 10.53 (9.79–11.27) | 0.0005 |
| Anti‐anxiety or anti‐depressant | No: 9.15 (8.78–9.51) | |
| Medications for sleep | Yes: 11.82 (11.05–12.60) | <0.001 |
| Medications for sleep | No: 9.09 (8.75–9.45) | |
| Benzodiazepines or zolpidem | Yes: 12.06 (11.03–13.09) | <0.0001 |
| Benzodiazepines or zolpidem | No: 9.21 (8.87–9.56) | |
| Melatonin | Yes: 11.23 (10.22–12.24) | 0.0064 |
| Melatonin | No: 9.35 (9.01–9.69) | |
| Vitamin D | Yes: 9.77 (9.16–10.38) | 0.26 |
| Vitamin D | No: 9.35 (8.96–9.75) | |
| 5ASA medication | Yes: 9.25 (8.70–9.80) | 0.35 |
| 5ASA medication | No: 9.58 (9.17–9.99) | |
| Corticosteroids | Yes: 11.09 (9.97–12.21) | 0.0014 |
| Corticosteroids | No: 9.29 (8.95–9.64) | |
| Immunomodulators | Yes: 9.47 (8.90–10.04) | 0.99 |
| Immunomodulators | No: 9.47 (9.06–8.87) | |
| Thiopurine | Yes: 9.03 (8.39–9.66) | 0.0942 |
| Thiopurine | No: 9.65 (9.26–10.04) | |
| Methotrexate | Yes: 11.07 (9.84–12.29) | 0.0051 |
| Methotrexate | No: 9.33 (8.99–9.68) | |
| Biologics | Yes: 9.33 (8.89–9.78) | 0.38 |
| Biologics | No: 9.63 (9.13–10.13) | |
| Anti‐TNF | Yes: 9.51 (8.92–10.09) | 0.88 |
| Anti‐TNF | No: 9.45 (9.05–9.86) | |
Logistic regression was performed for outcome of poor sleep (PSQI >5) (see Table 3), with increased odds of poor sleep seen in those on opioids, medications for sleep including zolpidem and benzodiazepines, medications for anxiety or depression, corticosteroids, vitamin D, methotrexate, and infliximab but not other biologics. No medication was associated with decreased odds of poor sleep. All those on melatonin had poor sleep. Considering PSQI subscales, infliximab was associated with higher sleep disturbance scores and higher scores for needing medications for sleep (see Table S1). Considering a subgroup of those not on any medications for sleep, infliximab remained associated with poor sleep. Methotrexate had higher daytime dysfunction scores, worse sleep efficiency scores, and worse sleep quality scores (see Table S1).
**Table 3**
| Medication | Poor sleep |
| --- | --- |
| Opioids | 6.95 (2.49–19.37) P < 0.001 |
| Anti‐anxiety or anti‐depressant | 1.72 (1.03–2.85) P = 0.035 |
| Medications for sleep | 13.88 (3.36–57.31) P < 0.001 |
| 5ASA medication | 1.16 (0.77–1.76) P = 0.47 |
| Vitamin D | 1.98 (1.22–3.23) P = 0.006 |
| Corticosteroids | 2.69 (1.13–6.45) P = 0.026 |
| Immunomodulators | 1.28 (0.85–1.92) P = 0.23 |
| Methotrexate | 3.34 (1.17–9.52) P = 0.024 |
| Thiopurine | 0.97 (0.63–1.49) P = 0.89 |
| Biologics | 1.43 (0.95–2.10) P = 0.067 |
| Adalimumab | 0.85 (0.49–1.46) P = 0.56 |
| Infliximab | 2.02 (1.11–3.69) P = 0.022 |
| Vedolizumab | 0.90 (0.047–1.76) P = 0.77 |
| Ustekinumab | 1.54 (0.80–2.97) P = 0.19 |
| Tofacitinib | 0.64 (0.12–3.52) P = 0.61 |
Combinations of IBD medications were also considered (see Table S2). All of the cohort on opioids and either methotrexate or infliximab had poor sleep. The combination of methotrexate and infliximab did not reach significance for an association with poor sleep ($$P \leq 0.094$$). There was no association with poor sleep seen for combinations of aminosalicyates, biologics, and immunomodulators.
## IBD disease activity
IBD disease activity was assessed using the Harvey Bradshaw Index in the case of Crohn's disease with HBI >5 considered active disease, 36 and the Simple Clinical Colitis Activity Index (SCCAI) in the case of ulcerative colitis. An SCCAI >2 was considered active disease. 37
## Physical activity
Physical activity was assessed using the international physical activity questionnaire short form (IPAQ‐SF). 38 This allows the calculation of metabolic equivalent of task (MET) values over a one‐week period of walking, moderate and vigorous activity, along with sitting time.
Physical activity as measured by total METs, sitting time, and vigorous METs was not associated with any sleep quality measure. Further analysis was consequently not undertaken.
## Anxiety and depression
Anxiety was assessed using the generalized anxiety disorder 7‐item scale (GAD‐7) 39 with a score over 10 used to indicate clinically significant anxiety. The Patient Health Questionnaire 9 (PHQ‐9) was used to assess depression with a score over 15 used to indicate clinically significant depression. 40
## Statistical analysis
Statistical analysis was performed using Stata SE 16 (StataCorp, College Station, TX, USA). Inadequate completion of a score or index led to that result not being included. For normally distributed variables, mean and SD were reported, with comparisons made using the Student t‐test. For non‐normally distributed variables, median and interquartile range (IQR) were reported, with comparisons made using the Mann–Whitney U test. For categorical data, Pearson's χ 2 test was used or Fisher's exact test when appropriate. Logistic regression was performed for an outcome of poor sleep (PSQI >5). Logistic regression was used to calculate adjusted odds ratios for known IBD activity, anxiety, and depression. A multivariate logistic regression model was built for outcomes of poor sleep including demographic variables. This was model optimized by sequentially adding and removing variables to maximize the likelihood function. A generalized linear model was also constructed for outcome of raw PSQI score with univariate and multivariate regression performed with this optimized by the Bayesian information criterion.
## Results
There were 544 participants who completed the questionnaire. The completion rate for the survey was $93\%$. Given the method of survey distribution, we are unable to estimate the response rate. The mean age was 42 years (SD 13), $61\%$ had Crohn's disease, and median disease duration was 10 years (IQR 3–17). The mean HBI was 7.2 (3.1) and SCCAI was 7.2 (2.8), with clinically active IBD in $64\%$. IBD‐related medications included biologics in $54.6\%$ of the cohort, immunomodulators in $37.1\%$, 5ASA in $35.4\%$, corticosteroids in $10.1\%$, and immunomodulator in combination with a biologic in $20.5\%$ (see Table 1).
**Table 1**
| n | 544 |
| --- | --- |
| Age, mean (SD) | 42 (13) |
| Female gender, n | 436 |
| Weight (kg), mean (SD) | 78.9 (20.4) |
| Height (cm), mean (SD) | 167.7 (8.9) |
| Crohn's disease, n | 333 |
| Ulcerative colitis, n | 218 |
| Indeterminate colitis, n | 15 |
| Disease duration (years, median [IQR]) | 10 (3–17) |
| Previous surgery for IBD, n | 183 |
| Corticosteroids | Corticosteroids |
| Budesonide, n (%) | 11 (2) |
| Prednisolone, n (%) | 44 (8) |
| Biologics | Biologics |
| Adalimumab, n (%) | 79 (14) |
| Infliximab, n (%) | 95 (17) |
| Ustekinumab, n (%) | 67 (12) |
| Vedolizumab, n (%) | 50 (9) |
| Tofacitinib, n (%) | 6 (1) |
| Immunomodulator | Immunomodulator |
| Azathioprine, n (%) | 105 (19) |
| Mercaptopurine, n (%) | 53 (10) |
| Methotrexate, n (%) | 44 (8) |
| Aminosalicyate | Aminosalicyate |
| Mesalazine, n (%) | 172 (31) |
| Sulfasalazine, n (%) | 21 (4) |
| Other IBD medication | Other IBD medication |
| Bactrim, n (%) | 1 (0.01) |
| Cyclosporine, n (%) | 0 |
| Tacrolimus, n (%) | 2 (0.03) |
| Immunomodulator and biologic, n (%) | 112 (20) |
| Aminosalicyate and biologic, n (%) | 50 (9) |
| Aminosalicyate, immunomodulator and biologic, n (%) | 28 (5) |
| Vitamin D, n | 152 (28) |
| Medications for sleep | Medications for sleep |
| Melatonin, n (%) | 34 (6) |
| Benzodiazepines or zolpidem, n (%) | 49 (9) |
| Medications for depression or anxiety, n (%) | 128 (23) |
| Opioids, n (%) | 78 (14) |
| Overnight shift work, n (%) | 31 (6) |
## Clinically active IBD
Clinically active IBD was defined as SCCAI >2 or HBI >5, mean SCCAI was 5.7 (4.1), and mean HBI was 5.7 (4.2). Clinically active IBD was associated with poor sleep (see Table S3). Logistic regression was used to calculate adjusted odds ratios by IBD activity for outcomes of poor sleep (see Table 4). Adjusted odd ratios were no longer significant for corticosteroids and methotrexate. Opioids, infliximab, medications for sleep, and vitamin D remained significantly associated with increased odds of poor sleep.
**Table 4**
| Medication | Active IBD | Significant depression | Significant anxiety |
| --- | --- | --- | --- |
| Opioids | 6.19 (2.19–17.53) P = 0.001 | 7.27 (2.58–20.41) P < 0.001 | 7.33 (2.60–20.63) P < 0.001 |
| Infliximab | 2.02 (1.08–3.77) P = 0.028 | 2.24 (1.21–4.12) P = 0.010 | 2.19 (1.18–4.06) P = 0.013 |
| Methotrexate | 2.68 (0.92–7.82) P = 0.072 | 3.21 (1.1–9.28) P = 0.027 | 3.48 (1.20–10.07) P = 0.021 |
| Corticosteroids | 2.39 (0.97–5.87) P = 0.056 | 2.39 (0.98–5.82) P = 0.054 | 2.14 (0.87–5.26) P = 0.096 |
| Medications for sleep | 12.13 (2.90–50.69) P = 0.001 | 14.7 (3.55–61.09) P < 0.001 | 13.40 (3.22–55.8) P < 0.001 |
| Medications for anxiety or depression | 1.29 (0.76–2.19) P = 0.34 | 1.35 (0.80–2.29) P = 0.25 | 1.37 (0.81–2.32) P = 0.23 |
| Vitamin D | 1.87 (1.13–3.10) P = 0.015 | 1.89 (1.15–3.10) P = 0.012 | 1.97 (1.19–3.24) P = 0.008 |
## Depression
Clinically significant depression (PHQ‐9 >15) was associated with poor sleep, (see Table S3). Logistic regression was used to calculate adjusted odds ratios by depression (PHQ‐9 >15) for outcomes of poor sleep (see Table 4). After adjustment, corticosteroids were no longer significant ($$P \leq 0.054$$), and other medications remained significantly associated with poor sleep.
## Anxiety
Clinically significant anxiety (GAD‐7 >10) was associated with poor sleep (see Table S3). Logistic regression was used to calculate adjusted odds ratios by anxiety (GAD‐7 >10) for outcomes of poor sleep (see Table 4). After adjustment, corticosteroids were no longer significant ($$P \leq 0.096$$), and other medications remained significantly associated with poor sleep.
## Multivariate regression
A multivariate model was constructed for outcome of poor sleep that included medications—opioids, medications for sleep, Infliximab, and vitamin D. The model also included demographic variables and IBD‐related variables such as disease duration (for univariate logistic regression see Table S4). In the final model (see Table 5), opioid usage and medications for sleep remained associated with increased odds of poor sleep. Infliximab and vitamin D were not included in the final model. Other variables in the final model included body weight, IBD disease duration, clinically significant anxiety, clinically significant depression and clinically active IBD.
**Table 5**
| Variable | Odds ratio | Confidence interval | P value |
| --- | --- | --- | --- |
| Opioids | 3.08 | 1.04–9.122 | 0.041 |
| Benzodiazepines or zolpidem | 9.21 | 2.08–40.86 | 0.003 |
| Weight | 1.02 | 1.01–1.03 | <0.001 |
| IBD disease duration | 1.02 | 1.00–1.04 | 0.042 |
| Clinically significant anxiety | 3.82 | 1.88–7.78 | <0.001 |
| Clinically significant depression | 3.67 | 1.21–11.08 | 0.021 |
| Active IBD | 2.56 | 1.64–4.00 | <0.001 |
Infliximab was not significantly associated with poor sleep when adjusted for by body weight (see Table S5). People on infliximab had a higher body weight than the remainder of the cohort (79.56 vs 72.45, $$P \leq 0.024$$). Infliximab remained significantly associated with poor sleep when adjusted by other variables in the final model excluding body weight (Table S5). This was similarly observed with vitamin D with those on vitamin D having a higher body weight than the remainder of the cohort (80.43 vs 71.19, $$P \leq 0.0005$$). Vitamin D remained significantly associated with poor sleep when adjusted by the other variables in the final model excluding body weight (see Table S5). Sub scores from IBD clinical activity were considered for abdominal pain and nocturnal diarrhea. Opioids were associated with abdominal pain ($P \leq 0.001$) but not nocturnal diarrhea ($$P \leq 0.19$$). Adjusted odds ratio for poor sleep for those on opioids remained significant after adjustment for abdominal pain. Opioids were associated with longer IBD disease duration (14.6 [11.8–17.5] vs 11.5 [10.6–12.4], $$P \leq 0.014$$), and higher SCCAI scores ($P \leq 0.012$) or HBI scores ($P \leq 0.0001$).
A generalized linear model was constructed for outcome of PSQI score with univariate (see Table S6) and multivariate regression performed (see Table S7). With respect to IBD medications, the univariate regression was again significant for methotrexate and corticosteroids with increased odds of poor sleep but not infliximab and vitamin D. The final multivariate model included methotrexate in addition to opioids and medications for sleep including melatonin.
## Discussion
Here we have described the results of an online questionnaire demonstrating a relationship between IBD medications and sleep quality in people with IBD. Opioids, commonly prescribed in IBD, were associated with increased odds of poor sleep as part of a multivariate model including clinically active IBD, body weight, depression, anxiety, IBD disease duration, and medications for sleep. Infliximab and vitamin D were associated with poor sleep, but this appeared to be confounded by body weight, with both medications not included in the final multivariate model. Methotrexate was associated with higher PSQI scores. This study builds on previous work that did not show any significant relationship between sleep quality and biologics or immunomodulators. 25 Chronic opioid usage in people with IBD has been associated with increased all‐cause mortality, 41, 42 worse IBD outcomes such as infection, 43 worse quality of life, 44 and increased health care utilization. 45 In our population, opioids were associated with worse sleep quality. Opioids are known to alter sleep architecture 46 and are associated with sleep disordered breathing, 47 in particular central sleep apnoea. 48 We also note that opioids may be a marker of more severe IBD. 49 In our study, opioids were associated with longer disease duration and higher clinical disease activity scores. Opioids remained associated with poor sleep following adjustment for abdominal pain; however, it is possible that other types of pain contributed to sleep quality that was not accounted for.
Infliximab has been associated with weight gain in people with IBD, 50, 51, 52, 53, 54, 55 with the suggestion that infliximab may inhibit leptin production. 56 Infliximab‐related weight gain has also been observed in cohorts of people with rheumatoid arthritis 57, 58, 59 and psoriasis. 60, 61, 62 Vitamin D deficiency has been associated with obesity, 63, 64 although to the authors' knowledge there is no known association between vitamin D replacement and weight gain. 65 Increased body weight is a risk factor for sleep apnoea 66 and perhaps the associated weight gain from infliximab or vitamin D deficiency increases the likelihood of sleep apnea and consequently more likely to have poor sleep. Vitamin D deficiency has been associated with increased risk of sleep disorders 67 and supplementation has been associated with improvement in sleep quality. 68 Corticosteroids, known to cause sleep disturbances, 69 were associated with worse sleep quality scores and poor sleep. This replicates previous work showing worse sleep quality scores 15, 23 in those on corticosteroids. The association with poor sleep was confounded by firstly IBD activity and also mental health scores, of which corticosteroids are well known to influence. 70 Methotrexate was associated with higher PSQI scores on multivariate regression but not increased odds of poor sleep on multivariate regression. This may be due to the small number of participants on methotrexate ($8\%$) and consequent vulnerability to some yet unidentified bias. Methotrexate is associated with fatigue, 71, 72 which commonly limits its use. Associations studies in a rheumatoid arthritis population have not demonstrated a relationship between sleep quality and methotrexate; 73 however, in other prospective studies, introduction of methotrexate did not lead to any improvement in sleep quality—unlike introduction of TNF‐a inhibitors. 74, 75 Limitations of this study include selection bias as a result of the use of an online questionnaire that may attract people with sleep problems. The rate of poor sleep in this cohort ($75\%$) was higher than that reported in a recent meta‐analysis on the prevalence of poor sleep in IBD 10 ($56\%$), although a number of other studies have reported higher rates of poor sleep 15, 21, 25, 76, 77 than seen in our cohort. Similarly, the form of survey and method of recruitment is likely responsible for the predominantly female cohort. Reporting bias may also be significant, noting a study of people with Crohn's disease reported worse sleep quality than that observed by objective measures. 26 The absence of objective measures of sleep quality and objective IBD activity is also considered a limitation. Further studies should consider objective measures of IBD activity and sleep quality. Studies incorporating mental health interventions in those with poor sleep should be pursued. Consideration should also be given to examining the relationship between serum levels of vitamin D and sleep quality in an IBD population.
## Conclusions
A large survey of people with IBD has shown that opioids are associated with increased odds of poor sleep. Infliximab and vitamin D usage was associated with poor sleep; however, this was confounded by higher body weight and consequently perhaps increased rates of sleep apnea. Further studies are required to confirm these results and should incorporate objective measures of sleep quality and IBD activity.
## Data availability statement
The data underlying this article are available upon request to the author.
## References
1. Daghlas I, Dashti H, Lane J. **Sleep duration and myocardial infarction**. *J. Am. Coll. Cardiol.* (2019) **74** 1304-14. PMID: 31488267
2. Deng H, Tam T, Zee B. **Short sleep duration increases metabolic impact in healthy adults: a population‐based cohort study**. *Sleep* (2017) **40** 133-144
3. Cappuccio F, D'Elia L, Strazzullo P, Miller MA. **Sleep duration and all‐cause mortality: a systematic review and meta‐analysis of prospective studies**. *Sleep* (2010) **33** 585-92. PMID: 20469800
4. Hillman D, Mitchell S, Streatfeild J, Burns C, Bruck D, Pezzullo L. **The economic cost of inadequate sleep**. *Sleep* (2018) **41** 1-13
5. Orr WC, Fass R, Sundaram SS, Scheimann AO. **The effect of sleep on gastrointestinal functioning in common digestive diseases**. *Lancet Gastroenterol. Hepatol.* (2020) **5** 616-24. PMID: 32416862
6. Vgontzas AN, Zoumakis E, Bixler EO. **Adverse effects of modest sleep restriction on sleepiness, performance, and inflammatory cytokines**. *J. Clin. Endocrinol. Metab.* (2004) **89** 2119-26. PMID: 15126529
7. Shoham SH, Davenne DA, Cady AB, Dinarello CA, Krueger JM. **Recombinant tumor necrosis factor and interleukin 1 enhance slowwave sleep**. *Am. J. Physiol.* (1987) **253** R142-9. PMID: 3496800
8. Majde JA, Krueger JM. **Links between the innate immune system and sleep**. *J. Allergy Clin. Immunol.* (2005) **116** 1188-98. PMID: 16337444
9. Ananthakrishnan AN. **Environmental risk factors for inflammatory bowel diseases: a review**. *Dig. Dis. Sci.* (2015) **60** 290-8. PMID: 25204669
10. Barnes A, Mountifield R, Baker J. **A systematic review and meta‐analysis of the prevalence of poor sleep in inflammatory bowel disease**. *Sleep Advances* (2022) **3**
11. Hao G, Zhu B, Li Y, Hou L. **Sleep quality and disease activity in patients with inflammatory bowel disease: a systematic review and meta‐analysis**. *Sleep Med.* (2020) **75** 301-8. PMID: 32947172
12. Ballesioa A, Zagariaa A, Baccinib F, Micheli F, Di Nardo G, Lombardo C. **A meta‐analysis on sleep quality in inflammatory bowel disease**. *Sleep Med. Rev.* (2021) **60** 301-8
13. Barnes A, Mountifield R, Baker J, Spizzo P, Bampton P, Mukherjee S. **Systematic review and meta‐analysis of sleep quality in inactive inflammatory bowel disease**. *JGH Open* (2022) **6** 738-44. PMID: 36406652
14. Michalopoulos G, Vrakas S, Makris K, Tzathas C. **Association of sleep quality and mucosal healing in patients with inflammatory bowel disease in clinical remission**. *Ann. Gastroenterol.* (2018) **31** 211-16. PMID: 29507468
15. Ali T, Madhoun M, Orr W, Rubin DT. **Assessment of the relationship between quality of sleep and disease activity in inflammatory bowel disease patients**. *Inflamm. Bowel Dis.* (2013) **19** 2440-3. PMID: 23945186
16. Sochal M, Małecka‐Panas E, Gabryelska A. **Determinants of sleep quality in inflammatory bowel diseases**. *J. Clin. Med.* (2020) **9** 2921. PMID: 32927760
17. Hood M, Wilson R, Gorenz A. **Sleep quality in ulcerative colitis: associations with inflammation**. *Psychol. Distress Qual. Life.* (2018) **25** 517-25
18. Marinelli C, Savarino E, Marsilio I. **Sleep disturbance in Inflammatory bowel disease: prevalence and risk factors—a cross‐sectional study**. *Sci. Rep.* (2020) **10** 507. PMID: 31949257
19. Garcia Calvo E, DurÀ Gil M, Velayos Jimenez B. **Prevalence and factors associated with poor sleep quality in inflammatory bowel disease outpatients**. *Revista espanola de enfermedades digestivas* (2020) **113** 512-18
20. Stevens B, Borren N, Velonias G. **Vedolizumab therapy is associated with an improvement in sleep quality and mood in inflammatory bowel diseases**. *Dig. Dis. Sci.* (2017) **62** 206
21. Gîlc‐Blanariu G, Ștefnescu G, Trifan A. **Sleep impairment and psychological distress among patients with inflammatory bowel disease‐beyond the obvious**. *J. Clin. Med.* (2020) **9** 2304. PMID: 32698475
22. Wilson R, Stevens B, Guo A. **High C‐reactive protein is associated with poor sleep quality independent of nocturnal symptoms in patients with inflammatory bowel disease**. *Dig. Dis. Sci.* (2015) **60** 43
23. Ananthakrishnan A, Long M, Martin C, Sandler RS, Kappelman MD. **Sleep disturbance and risk of active disease in patients with Crohn's disease and ulcerative colitis**. *Clin. Gastroenterol. Hepatol.* (2013) **11** 965-71. PMID: 23376797
24. Dolezal BA, Neufeld EV, Boland DM, Martin JL, Cooper CB. **Interrelationship between sleep and exercise: a systematic review**. *Adv Prev Med* (2017) **2017**. PMID: 28458924
25. Lee A, Kraemer D, Kanar O, Berry AC, Smotherman C, Eid E. **Immunomodulator and biologic agent effects on sleep quality in patients with inflammatory bowel disease**. *Oschner J.* (2018) **18** 80
26. Iskandar H, Linan E, Patel A. **Self‐reported sleep disturbance in Crohn's disease is not confirmed by objective sleep measures**. *Sci. Rep.* (2020) **10** 1980. PMID: 32029832
27. Uemura R, Fujiwara Y, Iwakura N. **Sleep disturbances in Japanese patients with inflammatory bowel disease and their impact on disease flare**. *Springer plus* (2016) **13** 1792
28. Rockstrom MD, Chen L, Taishi P. **Tumor necrosis factor alpha in sleep regulation**. *Sleep Med. Rev.* (2018) **40** 69-78. PMID: 29153862
29. Zamarrón C, Maceiras F, Mera A, Gomez‐Reino JJ. **Effect of the first infliximab infusion on sleep and alertness in patients with active rheumatoid arthritis**. *Ann. Rheum. Dis.* (2004) **63** 88-90. PMID: 14672898
30. Papamichael K, Lin S, Moore M, Papaioannou G, Sattler L, Cheifetz AS. **Infliximab in inflammatory bowel disease**. *Ther. Adv. Chronic Dis.* (2019) **10**. PMID: 30937157
31. Wolfe F, Michaud K, Li T. **Sleep disturbance in patients with rheumatoid arthritis: evaluation by medical outcomes study and visual analog sleep scales**. *J. Rheumatol.* (2006) **33** 1942-51. PMID: 16960928
32. Wasan SK, Kane SV. **Adalimumab for the treatment of inflammatory bowel disease**. *Expert Rev. Gastroenterol. Hepatol.* (2011) **5** 679-84. PMID: 22017695
33. Strober BE, Sobell JM, Duffin KC. **Sleep quality and other patient‐reported outcomes improve after patients with psoriasis with suboptimal response to other systemic therapies are switched to adalimumab: results from PROGRESS, an open‐label Phase IIIB trial**. *Br. J. Dermatol.* (2012) **167** 1374-81. PMID: 22897348
34. Rudwaleit M, Gooch K, Michel B. **Adalimumab improves sleep and sleep quality in patients with active ankylosing spondylitis**. *J. Rheumatol.* (2011) **38** 79-86. PMID: 20952469
35. Buysse DJ, Reynolds CF, Monk TH. **The Pittsburgh Sleep Quality Index: a new instrument for psychiatric practice and research**. *Psychiatry Res.* (1989) **28** 193-213. PMID: 2748771
36. Harvey RF, Bradshaw JM. **A simple index of Crohn's‐disease activity**. *Lancet* (1980) **8** 514
37. Walmsley RS, Ayres RC, Pounder RE. **A simple clinical colitis activity index**. *Gut* (1998) **43** 29-32. PMID: 9771402
38. Craig CL, Marshall AL, Sjöström M. **International physical activity questionnaire: 12‐country reliability and validity**. *Med. Sci. Sports Exerc.* (2003) **35** 1381-95. PMID: 12900694
39. Spitzer RL, Kroenke K, Williams JB. **A brief measure for assessing generalized anxiety disorder: the GAD‐7**. *Arch. Intern. Med.* (2006) **166** 1092-7. PMID: 16717171
40. Kroenke K, Spitzer RL, Williams JB. **The PHQ‐9: validity of a brief depression severity measure**. *J. Gen. Intern. Med.* (2001) **16** 606-13. PMID: 11556941
41. Burr NE, Smith C, West R, Hull MA, Subramanian V. **Increasing prescription of opiates and mortality in patients with inflammatory bowel diseases in England**. *Clin. Gastroenterol. Hepatol.* (2018) **16** 534-541.e6. PMID: 29079223
42. Targownik LE, Nugent Z, Singh H, Bugden S, Bernstein CN. **The prevalence and predictors of opioid use in inflammatory bowel disease: a population‐based analysis**. *Am. J. Gastroenterol.* (2014) **109** 1613-20. PMID: 25178702
43. Lichtenstein GR, Feagan BG, Cohen RD. **Serious infections and mortality in association with therapies for Crohn's disease: TREAT registry**. *Clin. Gastroenterol. Hepatol.* (2006) **4** 621-30. PMID: 16678077
44. Sanford D, Thornley P, Teriaky A, Chande N, Gregor J. **Opioid use is associated with decreased quality of life in patients with Crohn's disease**. *Saudi J. Gastroenterol.* (2014) **20** 182-7. PMID: 24976282
45. Click B, Ramos Rivers C, Koutroubakis IE. **Demographic and clinical predictors of high healthcare use in patients with inflammatory bowel disease**. *Inflamm. Bowel Dis.* (2016) **22** 1442-9. PMID: 26950309
46. Dimsdale JE, Norman D, DeJardin D. **The effect of opioids on sleep architecture**. *J. Clin. Sleep Med.* (2007) **3** 33-6. PMID: 17557450
47. Teichtahl H, Prodromidis A, Miller B, Cherry G, Kronborg I. **Sleep‐disordered breathing in stable methadone programme patients: a pilot study**. *Addiction* (2001) **96** 395-403. PMID: 11255580
48. Wang D, Teichtahl H, Drummer O. **Central sleep apnea in stable methadone maintenance treatment patients**. *Chest* (2005) **128** 1348-56. PMID: 16162728
49. Hanson KA, Loftus EV, Harmsen WS. **Clinical features and outcome of patients with inflammatory bowel disease who use narcotics: a case‐control study**. *Inflamm. Bowel Dis.* (2009) **15** 772-7. PMID: 19107782
50. Lepp J, Höög C, Forsell A, Fyrhake U, Lördal M, Almer S. **Rapid weight gain in infliximab treated Crohn's disease patients is sustained over time: real‐life data over 12 months**. *Scand. J. Gastroenterol.* (2020) **55** 1411-18. PMID: 33108226
51. Wiese D, Lashner B, Seidner D. **Measurement of nutrition status in Crohn's disease patients receiving infliximab therapy**. *Nutr. Clin. Pract.* (2008) **23** 551-6. PMID: 18849561
52. Nakahigashi M, Yamamoto T. **Increases in body mass index during infliximab therapy in patients with Crohn's disease: an open label prospective study**. *Cytokine* (2011) **56** 531-5. PMID: 21820319
53. Diamanti A, Basso MS, Gambarara M. **Positive impact of blocking tumor necrosis factor alpha on the nutritional status in pediatric Crohn's disease patients**. *Int. J. Colorectal Dis.* (2009) **24** 19-25. PMID: 18797887
54. Vadan R, Gheorghe LS, Constantinescu A, Gheorghe C. **The prevalence of malnutrition and the evolution of nutritional status in patients with moderate to severe forms of Crohn's disease treated with Infliximab**. *Clin. Nutr.* (2011) **30** 86-91. PMID: 20719413
55. Kaazan P, Tan Z, Bednarz J, Hughes J, Grafton R, Andrews JM. **Infliximab‐related weight gain in inflammatory bowel disease: associations and financial impacts**. *Intern. Med. J.* (2020) **50** 1134-8. PMID: 32929820
56. Franchimont D, Roland S, Gustot T. **Impact of infliximab on serum leptin levels in patients with Crohn's disease**. *J. Clin. Endocrinol. Metab.* (2005) **90** 3510-6. PMID: 15784704
57. Engvall IL, Tengstrand B, Brismar K, Hafström I. **Infliximab therapy increases body fat mass in early rheumatoid arthritis independently of changes in disease activity and levels of leptin and adiponectin: a randomised study over 21 months**. *Arthritis Res. Ther.* (2010) **12**. PMID: 20964833
58. Marcora SM, Chester KR, Mittal G, Lemmey AB, Maddison PJ. **Randomized phase 2 trial of anti‐tumor necrosis factor therapy for cachexia in patients with early rheumatoid arthritis**. *Am. J. Clin. Nutr.* (2006) **84** 1463-72. PMID: 17158431
59. Metsios GS, Stavropoulos‐Kalinoglou A, Douglas KM. **Blockade of tumour necrosis factor‐alpha in rheumatoid arthritis: effects on components of rheumatoid cachexia**. *Rheumatology* (2007) **46** 1824-7. PMID: 18032540
60. Mahé E, Reguiai Z, Barthelemy H. **Evaluation of risk factors for body weight increment in psoriatic patients on infliximab: a multicentre, cross‐sectional study**. *J. Eur. Acad. Dermatol. Venereol.* (2014) **28** 151-9. PMID: 23279264
61. Saraceno R, Schipani C, Mazzotta A. **Effect of anti‐tumor necrosis factor‐alpha therapies on body mass index in patients with psoriasis**. *Pharmacol. Res.* (2008) **57** 290-5. PMID: 18400510
62. Tan E, Baker C, Foley P. **Weight gain and tumour necrosis factor‐alpha inhibitors in patients with psoriasis**. *Australas. J. Dermatol.* (2013) **54** 259-63. PMID: 24164178
63. Walsh JS, Bowles S, Evans AL. **Vitamin D in obesity**. *Curr. Opin. Endocrinol. Diabetes Obes.* (2017) **24** 389-94. PMID: 28915134
64. Wortsman J, Matsuoka LY, Chen TC, Lu Z, Holick MF. **Decreased bioavailability of vitamin D in obesity**. *Am. J. Clin. Nutr.* (2000) **72** 690-3. PMID: 10966885
65. Duan L, Han L, Liu Q, Zhao Y, Wang L, Wang Y. **Effects of vitamin D supplementation on general and central obesity: results from 20 randomized controlled trials involving apparently healthy populations**. *Ann. Nutr. Metab.* (2020) **76** 153-64. PMID: 32645694
66. Peppard PE, Young T, Barnet JH, Palta M, Hagen EW, Hla KM. **Increased prevalence of sleep‐disordered breathing in adults**. *Am. J. Epidemiol.* (2013) **177** 1006-14. PMID: 23589584
67. Gao Q, Kou T, Zhuang B, Ren Y, Dong X, Wang Q. **The association between vitamin D deficiency and sleep disorders: a systematic review and meta‐analysis**. *Nutrients* (2018) **10** 1-13
68. Abboud M. **Vitamin D supplementation and sleep: a systematic review and meta‐analysis of intervention studies**. *Nutrients* (2022) **14** 1076. PMID: 35268051
69. Szmyd B, Rogut M, Białasiewicz P, Gabryelska A. **The impact of glucocorticoids and statins on sleep quality**. *Sleep Med. Rev.* (2021) **55**. PMID: 33010620
70. Warrington TP, Bostwick JM. **Psychiatric adverse effects of corticosteroids**. *Mayo Clin. Proc.* (2006) **81** 1361-7. PMID: 17036562
71. Nowell WB, Karis E, Gavigan K. **Patient‐reported nausea and fatigue related to methotrexate: a prospective, self‐controlled study in the ArthritisPower(®) registry**. *Rheumatol. Ther.* (2022) **9** 207-21. PMID: 34843092
72. Vaysse T, Carbonnel F. **Methotrexate in IBD: the return of the prodigal son**. *J. Crohn's Colitis* (2015) **9** 303-4. PMID: 25814523
73. Grabovac I, Haider S, Berner C. **sleep quality in patients with rheumatoid arthritis and associations with pain, disability, disease duration, and activity**. *J. Clin. Med.* (2018) **7** 7. PMID: 29300345
74. Detert J, Dziurla R, Hoff P. **Effects of treatment with etanercept versus methotrexate on sleep quality, fatigue and selected immune parameters in patients with active rheumatoid arthritis**. *Clin. Exp. Rheumatol.* (2016) **34** 848-56. PMID: 27385076
75. Krajewska‐Włodarczyk M, Owczarczyk‐Saczonek A, Placek W. **Sleep disorders in patients with psoriatic arthritis and psoriasis**. *Reumatologia* (2018) **56** 301-6. PMID: 30505012
76. Sofia M, Lipowska A, Zmeter N, Perez E, Kavitt R, Rubin DT. **Poor sleep quality in Crohn's disease is associated with disease activity and risk for hospitalization or surgery**. *Inflammatory Bowel Diseases* (2020) **26** 1251-9. PMID: 31820780
77. Zhang Y, Pi B, Xu X. **Sleep characteristics and influencing factors of sleep quality in patients with inflammatory bowel disease‐peripheral arthritis**. *Front. Med.* (2019) **6** 190
|
---
title: Human germline heterozygous gain-of-function STAT6 variants cause severe allergic
disease
authors:
- Mehul Sharma
- Daniel Leung
- Mana Momenilandi
- Lauren C.W. Jones
- Lucia Pacillo
- Alyssa E. James
- Jill R. Murrell
- Selket Delafontaine
- Jesmeen Maimaris
- Maryam Vaseghi-Shanjani
- Kate L. Del Bel
- Henry Y. Lu
- Gilbert T. Chua
- Silvia Di Cesare
- Oriol Fornes
- Zhongyi Liu
- Gigliola Di Matteo
- Maggie P. Fu
- Donato Amodio
- Issan Yee San Tam
- Gavin Shueng Wai Chan
- Ashish A. Sharma
- Joshua Dalmann
- Robin van der Lee
- Géraldine Blanchard-Rohner
- Susan Lin
- Quentin Philippot
- Phillip A. Richmond
- Jessica J. Lee
- Allison Matthews
- Michael Seear
- Alexandra K. Turvey
- Rachael L. Philips
- Terri F. Brown-Whitehorn
- Christopher J. Gray
- Kosuke Izumi
- James R. Treat
- Kathleen H. Wood
- Justin Lack
- Asya Khleborodova
- Julie E. Niemela
- Xingtian Yang
- Rui Liang
- Lin Kui
- Christina Sze Man Wong
- Grace Wing Kit Poon
- Alexander Hoischen
- Caspar I. van der Made
- Jing Yang
- Koon Wing Chan
- Jaime Sou Da Rosa Duque
- Pamela Pui Wah Lee
- Marco Hok Kung Ho
- Brian Hon Yin Chung
- Huong Thi Minh Le
- Wanling Yang
- Pejman Rohani
- Ali Fouladvand
- Hassan Rokni-Zadeh
- Majid Changi-Ashtiani
- Mohammad Miryounesi
- Anne Puel
- Mohammad Shahrooei
- Andrea Finocchi
- Paolo Rossi
- Beatrice Rivalta
- Cristina Cifaldi
- Antonio Novelli
- Chiara Passarelli
- Stefania Arasi
- Dominique Bullens
- Kate Sauer
- Tania Claeys
- Catherine M. Biggs
- Emma C. Morris
- Sergio D. Rosenzweig
- John J. O’Shea
- Wyeth W. Wasserman
- H. Melanie Bedford
- Clara D.M. van Karnebeek
- Paolo Palma
- Siobhan O. Burns
- Isabelle Meyts
- Jean-Laurent Casanova
- Jonathan J. Lyons
- Nima Parvaneh
- Anh Thi Van Nguyen
- Caterina Cancrini
- Jennifer Heimall
- Hanan Ahmed
- Margaret L. McKinnon
- Yu Lung Lau
- Vivien Béziat
- Stuart E. Turvey
journal: The Journal of Experimental Medicine
year: 2023
pmcid: PMC10037107
doi: 10.1084/jem.20221755
license: CC BY 4.0
---
# Human germline heterozygous gain-of-function STAT6 variants cause severe allergic disease
## Abstract
Sharma et al. define a new primary atopic disorder caused by heterozygous gain-of-function variants in STAT6. This results in severe, early-onset allergies, and is seen in 16 patients from 10 families. Anti–IL-4Rα antibody and JAK inhibitor treatment were highly effective.
STAT6 (signal transducer and activator of transcription 6) is a transcription factor that plays a central role in the pathophysiology of allergic inflammation. We have identified 16 patients from 10 families spanning three continents with a profound phenotype of early-life onset allergic immune dysregulation, widespread treatment-resistant atopic dermatitis, hypereosinophilia with esosinophilic gastrointestinal disease, asthma, elevated serum IgE, IgE-mediated food allergies, and anaphylaxis. The cases were either sporadic (seven kindreds) or followed an autosomal dominant inheritance pattern (three kindreds). All patients carried monoallelic rare variants in STAT6 and functional studies established their gain-of-function (GOF) phenotype with sustained STAT6 phosphorylation, increased STAT6 target gene expression, and TH2 skewing. Precision treatment with the anti–IL-4Rα antibody, dupilumab, was highly effective improving both clinical manifestations and immunological biomarkers. This study identifies heterozygous GOF variants in STAT6 as a novel autosomal dominant allergic disorder. We anticipate that our discovery of multiple kindreds with germline STAT6 GOF variants will facilitate the recognition of more affected individuals and the full definition of this new primary atopic disorder.
## Introduction
Asthma and related atopic diseases, including atopic dermatitis, food allergy, allergic rhinitis, and eosinophilic gastrointestinal diseases, are estimated to affect ∼$20\%$ of the global population imposing immense health and economic burdens (Dierick et al., 2020). Identifying human single-gene defects that lead to severe allergic disease—so-called primary atopic disorders (PADs)—is a powerful strategy to define the cellular and molecular mechanisms that drive human allergic inflammation (Lyons and Milner, 2018; Vaseghi-Shanjani et al., 2021). Identifying new PADs accelerates the diagnosis and treatment of affected individuals and can uncover new molecular targets for preventing and treating common allergic disease.
Currently there are only a few known inborn errors of immunity (IEIs) underlying severe allergic disease (Milner, 2020). Indeed, most cases are of unknown etiology, particularly those that are isolated or sporadic. In this study, we describe a novel human PAD caused by germline heterozygous gain-of-function (GOF) variants in the gene STAT6 found in 16 individuals from 10 unrelated families spanning three continents. Signal transducer and activator of transcription 6 (STAT6) is the main transcription factor that mediates the biological effects of IL-4, a key cytokine necessary for type 2 differentiation of T cells, B cell survival, proliferation, and class switching to IgE (Goenka and Kaplan, 2011; Takeda et al., 1996; Villarino et al., 2020; Villarino et al., 2017), as well as that of IL-13, a cytokine linked to anaphylaxis (Gowthaman et al., 2019). Affected individuals experienced early-onset severe, sometimes fatal, multisystem allergic disease which was refractory to conventional treatments. Notably, precision therapeutics aimed at targeting exaggerated STAT6 signaling were beneficial in those who received them.
## Identification of 16 patients from 10 families with severe early-onset allergic disease heterozygous for rare damaging STAT6 variants
We studied 16 patients from 10 kindreds with severe early-onset allergic disease spanning three continents. Patients were identified by their expert clinicians as candidates for genetic assessment based on their extreme phenotype and, in some cases, their family history (see clinical narratives in Data S1; Tables S4 and S5; and Fig. 1 A). The patients were from diverse ethnicities, specifically European (Kindred D, F, and J), Middle Eastern (Kindred A and C), Hispanic (Kindred B), South Asian (Kindred H), East Asian (Kindred E), and Southeast Asian (Kindred Y). The cases were either sporadic (seven kindreds) or affected multiple individuals of either sex over different generations consistent with autosomal dominant (AD) inheritance (Kindreds C, F, and J). All patients carried monoallelic rare variants in STAT6 (NM_001178079.2). The consensus negative selection score of STAT6 reveals a negative selection score that overlaps with known AD IEIs (Rapaport et al., 2021), consistent with the AD inheritance pattern observed in Kindreds C, F, and J (Fig. 1 B). In addition, and also consistent with an AD disorder, by sequencing both healthy parents (when available) we established that the STAT6 mutation was de novo in Patient 2 (P2), P5, P10, and P12 (from Kindreds B, D, G, and I, respectively; Fig. 1 A). The disease was fully penetrant in the families studied, as all STAT6 variant carriers were affected. None of the variants have previously been reported in population databases (Fig. 1 C; i.e., gnomAD). All the variants were private to the studied kindreds, except the p.D419G variant which was common in Kindreds A and E. Pathogenicity prediction models identify all of these variants to be pathogenic, evidenced by high pathogenic CADD (Rentzsch et al., 2021), SIFT (Sim et al., 2012), and Polyphen-2 (Adzhubei et al., 2010) scores (Tables S4 and S5). Remarkably, nine patients from six kindreds carried a variant affecting amino acid D419. Importantly, variants leading to amino acid changes at p.D419, p.D519, and p.P643 can be found in the Catalogue of Somatic Mutations in Cancer (COSMIC) database as recurrent somatic variants in lymphoma with some experimental evidence for a GOF phenotype for variants at p.D419 (Yildiz et al., 2015; Zamò et al., 2018; Fig. S1 A). The reported variants lie in different protein domains of STAT6, including the DNA-binding domain (p.E382 and p.D419), the linker domain (p.D519), and the SH2 domain (p.K595), while p.P643 lies in close proximity to the p.Y641 phosphorylation site (Fig. 1, D–E). Although the variants were located within different domains of the STAT6 protein, modeling of STAT6 interacting with DNA reveals that all the identified variants (with the exception of p.P643) lie near the protein–DNA interface and result in amino acids changes leading to increased electro-positivity at physiological pH (Fig. 1 E). Notably, E382 and D419 are located in regions responsible for protein–DNA recognition (Li et al., 2016). Changes in these variants decrease the electro-negativity of the protein near the DNA-binding interface and are predicted to enhance STAT6 binding to DNA (Fig. 1 E and Fig. S1 B). In aggregate, these data suggest that the STAT6 germline monoallelic variants identified in the patients underlie severe allergic disease by a GOF mechanism.
**Figure 1.:** *16 patients with severe allergic disease and STAT6 variants in different protein domains. (A) Family pedigree of the 16 patients from 10 different families. Filled symbols = affected individual; unfilled symbols = unaffected individual. (B) Consensus negative selection (CoNeS) score for STAT6 in relation to the score for known IEI genes reported with inheritance pattern of either AD, AR, or both (AD + AR). (C) Frequency and CADD score for missense (black) and predicted LOF (pLOF, blue) STAT6 variants reported in a public database and STAT6 variants reported in our patient cohort (red). The dotted line corresponds to the mutation significance cutoff (MSC). (D) Schematic illustrating the protein domains of STAT6. Amino acid location of the variants shown are highlighted, with the length of the bar corresponding to the number of patients reported with variants at that site. (E) Structural model of the DNA-STAT6 homodimer complex showing location of the different STAT6 variants in relation to the DNA-binding interface.* **Figure S1.:** *Pathogenic STAT6 germline variants lie in different protein domains and are frequently identified as somatic variants. (A) Somatic mutation counts for different amino acid changed as reported by COSMIC for STAT6. Red highlighted changes are those germline variants also identified in our cohort that cause severe allergic disease. (B) Structural model of the DNA-STAT6 homodimer complex showing location of the different STAT6 variants in relation to the DNA-binding interface. Specifically, zoom-ins for variants at each location are shown in relation to previously described variants known to affect STAT6 function.*
## Unifying clinical features of the 16 patients with severe allergic disease
The patients in the cohort were aged from 3 to 60 yr. Full clinical narratives are provided in Data S1, and their clinical features are summarized in Fig. 2 A. All had severe allergic disease which began in their first year of life. Severe, treatment-resistant atopic dermatitis ($\frac{15}{16}$) and food allergies ($\frac{15}{16}$) were the most common clinical manifestations, followed by asthma ($\frac{11}{16}$) and eosinophilic gastrointestinal disease ($\frac{10}{16}$) and severe episodes of anaphylaxis ($\frac{9}{16}$). Clinical laboratory testing was notable for eosinophilia and markedly elevated serum IgE levels (Fig. 2, B and C). Other aspects of the clinical laboratory and immunological work up were largely unremarkable, although clinical hallmarks of chronic systemic inflammation were present in some patients (i.e., elevations in white blood cell counts, platelets, and serum immunoglobulin levels). T, B, and natural killer (NK) cell numbers were all typically in the normal range (Fig. S2). Clinical biopsies confirmed the presence of eosinophils in the skin and gastrointestinal tract (Fig. 2, E–G). Endoscopic visualization of the esophagus revealed trachealization and furrowing consistent with eosinophilic esophagitis (Bolton et al., 2018; Fig. 2, H and I).
**Figure 2.:** *Major clinical features of the 16 patients. (A) Tabulation and comparison of the clinical phenotype for 16 patients. Please note blood eosinophil and IgE values were only available for 15 patients. (B) IgE concentration in whole blood for 15 out of the 16 patients. Shaded area represents IgE < 240 µg/liter, which is the typical upper limit of normal. (C) Eosinophil count in whole blood for 15 out of the 16 patients. Shaded area represents counts <0.5 × 109/liter, which is the typical upper limit of normal. The horizontal broken line denotes an eosinophil count of 1.5 × 109/liter, since hypereosinophilic syndrome is traditionally defined as peripheral blood eosinophilia >1.5 × 109/liter persisting ≥6 mo. (D) Photograph of widespread and severe atopic disease. (E) Photomicrograph of the skin biopsy showing marked pseudoepitheliomatous hyperplasia with acanthosis, hyperkeratosis, and focal parakeratosis, suggestive of lichen simplex chronicus (H&E stain, original magnification 2×). Moderate chronic inflammation within the papillary dermis in which scattered eosinophils (white arrows) are conspicuous (inset, H&E stain; original magnification, 40×). (F) Photomicrograph of duodenal biopsy showing abundant eosinophils (white arrows) amongst lymphocytes (H&E stain; original magnification, 40×). (G) Photomicrograph of gastric antral biopsy showing abundant infiltrate of eosinophils (arrows) amongst lymphocytes and plasma cells (H&E stain; original magnification, 40×). (H and I) Endoscopic images showing (H) furrowing and (I) trachealization in the middle esophagus, suggestive of eosinophilic esophagitis.* **Figure S2.:** *Complete blood counts and immunological workup of patients with pathogenic STAT6 variants. (A–G) Complete blood count for 15 out of the 16 patients and age-based references (orange-shaded area) for the following populations: (A) hemoglobin, (B) platelets, (C) white blood cells, (D) lymphocytes, (E) neutrophils, (F) basophils, and (G) monocytes. (H–L) Immunological workup for 15 out of the 16 patients showing age-based references (orange-shaded area) and populations quantification for: (H) CD3+ T cells, (I) CD4+ CD3+ T cells, (J) CD8+ CD3+ T cells, (K) NK cells, and (L) CD19+ B cells. (M–O) Immunoglobulin concentrations for 15 out of the 16 patients showing age-based references (orange-shaded area): (M) IgA, (N) IgM, and (O) IgA.*
In addition to atopic disease, half of the patients presented with recurrent skin, respiratory, and viral infections, although there were no deep-seated or fatal infections. Short stature (less than third percentile for age) was a common feature ($\frac{7}{16}$), and $\frac{5}{16}$ had other skeletal issues such as pathologic fractures and generalized hypermobility. Finally, two of the patients died from their disease; one from anaphylaxis at the age of 20 yr and another from a cerebral aneurysm at age 35 yr. These clinical presentations highlight the severity of the multi-system disease found in this patient cohort.
## Functional analysis of the STAT6 variants confirms their GOF activity
To assess the functional impact of the STAT6 variants, we selected HEK293 cells as our model system, as these cells lack endogenous STAT6 but express other components of the IL-4R pathway (Fig. 3 A; Mikita et al., 1996). HEK293 cells were transfected with each of the 10 patient STAT6 variants along with three different controls: WT STAT6, a predicted damaging STAT6 population variant found in the gnomAD healthy population database (p.A321V), and a biochemically inactive STAT6 variant (p.Y641F; Wick and Berton, 2000). To investigate STAT6 transcription factor activity, we conducted luciferase assays with three different reporter plasmids containing STAT6 binding sites (Li et al., 2016). While there were some difference related to the specific combinations of reporter plasmids and patient variants, there was evidence of GOF activity with all STAT6 patient variants resulting in higher promoter activity at baseline and after stimulation compared to the controls (Fig. 3 B and Fig. S3, A and B). The phosphorylation status of STAT6 variants was also quantified by flow cytometry and was confirmed to be higher at baseline and after stimulation compared to WT (Fig. 3, C and D; and Fig. S3 C). STAT6 phosphorylation was not detectable by flow cytometry for the p.P643R variant; however, increased baseline phosphorylation was confirmed by immunoblotting (Fig. 3 E and Fig. S3, D and E). This may point to a conformational change in tertiary structure of STAT6 near the phosphorylation site p.Y641 for this variant that rendered it inaccessible to the flow cytometry antibody.
**Figure 3.:** *STAT6 variants lead to increased STAT6 activity in HEK293 cells and Jurkat T cells. (A) Schematic illustrating classical IL-4–mediated STAT6 activation, dimerization, and phosphorylation. (B) Luciferase assay of STAT6 activity on a plasmid containing a 4× STAT6 binding site (TTCCCAAGAA; the underlined bases represent two half-sites for STAT6-specific binding) for WT-, different STAT6 variant–transfected HEK293 cells before and after stimulation with IL-4 (0.02 ng/ml for 4 h); n = 3. (C) Phospho-STAT6 (Y641) expression in WT- and STAT6 variant–transfected HEK293 cells before and after treatment with IL-4 (10 ng/ml for 30 min). Gating strategy for pSTAT6+ cells can be found in Fig. S3 C. (D) Quantification of C; n = 4. (E) Immunoblot in HEK293 cells transfected with WT-, inactive- (p.Y641F), p.P643R-, and p.D419G-STAT6 variants for pSTAT6, and Myc-tag before and after treatment with IL-4 (10 ng/ml for 30 min); n = 3. Full-length immunoblot for this can be found in Fig. S3, D and E. (F) Principal component analysis (PCA) comparing unstimulated and stimulated (100 ng/ml IL-4 for 4 h) WT (green), p.E382Q (blue), and p.D419G (purple) STAT6-transduced Jurkat T cells. Individual symbols represent technical replicates of one transduced pool for each genotype. PC1 and PC2 contribution is shown in brackets. (G) Normalized counts comparing stimulated WT (green) vs. p.E382Q (blue) or p.D419G (purple), for IL4R, CISH, and XBP1. (H) Heatmap representation of normalized counts of a transcription set defined as IL-4 targets in transduced Jurkat T cells. (I and J) Asterisk indicates adjusted P value <0.05. GSEA plots for (I) curated STAT6 target genes comparing WT vs. either p.E382Q (blue) or p.D419G (purple) at baseline, or (J) IL-4-TH2 targets genes comparing WT vs. either p.E382Q (blue) or p.D419G (purple) after stimulation with IL-4. Normalized enrichment score and adjusted P value are shown. Source data are available for this figure: SourceData F3.* **Figure S3.:** *In vitro assays demonstrate that STAT6 variants lead to increased STAT6 activity. (A and B) Luciferase assay of STAT6 activity on a plasmid containing (A) CCL26 promoter and (B) FcεR2 promoter for WT-, different STAT6-variant transfected HEK293 cells before and after stimulation with IL-4 (100 ng/ml for 40 h), n = 3. (C) Gating strategy for determining % positive HEK293 pSTAT6 cells: dot plot for fluoresence minus one (FMO) is presented and was used for establishing pSTAT6+ cells. (D and E) Full-length immunoblots of the cropped immunoblots shown in Fig. 3 E, showing HEK293 cells transfected with WT-, inactive- (p.Y641F), p.P643R-, and p.D419G- STAT6 variants for (D) Myc-tag and β-actin, as well as (E) pSTAT6 before and after treatment with IL-4 (10 ng/ml for 30 min). (F) Significantly upregulated (i) and downregulated (ii) genes upon IL-4 treatment in WT (green), p.E382Q (blue), and p.D419G (purple) in Jurkat cells as shown through Venn diagram. (G) Sample level enrichment analyses of significantly enriched immune pathways from MSigDB Hallmark in unstimulated and IL-4–stimulated samples, comparing WT vs. either p.E382Q or p.D419G. Heatmap is normalized across the rows and shown as relative expression.*
We next evaluated if the increased promoter activity leads to global transcriptomic changes. As transcriptomic studies on HEK293 cells after IL-4 stimulation have been challenging to interpret (Yildiz et al., 2015), we stably expressed p.E382Q and p.D419G STAT6 by lentiviral transduction in Jurkat T cells, which express endogenous STAT6 and serve as a model of heterozygosity (Kim et al., 2012). Here, we observed that WT-, p.E382Q-, and p.D419G-transduced cells clustered separately from each other both at baseline and after stimulation with IL-4 (Fig. 3 F). *Differential* gene expression analysis revealed significantly increased transcript abundance of known STAT6 target genes, including IL4R (Goenka and Kaplan, 2011), CISH (Yildiz et al., 2015), and XBP1 (Bettigole et al., 2015) in p.E382Q and p.D419G transduced cells when compared to WT transduced control (Fig. 3 G). Interestingly, p.E382Q and p.D419G had 67 and 80 uniquely increased hits, which did not overlap with WT nor with each other (Fig. 3 H, Fig. S3 F, and Table S6). This suggests that the altered activity of both p.E382Q and p.D419G is not restricted to enhanced activity of known STAT6 targets alone. Further investigation through gene set enrichment analyses (GSEA) showed increased enrichment in IL-4-STAT6 target genes at baseline (Fig. 3 I and Table S7), increased enrichment in T helper 2 (TH2) drivers after stimulation (Elo et al., 2010; Fig. 3 J and Table S7), and increased enrichment in proliferation pathways for p.D419G consistent with its known oncogenic activity (Ritz et al., 2009; Tate et al., 2019; Yildiz et al., 2015; Fig. S3 G).
## Patients with GOF STAT6 variants have slower STAT6 dephosphorylation kinetics after IL-4 stimulation and an enhanced TH2 signature
To further investigate the role of STAT6 GOF variants in primary cells, STAT6 phosphorylation was quantified in patient samples. Unexpectedly, we found no differences in the percentage of phospho-STAT6 positive cells between patients and healthy controls after IL-4 stimulation over a 60 min time course nor after stimulation with different doses of IL-4 (Fig. S4, A and B). However, differences emerged when we monitored the kinetics of STAT6 dephosphorylation after stimulation (Fig. 4, A and B; and Fig. S4 C). Specifically, washing out of IL-4 led to slower dephosphorylation kinetics of STAT6 in most patient cells compared to healthy controls (Fig. 4, A and B; and Fig. S4 D), supporting a GOF mechanism in patient lymphocytes. We did note that one of our kindreds did not display delayed dephosphorylation (Fig. S4 D), suggesting that this might not be the only GOF mechanism. Indeed, increased STAT6 activity without phosphorylation has previously been reported in follicular lymphoma research studying the p.D419 mutational hotspot (Yildiz et al., 2015).
**Figure S4.:** *Measure of STAT6 activity in patient primary lymphocytes. (A) 1-h time course to measure phosphorylation of STAT6 in different populations of lymphocytes from five patients (red) and one healthy control (blue) after stimulation with IL-4 (10 ng/ml). (B) Dose response in LCLs of patient one (red) vs. one healthy control (blue) after stimulation of cells with various doses of IL-4 15 min. (C) Gating strategy to determine % pSTAT6 positive cells in LCLs: dot plot for FMO is presented and was used for establishing pSTAT6+ cells. (D) Histograms showing phosphorylation of STAT6 in healthy control (blue) and patients with genotype p.D419Y (red, n = 2), p.D519H (purple, n = 4), p.D419N (pink, n = 1), and healthy controls (blue, n = 5) in T cell blasts that were stimulated with IL-4 (10 ng/ml) for 15 min, washed with PBS, and subsequently incubated in IL-4–free media for 60 min. Quantification of pSTAT6+ cells is presented and normalized to max stimulation (noted at 15 min). Two-way ANOVA followed by Šídák’s multiple comparisons was conducted. **, P < 0.01. (E) Readout of 92 biomarkers for P5 using throughput Olink proteomics. Eight healthy control distribution are shown as a violin plot in blue. The patient is shown as a red circle. Key cytokines, IL-4 and IL-13, are highlighted in yellow. (F) T helper cell distribution for nine patients (red) and 15 age-matched healthy controls (blue) each. (G) Transcriptomic comparison of naive CD4+ and naive CD8+ T cells between P6 and one healthy control measured through scRNAseq. Red genes are enriched in patient; blue genes are enriched in healthy control. The two dotted lines are the P value and adjusted P value respectively. (H) Quantification of % CD23 positive cells in naive, non-class switched memory, and class-switched memory B cells between patients (red, n = 7) and healthy controls (blue, n = 9) after stimulation with IL-4 (10 ng/ml) for 20 h. Unpaired t test. *, P < 0.05; **, P < 0.01.* **Figure 4.:** *Primary lymphocytes of STAT6 GOF patients display higher STAT6 activity and TH2 skewing. (A) Histograms showing phosphorylation of STAT6 in healthy control (blue) and patient (red) LCLs that were stimulated with IL-4 (10 ng/ml) for 15 min, washed with PBS and subsequently incubated in IL-4–free media for 15, 30, and 60 min. Gating strategy for pSTAT6+ cells can be found in Fig. S4 C. (B) Quantification of pSTAT6+ cells from three separate experiments done in A, multiple unpaired t test corrected for multiple comparison using the Benjamini–Hochberg method. ***, P < 0.001. (C) Frequency of IL5+, IL13+, and IL4+ cells in memory CD4+ T cells of one representative patient, along with one representative healthy control. (D) Quantification of C showing IL5+, IL13+, and IL4+ cells in patients along with 15 age-matched healthy controls. **, P < 0.01; ***, P < 0.001. (E) Uniform manifold approximation and projection (UMAP) visualization of scRNAseq done on enriched T cells comparing one patient with one age-matched healthy control. (F) Dot plot showing expression of selected differentially expressed genes (adjusted P value < 0.05) observed in scRNAseq between patient and healthy control and associated with T cells, B cells, monocytes, or dendritic cells.*
Given that the STAT6 axis is critical for the differentiation of TH2 cells (Kaplan et al., 1996), a subset of CD4+ helper T cells that is a major contributor to the pathogenesis of allergic disease, we next investigated TH2 signatures in these patients. Patients showed skewing towards TH2 pathway activity compared to healthy controls based on higher levels of the TH2 cytokines IL-5, IL-13, and IL-4 as measured by flow cytometry (Fig. 4, C and D), or through transcriptomic signature by single-cell RNA sequencing (scRNAseq; Fig. 4 E). High throughput proteomics also validated the increased IL-4 expression, but not high IL-13 expression (Fig. S4 E). Differences in proportions of other subsets of helper T cells were restricted to higher IL-21+ cells in patient memory CD4+ T cells (Fig. S4 F). scRNAseq showed that patient B cells expressed high IGHE and low IGHG3 (Fig. 4 F), reflecting patterns opposite of those seen in STAT6 knockout mice (Shimoda et al., 1996; Sulczewski et al., 2021), and CD4+ T cells express high GATA3. scRNAseq further demonstrated that IL4R, a gene encoding a key receptor that mediates STAT6 activation, was upregulated in all B and T cell subsets (Fig. 4 F and Fig. S4 G).
Using flow cytometry, we confirmed that IL-4Rα expression was significantly higher on both naive and memory CD4+ T cells of seven patients from three different kindreds compared to nine healthy controls (Fig. 5 A). IL-4Rα expression was also found to be higher in non-class switched memory and class switched memory B cells of unstimulated patient primary cells (Fig. 5 B). These findings suggest higher baseline activity of STAT6 in patient cells driving IL-4Rα expression similar to that seen in our Jurkat model (Fig. 3, G–J; and Tables S6 and S7). Finally, we measured the expression of CD23 (the low-affinity IgE receptor, FcεRII) which is known to be upregulated by STAT6 (Fig. S3 B; Kneitz et al., 2000) and we found higher expression of CD23 on all subsets of patient B cells at baseline (Fig. 5 B) and following stimulation with IL-4 (Fig. S4 H). Taken together, these experiments conducted using primary patient cells further confirm the STAT6 GOF phenotype.
**Figure 5.:** *Primary lymphocytes of STAT6 GOF patients display high expression of STAT6 target genes. (A) Expression of IL4Rα in naive and memory CD4+ cells is quantified as % positive cells in primary patient cells (n = 7, red) and healthy controls (n = 9, blue). Gating strategy for naive and memory and CD4+ is presented along with a dot plot for a fluoresence minus one (FMO) IL-4Rα sample to display IL-4Rα+ gating, as well as a representative dot plot for a patient and healthy control. (B) Expression of CD23 and IL4Ra in naive, non-class switched memory and memory B cells is quantified as % positive cells in primary patient cells (n = 7, red) and healthy controls (n = 9, blue). Gating strategy for B cell subsets is presented along with a dot plot for an FMO IL-4Rα sample to display IL-4Rα+ gating, as well as a representative dot plot for a patient and healthy control. Unpaired t test. **, P < 0.01; ***, P < 0.001.*
## JAK inhibitors and IL-4Rα monoclonal antibody can be used as potential therapeutics for patients with STAT6 GOF variants
Due to the severity of the multi-system allergic disease experienced by the patients in our cohort, various treatment approaches were used over the years. Corticosteroids, administered topically and systemically, were the mainstay of treatment for most patients. Unfortunately, even high doses of corticosteroids were unable to control the allergic inflammation and they were responsible for many side-effects including cataracts and osteoporosis. Other treatments used in this cohort that proved ineffective included topical tacrolimus, oral methotrexate, and mepolizumab (an anti–IL-5 monoclonal antibody).
Having demonstrated that heterozygous STAT6 variants lead to higher STAT6 activity and TH2 immunological skewing, we tested in vitro whether targeting the STAT6 pathway could be clinically beneficial. Based on our findings of higher phosphorylation of STAT6, and higher expression of IL-4Rα, we selected the JAK inhibitors, ruxolitinib and tofacitinib, and the anti–IL-4Rα antibody, dupilumab, as drugs to test in vitro as they are all used clinically for treatment of allergic disease (Bacharier et al., 2021; Beck et al., 2014; Bissonnette et al., 2016; Kim et al., 2020). We demonstrated that both ruxolitinib and tofacitinib effectively decreased the patient-specific enhanced STAT6 phosphorylation/activation in HEK293 cells at baseline and after IL-4 stimulation, whereas dupilumab inhibited IL-4 mediated increase in STAT6 activity (Fig. 6 A and Fig. S5 A). We further confirmed in patient primary cells that tofacitinib accelerated STAT6 dephosphorylation following IL-4 stimulation (Fig. 6 B). These data suggest the potential clinical benefit of directly targeting the IL-4/STAT6 pathway in patients with STAT6 GOF variants.
**Figure 6.:** *JAK inhibitors and IL-4Rα antibody can be used as potential therapeutics for patients with GOF STAT6 variants. (A) Quantification of phospho-STAT6 expression in transfected HEK293 cells left untreated (black) pre-treated with ruxolitinib (10 μM, 1 h; pink), tofacitinib (10 μM, 1 h; green), or dupilumab (10 nM, 1 h; blue), before and after stimulation with IL-4 (10 ng/ml, 30 min). Individual points represent separate transfections for each genotype (n = 4). Gating strategy for pSTAT6+ cells can be found in Fig. S3 C. One-way ANOVA and Tukey’s post-hoc test. *, P < 0.05; **, P < 0.01; ***, P < 0.001. (B) Quantification of pSTAT6+ cells in patient (red) and healthy control (blue) LCLs stimulated with IL-4 (10 ng/ml for 15 min), washed and incubated in tofacitinib (10 μM) for 15, 30, and 60 min. Dotted translucent lines are indicative of no tofacitinib treatment (Fig. 4 B); n = 1. Gating strategy for pSTAT6+ cells can be found in Fig. S4 C. (C) Cell type proportion gene signature as determined by the software Cibersort, in a patient undergoing dupilumab treatment for 2 yr and five healthy controls. Cell labels are listed on the right. (D) Donut plot showing frequencies of CD4+ T helper subsets in one patient, an age-matched healthy control (Fig. 4 E), and a 2-yr post-dupilumab treatment patient sample as measured by scRNAseq on enriched T cells. Frequency of TH2 cells is quantified in the donut plots of the different samples. (E) Violin plots showing expression of IL4R in the patient (red), healthy control (blue), and a 2-yr post-dupilumab sample (green). (F) Eczema scoring, EASI and SCORAD, for two patients after treatment with multiple doses of dupilumab. (G and H) Photographs of hands showing (G) the severity of atopic dermatitis before and (H) the improvement after dupilumab treatment.* **Figure S5.:** *STAT6 activity can be therapeutically targeted and can resolve clinical disease severity. (A) Quantification of luciferase assay in HEK293 transfected cells pre-treated with ruxolitinib (10 μM, 1 h), tofacitinib (10 μM, 1 h), or dupilumab (10 nM, 1 h), before and after stimulation with IL-4 (0.02 ng/ml, 4 h). n = 4. One-way ANOVA and Tukey’s post-hoc test. *, P < 0.05; **, P < 0.01; ***, P < 0.001. (B) Eosinophil counts before and following initiation of treatment with dupilumab are presented. Dots in red corresponds to transcriptomic data from this patient presented in Fig. 6 C. (C) PCA comparing whole blood bulk RNAseq of P6 before treatment with dupilumab and four time points after treatment, alongside five healthy controls. (D) Heatmap signatures of differentially expressed genes comparing pre-treatment patient samples against five healthy controls. Genes are row normalized. (E) Key genes, previously described to be biomarkers for allergic disease (Lemonnier et al., 2020) in whole blood RNA are presented for the patient samples. Gray shaded area is the range for the expression of these genes in five healthy controls.*
Once their STAT6 GOF variant was identified, three of the patients were started on dupilumab and all showed significant clinical improvement. P6, who has been on treatment with dupilumab for over 2 yr, serves as an illustrative example. Mirroring peripheral blood eosinophil counts (Fig. S5 B), repeated whole blood RNAseq showed global transcriptomic changes that were suggestive of mildly increased eosinophilic and allergic gene signatures after 38 d, followed by a shift of the transcriptome towards healthy controls after 123 and 492 d, respectively (Fig. 6 C and Fig. S5, C–E). scRNAseq confirmed a decrease in TH2 gene signatures 2 yr following initiation of dupilumab, accompanied by a decrease in the expression of IL-4R on both naive CD4+ and CD8+ T cells (Fig. 6, D–E). Clinically, these changes were accompanied by dramatic increase in growth velocity, improved skin condition as quantified by SCORAD (SCORing atopic dermatitis) and EASI (eczema area and severity index) scores, and the ability to wean from oral corticosteroids (Fig. 6, F–H). Similarly, P1 experienced remarkable clinical benefits with dupilumab with improved skin inflammation (Fig. 6 D), resolution of pruritus, and the ability to discontinue oral daily high dose corticosteroids. In addition to resolution of skin inflammation with dupilumab, P2 was able to discontinue swallowed budesonide without a flare in the severity of her eosinophilic esophagitis. Our preclinical data also suggested that JAK inhibitors may be beneficial (Fig. 6, A and B), and P4 had received tofacitinib (5 mg/d) for 2 mo at the time this manuscript was finalized. His initial response to tofacitinib was encouraging with less dysphagia, less esophageal food impaction, improved endoscopic appearance of the esophagus, and a marked reduction in the number of intraepithelial eosinophils.
## Discussion
We present a combination of clinical, genetic, molecular, and transcriptional evidence of a new human disorder caused by germline AD GOF STAT6 variants in 16 patients with life-long severe allergic disease. These variants led to sustained STAT6 phosphorylation, increased STAT6 target gene expression, and TH2 skewed transcriptional profile. Importantly, we demonstrate in three patients that dupilumab treatment is a highly effective targeted therapeutic option, improving both clinical manifestations of disease and immunological biomarkers.
Although the full phenotype(s) of individuals with GOF STAT6 variants will only be uncovered through the identification of additional affected individuals, we propose to classify human germline AD GOF STAT6 syndrome as a PAD (Lyons and Milner, 2018; Milner, 2020; Vaseghi-Shanjani et al., 2021). Based on our study, possible clinical “red flags” for this new disorder include: (i) early life onset; (ii) peripheral blood eosinophilia; (iii) elevated serum IgE; (iii) widespread, treatment-resistant atopic dermatitis; (iv) multiple food and drug allergies; (v) severe (and even fatal) anaphylaxis; (vi) recurrent skin and respiratory infections; (vii) eosinophilic gastrointestinal disorder, including eosinophilic esophagitis; (viii) asthma; (ix) allergic rhinoconjunctivitis; (x) short stature; and possibly (xi) vascular malformations of the brain.
STAT6 is intimately linked to the biology of allergic inflammation. The central and most studied role of STAT6 is in mediating the biological effects of IL-4, a cytokine necessary for TH2 differentiation, B cell survival, proliferation, and class switching to IgE (Elo et al., 2010; Takeda et al., 1996), as well as in driving M2 macrophage polarization (Ginhoux et al., 2016). In T cells, STAT6 activation induces the expression of GATA3, the master regulator of TH2 differentiation, which in turn enhances expression of IL-4, IL-5, and IL-13, cytokines necessary for promoting allergic responses by activating mast cells and eosinophils (Sloka et al., 2011). The presence of greater TH2 cell populations, or TH2 cells producing copious amounts of IL-4/IL-5/IL-13, could be a driver of the observed allergic phenotype presented in our patients. Elevated IgE in partnership with mast cells is important for both acute and chronic manifestations of allergic disorders and can be an additional driver of the allergic diathesis (Galli and Tsai, 2012). STAT6 hyperactivation in airway epithelial cells and resident dendritic cells can further create an environment favoring asthma and chronic lung disease, as this would induce production of chemokines that promote TH2 cells and eosinophil recruitment (Matsukura et al., 2001; Medoff et al., 2009). *Population* genetics provide further support for the central role that STAT6 plays in the development of human allergic disease. Multiple independent genome-wide association studies (GWAS) have found that polymorphisms in STAT6 associate with many allergic conditions (Table 1). Our study expands this appreciation of the role of STAT6 in human disease by establishing that heterozygous GOF variants cause a monogenic form of severe allergic disease.
**Table 1.**
| Phenotype | Number of published associations | References |
| --- | --- | --- |
| Asthmaa | 14 | Daya et al., 2019; Demenais et al., 2018; Ferreira et al., 2019; Han et al., 2020; Johansson et al., 2019; Olafsdottir et al., 2020; Pickrell et al., 2016; Pividori et al., 2019; Sakaue et al., 2021; Shrine et al., 2019; Valette et al., 2021; Zhu et al., 2020; Zhu et al., 2018; Zhu et al., 2019 |
| Eosinophil count | 7 | Astle et al., 2016; Chen et al., 2020; Höglund et al., 2022; Kachuri et al., 2021; Kichaev et al., 2019; Sakaue et al., 2021; Vuckovic et al., 2020 |
| Allergic disease | 3 | Ferreira et al., 2017; Ferreira et al., 2020; Zhu et al., 2018 |
| Atopic dermatitis/eczema | 3 | Johansson et al., 2019; Kichaev et al., 2019; Tanaka et al., 2021 |
| Serum IgE level | 2 | Daya et al., 2021; Granada et al., 2012 |
| Allergic sensitization | 2 | Bønnelykke et al., 2013; Waage et al., 2018 |
| Allergic rhinitis | 1 | Johansson et al., 2019 |
| Eosinophilic gastrointestinal disorders | 1 | Sleiman et al., 2014 |
The fatal cerebral aneurysm in P10 (p.E382Q) was not clinically anticipated, but it is possible that the STAT6 GOF variant also increased the risk of developing cerebral aneurysms. Indeed, P1 (p.D419G) also had multiple rare anatomical variants in the arteries of the Circle of Willis. Intracranial aneurysms have been reported in patients with both STAT3 loss-of-function (LOF) and STAT1 GOF (Chandesris et al., 2012; Dadak et al., 2017; Toubiana et al., 2016). Increased activation of other STAT family members, including STAT2, STAT3, and STAT5 have also been observed in human abdominal aortic aneurysms (STAT6 was not evaluated), although it is not clear whether enhanced STAT phosphorylation contributes to aneurysms or is the result of inflammation caused by aneurysms (Liao et al., 2012). As more individuals with STAT6 GOF variants are identified, the possibility of cerebral vascular anomalies warrants investigation.
It is noteworthy that the oldest patient in this cohort, P7 (p.D419H), experienced recurrent B cell lymphoma—follicular lymphoma at 49 yr and diffuse large B cell lymphoma at age 60 yr. Activating somatic mutations in STAT6 are well documented in B cell lymphoma with amino acid D419 being a particular mutational hotspot (Ritz et al., 2009; Tate et al., 2019; Yildiz et al., 2015). The patient’s p.D419H variant has been reported multiple times as a somatic mutation in COSMIC, as have other variants found in our patient cohort (i.e., p.D419, p.D519, and p.P643). More patients will need to be identified and followed to fully define the phenotype caused by germline STAT6 GOF variants, but it is biologically plausible that these patients may be at higher risk of developing B cell malignancies warranting enhanced clinical vigilance.
A GOF STAT6 model (designated STAT6VT) has previously been described in vitro (Daniel et al., 2000) and has been used to study chronic atopic dermatitis in mouse models (Bruns et al., 2003; DaSilva-Arnold et al., 2018). STAT6VT has the substitution of two amino acid residues, at positions 547 and 548, in the SH2 domain resulting in a STAT6 mutant that is constitutively active in an IL-4–independent manner and is unresponsive to IL-4 stimulation (Daniel et al., 2000). The humans we identified with STAT6 GOF variants and STAT6VT mice share a number of key features of the allergic diathesis, including elevated serum IgE and chronic atopic dermatitis. Very recently, a report was published describing a father and his two sons with severe allergic disease who were all heterozygous for the GOF STAT6 variant p.E377K (Suratannon et al., 2022). This new family shares many of the features we report in our cohort of 10 families (Fig. 2 A), further emphasizing that patients with early onset severe allergic disease should be assessed for underlying monogenic gene defects, including STAT6.
There is now a growing list of human single gene defects that cause the classic hyper-IgE phenotypic triad of eczema, recurrent skin and lung infections, and elevated serum IgE (Freeman and Milner, 2020; Vaseghi-Shanjani et al., 2021; Zhang et al., 2018). AD hyper-IgE syndrome caused by dominant negative variants in STAT3 (i.e., Job’s syndrome or STAT3 LOF) is the best characterized of these conditions, but this list of disorders also includes other AD (IL6ST; Beziat et al., 2020) and autosomal recessive (AR; ZNF341 [Béziat et al., 2018], IL6ST [Shahin et al., 2019]) disorders (Bergerson and Freeman, 2019; Minegishi, 2021). Notably, the patients we identified with STAT6 GOF variants did exhibit some of the extra-immunological features typical of STAT3 LOF, specifically hyperextensible joints, pathologic fractures, and vascular anomalies (Bergerson and Freeman, 2019).
Beyond defining the phenotype of STAT6 GOF, we also present laboratory and clinical evidence supporting the effectiveness of dupilumab and tofacitinib treatment in these patients. It was notable that the three patients (P1, P2, and P6) who received dupilumab have experienced dramatic improved atopic dermatitis and could be weaned off systemic corticosteroids. P6, who had short stature and delayed bone age before starting the biologic agent, experienced rapid height and weight gain following initiation of dupilumab. In addition to the documented clinical benefits of dupilumab therapy in patients with STAT6 GOF, we also present early data suggesting that the JAK inhibitors, ruxolitinib or tocafitinib, may be effective in this patient population.
While this study has many strengths, notably the extreme allergic phenotype of the 16 patients combined with in-depth functional assessment of their STAT6 variants; because of the global nature of our cohort, the study does have limitations. First, patients were identified by their local expert clinicians as candidates for genetic assessment based on their extreme allergic phenotype and, in some cases, their family history. As a result, we do not have prospectively defined inclusion criteria. Second, the global nature of the cohort and variation in local access to medications such as dupilumab limited our ability to run the same assays on pre-treatment primary cells from all patients. Despite these limitations, our study does identify GOF variants in STAT6 as a novel monogenic allergic disorder. We also present clinical and single cell evidence of the effectiveness of dupilumab in STAT6 GOF patients. We anticipate that this discovery will facilitate the recognition and targeted treatment of more affected individuals and, with time, a full definition of the human genotype-phenotype relationship caused by germline human STAT6 GOF variants will emerge, including understanding the risk of lymphoma.
Based on our findings reported in this study, we suggest that heterozygous GOF variants in STAT6 be added to the list of AD causes of the hyper-IgE phenotype. While each of the conditions known to cause a hyper-IgE phenotype has some specific clinical features (e.g., viral skin infections are a defining feature of DOCK8 deficiency; Biggs et al., 2017; Chu et al., 2012), there is considerable clinical overlap and clinically approved testing of these pathways is rarely available. Therefore, we recommend genetic testing as the most efficient initial diagnostic approach to patients who experience severe allergic disease beginning very early in life. Finally, we demonstrate that dupilumab and JAK inhibition may be an effective targeted treatment options for patients with GOF STAT6 variants.
## Ethical considerations
All procedures performed in the study were in accordance with the ethical standards of the institutional research committee and with the 1964 Helsinki declaration and its later amendments or comparable ethical standards. All study participants and/or their parents/guardians provided written informed consent. Research study protocols were approved by local institutions, specifically: The University of British Columbia Clinical Research Ethics Board (H09-01228, H15-00641), University College London Research Ethics Committee (04/Q$\frac{0501}{119}$, 06/Q$\frac{0508}{16}$), University of Hong Kong Institutional Review Board (UW 08-301), National Institutes of Health Institutional Review Board (NCT01164241), Children’s Hospital of Philadelphia Institutional Review Board [19-016617], Children’s Hospital Bambino Gesù Ethical Committee (1702_OPBG_2018).
## Identification of STAT6 variant via next-generation sequencing
Based on local availability, research or clinical next-generation sequencing of the genomic DNA was performed using either whole exome or a targeted panel approach (as described previously; Béziat et al., 2021; Campbell et al., 2022; Chovanec et al., 2022; Hebert et al., 2022; Murrell et al., 2022; Tarailo-Graovac et al., 2016). After bioinformatic analysis, de novo and inherited STAT6 variants that were predicted to be damaging and that segregated with disease were identified in each family (Tables S4 and S5).
## Generation and expression of STAT6 variant plasmids
STAT6 variants described in this study were generated through site-directed mutagenesis (SDM) for transfection purposes. Expression of WT STAT6 or STAT6 variants were induced transiently in HEK293 cells using lipofectamine, or stably in Jurkat T cells using a lenti-viral approach. See supplemental methods at the end of the PDF for details.
## Luciferase reporter assays
Luciferase reporter plasmids encoding a 4× STAT6 binding site (TTCCCAAGAA; the underlined bases represent the two half-sites for STAT6-specific binding), encoding the promoter site for CCL26, and encoding the promoter site for FCER2 were used to assess WT and variant STAT6 promoter activity (Li et al., 2016; Yildiz et al., 2015). See supplemental methods at the end of the PDF for details.
## Flow cytometry
(a) Phospho-flow cytometry: STAT6 phosphorylation was determined via phospho-flow cytometry for STAT6-transfected HEK293 cells, peripheral blood mononuclear cells (PBMCs), T cell blasts, and EBV-transformed lymphoblastoid B cell lines (LCLs). ( b) Intracellular cytokine staining: Intracellular cytokine staining was conducted on nine patient PBMCs, alongside 15 healthy controls, to study CD4+ T helper subsets as previously described (Sharma et al., 2022). ( c) CD23 and IL-4Rα expression was studied on seven patient PBMCs and nine healthy control PBMCs. See supplemental methods at the end of the PDF for details.
## Immunoblotting
Immunoblotting was conducted as previously described (Sharma et al., 2022) to assess the phosphorylation status of p.P643R STAT6 variant, as phosphorylation of this variant could not be detected via flow cytometry, using an antibody against the tyrosine 641 phosphorylation site. See supplemental methods at the end of the PDF for details.
## RNAseq
(a) Jurkat cells: To model transcriptomics changes caused by STAT6 variants, Jurkat T cells were transduced with either c.1144G>C, p.E382Q, c.1256A>G, p.D419G, or WT STAT6. The cells were either left unstimulated or stimulated with 100 ng/ml of IL-4 for 4 h and subsequently used for RNA extraction and sequencing. ( b) Whole blood: Bulk RNAseq was done on 10 samples: one patient sample before dupilumab treatment initiation, four patient samples after dupilumab treatment initiation, and five healthy controls. ( c) scRNAseq: Performed on PBMCs and enriched T cells from the patient sample before and 2 yr after dupilumab treatment, along with one age-matched healthy control. See supplemental methods at the end of the PDF for details.
## Histology
Formalin-fixed, paraffin-embedded gastric, duodenal, and skin tissue were sectioned and subjected to H&E staining.
## Structural modeling
The effects of the STAT6 variants on the protein function and structure were analyzed using three-dimensional models. SWISS-MODEL (Waterhouse et al., 2018) was used to model the variants based on a template structure of the human STAT6 transcription factor solved as a homodimer and in complex with DNA (PDB: 4Y5W, resolution: 3.1 Å, chains A, C, M, and N; Li et al., 2016). Structures were visualized with UCSF Chimera (Pettersen et al., 2004).
## Online supplemental material
Clinical narratives for each patient are presented as Data S1. Fig. S1 is a detailed structural model showing the DNA and STAT6 variant interface. Fig. S2 shows complete blood count and the immunological workup of all the patients. Fig. S3 shows additional in vitro data confirming the GOF nature of the STAT6 variants. Further workup of the primary patient cells is shown in Fig. S4. Additional IL4Rα antibody and JAK inhibitor treatment data of cells and patients is presented in Fig. S5. Table S1 lists primers used for site-directed mutagenesis. Table S2 lists antibodies used for phospho-flow on different immune subsets. Table S3 lists antibodies used for TH phenotyping in patient PBMCs. Pathogenicity prediction of the variants are presented in Tables S4 and S5. Tables S6 and S7 are gene lists from GSEA analysis shown in Fig. 3. Supplemental methods are listed at the end of the PDF.
## References
1. Adzhubei I.A., Schmidt S., Peshkin L., Ramensky V.E., Gerasimova A., Bork P., Kondrashov A.S., Sunyaev S.R.. **A method and server for predicting damaging missense mutations**. *Nat. Methods* (2010) **7** 248-249. DOI: 10.1038/nmeth0410-248
2. Aran D., Looney A.P., Liu L., Wu E., Fong V., Hsu A., Chak S., Naikawadi R.P., Wolters P.J., Abate A.R.. **Reference-based analysis of lung single-cell sequencing reveals a transitional profibrotic macrophage**. *Nat. Immunol.* (2019) **20** 163-172. DOI: 10.1038/s41590-018-0276-y
3. Astle W.J., Elding H., Jiang T., Allen D., Ruklisa D., Mann A.L., Mead D., Bouman H., Riveros-Mckay F., Kostadima M.A.. **The allelic landscape of human blood cell trait variation and links to common complex disease**. *Cell* (2016) **167** 1415-1429.e19. DOI: 10.1016/j.cell.2016.10.042
4. Bacharier L.B., Maspero J.F., Katelaris C.H., Fiocchi A.G., Gagnon R., de Mir I., Jain N., Sher L.D., Mao X., Liu D.. **Dupilumab in children with uncontrolled moderate-to-severe asthma**. *N. Engl. J. Med.* (2021) **385** 2230-2240. DOI: 10.1056/NEJMoa2106567
5. Beck L.A., Thaçi D., Hamilton J.D., Graham N.M., Bieber T., Rocklin R., Ming J.E., Ren H., Kao R., Simpson E.. **Dupilumab treatment in adults with moderate-to-severe atopic dermatitis**. *N. Engl. J. Med.* (2014) **371** 130-139. DOI: 10.1056/NEJMoa1314768
6. Bergerson J.R.E., Freeman A.F.. **An update on syndromes with a hyper-IgE phenotype**. *Immunol. Allergy Clin. North Am.* (2019) **39** 49-61. DOI: 10.1016/j.iac.2018.08.007
7. Bettigole S.E., Lis R., Adoro S., Lee A.H., Spencer L.A., Weller P.F., Glimcher L.H.. **The transcription factor XBP1 is selectively required for eosinophil differentiation**. *Nat. Immunol.* (2015) **16** 829-837. DOI: 10.1038/ni.3225
8. Béziat V., Li J., Lin J.X., Ma C.S., Li P., Bousfiha A., Pellier I., Zoghi S., Baris S., Keles S.. **A recessive form of hyper-IgE syndrome by disruption of ZNF341-dependent STAT3 transcription and activity**. *Sci. Immunol.* (2018) **3** 3. DOI: 10.1126/sciimmunol.aat4956
9. Béziat V., Rapaport F., Hu J., Titeux M., Bonnet des Claustres M., Bourgey M., Griffin H., Bandet É., Ma C.S., Sherkat R.. **Humans with inherited T cell CD28 deficiency are susceptible to skin papillomaviruses but are otherwise healthy**. *Cell* (2021) **184** 3812-3828.e30. DOI: 10.1016/j.cell.2021.06.004
10. Beziat V., Tavernier S.J., Chen Y.H., Ma C.S., Materna M., Laurence A., Staal J., Aschenbrenner D., Roels L., Worley L.. **Dominant-negative mutations in human IL6ST underlie hyper-IgE syndrome**. *J. Exp. Med.* (2020) **217**. DOI: 10.1084/jem.20191804
11. Biggs C.M., Keles S., Chatila T.A.. **DOCK8 deficiency: Insights into pathophysiology, clinical features and management**. *Clin. Immunol.* (2017) **181** 75-82. DOI: 10.1016/j.clim.2017.06.003
12. Bissonnette R., Papp K.A., Poulin Y., Gooderham M., Raman M., Mallbris L., Wang C., Purohit V., Mamolo C., Papacharalambous J., Ports W.C.. **Topical tofacitinib for atopic dermatitis: A phase IIa randomized trial**. *Br. J. Dermatol.* (2016) **175** 902-911. DOI: 10.1111/bjd.14871
13. Bolton S.M., Kagalwalla A.F., Wechsler J.B.. **Eosinophilic esophagitis in children: Endoscopic findings at diagnosis and post-intervention**. *Curr. Gastroenterol. Rep.* (2018) **20** 4. DOI: 10.1007/s11894-018-0607-z
14. Bønnelykke K., Matheson M.C., Pers T.H., Granell R., Strachan D.P., Alves A.C., Linneberg A., Curtin J.A., Warrington N.M., Standl M.. **Meta-analysis of genome-wide association studies identifies ten loci influencing allergic sensitization**. *Nat. Genet.* (2013) **45** 902-906. DOI: 10.1038/ng.2694
15. Bruns H.A., Schindler U., Kaplan M.H.. **Expression of a constitutively active Stat6 in vivo alters lymphocyte homeostasis with distinct effects in T and B cells**. *J. Immunol.* (2003) **170** 3478-3487. DOI: 10.4049/jimmunol.170.7.3478
16. Campbell T.M., Liu Z., Zhang Q., Moncada-Velez M., Covill L.E., Zhang P., Alavi Darazam I., Bastard P., Bizien L., Bucciol G.. **Respiratory viral infections in otherwise healthy humans with inherited IRF7 deficiency**. *J. Exp. Med.* (2022) **219**. PMID: 35670811
17. Chandesris M.O., Azarine A., Ong K.T., Taleb S., Boutouyrie P., Mousseaux E., Romain M., Bozec E., Laurent S., Boddaert N.. **Frequent and widespread vascular abnormalities in human signal transducer and activator of transcription 3 deficiency**. *Circ. Cardiovasc. Genet.* (2012) **5** 25-34. DOI: 10.1161/CIRCGENETICS.111.961235
18. Chen M.-H., Raffield L.M., Mousas A., Sakaue S., Huffman J.E., Moscati A., Trivedi B., Jiang T., Akbari P., Vuckovic D.. **Trans-ethnic and ancestry-specific blood-cell genetics in 746,667 individuals from 5 global populations**. *Cell* (2020) **182** 1198-1213.e14. DOI: 10.1016/j.cell.2020.06.045
19. Chen Y., Lun A.T., Smyth G.K.. **From reads to genes to pathways: Differential expression analysis of RNA-seq experiments using rsubread and the edgeR quasi-likelihood pipeline**. *F1000 Res.* (2016) **5** 1438
20. Chovanec J., Tunc I., Hughes J., Halstead J., Mateja A., Liu Y., O’Connell M., Kim J., Park Y.H., Wang Q.. **Genetically determining individualized clinical reference ranges for the biomarker tryptase can limit unnecessary procedures and unmask myeloid neoplasms**. *Blood Adv.* (2022) bloodadvances.2022007936. DOI: 10.1182/bloodadvances.2022007936
21. Chu E.Y., Freeman A.F., Jing H., Cowen E.W., Davis J., Su H.C., Holland S.M., Turner M.L.. **Cutaneous manifestations of DOCK8 deficiency syndrome**. *Arch. Dermatol.* (2012) **148** 79-84. DOI: 10.1001/archdermatol.2011.262
22. Dadak M., Jacobs R., Skuljec J., Jirmo A.C., Yildiz Ö., Donnerstag F., Baerlecken N.T., Schmidt R.E., Lanfermann H., Skripuletz T.. **Gain-of-function STAT1 mutations are associated with intracranial aneurysms**. *Clin. Immunol.* (2017) **178** 79-85. DOI: 10.1016/j.clim.2017.01.012
23. Daniel C., Salvekar A., Schindler U.. **A gain-of-function mutation in STAT6**. *J. Biol. Chem.* (2000) **275** 14255-14259. DOI: 10.1074/jbc.C000129200
24. DaSilva-Arnold S.C., Thyagarajan A., Seymour L.J., Yi Q., Bradish J.R., Al-Hassani M., Zhou H., Perdue N.J., Nemeth V., Krbanjevic A.. **Phenotyping acute and chronic atopic dermatitis-like lesions in Stat6VT mice identifies a role for IL-33 in disease pathogenesis**. *Arch. Dermatol. Res.* (2018) **310** 197-207. DOI: 10.1007/s00403-018-1807-y
25. Daya M., Cox C., Acevedo N., Boorgula M.P., Campbell M., Chavan S., Cho M.H., David G.L., Kachroo P., Lasky-Su J.. **Multiethnic genome-wide and HLA association study of total serum IgE level**. *J. Allergy Clin. Immunol.* (2021) **148** 1589-1595. DOI: 10.1016/j.jaci.2021.09.011
26. Daya M., Rafaels N., Brunetti T.M., Chavan S., Levin A.M., Shetty A., Gignoux C.R., Boorgula M.P., Wojcik G., Campbell M.. **Association study in African-admixed populations across the Americas recapitulates asthma risk loci in non-African populations**. *Nat. Commun.* (2019) **10** 880. DOI: 10.1038/s41467-019-08469-7
27. Demenais F., Margaritte-Jeannin P., Barnes K.C., Cookson W.O.C., Altmüller J., Ang W., Barr R.G., Beaty T.H., Becker A.B., Beilby J.. **Multiancestry association study identifies new asthma risk loci that colocalize with immune-cell enhancer marks**. *Nat. Genet.* (2018) **50** 42-53. DOI: 10.1038/s41588-017-0014-7
28. Dierick B.J.H., van der Molen T., Flokstra-de Blok B.M.J., Muraro A., Postma M.J., Kocks J.W.H., van Boven J.F.M.. **Burden and socioeconomics of asthma, allergic rhinitis, atopic dermatitis and food allergy**. *Expert Rev. Pharmacoecon. Outcomes Res.* (2020) **20** 437-453. DOI: 10.1080/14737167.2020.1819793
29. Elo L.L., Järvenpää H., Tuomela S., Raghav S., Ahlfors H., Laurila K., Gupta B., Lund R.J., Tahvanainen J., Hawkins R.D.. **Genome-wide profiling of interleukin-4 and STAT6 transcription factor regulation of human Th2 cell programming**. *Immunity* (2010) **32** 852-862. DOI: 10.1016/j.immuni.2010.06.011
30. Ferreira M.A., Vonk J.M., Baurecht H., Marenholz I., Tian C., Hoffman J.D., Helmer Q., Tillander A., Ullemar V., van Dongen J.. **Shared genetic origin of asthma, hay fever and eczema elucidates allergic disease biology**. *Nat. Genet.* (2017) **49** 1752-1757. DOI: 10.1038/ng.3985
31. Ferreira M.A.R., Mathur R., Vonk J.M., Szwajda A., Brumpton B., Granell R., Brew B.K., Ullemar V., Lu Y., Jiang Y.. **Genetic architectures of childhood- and adult-onset asthma are partly distinct**. *Am. J. Hum. Genet.* (2019) **104** 665-684. DOI: 10.1016/j.ajhg.2019.02.022
32. Ferreira M.A.R., Vonk J.M., Baurecht H., Marenholz I., Tian C., Hoffman J.D., Helmer Q., Tillander A., Ullemar V., Lu Y.. **Age-of-onset information helps identify 76 genetic variants associated with allergic disease**. *PLoS Genet.* (2020) **16**. DOI: 10.1371/journal.pgen.1008725
33. Freeman A.F., Milner J.D.. **The child with elevated IgE and infection susceptibility**. *Curr. Allergy Asthma Rep.* (2020) **20** 65. DOI: 10.1007/s11882-020-00964-y
34. Fung S.Y., Lu H.Y., Sharma M., Sharma A.A., Saferali A., Jia A., Abraham L., Klein T., Gold M.R., Noterangelo L.D.. **MALT1-Dependent cleavage of HOIL1 modulates canonical NF-κB signaling and inflammatory responsiveness**. *Front. Immunol.* (2021) **12** 749794. DOI: 10.3389/fimmu.2021.749794
35. Galli S.J., Tsai M.. **IgE and mast cells in allergic disease**. *Nat. Med.* (2012) **18** 693-704. DOI: 10.1038/nm.2755
36. Ginhoux F., Schultze J.L., Murray P.J., Ochando J., Biswas S.K.. **New insights into the multidimensional concept of macrophage ontogeny, activation and function**. *Nat. Immunol.* (2016) **17** 34-40. DOI: 10.1038/ni.3324
37. Goenka S., Kaplan M.H.. **Transcriptional regulation by STAT6**. *Immunol. Res.* (2011) **50** 87-96. DOI: 10.1007/s12026-011-8205-2
38. Gowthaman U., Chen J.S., Zhang B., Flynn W.F., Lu Y., Song W., Joseph J., Gertie J.A., Xu L., Collet M.A.. **Identification of a T follicular helper cell subset that drives anaphylactic IgE**. *Science* (2019) **365** 365. DOI: 10.1126/science.aaw6433
39. Granada M., Wilk J.B., Tuzova M., Strachan D.P., Weidinger S., Albrecht E., Gieger C., Heinrich J., Himes B.E., Hunninghake G.M.. **A genome-wide association study of plasma total IgE concentrations in the Framingham Heart Study**. *J. Allergy Clin. Immunol.* (2012) **129** 840-845.e21. DOI: 10.1016/j.jaci.2011.09.029
40. Hafemeister C., Satija R.. **Normalization and variance stabilization of single-cell RNA-seq data using regularized negative binomial regression**. *Genome Biol.* (2019) **20** 296. DOI: 10.1186/s13059-019-1874-1
41. Han Y., Jia Q., Jahani P.S., Hurrell B.P., Pan C., Huang P., Gukasyan J., Woodward N.C., Eskin E., Gilliland F.D.. **Genome-wide analysis highlights contribution of immune system pathways to the genetic architecture of asthma**. *Nat. Commun.* (2020) **11** 1776. DOI: 10.1038/s41467-020-15649-3
42. Hebert A., Simons A., Schuurs-Hoeijmakers J.H.M., Koenen H.J.P.M., Zonneveld-Huijssoon E., Henriet S.S.V., Schatorjé E.J.H., Hoppenreijs E.P.A.H., Leenders E.K.S.M., Janssen E.J.M.. **Trio-based whole exome sequencing in patients with suspected sporadic inborn errors of immunity: A retrospective cohort study**. *Elife* (2022) **11**. DOI: 10.7554/eLife.78469
43. Höglund J., Hadizadeh F., Ek W.E., Karlsson T., Johansson Å.. **Gene-based variant analysis of whole-exome sequencing in relation to eosinophil count**. *Front. Immunol.* (2022) **13** 862255. DOI: 10.3389/fimmu.2022.862255
44. Johansson Å., Rask-Andersen M., Karlsson T., Ek W.E.. **Genome-wide association analysis of 350 000 Caucasians from the UK Biobank identifies novel loci for asthma, hay fever and eczema**. *Hum. Mol. Genet.* (2019) **28** 4022-4041. DOI: 10.1093/hmg/ddz175
45. Kachuri L., Jeon S., DeWan A.T., Metayer C., Ma X., Witte J.S., Chiang C.W.K., Wiemels J.L., de Smith A.J.. **Genetic determinants of blood-cell traits influence susceptibility to childhood acute lymphoblastic leukemia**. *Am. J. Hum. Genet.* (2021) **108** 1823-1835. DOI: 10.1016/j.ajhg.2021.08.004
46. Kaplan M.H., Schindler U., Smiley S.T., Grusby M.J.. **Stat6 is required for mediating responses to IL-4 and for development of Th2 cells**. *Immunity* (1996) **4** 313-319. DOI: 10.1016/S1074-7613(00)80439-2
47. Kichaev G., Bhatia G., Loh P.-R., Gazal S., Burch K., Freund M.K., Schoech A., Pasaniuc B., Price A.L.. **Leveraging polygenic functional enrichment to improve GWAS power**. *Am. J. Hum. Genet.* (2019) **104** 65-75. DOI: 10.1016/j.ajhg.2018.11.008
48. Kim B.S., Howell M.D., Sun K., Papp K., Nasir A., Kuligowski M.E., Investigators I.S.. **Treatment of atopic dermatitis with ruxolitinib cream (JAK1/JAK2 inhibitor) or triamcinolone cream**. *J. Allergy Clin. Immunol.* (2020) **145** 572-582. DOI: 10.1016/j.jaci.2019.08.042
49. Kim Y., Kwon E.K., Jeon J.H., So I., Kim I.G., Choi M.S., Kim I.S., Choi J.K., Jung J.U., Cho N.H.. **Activation of the STAT6 transcription factor in Jurkat T-cells by the herpesvirus saimiri Tip protein**. *J. Gen. Virol.* (2012) **93** 330-340. DOI: 10.1099/vir.0.036087-0
50. Kneitz C., Goller M., Seggewiss R., Yaman A., Serfling E., Tony H.P.. **STAT6 and the regulation of CD23 expression in B-chronic lymphocytic leukemia**. *Leuk. Res.* (2000) **24** 331-337. DOI: 10.1016/S0145-2126(99)00191-5
51. Kulpa D.A., Talla A., Brehm J.H., Ribeiro S.P., Yuan S., Bebin-Blackwell A.G., Miller M., Barnard R., Deeks S.G., Hazuda D.. **Differentiation into an effector memory phenotype potentiates HIV-1 latency reversal in CD4**. *J. Virol.* (2019) **93** 93. DOI: 10.1128/JVI.00969-19
52. Kutner R.H., Zhang X.Y., Reiser J.. **Production, concentration and titration of pseudotyped HIV-1-based lentiviral vectors**. *Nat. Protoc.* (2009) **4** 495-505. DOI: 10.1038/nprot.2009.22
53. Lemonnier N., Melén E., Jiang Y., Joly S., Ménard C., Aguilar D., Acosta-Perez E., Bergström A., Boutaoui N., Bustamante M.. **A novel whole blood gene expression signature for asthma, dermatitis, and rhinitis multimorbidity in children and adolescents**. *Allergy* (2020) **75** 3248-3260. DOI: 10.1111/all.14314
54. Li J., Rodriguez J.P., Niu F., Pu M., Wang J., Hung L.W., Shao Q., Zhu Y., Ding W., Liu Y.. **Structural basis for DNA recognition by STAT6**. *Proc. Natl. Acad. Sci. USA* (2016) **113** 13015-13020. DOI: 10.1073/pnas.1611228113
55. Liao M., Xu J., Clair A.J., Ehrman B., Graham L.M., Eagleton M.J.. **Local and systemic alterations in signal transducers and activators of transcription (STAT) associated with human abdominal aortic aneurysms**. *J. Surg. Res.* (2012) **176** 321-328. DOI: 10.1016/j.jss.2011.05.041
56. Lu H.Y., Sharma M., Sharma A.A., Lacson A., Szpurko A., Luider J., Dharmani-Khan P., Shameli A., Bell P.A., Guilcher G.M.T.. **Mechanistic understanding of the combined immunodeficiency in complete human CARD11 deficiency**. *J. Allergy Clin. Immunol.* (2021) **148** 1559-1574.e13. DOI: 10.1016/j.jaci.2021.04.006
57. Lundberg M., Eriksson A., Tran B., Assarsson E., Fredriksson S.. **Homogeneous antibody-based proximity extension assays provide sensitive and specific detection of low-abundant proteins in human blood**. *Nucleic Acids Res.* (2011) **39**. DOI: 10.1093/nar/gkr424
58. Lyons J.J., Milner J.D.. **Primary atopic disorders**. *J. Exp. Med.* (2018) **215** 1009-1022. DOI: 10.1084/jem.20172306
59. Matsukura S., Stellato C., Georas S.N., Casolaro V., Plitt J.R., Miura K., Kurosawa S., Schindler U., Schleimer R.P.. **Interleukin-13 upregulates eotaxin expression in airway epithelial cells by a STAT6-dependent mechanism**. *Am. J. Respir. Cell Mol. Biol.* (2001) **24** 755-761. DOI: 10.1165/ajrcmb.24.6.4351
60. Medoff B.D., Seung E., Hong S., Thomas S.Y., Sandall B.P., Duffield J.S., Kuperman D.A., Erle D.J., Luster A.D.. **CD11b**. *J. Immunol.* (2009) **182** 623-635. DOI: 10.4049/jimmunol.182.1.623
61. Mikita T., Campbell D., Wu P., Williamson K., Schindler U.. **Requirements for interleukin-4-induced gene expression and functional characterization of Stat6**. *Mol. Cell. Biol.* (1996) **16** 5811-5820. DOI: 10.1128/MCB.16.10.5811
62. Milner J.D.. **Primary atopic disorders**. *Annu. Rev. Immunol.* (2020) **38** 785-808. DOI: 10.1146/annurev-immunol-042718-041553
63. Minegishi Y.. **Hyper-IgE syndrome, 2021 update**. *Allergol. Int.* (2021) **70** 407-414. DOI: 10.1016/j.alit.2021.07.007
64. Monaco G., Lee B., Xu W., Mustafah S., Hwang Y.Y., Carre C., Burdin N., Visan L., Ceccarelli M., Poidinger M.. **RNA-seq signatures normalized by mRNA abundance allow absolute deconvolution of human immune cell types**. *Cell Rep* (2019) **26** 1627-1640 e1627. DOI: 10.1016/j.celrep.2019.01.041
65. Murrell J.R., Nesbitt A.M.I., Baker S.W., Pechter K.B., Balciuniene J., Zhao X., Denenberg E.H., DeChene E.T., Wu C., Jayaraman P.. **Molecular diagnostic outcomes from 700 cases: What can we learn from a retrospective analysis of clinical exome sequencing?**. *J. Mol. Diagn.* (2022) **24** 274-286. DOI: 10.1016/j.jmoldx.2021.12.002
66. Newman A.M., Liu C.L., Green M.R., Gentles A.J., Feng W., Xu Y., Hoang C.D., Diehn M., Alizadeh A.A.. **Robust enumeration of cell subsets from tissue expression profiles**. *Nat. Methods* (2015) **12** 453-457. DOI: 10.1038/nmeth.3337
67. Novershtern N., Subramanian A., Lawton L.N., Mak R.H., Haining W.N., McConkey M.E., Habib N., Yosef N., Chang C.Y., Shay T.. **Densely interconnected transcriptional circuits control cell states in human hematopoiesis**. *Cell* (2011) **144** 296-309. DOI: 10.1016/j.cell.2011.01.004
68. Olafsdottir T.A., Theodors F., Bjarnadottir K., Bjornsdottir U.S., Agustsdottir A.B., Stefansson O.A., Ivarsdottir E.V., Sigurdsson J.K., Benonisdottir S., Eyjolfsson G.I.. **Eighty-eight variants highlight the role of T cell regulation and airway remodeling in asthma pathogenesis**. *Nat. Commun.* (2020) **11** 393. DOI: 10.1038/s41467-019-14144-8
69. Pettersen E.F., Goddard T.D., Huang C.C., Couch G.S., Greenblatt D.M., Meng E.C., Ferrin T.E.. **UCSF Chimera--a visualization system for exploratory research and analysis**. *J. Comput. Chem.* (2004) **25** 1605-1612. DOI: 10.1002/jcc.20084
70. Pickrell J.K., Berisa T., Liu J.Z., Ségurel L., Tung J.Y., Hinds D.A.. **Detection and interpretation of shared genetic influences on 42 human traits**. *Nat. Genet.* (2016) **48** 709-717. DOI: 10.1038/ng.3570
71. Pividori M., Schoettler N., Nicolae D.L., Ober C., Im H.K.. **Shared and distinct genetic risk factors for childhood-onset and adult-onset asthma: Genome-wide and transcriptome-wide studies**. *Lancet Respir. Med.* (2019) **7** 509-522. DOI: 10.1016/S2213-2600(19)30055-4
72. Rapaport F., Boisson B., Gregor A., Béziat V., Boisson-Dupuis S., Bustamante J., Jouanguy E., Puel A., Rosain J., Zhang Q.. **Negative selection on human genes underlying inborn errors depends on disease outcome and both the mode and mechanism of inheritance**. *Proc. Natl. Acad. Sci. USA* (2021) **118** 118. DOI: 10.1073/pnas.2001248118
73. Rentzsch P., Schubach M., Shendure J., Kircher M.. **CADD-Splice-improving genome-wide variant effect prediction using deep learning-derived splice scores**. *Genome Med.* (2021) **13** 31. DOI: 10.1186/s13073-021-00835-9
74. Ritchie M.E., Phipson B., Wu D., Hu Y., Law C.W., Shi W., Smyth G.K.. **Limma powers differential expression analyses for RNA-sequencing and microarray studies**. *Nucleic Acids Res.* (2015) **43**. DOI: 10.1093/nar/gkv007
75. Ritz O., Guiter C., Castellano F., Dorsch K., Melzner J., Jais J.P., Dubois G., Gaulard P., Möller P., Leroy K.. **Recurrent mutations of the STAT6 DNA binding domain in primary mediastinal B-cell lymphoma**. *Blood* (2009) **114** 1236-1242. DOI: 10.1182/blood-2009-03-209759
76. Robinson M.D., Smyth G.K.. **Moderated statistical tests for assessing differences in tag abundance**. *Bioinformatics* (2007) **23** 2881-2887. DOI: 10.1093/bioinformatics/btm453
77. Sakaue S., Kanai M., Tanigawa Y., Karjalainen J., Kurki M., Koshiba S., Narita A., Konuma T., Yamamoto K., Akiyama M.. **A cross-population atlas of genetic associations for 220 human phenotypes**. *Nat. Genet.* (2021) **53** 1415-1424. DOI: 10.1038/s41588-021-00931-x
78. Shahin T., Aschenbrenner D., Cagdas D., Bal S.K., Conde C.D., Garncarz W., Medgyesi D., Schwerd T., Karaatmaca B., Cetinkaya P.G.. **Selective loss of function variants in cause Hyper-IgE syndrome with distinct impairments of T-cell phenotype and function**. *Haematologica* (2019) **104** 609-621. DOI: 10.3324/haematol.2018.194233
79. Sharma M., Fu M.P., Lu H.Y., Sharma A.A., Modi B.P., Michalski C., Lin S., Dalmann J., Salman A., Del Bel K.L.. **Human complete NFAT1 deficiency causes a triad of joint contractures, osteochondromas, and B-cell malignancy**. *Blood* (2022) **140** 1858-1874. DOI: 10.1182/blood.2022015674
80. Shimoda K., van Deursen J., Sangster M.Y., Sarawar S.R., Carson R.T., Tripp R.A., Chu C., Quelle F.W., Nosaka T., Vignali D.A.. **Lack of IL-4-induced Th2 response and IgE class switching in mice with disrupted Stat6 gene**. *Nature* (1996) **380** 630-633. DOI: 10.1038/380630a0
81. Shrine N., Portelli M.A., John C., Soler Artigas M., Bennett N., Hall R., Lewis J., Henry A.P., Billington C.K., Ahmad A.. **Moderate-to-severe asthma in individuals of European ancestry: A genome-wide association study**. *Lancet Respir. Med.* (2019) **7** 20-34. DOI: 10.1016/S2213-2600(18)30389-8
82. Sim N.L., Kumar P., Hu J., Henikoff S., Schneider G., Ng P.C.. **SIFT web server: Predicting effects of amino acid substitutions on proteins**. *Nucleic Acids Res.* (2012) **40** W452-W457. DOI: 10.1093/nar/gks539
83. Sleiman P.M.A., Wang M.-L., Cianferoni A., Aceves S., Gonsalves N., Nadeau K., Bredenoord A.J., Furuta G.T., Spergel J.M., Hakonarson H.. **GWAS identifies four novel eosinophilic esophagitis loci**. *Nat. Commun.* (2014) **5** 5593. DOI: 10.1038/ncomms6593
84. Sloka S., Silva C., Wang J., Yong V.W.. **Predominance of Th2 polarization by vitamin D through a STAT6-dependent mechanism**. *J. Neuroinflammation* (2011) **8** 56. DOI: 10.1186/1742-2094-8-56
85. Sulczewski F.B., Martino L.A., da Silva Almeida B., Yamamoto M.M., Rosa D.S., Boscardin S.B.. **STAT6 signaling pathway controls germinal center responses promoted after antigen targeting to conventional type 2 dendritic cells**. *Curr. Res. Immunol.* (2021) **2** 120-131. DOI: 10.1016/j.crimmu.2021.08.001
86. Suratannon N., Ittiwut C., Dik W.A., Ittiwut R., Meesilpavikkai K., Israsena N., Ingrungruanglert P., Dalm V.A.S.H., van Daele P.L.A., Sanpavat A.. **A germline STAT6 gain-of-function variant is associated with early-onset allergies**. *J. Allergy Clin. Immunol.* (2022) **151** 565-571.e9. PMID: 36216080
87. Takeda K., Tanaka T., Shi W., Matsumoto M., Minami M., Kashiwamura S., Nakanishi K., Yoshida N., Kishimoto T., Akira S.. **Essential role of Stat6 in IL-4 signalling**. *Nature* (1996) **380** 627-630. DOI: 10.1038/380627a0
88. Tanaka N., Koido M., Suzuki A., Otomo N., Suetsugu H., Kochi Y., Tomizuka K., Momozawa Y., Kamatani Y., Ikegawa S.. **Eight novel susceptibility loci and putative causal variants in atopic dermatitis**. *J. Allergy Clin. Immunol.* (2021) **148** 1293-1306. DOI: 10.1016/j.jaci.2021.04.019
89. Tarailo-Graovac M., Shyr C., Ross C.J., Horvath G.A., Salvarinova R., Ye X.C., Zhang L.H., Bhavsar A.P., Lee J.J., Drögemöller B.I.. **Exome sequencing and the management of neurometabolic disorders**. *N. Engl. J. Med.* (2016) **374** 2246-2255. DOI: 10.1056/NEJMoa1515792
90. Tate J.G., Bamford S., Jubb H.C., Sondka Z., Beare D.M., Bindal N., Boutselakis H., Cole C.G., Creatore C., Dawson E.. **COSMIC: The Catalogue of somatic mutations in cancer**. *Nucleic Acids Res.* (2019) **47** D941-D947. DOI: 10.1093/nar/gky1015
91. Toubiana J., Okada S., Hiller J., Oleastro M., Lagos Gomez M., Aldave Becerra J.C., Ouachée-Chardin M., Fouyssac F., Girisha K.M., Etzioni A.. **Heterozygous STAT1 gain-of-function mutations underlie an unexpectedly broad clinical phenotype**. *Blood* (2016) **127** 3154-3164. DOI: 10.1182/blood-2015-11-679902
92. Valette K., Li Z., Bon-Baret V., Chignon A., Bérubé J.-C., Eslami A., Lamothe J., Gaudreault N., Joubert P., Obeidat M.. **Prioritization of candidate causal genes for asthma in susceptibility loci derived from UK Biobank**. *Commun. Biol.* (2021) **4** 700. DOI: 10.1038/s42003-021-02227-6
93. Vaseghi-Shanjani M., Smith K.L., Sara R.J., Modi B.P., Branch A., Sharma M., Lu H.Y., James E.L., Hildebrand K.J., Biggs C.M., Turvey S.E.. **Inborn errors of immunity manifesting as atopic disorders**. *J. Allergy Clin. Immunol.* (2021) **148** 1130-1139. DOI: 10.1016/j.jaci.2021.08.008
94. Villarino A.V., Gadina M., O'Shea J.J., Kanno Y.. **SnapShot: Jak-STAT signaling II**. *Cell* (2020) **181** 1696-1696.e1691. DOI: 10.1016/j.cell.2020.04.052
95. Villarino A.V., Kanno Y., O’Shea J.J.. **Mechanisms and consequences of Jak-STAT signaling in the immune system**. *Nat. Immunol.* (2017) **18** 374-384. DOI: 10.1038/ni.3691
96. Vuckovic D., Bao E.L., Akbari P., Lareau C.A., Mousas A., Jiang T., Chen M.-H., Raffield L.M., Tardaguila M., Huffman J.E.. **The polygenic and monogenic basis of blood traits and diseases**. *Cell* (2020) **182** 1214-1231.e11. DOI: 10.1016/j.cell.2020.08.008
97. Waage J., Standl M., Curtin J.A., Jessen L.E., Thorsen J., Tian C., Schoettler N., Flores C., Abdellaoui A., Ahluwalia T.S.. **Genome-wide association and HLA fine-mapping studies identify risk loci and genetic pathways underlying allergic rhinitis**. *Nat. Genet.* (2018) **50** 1072-1080. DOI: 10.1038/s41588-018-0157-1
98. Waterhouse A., Bertoni M., Bienert S., Studer G., Tauriello G., Gumienny R., Heer F.T., de Beer T.A.P., Rempfer C., Bordoli L.. **SWISS-MODEL: Homology modelling of protein structures and complexes**. *Nucleic Acids Res.* (2018) **46** W296-W303. DOI: 10.1093/nar/gky427
99. Wick K.R., Berton M.T.. **IL-4 induces serine phosphorylation of the STAT6 transactivation domain in B lymphocytes**. *Mol. Immunol.* (2000) **37** 641-652. DOI: 10.1016/S0161-5890(00)00088-2
100. Yildiz M., Li H., Bernard D., Amin N.A., Ouillette P., Jones S., Saiya-Cork K., Parkin B., Jacobi K., Shedden K.. **Activating STAT6 mutations in follicular lymphoma**. *Blood* (2015) **125** 668-679. DOI: 10.1182/blood-2014-06-582650
101. Zamò A., Pischimarov J., Horn H., Ott G., Rosenwald A., Leich E.. **The exomic landscape of t(14;18)-negative diffuse follicular lymphoma with 1p36 deletion**. *Br. J. Haematol.* (2018) **180** 391-394. DOI: 10.1111/bjh.15041
102. Zhang Q., Boisson B., Béziat V., Puel A., Casanova J.L.. **Human hyper-IgE syndrome: Singular or plural?**. *Mamm. Genome* (2018) **29** 603-617. DOI: 10.1007/s00335-018-9767-2
103. Zhu Z., Guo Y., Shi H., Liu C.-L., Panganiban R.A., Chung W., O’Connor L.J., Himes B.E., Gazal S., Hasegawa K.. **Shared genetic and experimental links between obesity-related traits and asthma subtypes in UK Biobank**. *J. Allergy Clin. Immunol.* (2020) **145** 537-549. DOI: 10.1016/j.jaci.2019.09.035
104. Zhu Z., Lee P.H., Chaffin M.D., Chung W., Loh P.-R., Lu Q., Christiani D.C., Liang L.. **A genome-wide cross-trait analysis from UK Biobank highlights the shared genetic architecture of asthma and allergic diseases**. *Nat. Genet.* (2018) **50** 857-864. DOI: 10.1038/s41588-018-0121-0
105. Zhu Z., Zhu X., Liu C.-L., Shi H., Shen S., Yang Y., Hasegawa K., Camargo C.A., Liang L.. **Shared genetics of asthma and mental health disorders: A large-scale genome-wide cross-trait analysis**. *Eur. Respir. J.* (2019) **54** 1901507. DOI: 10.1183/13993003.01507-2019
|
---
title: Concordance and Discrepancies Among 5 Creatinine-Based Equations for Assessing
Estimated Glomerular Filtration Rate in Older Adults
authors:
- Giorgi Beridze
- Davide L. Vetrano
- Alessandra Marengoni
- Lu Dai
- Juan-Jesús Carrero
- Amaia Calderón-Larrañaga
journal: JAMA Network Open
year: 2023
pmcid: PMC10037147
doi: 10.1001/jamanetworkopen.2023.4211
license: CC BY 4.0
---
# Concordance and Discrepancies Among 5 Creatinine-Based Equations for Assessing Estimated Glomerular Filtration Rate in Older Adults
## Key Points
### Question
Can the 5 most common estimated glomerular filtration rate (eGFR) equations be used interchangeably among older adults?
### Findings
In this cohort study including 3094 participants, different creatinine-based equations provided divergent estimates of eGFR, with the Berlin Initiative Study equation having the highest prognostic accuracy for 15-year all-cause mortality. The differences between equations were not consistent across levels of calf circumference, body mass index, and age, suggesting that these factors may be potential sources of the observed discrepancies.
### Meaning
These results suggest that clinicians and researchers should carefully consider their choice of equation when monitoring kidney function in old age.
## Abstract
This cohort study of older Swedish adults examines the concordance among 5 common equations for determining estimated glomerular filtration rate for older adults and considers possible sources for discrepancies between measures.
### Importance
There is uncertainty as to which estimated glomerular filtration rate (eGFR) equation should be used among older adults.
### Objective
To compare the 5 most commonly used creatinine-based eGFR equations in older adults, quantifying the concordance among the equations, comparing their discriminative capacity in regards to 15-year mortality, and identifying sources of potential discrepancies.
### Design, Setting, and Participants
This cohort study used data from the Swedish National Study on Aging and Care in Kungsholmen (SNAC-K), a longitudinal study of adults aged 60 years or older in Sweden. Participants were recruited between 2001 and 2004 and followed up for mortality until December 2016. Participants missing creatinine values were excluded. Data were originally analyzed March through July 2022, and were rerun in January 2023.
### Exposures
Five creatinine-based equations were considered: Modification of Diet in Renal Disease (MDRD), 2009 Chronic Kidney Disease Epidemiological Collaboration (CKD-EPI), Revised Lund-Malmö (RLM), Berlin Initiative Study (BIS), and European Kidney Function Consortium (EKFC).
### Main Outcomes and Measures
Concordance between equations was quantified using Cohen κ. Discriminative capacity for mortality was quantified using area under the receiver operating characteristic curve (AUC) and the Harrel C statistic. Calf circumference, body mass index (BMI), and age were explored as correlates of discrepancies.
### Results
The study sample consisted of 3094 older adults (1972 [$63.7\%$] female; median [IQR] age, 72 [66-81] years). Cohen κ between dyads of equations ranged from 0.42 to 0.91, with poorest concordance between MDRD and BIS, and best between RLM and EKFC. MDRD and CKD-EPI provided higher estimates of GFR compared with the other equations. The best mix of AUC and Harrel C statistic was observed for BIS (0.80 and 0.73, respectively); however, the prognostic accuracy for death decreased among those aged over 78 years and those with low calf circumference. Differences between equations were inconsistent across levels of calf circumference, BMI, and age.
### Conclusions and Relevance
In this cohort study, we found that eGFR equations were not interchangeable when assessing kidney function. BIS outperformed other equations in predicting mortality; however, its discriminative capacity was reduced in subgroup analyses. Clinicians should consider these discrepancies when monitoring kidney function in old age.
## Introduction
Chronic kidney disease (CKD), a progressive loss of kidney function, is prevalent among $9.1\%$ of the global population and $28\%$ of older adults.1,2 It presents both as a direct cause of mortality as well as a consequence and risk factor of several diseases, and up to 2.6 million deaths worldwide in 2017 were estimated to have been attributable to CKD.1 Timely detection and management of kidney function deterioration can reduce progression of CKD to more severe forms, including end-stage kidney disease. However, the measurement of glomerular filtration rate (GFR), the criterion standard for assessing kidney function, is cumbersome and unfeasible for screening purposes. Indeed, obtaining measures of GFR using the clearance of exogenous filtration markers, such as iohexol, is complex and not routinely performed outside specialized settings.
The Modification of Diet in Renal Disease (MDRD) and Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) equations have been used over the last decades to obtain measures of estimated GFR (eGFR) using serum creatinine values and routinely available data such as age, sex, and ethnicity.3,4,5 Even if CKD-EPI remains the recommended equation by Kidney Disease: Improving Global Outcomes (KDIGO) work group for all adults,6 the underrepresentation of older adults in the original population ($13\%$ were older than 65 years) has yielded criticism. The transition into old age is associated with changes in body composition, including lower muscle mass and bone density as well as increased fat mass.7 As such, older adults with sarcopenia, a condition characterized by loss of muscle mass and function, may present with low serum creatinine even in the presence of a significant decline in GFR.8 In order to address this limitation, recently the Berlin Initiative Equation (BIS) was developed and specifically tailored for older adults.9 Other equations, such as the Revised Lund-Malmö (RLM) equation developed in Sweden, as well as the European Kidney Function Consortium (EKFC), an equation for the full-age spectrum, have also shown good accuracy across age and GFR intervals in validation cohorts.10,11 Several studies with different sets of equations have attempted to identify the least biased equation for older populations; however, results have been mixed and no consensus has been reached.12 Additionally, given the array of eGFR equations available, quantifying the concordance between equations that are commonly used in clinical practice and research is of great importance. Misclassifying individuals with CKD may delay treatment or, conversely, subject the patient to unnecessary treatment. Moreover, poor concordance among equations hinders the comparability of existing studies providing estimates of CKD prevalence or of associations between eGFR and negative health outcomes, and prevents the synthesis of the large body of information available on the topic. Several studies have reported poor concordance between the newer equations and older, traditionally used equations, and the need for further research has been highlighted.13,14,15 To the best of our knowledge, no study to date has quantified the concordance and explored sources of discrepancies among the aforementioned 5 eGFR equations in a large, population-based cohort of older adults with and without impaired kidney function. Therefore, we aimed to: [1] quantify the concordance among MDRD, CKD-EPI, RLM, BIS, and EKFC equations in the classification of GFR categories; [2] compare the discriminative capacity of 5 eGFR equations in regards to long-term mortality risk; and [3] explore the role of low muscle mass, low body mass index (BMI), and older age as potential sources of discrepancies among the equations.
## Data Source and Sample Selection
This study is based on data from the Swedish National study on Aging and Care in Kungsholmen (SNAC-K),16 an ongoing longitudinal, community-based study of randomly sampled adults aged 60 years and above living in the *Kungsholmen area* of Stockholm, Sweden. Eligible participants who attended the baseline examination between 2001 and 2004 have been followed up regularly: every 6 years before the age of 78 years and every 3 years thereafter. At each study visit, participants undergo thorough clinical examinations, assessments, and interviews by physicians, trained nurses, and psychologists. Participants are also tracked in several national registers, namely the Swedish National Patient Register (specialist care medical history) and the Swedish Cause of Death Register (vital status). Out of the 3363 individuals ($73.3\%$ response rate) who attended the SNAC-K baseline examination between 2001 and 2004, 269 ($8.0\%$) were excluded due to missing serum creatinine, resulting in a final sample of 3094 older adults.
SNAC-K was approved by the Regional Ethical Review Board in Stockholm, and written informed consent was obtained from participants or their next of kin. The study is reported in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines for cohort studies.
## Kidney Function Assessment
Serum creatinine was measured at St. Göran’s hospital laboratory in Stockholm, where creatinine measurement was not standardized to isotope dilution mass spectrometry at the time of data collection. Thus, creatinine values were reduced by $5\%$ before calculating eGFR.17 Creatinine-based eGFR was calculated using MDRD, CKD-EPI, RLM, BIS, and EKFC (full equations available in eTable 1 in Supplement 1). Participants were grouped into eGFR categories from the KDIGO guidelines18: G1, eGFR 90 mL/min/1.73 m2 or higher; G2, 89.9 to 60 mL/min/1.73 m2; G3a, 59.9 to 45 mL/min/1.73 m2; G3b, 44.9 to 30 mL/min/1.73 m2; and G4-5, below 30 mL/min/1.73 m2.
## Vital Status and Covariates
Information about the vital status of participants was available from the Swedish Cause of Death Register from start of follow-up until December 2016. Covariates included age (continuous), sex (male or female), highest attained education (elementary, high school, or university or above), BMI (continuous), and smoking (never, former, current). BMI (calculated as weight in kilograms divided by height in meters squared) was categorized as underweight (below 23), normal weight (23 to 30), and overweight (above 30) based on suggested cutoffs from a large meta-analysis conducted among older adults.19 Information on chronic conditions (diabetes, heart failure, cancer, hypertension) was ascertained based on clinical examinations, lab blood tests, medications, patient history, and inpatient and outpatient medical records. In short, the SNAC-K physicians combined information from all sources to create a list of International Statistical Classification of Diseases and Related Health Problems, Tenth Revision (ICD-10) diagnoses for each participant. Those ICD-10 codes that reflect chronic health problems were later grouped into 60 broader categories of chronic conditions. Further information about the operationalization of chronic diseases in SNAC-K is available elsewhere.20 Calf circumference was used as a proxy for muscle mass. Low muscle mass was defined as having a calf circumference less than the 20th sex-specific percentile (below 32 cm for female and 34 cm for male individuals).21 Race and ethnicity were not considered due to data not being available.
## Statistical Analysis
Demographic and clinical baseline characteristics of participants were reported as median values for continuous variables and percentages for categorical variables. Cohen κ was used to quantify the concordance among different equations in the classification of GFR categories using the following cutoffs: below 0.20, poor; 0.21 to 0.40, fair; 0.41 to 0.60, moderate; 0.61 to 0.80, good; and 0.8 to 1.0, excellent.22 Areas under receiver operating characteristic curves (AUC) and Harrel C statistics were obtained from crude logistic regression and Cox regression models to assess how each equation predicted all-cause 15-year cumulative mortality and mortality rates. Analyses were conducted in the overall sample as well as among those with low muscle mass, those with low BMI, and those aged 78 years or older. Using the best-performing equation as reference, Bland-Altman plots were generated to plot the difference of pairs of equations against the mean of their estimates. Linear regression was used to plot the differences between the best-performing and other equations against calf circumference, BMI, and age, which were modeled using cubic splines with 5 data-driven knots. Wald tests were used to test for statistical significance. Multivariable Cox regression models were used to assess the association between eGFR (modeled using cubic splines with 5 knots prespecified at 30, 45, 60, 75, and 90) and 15-year mortality, adjusting for age, sex, education, diabetes, heart failure, cancer, hypertension, smoking, BMI, and calf circumference. Proportional hazards assumptions were tested on the basis of Schoenfeld residuals.
Statistical analysis and data visualization were performed using Stata version 17.0 (StataCorp) and Graphpad Prism 9 (GraphPad Software). Significance level was set at α <.05.
## Results
Almost two-thirds of the 3094 participants (1972 [$63.7\%$]) were female, with a median (IQR) age of 72 [66-81] years (Table 1). Over a third (1046 [$33.8\%$]) of the sample attained university education. Median (IQR) eGFR ranged from 61 [51-71] mL/min/1.73 m2 (calculated using BIS) to 69 [59-78] mL/min/1.73 m2 (calculated using CKD-EPI). Slightly over half (1559 participants [$50.4\%$]) of the study population survived until the end of follow-up on December 31, 2016. Compared with the 3094 participants included in the study, the 269 excluded individuals with missing data on creatinine were more likely to be older, male, have lower education, lower calf circumference, and a higher prevalence of hypertension and heart failure (eTable 2 in Supplement 1).
**Table 1.**
| Characteristic | Total, No. (%) (N = 3094)a |
| --- | --- |
| Age, median (IQR), y | 72 (66-81) |
| Sex | |
| Male | 1122 (36.3) |
| Female | 1972 (63.7) |
| Highest attained education | |
| Elementary | 506 (16.4) |
| High school | 1539 (49.8) |
| University | 1046 (33.8) |
| BMI | |
| Underweight (<23) | 744 (25.4) |
| Normal weight (23-30) | 1809 (61.8) |
| Overweight (>30) | 374 (12.8) |
| Smoking status | |
| Never | 1444 (47.0) |
| Former | 1179 (38.4) |
| Current | 447 (14.6) |
| Calf circumference, median (IQR), cm | 36 (34-38) |
| Diabetes | 276 (8.9) |
| Cancer | 272 (8.8) |
| Heart failure | 296 (9.6) |
| Hypertension | 2149 (69.5) |
| Creatinine, median (IQR), μmol/L | 82 (73-92) |
| eGFR, median (IQR), mL/min/1.73 m2 | |
| MDRD | 67 (59-75) |
| CKD-EPI | 69 (59-78) |
| RLM | 62 (53-70) |
| BIS | 61 (51-71) |
| EKFC | 63 (53-72) |
When considering the concordance among equations with regards to classifying eGFR categories and category distribution for each equation, Cohen κ between dyads of equations ranged from 0.42 to 0.91, with the poorest concordance between MDRD-BIS and best between RLM-EKFC (Figure 1). The distribution of eGFR categories varied across equations, with MDRD and CKD-EPI placing more individuals in G1 and G2 categories compared with the other 3 equations. The prevalence of G3 and above was $29.2\%$ using MDRD, $28.0\%$ using CKD-EPI, $43.2\%$ using RLM, $47.7\%$ using BIS, and $42.3\%$ using EKFC. The concordance between dyads including MDRD or CKD-EPI, and RLM or BIS or EKFC further decreased among the subgroups of participants aged 78 years or older and those with low muscle mass, with the poorest concordance between MDRD and BIS among those aged 78 years or older (Cohen κ = 0.17) (eFigure 1 in Supplement 1). RLM, BIS, and EKFC maintained good or excellent concordance with each other in all subgroups.
**Figure 1.:** *Concordance Among Equations in the Staging of Chronic Kidney Disease (CKD) and Distribution of Participants Across Estimated Glomerular Filtration Rate (eGFR) CategoriesG1 represents eGFR ≥90 mL/min/1.73 m2; G2, 89.9-60 mL/min/1.73 m2; G3a, 59.9-45 mL/min/1.73 m2; G3b, 44.9-30 mL/min/1.73 m2; G4-5, <30 mL/min/1.73 m2. BIS indicates Berlin Initiative Study; CKD-EPI, Chronic Kidney Disease Epidemiological Collaboration; EKFC, European Kidney Function Consortium; MDRD, Modification of Diet in Renal Disease; RLM, Revised Lund-Malmö.*
The best mix of AUC and Harrel C statistic was observed for BIS in the overall sample as well as in the subgroup analyses (Table 2). The prognostic accuracy was decreased among age 78 years or older and low muscle mass subgroups, but remained similar among those with low BMI.
**Table 2.**
| Equation | Overall (n = 3094) | Overall (n = 3094).1 | Low muscle mass (n = 407)b | Low muscle mass (n = 407)b.1 | Low BMI (n = 744)c | Low BMI (n = 744)c.1 | Age ≥78 y (n = 1369) | Age ≥78 y (n = 1369).1 |
| --- | --- | --- | --- | --- | --- | --- | --- | --- |
| Equation | AUC (95% CI) | Harrel C (95% CI) | AUC (95% CI) | Harrel C (95% CI) | AUC (95% CI) | Harrel C (95% CI) | AUC (95% CI) | Harrel C (95% CI) |
| MDRD | 0.66 (0.64-0.68) | 0.62 (0.61-0.64) | 0.72 (0.66-0.78) | 0.61 (0.58-0.65) | 0.66 (0.62-0.70) | 0.62 (0.60-0.65) | 0.58 (0.54-0.61) | 0.56 (0.55-0.58) |
| CKD-EPI | 0.72 (0.70-0.74) | 0.67 (0.66-0.69) | 0.78 (0.72-0.83) | 0.64 (0.61-0.68) | 0.73 (0.70-0.77) | 0.67 (0.65-0.70) | 0.60 (0.57-0.64) | 0.58 (0.56-0.60) |
| RLM | 0.78 (0.76-0.79) | 0.71 (0.70-0.73) | 0.81 (0.76-0.87) | 0.67 (0.64-0.70) | 0.79 (0.76-0.82) | 0.72 (0.69-0.74) | 0.63 (0.60-0.67) | 0.60 (0.58-0.62) |
| BIS | 0.80 (0.78-0.81) | 0.73 (0.72-0.74) | 0.82 (0.77-0.87) | 0.68 (0.64-0.71) | 0.81 (0.78-0.84) | 0.73 (0.71-0.75) | 0.64 (0.61-0.68) | 0.61 (0.59-0.62) |
| EKFC | 0.76 (0.74-0.77) | 0.70 (0.69-0.71) | 0.80 (0.75-0.86) | 0.66 (0.63-0.69) | 0.77 (0.74-0.81) | 0.70 (0.68-0.73) | 0.62 (0.58-0.65) | 0.59 (0.57-0.61) |
Bland-Altman plots were used to compare BIS to the other 4 equations (eFigure 2 in Supplement 1). On average, MDRD and CKD-EPI provided higher estimates of eGFR by 6.0 ($95\%$ limits of agreement [LoA], −6.9 to 18.9) and 7.4 ($95\%$ LoA, −1.1 to 15.9) mL/min/1.73 m2 compared with BIS, respectively. Both equations showed greater differences at higher levels of eGFR. EKFC provided higher estimates of eGFR by 1.6 ($95\%$ LoA, −3.8 to 6.9) while the average bias between RLM and BIS was close to zero (0.1; $95\%$ LoA, −4.5 to 4.8).
Results from linear regressions using cubic splines showed that calf circumference, BMI, and age were significantly associated with the differences among all pairs of equations (all $P \leq .001$). The differences were more pronounced toward the lower values of calf circumference and BMI, of higher age, and of larger magnitude for MDRD and CKD-EPI compared with RLM and EKFC (Figure 2). The threshold at which the risk of mortality increases was lower for RLM, BIS, and EKFC compared with MDRD and CKD-EPI (Figure 3).
**Figure 2.:** *Discrepancies Between Estimated Glomerular Filtration Rate (eGFR) EquationsBIS indicates Berlin Initiative Study; MDRD, Modification of Diet in Renal Disease; CKD-EPI, Chronic Kidney Disease Epidemiological Collaboration; RLM, Revised Lund-Malmö; EKFC, European Kidney Function Consortium. Shaded areas represent 95% CIs.* **Figure 3.:** *Associations Between Estimated Glomerular Filtration Rate (eGFR) Equations and 15-year MortalityAll models adjusted for age, sex, education, diabetes, heart failure, cancer, hypertension, smoking, BMI, calf circumference. Low muscle mass includes individuals with calf circumference less than the 20th sex-specific percentile; low BMI, <23. BIS indicates Berlin Initiative Study; BMI, body mass index (calculated as weight in kilograms divided by height in meters squared); CKD-EPI, Chronic Kidney Disease Epidemiological Collaboration; EKFC, European Kidney Function Consortium; MDRD, Modification of Diet in Renal Disease; RLM, Revised Lund-Malmö.*
## Discussion
Our findings highlight a poor concordance among eGFR equations in this community-based cohort of Swedish older adults. Equations that were originally developed in populations where older adults were well-represented (RLM, BIS, and EKFC) showed good concordance with each other and poor concordance with equations that included fewer older adults in their development cohorts (MDRD and CKD-EPI). Both absolute differences in eGFR and concordance in CKD staging remained consistent across levels of muscle mass, body mass, and age for RLM, BIS, and EKFC. Of these equations, BIS showed the best mortality prediction, but predictive accuracy was poorer among participants aged 78 years or older and those with low muscle mass (ie, those commonly suffering from sarcopenia and underweight).
Kidney function may decline across the lifespan, and there has been a decades-long debate on how to separate age-related declines from disease-related declines. Some experts suggest that eGFR thresholds to define CKD should be age specific,23,24 but an age-specific definition of CKD is not yet adopted by clinical guidelines.25 Given that equations tend to perform best in populations similar to their development cohorts, it is not surprising that BIS, RLM, and EKFC showed better concordance with each other than with MDRD and CKD-EPI equations, and that these discordances were further amplified in subgroup analyses. Indeed, the mean age of participants was 63 in the RLM development cohort11 and 78.5 in the BIS cohort,26 while EKFC included several older adult cohorts.10 In contrast, older adults were far less represented in the CKD-EPI and MDRD development cohorts (mean ages of 47 and 50.6, respectively3,27), and the latter was developed in a cohort of CKD patients.
In addition to physiological changes in kidney function, old age is associated with changes in non–GFR-related determinants of creatinine levels. Given that the main source of serum creatinine variation is its generation after muscle breakdown, creatinine values may remain within the normal range despite a significantly impaired kidney function among individuals with low muscle mass. This could explain the large discrepancies we observed between equations at the extremes of BMI and muscle mass, as being underweight and having sarcopenia are prevalent conditions in old age that affect the metabolism of kidney function biomarkers, and equations that extrapolate findings from younger adults may not be able to fully account for such changes.
In our study, BIS outperformed all other equations in predicting mortality over the 15-year follow-up period, similar to other studies where BIS was compared with CKD-EPI.28,29 This was not surprising as, on the one hand, GFR is known to be associated with mortality,30 and on the other hand, several studies conducted in similar populations have identified the BIS equation as the least biased equation when compared with the criterion standard measured GFR (mGFR).12 However, even for the best-performing equation, the discriminative capacity was fair at best, particularly in subgroup analyses. This was somewhat expected given that eGFR equations were not created or validated to predict mortality, and that in our cohort of community-dwelling older adults, the prevalence of CKD is lower, and several other causes of death are competing with CKD and its complications.
The small differences observed in mortality prediction may not be as clinically relevant as the potentially large misclassification of CKD stages between different equations. In our study, CKD-EPI and MDRD, which on average provided higher estimates of GFR, had good concordance with each other but at best moderate concordance with the other 3 equations. Similar to our findings, a study conducted among 828 community-dwelling older adults aged 65 years or older in Italy found poor concordance between CKD-EPI and BIS equations (κ = 0.39) in the overall sample, and higher relative misclassification among participants aged 80 years or older and those with limitations in basic activities of daily living.13 Likewise, a study conducted among cardiology unit patients in Italy reported moderate concordance between MDRD and BIS, and CKD-EPI and BIS among older participants.28 On the other hand, a French study comparing CKD-EPI with RLM and BIS did not find significant differences between equations when comparing them to mGFR obtained by measuring iohexol clearance.31 The conflicting findings could be due to the differences in study populations; the French study used a clinical sample of older adults with suspected or confirmed kidney function impairment, whereas our study population included healthier participants sampled from the population. Another explanation could be that our population more closely resembled the development cohorts of the RLM and BIS equations.
Given that different CKD stages can trigger different therapeutic approaches not only in the context of CKD itself but also its comorbidities, these discordances may further complicate the health care needs encountered by older adults. Indeed, $89\%$ of the SNAC-K population suffers from multimorbidity (defined as the presence of 2 or more chronic diseases),20 and at least a third of the multimorbid population is on polypharmacy.32 Several commonly used medications, such as methotrexate and metformin, should not be administered to those with advanced CKD as per the guidelines, and careful consideration of dosage and alternatives is needed for other antidiabetic, cardiovascular, and anticonvulsive medications with kidney clearance.33 Thus, diverging classifications of older peoples’ kidney function have direct and important clinical implications, and validation studies against measured GFR among the oldest patients and those with sarcopenia are highly warranted.
## Limitations
Some limitations of our study need to be highlighted. Measured GFR was not available in our study, thus we were limited to making relative comparisons between equations. We used mortality as an outcome due to its high relevance as a biological endpoint and the availability of comprehensive register data; however, the equations under study were not designed to directly assess mortality risk. We could not differentiate between transient and chronic declines in kidney function due to a single measurement of creatinine available in our study. However, this is unlikely to significantly affect the relative comparisons of eGFR equations, which was the main aim of this paper. Finally, the generalizability of our findings may be limited due to SNAC-K participants being healthier and more affluent than the average older person in Sweden. On the other hand, a major strength of the present study is that it is based on a large population-based sample of older adults with and without kidney function impairment and with comprehensive sociodemographic and clinical information available. Additional strengths include the long follow-up time and that all blood samples were collected and analyzed at a single lab in Stockholm, reducing random variability in creatinine assessment.
## Conclusions
In this population-based cohort study of Swedish older adults, we found that not all eGFR equations were interchangeable when assessing kidney function. BIS outperformed other equations with respect to mortality discrimination; however, the overall predictive capacity was only fair and further reduced in subgroups of individuals aged 78 years or older and those with low muscle mass. Validation studies against measured GFR are highly warranted in these subgroups. Until then, clinicians should consider the potential discrepancies between different eGFR equations when monitoring kidney function in old age. Accordingly, researchers should carefully consider the choice of equations in their studies.
## References
1. Bikbov B, Purcell CA, Levey AS. **Global, regional, and national burden of chronic kidney disease, 1990-2017: a systematic analysis for the Global Burden of Disease Study 2017**. *Lancet* (2020.0) **395** 709-733. DOI: 10.1016/S0140-6736(20)30045-3
2. Gasparini A, Evans M, Coresh J. **Prevalence and recognition of chronic kidney disease in Stockholm healthcare**. *Nephrol Dial Transplant* (2016.0) **31** 2086-2094. DOI: 10.1093/ndt/gfw354
3. Levey AS, Stevens LA, Schmid CH. **A new equation to estimate glomerular filtration rate**. *Ann Intern Med* (2009.0) **150** 604-612. DOI: 10.7326/0003-4819-150-9-200905050-00006
4. Levey AS, Coresh J, Greene T. **Using standardized serum creatinine values in the modification of diet in renal disease study equation for estimating glomerular filtration rate**. *Ann Intern Med* (2006.0) **145** 247-254. DOI: 10.7326/0003-4819-145-4-200608150-00004
5. Inker LA, Eneanya ND, Coresh J. **New creatinine- and cystatin C-based equations to estimate GFR without race**. *N Engl J Med* (2021.0) **385** 1737-1749. DOI: 10.1056/NEJMoa2102953
6. Levin A, Stevens PE, Bilous RW. **Kidney disease: Improving global outcomes (KDIGO) CKD work group. KDIGO 2012 clinical practice guideline for the evaluation and management of chronic kidney disease**. *Kidney Int Suppl* (2013.0) **3** 1-150. DOI: 10.1038/kisup.2012.73
7. St-Onge MP, Gallagher D. **Body composition changes with aging: the cause or the result of alterations in metabolic rate and macronutrient oxidation?**. *Nutrition* (2010.0) **26** 152-155. DOI: 10.1016/j.nut.2009.07.004
8. Dharmarajan TS, Yoo J. **Estimated glomerular filtration rate and muscle mass in older patients: diagnostic accuracy of creatinine-based equations and implications in practice**. *J Am Med Dir Assoc* (2020.0) **21** 566-567. DOI: 10.1016/j.jamda.2020.01.098
9. Alshaer IM, Kilbride HS, Stevens PE. **External validation of the Berlin equations for estimation of GFR in the elderly**. *Am J Kidney Dis* (2014.0) **63** 862-865. DOI: 10.1053/j.ajkd.2014.01.013
10. Pottel H, Björk J, Courbebaisse M. **Development and validation of a modified full age spectrum creatinine-based equation to estimate glomerular filtration rate: a cross-sectional analysis of pooled data**. *Ann Intern Med* (2021.0) **174** 183-191. DOI: 10.7326/M20-4366
11. Nyman U, Grubb A, Larsson A. **The revised Lund-Malmö GFR estimating equation outperforms MDRD and CKD-EPI across GFR, age and BMI intervals in a large Swedish population**. *Clin Chem Lab Med* (2014.0) **52** 815-824. DOI: 10.1515/cclm-2013-0741
12. Raman M, Middleton RJ, Kalra PA, Green D. **Estimating renal function in old people: an in-depth review**. *Int Urol Nephrol* (2017.0) **49** 1979-1988. DOI: 10.1007/s11255-017-1682-z
13. Corsonello A, Pedone C, Bandinelli S, Ferrucci L, Antonelli Incalzi R. **Agreement between Chronic Kidney Disease Epidemiological Collaboration and Berlin Initiative Study equations for estimating glomerular filtration rate in older people: the Invecchiare in Chianti (Aging in Chianti Region) study**. *Geriatr Gerontol Int* (2017.0) **17** 1559-1567. DOI: 10.1111/ggi.12932
14. Guo M, Niu JY, Ye XW. **Evaluation of various equations for estimating renal function in elderly Chinese patients with type 2 diabetes mellitus**. *Clin Interv Aging* (2017.0) **12** 1661-1672. DOI: 10.2147/CIA.S140289
15. Zafari N, Lotfaliany M, O’Keefe GJ. **Performance of 4 creatinine-based equations in assessing glomerular filtration rate in adults with diabetes**. *J Clin Endocrinol Metab* (2021.0) **106** e61-e73. DOI: 10.1210/clinem/dgaa722
16. 16Swedish National study on Aging and Care in Kungsholmen. SNAC-K website. Accessed August 15, 2022. https://www.snac-k.se/
17. Levey AS, Coresh J, Greene T. **Expressing the Modification of Diet in Renal Disease Study equation for estimating glomerular filtration rate with standardized serum creatinine values**. *Clin Chem* (2007.0) **53** 766-772. DOI: 10.1373/clinchem.2006.077180
18. Stevens PE, Levin A. **Evaluation and management of chronic kidney disease: synopsis of the kidney disease: improving global outcomes 2012 clinical practice guideline**. *Ann Intern Med* (2013.0) **158** 825-830. DOI: 10.7326/0003-4819-158-11-201306040-00007
19. Winter JE, MacInnis RJ, Wattanapenpaiboon N, Nowson CA. **BMI and all-cause mortality in older adults: a meta-analysis**. *Am J Clin Nutr* (2014.0) **99** 875-890. DOI: 10.3945/ajcn.113.068122
20. Calderón-Larrañaga A, Vetrano DL, Onder G. **Assessing and measuring chronic multimorbidity in the older population: a proposal for its operationalization**. (2017.0) **72** 1417-1423. DOI: 10.1093/gerona/glw233
21. Newman AB, Kupelian V, Visser M. **Sarcopenia: alternative definitions and associations with lower extremity function**. *J Am Geriatr Soc* (2003.0) **51** 1602-1609. DOI: 10.1046/j.1532-5415.2003.51534.x
22. Altman DG. *Practical Statistics for Medical Research* (1999.0)
23. Waas T, Schulz A, Lotz J. **Distribution of estimated glomerular filtration rate and determinants of its age dependent loss in a German population-based study**. *Sci Rep* (2021.0) **11** 10165. DOI: 10.1038/s41598-021-89442-7
24. Liu P, Quinn RR, Lam NN. **Accounting for age in the definition of chronic kidney disease**. *JAMA Intern Med* (2021.0) **181** 1359-1366. DOI: 10.1001/jamainternmed.2021.4813
25. Delanaye P, Jager KJ, Bökenkamp A. **CKD: a call for an age-adapted definition**. *J Am Soc Nephrol* (2019.0) **30** 1785-1805. DOI: 10.1681/ASN.2019030238
26. Schaeffner ES, Ebert N, Delanaye P. **Two novel equations to estimate kidney function in persons aged 70 years or older**. *Ann Intern Med* (2012.0) **157** 471-481. DOI: 10.7326/0003-4819-157-7-201210020-00003
27. Levey AS, Bosch JP, Lewis JB, Greene T, Rogers N, Roth D. **A more accurate method to estimate glomerular filtration rate from serum creatinine: a new prediction equation**. *Ann Intern Med* (1999.0) **130** 461-470. DOI: 10.7326/0003-4819-130-6-199903160-00002
28. Malavasi VL, Valenti AC, Ruggerini S. **Kidney function according to different equations in patients admitted to a cardiology unit and impact on outcome**. *J Clin Med* (2022.0) **11** 891. DOI: 10.3390/jcm11030891
29. Corsonello A, Pedone C, Bandinelli S, Ferrucci L, Antonelli Incalzi R. **Predicting survival of older community-dwelling individuals according to five estimated glomerular filtration rate equations: the InChianti study**. *Geriatr Gerontol Int* (2018.0) **18** 607-614. DOI: 10.1111/ggi.13225
30. Inker LA, Titan S. **Measurement and estimation of GFR for use in clinical practice: core curriculum 2021**. *Am J Kidney Dis* (2021.0) **78** 736-749. DOI: 10.1053/j.ajkd.2021.04.016
31. da Silva Selistre L, Rech DL, de Souza V, Iwaz J, Lemoine S, Dubourg L. **Diagnostic performance of creatinine-based equations for estimating glomerular filtration rate in adults 65 years and older**. *JAMA Intern Med* (2019.0) **179** 796-804. DOI: 10.1001/jamainternmed.2019.0223
32. Marengoni A, Akugizibwe R, Vetrano DL. **Patterns of multimorbidity and risk of disability in community-dwelling older persons**. *Aging Clin Exp Res* (2021.0) **33** 457-462. DOI: 10.1007/s40520-020-01773-z
33. Hartmann B, Czock D, Keller F. **Drug therapy in patients with chronic renal failure**. *Dtsch Arztebl Int* (2010.0) **107** 647-655. DOI: 10.3238/arztebl.2010.0647
|
---
title: 'Remote Monitoring of Physiology in People Living With Dementia: An Observational
Cohort Study'
journal: JMIR Aging
year: 2023
pmcid: PMC10037178
doi: 10.2196/43777
license: CC BY 4.0
---
# Remote Monitoring of Physiology in People Living With Dementia: An Observational Cohort Study
## Abstract
### Background
Internet of Things (IoT) technology enables physiological measurements to be recorded at home from people living with dementia and monitored remotely. However, measurements from people with dementia in this context have not been previously studied. We report on the distribution of physiological measurements from 82 people with dementia over approximately 2 years.
### Objective
Our objective was to characterize the physiology of people with dementia when measured in the context of their own homes. We also wanted to explore the possible use of an alerts-based system for detecting health deterioration and discuss the potential applications and limitations of this kind of system.
### Methods
We performed a longitudinal community-based cohort study of people with dementia using “Minder,” our IoT remote monitoring platform. All people with dementia received a blood pressure machine for systolic and diastolic blood pressure, a pulse oximeter measuring oxygen saturation and heart rate, body weight scales, and a thermometer, and were asked to use each device once a day at any time. Timings, distributions, and abnormalities in measurements were examined, including the rate of significant abnormalities (“alerts”) defined by various standardized criteria. We used our own study criteria for alerts and compared them with the National Early Warning Score 2 criteria.
### Results
A total of 82 people with dementia, with a mean age of 80.4 (SD 7.8) years, recorded 147,203 measurements over 958,000 participant-hours. The median percentage of days when any participant took any measurements (ie, any device) was $56.2\%$ (IQR $33.2\%$-$83.7\%$, range $2.3\%$-$100\%$). Reassuringly, engagement of people with dementia with the system did not wane with time, reflected in there being no change in the weekly number of measurements with respect to time (1-sample t-test on slopes of linear fit, $$P \leq .45$$). A total of $45\%$ of people with dementia met criteria for hypertension. People with dementia with α-synuclein–related dementia had lower systolic blood pressure; $30\%$ had clinically significant weight loss. Depending on the criteria used, $3.03\%$-$9.46\%$ of measurements generated alerts, at 0.066-0.233 per day per person with dementia. We also report 4 case studies, highlighting the potential benefits and challenges of remote physiological monitoring in people with dementia. These include case studies of people with dementia developing acute infections and one of a person with dementia developing symptomatic bradycardia while taking donepezil.
### Conclusions
We present findings from a study of the physiology of people with dementia recorded remotely on a large scale. People with dementia and their carers showed acceptable compliance throughout, supporting the feasibility of the system. Our findings inform the development of technologies, care pathways, and policies for IoT-based remote monitoring. We show how IoT-based monitoring could improve the management of acute and chronic comorbidities in this clinically vulnerable group. Future randomized trials are required to establish if a system like this has measurable long-term benefits on health and quality of life outcomes.
## Introduction
Dementia engenders a significant burden to patients, carers, and health care services. In the United Kingdom (UK), there are an estimated 850,000 people with dementia, a number expected to rise to over 2 million by 2051 [1]. Dementia care costs the National Health Service approximately £23 (US $27.88) billion per year. There is a pressing need for interventions that reduce the burden on health care services and carers.
In addition to cognitive and behavioral symptoms, dementia is commonly associated with long-term comorbidities, including hypertension, diabetes, malnutrition or unintentional weight loss, and heart disease [2-5]. Many such comorbidities are adverse factors in its progression [2,3,6-10] but are underrecognized and undertreated [11,12]. People with dementia are also at increased risk of hospital admission especially for infections and falls [3,13-15]. People with dementia are more likely to die during admissions, and over a third who go into hospital from home are discharged to a care home [16,17]. Abnormal physiological measurements are more common in people with dementia because of autonomic dysfunction, comorbidities, medication side effects, and acute illnesses [18-21]. People with dementia are also at increased risk of frailty [22]. The interaction between frailty and acute illnesses confers a multiplicative risk for significant morbidity and mortality [23]. A combination of poor premorbid function with atypical presentations and a reduced ability to describe and communicate symptoms drives the increased risk [24]. Recognizing and treating illnesses early leads to better outcomes, especially in the elderly [25].
Using “Internet of Things” (IoT) technology, physiological measurements can be recorded at home and transmitted automatically to caregivers [26]. Such technology can improve the monitoring and treatment of comorbidities and detect developing acute illness [26]. Higher temporal frequency of measurements in a more “naturalistic” setting can potentially provide more accurate, granular data on patients’ health. It also reduces the need for patients with reduced mobility to travel to access care. Therefore, IoT could improve the health and quality of life of people with dementia and reduce the burden on services [26]. Furthermore, by involving people with dementia in their own care, we can maximize empowerment regarding their own health [27]. This is vital for effective ethical care and can be in part enabled through technology [28]. There are also likely benefits to carers who can be involved directly in the ongoing assessments while having increased confidence to leave the people with dementia alone.
There is considerable interest from health and social care policy makers in systems that enable remote monitoring of physiological parameters in community-dwelling people with dementia [29,30], especially given the COVID-19 pandemic [31]. However, in-home monitoring creates new challenges for clinical practice. Guidelines for the frequency of remote measurements (eg, daily, weekly), definitions of clinically significant abnormalities (eg, the threshold heart rate [HR] of clinically relevant tachycardia), and algorithms for best-practice care are not established, unlike in the inpatient setting. The UK-wide National Early Warning Score 2 (NEWS, Table S1 in Multimedia Appendix 1) defines illness severity for hospitalized patients using a score aggregated from 6 physiological domains, where a higher subscore means the parameter is further from normal [32]. However, it is unclear whether NEWS, validated for inpatients, is suitable in the home setting. Also, little is known about the distribution of physiological measurements in people with dementia recorded in the community, which is crucial to establish when designing a system to detect abnormalities.
We are unaware of any published data from long-term studies employing IoT devices for physiological monitoring in the homes of people with dementia.
We have developed an IoT platform, “Minder,” that enables physiological measurements to be recorded at home and monitored remotely [33]. Here, we carried out an analysis of the physiology (HR, systolic blood pressure [SBP] and diastolic blood pressure [DBP], oxygen saturation, and body weight) recorded by a group of people with dementia at home. We report on the effectiveness of our Minder system, designed to detect abnormal measurements (“alerts”) and direct a clinical monitoring team. We also retrospectively applied the NEWS criteria to the data as a comparator. Finally, we present case studies highlighting the potential benefits of remotely monitoring people with dementia.
Our aims are to [1] characterize the physiology of people with dementia in their home setting, [2] test whether our system is sensitive to comorbidities and dementia subtypes, and [3] test how well NEWS-style alerts systems translate to community measurements.
## Study Design, Participants, and Recruitment
We are conducting an ongoing longitudinal community-based cohort study of people with dementia living at home using Minder, passive infrared sensing, and data analytics to enable remote health care monitoring [33]. Patients with an existing clinical diagnosis of dementia of any cause were recruited from primary care, adult social care services, and memory clinics across Surrey and Borders Partnership NHS Foundation Trust and Hammersmith and Fulham Partnership. People with dementia were enrolled with an associated “study partner,” defined as “a relative or friend who has known the people with dementia for at least 6 months.” A distinction was not made between “study partner” and “carer”; however, the average number of hours the study partners spent caring for their respective people with dementia was 5.4 (range 1-8) for the 40 study partners for whom we have this data. Full inclusion and exclusion criteria are listed in “Methods” in Multimedia Appendix 2. Owing to the developmental and exploratory nature of the ongoing study, the number of participants was not predetermined by a power calculation. The study is reported according to the STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) statement (Multimedia Appendix 3), guidelines for reporting observational studies [34].
## Ethics Approval
The study was approved by the Health Research Authority’s London-Surrey Borders Research Ethics Committee (19/LO/0102). All people with dementia and study partners provided written informed consent for participation and for their data to be included in publications.
## Study Procedures and Physiological Measurements
The study protocol and devices used were based on a previous trial [26]. That trial was co-designed with 20 people with dementia, carers, health care and social workers, and academics, who designed the system to be appropriate for use in people with dementia, in addition to data from an Alzheimer's Society survey on technology-enhanced care. The system was first tested in a laboratory setting and home mock-up scenario before being deployed in participants’ homes.
At baseline, people with dementia and their study partners completed demographics questionnaires and people with dementia completed the Standardized Mini-Mental State Examination (SMMSE). All people with dementia received up to 4 IoT medical devices to record 6 physiological measurements: a blood pressure machine for SBP and DBP, a pulse oximeter measuring oxygen saturation and HR, and body weight scales (provided by iHealth), and a thermometer (provided by Withings), and were asked to use each device once a day at any time. Measurements recorded by each device, annotated with a datetime stamp, were automatically transmitted immediately to a centralized secure server.
## Study Oversight and Minder Alert Criteria
All people with dementia and study partners provided written informed consent. To oversee the study, a monitoring team was established, operating from 9 AM to 5 PM daily, to respond to clinical or technical alerts. The monitoring team was supervised by a consultant psychiatrist, 2 consultant neurologists, a general practitioner, and an occupational therapist. The team had near real-time access to the physiological measurements via a clinical dashboard, which additionally alerted staff when observations met standardized criteria devised by the study team (Table S2 in Multimedia Appendix 4). For any alert, the monitoring team would follow a predefined flowchart to investigate the abnormality, beginning by attempting to contact the people with dementia or study partner (Figures S1-S4 in Multimedia Appendices 5-8, respectively).
## Statistical Analyses
All analyses were performed in MATLAB [35]. Data distributions were assessed for normality. Mean and SD are reported for *Gaussian data* (as per the Shapiro-Wilk test), whereas median and IQR are used for non-Gaussian data. We calculated descriptive statistics for baseline demographics and SMMSE, and grouped these data by dementia subtype, defined as Alzheimer disease (AD), vascular dementia (VD), and α-synuclein–associated dementias (ASyn), that is, combining participants with Parkinson disease dementia and Lewy body dementia.
We summarized the 24-hour timing of measurements by grouping timestamps into hourly bins. We calculated the overall frequency of recordings by dividing the number of days where a participant took at least 1 complete set (ie, all devices available) of measurements by the number of days of observation; we also counted the days when any measurement was taken. To examine whether the frequency of measurements changed over time, for each participant, we counted the number of measurements per week; fit a linear model to this time series; and, at the group level, tested whether the slopes of these fits were significantly different to zero (1-sample t test).
For subsequent analyses, measurement values were excluded as outliers if values were greater than 4 SD from the mean for that participant. For blood pressure, we calculated the proportion of people with dementia whose mean values met clinical criteria for hypertension (SBP/DBP ≥$\frac{135}{85}$ mg) [36] or hypotension (SBP/DBP ≤$\frac{90}{60}$ mm Hg) [37].
We charted the body weight of people with dementia over time, using a sliding window average of 5 values. To identify clinically significant weight loss (or gain) [38] in participants with at least 6 months of data, we identified averaged weights that were >$5\%$ different, in either direction, from the most recent value recorded at least 6 months previously.
We refer here to 2 different sets of alert thresholds: our own Minder thresholds and the established NEWS thresholds (Table S1 in Multimedia Appendix 1 and Table S2 in Multimedia Appendix 4). The Minder thresholds were used in real time during the study, to alert the monitoring team, whereas the NEWS thresholds were applied retrospectively to the data for comparison. When applying the subscore thresholds of the NEWS [32] to the data, we first removed any repeat measurements recorded within 60 seconds. For each domain within the NEWS criteria, a normal value is scored 0, but 1-3 when values meet predetermined criteria for abnormality (Table S1 in Multimedia Appendix 1), and these subscores are aggregated into a single NEWS score [32]. For each participant, we calculated the number of individual measurements that triggered Minder alerts and NEWS thresholds of 1+ (less abnormal) and 2+ (more abnormal). We then summarized the overall “burden” of alerts per day per participant.
We used the Spearman rank to test for correlations between physiological summary measures and baseline SMMSE scores and 1-way ANOVA to test for differences between the dementia subtypes (AD, VD, ASyn), with post hoc Tukey tests. Owing to the exploratory nature of the analysis, we did not correct for multiple comparisons.
## Case Studies
Four case studies, each based on a snapshot from an individual person with dementia, have been identified. These are included for the purpose of demonstrating the use of this monitoring system for detecting acute clinical events as well as chronic changes in physiological measures over time.
Here we report four case studies demonstrating the ability of the monitoring system to detect clinically relevant changes in physiological measurements. This provides evidence for the use of the system in picking up acute illness in a timely manner while also allowing clinicians to monitor chronic changes in individual patients.
**Figure 5:** *Case studies highlighting the potential benefits and challenges of remote physiological monitoring in people with dementia. Case studies are labeled according to descriptions in text. HR: heart rate; UTI: urinary tract infection.*
## Participant Characteristics, Dementia Subtypes, and Analysis Period
Data from 82 people with dementia were analyzed, with a mean age of 80.4 (SD 7.8, range 60.5-96.4) years at study entry; 36 ($44\%$) were women. Table 1 shows baseline participant characteristics including SMMSE scores (mean 23.0, SD 4.2), grouped by dementia subtypes—AD, VD, and ASyn. The medical history of people with dementia was accessed via their general practitioner (GP) records on enrollment: 1 had type 1 diabetes mellitus and 2 had type 2 diabetes mellitus. Although 18 records stated that they had a diagnosis of essential hypertension, 29 were on at least 1 medication with antihypertensive action (calcium channel blockers, angiotensin-converting enzyme inhibitors, β-blockers, α-blocker, angiotensin receptor blockers, and diuretics). There was a significant difference in age between dementia subtypes (1-way ANOVA F2,79=5.346, $$P \leq .007$$), with participants with AD older than those with ASyn (post hoc Tukey test, $$P \leq .004$$). There was no difference in baseline SMMSE scores between dementia subtypes (F2,78=2.324, $$P \leq .11$$). Recruitment to the study was ongoing throughout, and thus those included commenced the study at different points in the analysis period: April 1, 2019, to March 14, 2022 (1078 days; Figure 1A). The median number of days of observations per patient, defined as the days between the first and last recorded measurement, was 432.5 (IQR 164.9-764.0, range 15.8-1077.1) days, a total of 957,861 participant-hours. A total of 37 participants withdrew from the study during the analysis period, including 5 ($6\%$) who died. The most frequent reason for withdrawal was people with dementia moving to a care home. Full details of the withdrawals are reported in Table S3 in Multimedia Appendix 9. Given the nature of the population in this observational study, it was expected that a significant proportion would withdraw or pass away, as their condition progressed.
## Number, Timing, and Frequency of Physiological Measurements
There were 147,203 individual measurements recorded among the 82 participants. Measurements were most often recorded in the morning, with 8-9 AM the most frequent hour (Figure 1B), but there was wide between- and within-participant variability (Figure 1C). We defined participants’ frequency of physiological measurement recording in 2 ways. The median proportion of days of observation during which any participant took at least 1 full set of measurements was $13.9\%$ (IQR $1.2\%$-$33.1\%$, range $0\%$-$61.0\%$). The median percentage of days when people with dementia took at least 1 measurement using any device was $56.2\%$ (IQR $33.2\%$-$83.7\%$, range $2.3\%$-$100\%$) and did not differ between dementia subtypes (F2,79=0.944, $$P \leq .91$$). By either definition, there was no correlation between frequency and SMMSE (Spearman ρ $$P \leq .17$$ and $$P \leq .095$$, respectively). We also examined whether measurement frequency changed over the study, that is, that might reflect study fatigue or difficulties with devices. There was no change in the weekly number of measurements in people with dementia with respect to time (1-sample t test on slopes of linear fit, $$P \leq .45$$). There was also no correlation between any change in frequency (fitted slopes) and SMMSE ($$P \leq .34$$).
## Values of Physiological Measurements and Prevalence of Hypertension
Figure 1D-I shows for each measurement domain the distributions of participants’ mean values recorded during the study for people with dementia with more than 7 days of measurements. Grand means, calculated as the mean of all the within-subject means, were as follows: group mean HR 69.6 (SD 9.4, range 53.5-97.4) bpm, mean SBP 131.7 (SD 14.1, range 85.0-165.9) mm Hg, mean DBP 74.9 (SD 7.5, range 47.7-90.0) mm Hg, median temperature 36.4 (IQR 36.2-36.6, range 36.0-37.2) °C, median oxygen saturation $95.2\%$ (IQR $94.4\%$-$96.5\%$, range $86.7\%$-$97.5\%$), median body weight 71.4 (IQR 61.8-83.1, range 48.7-132.0) kg. Using typical clinical criteria [36,37], $45.4\%$ of people with dementia with available data had hypertension (within-subject mean either SBP/DBP ≥$\frac{135}{85}$ mg), and 1 had hypotension (within-subject mean either SBP/DBP ≤$\frac{90}{60}$ mm Hg; Figure 1D and E). Figure S5A-F in Multimedia Appendix 10 shows the distributions of within-participant SDs of values.
## Physiological Measurements Between Dementia Subtypes
We grouped the within-participant means and SDs by dementia subtypes (Figure 2A-D for SBP/DBP; see Figure S6 in Multimedia Appendix 11 for other domains). There was a significant difference across subtypes in mean SBP (F2,63=6.203, $$P \leq .003$$), SD of DBP (F2,63=3.790, $$P \leq .03$$, post hoc Tukey test results shown in Figure 2), and SD of oxygen saturation (F2,37=6.317, $$P \leq .004$$), with ASyn higher than AD participants (post hoc Tukey test, $$P \leq .005$$) and VD participants (post hoc Tukey test, $$P \leq .01$$).
**Figure 2:** *Within-participant means and SDs of physiological measurements in people with dementia, grouped by dementia subtypes. Only participants with more than 7 days of data are included. Results of post hoc Tukey test are shown where significant (P<.05). (A) Mean systolic blood pressure (SBP, mm Hg). (B) Mean diastolic blood pressure (DBP, mm Hg). (C) SD of SBP (mm Hg). (D) SD of DBP (mm Hg). AD: Alzheimer disease; ASyn: α-synuclein–associated disorders; bp: blood pressure; VD: vascular dementia.*
## Body Weight and Prevalence of Clinically Significant Weight Loss
Using the criteria of >$5\%$ change over 6 months [38] (Figure 3A), 13 ($28\%$) participants who recorded weight measurements for at least 6 months had at least 1 period of weight loss during the study; 15 ($33\%$) had weight gain. The median percentage change in body weight over the whole study was –$1.6\%$ (IQR –$3.4\%$ to $1.7\%$, range –$8.4\%$ to $17.6\%$; Figure 3B). There was no relationship between change in body weight and SMMSE ($$P \leq .54$$).
**Figure 3:** *Body weight over time in PwD. (A) Participants’ body weight over time. Each line represents a different PwD, and each point on the line represents the sliding window average value of 5 body weights. The blue segments indicate values that were >5% less than values 6 months previously, suggesting clinically significant weight loss [38]; the red segments indicate >5% weight gain. (B) Participants’ final body weight (average of 5 measurements) is expressed as a percentage of their baseline weight; bins for >5% weight change are colored blue (loss) and red (gain). PwD: people with dementia.*
## Physiological Measurements Generating Alerts
To inform the development of remote monitoring services, we calculated the prevalence of abnormalities generating “alerts” according to several criteria.
The prevalence of abnormalities recorded by people with dementia according to NEWS subscore criteria is shown in Figure 4A-D. In-home measurement domains were here treated independently, that is, not combined into a single NEWS score, because people with dementia did not necessarily record measures contemporaneously and because the full NEWS data, that is, respiratory rate and conscious level, were not captured. Instead, alerts could be generated for measurements in a single domain if they crossed either the 1+ or 2+ thresholds.
We retrospectively calculated the burden of alerts that would have been generated using NEWS subscore criteria, that is, the proportion and rate of measurements that exceeded different subscore thresholds (Figure 4). A total of $9.46\%$ of all measurements would generate an alert for meeting the criteria of a NEWS subscore of 1 or more (1+), and $3.03\%$ of measurements using a NEWS subscore of 2+. By comparison, the proportion using our Minder study criteria (Table S2 in Multimedia Appendix 4) was $7.88\%$. We summarized the alerts per participant per day to indicate the potential overall alert burden (Figure 4). The median frequency of alerts per day per participant was 0.233 (IQR 0.14-0.37; range 0-1.33) using NEWS 1+ and 0.066 (IQR 0.02-0.15; range 0-0.67) using NEWS 2+. Using our Minder criteria, the median frequency was 0.140 (IQR 0.07-0.25; range 0-1.17). There was no relationship between alerts per day, using either NEWS (1+ or 2+) or *Minder criteria* and SMMSE ($$P \leq .56$$, $$P \leq .54$$, $$P \leq .79$$, respectively) or dementia subtype ($$P \leq .92$$, $$P \leq .26$$, and $$P \leq .95$$).
**Figure 4:** *Frequency of physiological measurement alerts in people with dementia, by NEWS and Minder criteria. For each domain, each line with superimposed circles shows the minimum, maximum, and individual observations for each PwD, after first removing any measure recorded within 60 seconds of another in the same domain. The shaded areas correspond to a NEWS subscore of 1 (yellow), 2 (orange), or 3 (red). The histograms show the distribution of measurements across the group, in relation to the ranges for NEWS subscores, annotated with the percentage of measurements in each range. (A) Heart rate (beats per minute); (B) SBP (mm Hg); (C) temperature (°C); (D) oxygen saturation (%). Data were first filtered by removing any measure recorded within 60 seconds of another in the same domain. Alert for meeting the criteria of a NEWS subscore of 1 or more = NEWS 1+; alert for meeting the criteria of a NEWS subscore of 2 or more = NEWS 2+. (E) Data were labeled by a NEWS subscore of 0 (turquoise), 1+ (yellow), or 2+ (orange). Each horizontal row of circles (left) shows measurements for each participant, colored accordingly. Each subsequent column (left-right) shows the daily number of measurements (turquoise), and then the frequency of alerts using the criteria of NEWS 1+, NEWS 2+, and the Minder platform. (F) Histograms of the alerts per day per participant across the duration of the study, for criteria of NEWS 1+ (top), NEWS 2+ (middle), and the Minder platform (bottom). NEWS: National Early Warning Score 2; PwD: people with dementia; SBP: systolic blood pressure.*
## Study Scale
We have deployed IoT medical devices in the homes of a cohort of people with dementia, providing a rich data set of naturalistic physiological measurements. We believe that this is the first time the physiology of people with dementia has been recorded in this setting at such a scale and sustained period (approximately 150,000 measurements, approximately 1,000,000 participant-hours). We found a system of this nature to be realizable and effective in detecting acute and chronic physiological abnormalities.
## Principal Findings
We found that people with dementia recorded a full set of observations on $13.9\%$ of days; however, at least 1 measurement was taken on more than half the study days, on average. Although the data were often incomplete, with measurement in some domains more likely to be recorded than others, there was no decline in compliance over time. Overall, concerns that people with dementia are unlikely to remember or be unwilling to have measurements taken in the community appear misplaced. However, these findings show that serial naturalistic data in people with dementia obtained remotely are likely to be patchy compared with what is possible in a nurse-led inpatient setting. Although there was no correlation between compliance and SMMSE in our data, it is possible that in more advanced dementia, there may be a reduction in compliance. However, although people with dementia with a lower SMMSE may be less likely to remember to take measurements, they are more likely to have a carer to support measurements.
Our system has provided an opportunity to describe the hitherto poorly understood behavior of physiological measures over extended periods in an older adult, cognitively impaired population, and to consider how physiology relates to comorbidity. On the basis of the data collected in this study, we detected a high prevalence of hypertension, an important factor in dementia development [8,9]. We also detected physiological differences between dementia subtypes, with lower and more variable blood pressure seen in people with dementia with ASyn, in keeping with the known associated autonomic dysfunction [39]. Our case studies provide further evidence that, at the individual level, remote monitoring can detect symptomatic bradycardia, acute infections, and medication side effects.
The relationship between cognitive decline and physical health is complex. For example, regarding blood pressure, both hypo- and hypertension have been implicated in the progression of cognitive decline [40]. Furthermore, because they have been largely excluded from previous randomized trials [41], the value of treating hypertension in older adult, cognitively impaired patients is not established. Body weight, as a marker of nutritional status, also has an important but complex interaction with dementia. We did not find an association with SMMSE, but it is likely that a longer time course would be required to detect one.
IoT-based platforms like Minder represent a new paradigm for clinical measurement, that is, large-scale, long-term, sporadic, patient-initiated, and remotely recorded. Definitions and care pathways for significant abnormalities in this context are not well established. These are important differences versus the established settings of primary care (infrequent, supervised, in-person measurements), secondary care (high-frequency in-person multimodal monitoring of acutely unwell patients), and ambulatory monitoring (devices worn continuously for several days). When we applied the NEWS criteria, validated for hospital use, we found that the median rate of alerts generated was 0.066 or 0.233 per day per participant, depending on the threshold used, which spanned the rate (0.114) from our own criteria. These findings provide an indication of the potential workload that would be placed on remote monitoring service (approximately 100 alerts per day per 1000 patients). There is, however, potential to improve the clinical use of such alerts, for example, by using personalization, whereby thresholds are set according to patients’ own historical data and constantly updated in response to their measurements.
## Future Directions
Our study highlights the benefits and risks of remote monitoring systems. With such systems, there is the potential to detect developing acute illness, facilitating early intervention, improving outcomes, and avoiding hospital admission [42,43]. Remote physiological monitoring of this kind could identify trends over weeks to months, relating to, for example, comorbidities like hypertension, malnutrition, and drug side effects. Both timescales are pertinent in people with dementia who, less able to recognize and communicate when they become acutely unwell, are more likely to develop comorbidities.
We are currently scaling up the size of the cohort in the study to 200 people with dementia in the form described here. There are further plans to provide a cut-down version of the system to a cohort of 1000. A key part of the ongoing work is ascertaining, which features are most informative, in order to design a system that is scalable.
It remains to be established the measurable benefit a system like this can have on long-term health outcomes and quality of life. A previous randomized trial did not point toward benefit, but this may well be due to a piecemeal approach [44]. In fact, successful implementation will depend on systematic work understanding how best to use technology in the home.
## Limitations
We have identified the following limitations to our IoT system and to the analysis presented. First, we were limited in our ability to reliably characterize and record every change in patients’ medical status and medication over the course of the study. This has implications for interpretation of the physiological findings. The association between changes in physiological observations and medication may be addressed using data linkage with the patient electronic clinical record—something we are exploring as the study develops. Second, we did not have a control group of age-matched healthy participants to provide comparisons for measurement compliance and physiological values. The normative values are for vital signs are well established, but not necessarily in the context of elderly people, in their own homes. Third, our analysis of compliance was limited as we could not discern which recordings were initiated by people with dementia versus in response to contact by the monitoring team. We therefore have likely overestimated the frequency with which people with dementia recorded measurements. However, the value in allowing people with dementia to record their vital signs remotely withstands, even if they have had to be prompted. Fourth, our alert rates are instead likely to be underestimates, because we excluded extreme outliers and duplicate values and did not evaluate abnormalities relating to DBP or body weight.
## Conclusions
There is growing interest in establishing remote monitoring within care services, amplified by the COVID-19 pandemic and calls for “hospital and home” initiatives [29,30]. We believe that remote monitoring technology can be transformative for the health and social care of people with dementia. Future research must demonstrate the clinical use of remote monitoring and address how such technologies are best integrated with existing care. Together, our findings inform the development of technologies, pathways, and policies for remote monitoring of people with dementia.
## Data Availability
The data collected during the current study are available from the corresponding author on reasonable request. The institute intends to store data in a public repository on dementiasplatform.uk in due course.
## References
1. Prince M, Knapp M, Guerchet M, McCrone P, Prina M, Comas-Herrera A. **Dementia UK: Update, second edition**. *Alzheimer’s Society* (2014)
2. Bunn F, Burn AM, Goodman C, Robinson L, Rait G, Norton S, Bennett H, Poole M, Schoeman J, Brayne C. **Comorbidity and dementia: a mixed-method study on improving health care for people with dementia (CoDem)**. *Heal Serv Deliv Res* (2016) **4** 1-156
3. Bunn F, Burn AM, Goodman C, Rait G, Norton S, Robinson L, Schoeman J, Brayne C. **Comorbidity and dementia: a scoping review of the literature**. *BMC Med* (2014) **12** 192. DOI: 10.1186/s12916-014-0192-4
4. Roqué M, Salvà A, Vellas B. **Malnutrition in community-dwelling adults with dementia (NutriAlz Trial)**. *J Nutr Health Aging* (2013) **17** 295-299. DOI: 10.1007/s12603-012-0401-9
5. Albanese E, Taylor C, Siervo M, Stewart R, Prince MJ, Acosta D. **Dementia severity and weight loss: a comparison across eight cohorts. The 10/66 study**. *Alzheimers Dement* (2013) **9** 649-656. DOI: 10.1016/j.jalz.2012.11.014
6. Eriksson U, Bennet A, Gatz M, Dickman P, Pedersen N. **Nonstroke cardiovascular disease and risk of Alzheimer disease and dementia**. *Alzheimer Dis Assoc Disord* (2010) **24** 9. DOI: 10.1097/wad.0b013e3181d1b99b
7. Newman A, Fitzpatrick A, Lopez O, Jackson S, Lyketsos C, Jagust W, Ives D, Dekosky ST, Kuller LH. **Dementia and Alzheimer's disease incidence in relationship to cardiovascular disease in the cardiovascular health study cohort**. *J Am Geriatr Soc* (2005) **53** 1101-1107. DOI: 10.1111/j.1532-5415.2005.53360.x
8. Nash D, Fillit H. **Cardiovascular disease risk factors and cognitive impairment**. *Am J Cardiol* (2006) **97** 1262-1265. DOI: 10.1016/j.amjcard.2005.12.031
9. Ding J, Davis-Plourde KL, Sedaghat S, Tully PJ, Wang W, Phillips C, Pase MP, Himali JJ, Gwen Windham B, Griswold M, Gottesman R, Mosley TH, White L, Guðnason V, Debette S, Beiser AS, Seshadri S, Ikram MA, Meirelles O, Tzourio C, Launer LJ. **Antihypertensive medications and risk for incident dementia and Alzheimer's disease: a meta-analysis of individual participant data from prospective cohort studies**. *Lancet Neurol* (2020) **19** 61-70. DOI: 10.1016/s1474-4422(19)30393-x
10. Miller SL, Wolfe RR. **The danger of weight loss in the elderly**. *J Nutr Health Aging* (2008) **12** 487-491. DOI: 10.1007/bf02982710
11. Fox C, Smith T, Maidment I, Hebding J, Madzima T, Cheater F, Cross J, Poland F, White J, Young J. **The importance of detecting and managing comorbidities in people with dementia?**. *Age Ageing* (2014) **43** 741-743. DOI: 10.1093/ageing/afu101
12. Solomon A, Dobranici L, Kåreholt I, Tudose C, Lăzărescu M. **Comorbidity and the rate of cognitive decline in patients with Alzheimer dementia**. *Int J Geriatr Psychiatry* (2011) **26** 1244-1251. DOI: 10.1002/gps.2670
13. Wolf D, Rhein C, Geschke K, Fellgiebel A. **Preventable hospitalizations among older patients with cognitive impairments and dementia**. *Int Psychogeriatr* (2018) **31** 383-391. DOI: 10.1017/s1041610218000960
14. Bynum JPW, Rabins PV, Weller W, Niefeld M, Anderson GF, Wu AW. **The relationship between a dementia diagnosis, chronic illness, medicare expenditures, and hospital use**. *J Am Geriatr Soc* (2004) **52** 187-194. DOI: 10.1111/j.1532-5415.2004.52054.x
15. Scrutton J, Brancati CU. **Dementia and comorbidities ensuring parity of care**. *COPCV* (2016)
16. Sampson EL, Blanchard MR, Jones L, Tookman A, King M. **Dementia in the acute hospital: prospective cohort study of prevalence and mortality**. *Br J Psychiatry* (2009) **195** 61-66. DOI: 10.1192/bjp.bp.108.055335
17. Chidgey A. **Caring for people with dementia on hospital wards**. *Br J Neurosci Nurs* (2010) **6** 210-210. DOI: 10.12968/bjnn.2010.6.5.48677
18. Isik AT, Kocyigit SE, Smith L, Aydin AE, Soysal P. **A comparison of the prevalence of orthostatic hypotension between older patients with Alzheimer's disease, lewy body dementia, and without dementia**. *Exp Gerontol* (2019) **124** 110628. DOI: 10.1016/j.exger.2019.06.001
19. Chester JG, Rudolph JL. **Vital signs in older patients: age-related changes**. *J Am Med Dir Assoc* (2011) **12** 337-343. DOI: 10.1016/j.jamda.2010.04.009
20. Rowland JP, Rigby J, Harper AC, Rowland R. **Cardiovascular monitoring with acetylcholinesterase inhibitors: a clinical protocol**. *Adv Psychiatr Treat* (2007) **13** 178-184. DOI: 10.1192/apt.bp.106.002725
21. Franceschi M, Ferini-Strambi L, Minicucci F, Sferrazza-Papa A, Smirne S. **Signs of cardiac autonomic dysfunction during sleep in patients with Alzheimer's disease**. *Gerontology* (1986) **32** 327-334. DOI: 10.1159/000212811
22. Kulmala J, Nykänen I, Mänty M, Hartikainen S. **Association between frailty and dementia: a population-based study**. *Gerontology* (2013) **60** 16-21. DOI: 10.1159/000353859
23. Romero-Ortuno R, Wallis S, Biram R, Keevil V. **Clinical frailty adds to acute illness severity in predicting mortality in hospitalized older adults: an observational study**. *Eur J Intern Med* (2016) **35** 24-34. DOI: 10.1016/j.ejim.2016.08.033
24. Jarrett PG, Rockwood K, Carver P, Stolee D, Cosway S. **Illness presentation in elderly patients**. *Arch Intern Med* (1995) **155** 1060. DOI: 10.1001/archinte.1995.00430100086010
25. Alexander GL, Rantz M, Skubic M, Koopman R, Phillips L, Guevara R, Miller S. **Evolution of an early illness warning system to monitor frail elders in independent living**. *J Healthc Eng* (2011) **2** 337-363. DOI: 10.1260/2040-2295.2.3.337
26. Rostill H, Nilforooshan R, Morgan A, Barnaghi P, Ream E, Chrysanthaki T. **Technology integrated health management for dementia**. *Br J Community Nurs* (2018) **23** 502-508. DOI: 10.12968/bjcn.2018.23.10.502
27. van Corven CTM, Bielderman A, Wijnen M, Leontjevas R, Lucassen P, Graff M, Gerritsen DL. **Defining empowerment for older people living with dementia from multiple perspectives: a qualitative study**. *Int J Nurs Stud* (2021) **114** 103823. DOI: 10.1016/j.ijnurstu.2020.103823
28. Kenigsberg PA, Aquino JP, Bérard A, Brémond F, Charras K, Dening T, Droës RM, Gzil F, Hicks B, Innes A, Nguyen S, Nygård L, Pino M, Sacco G, Salmon E, van der Roest H, Villet H, Villez M, Robert P, Manera V. **Assistive technologies to address capabilities of people with dementia: from research to practice**. *Dementia (London)* (2019) **18** 1568-1595. DOI: 10.1177/1471301217714093
29. Martin M. **The role of remote monitoring in the future of the NHSX**. *NHSX* (2020)
30. **Data saves lives: reshaping health and social care with data**. *Department of Health and Social Care* (2022)
31. Pronovost PJ, Cole MD, Hughes RM. **Remote patient monitoring during COVID-19: an unexpected patient safety benefit**. *JAMA* (2022) **327** 1125-1126. DOI: 10.1001/jama.2022.2040
32. **National Early Warning Score (NEWS) 2: standardising the assessment of acute-illness severity in the NHS. Vol. 17, Updated report of a working party**. *Royal College of Physicians* (2017)
33. Enshaeifar S, Barnaghi P, Skillman S, Sharp D, Nilforooshan R, Rostill H. **A digital platform for remote healthcare monitoring**. (2020). DOI: 10.1145/3366424.3383541
34. von Elm E, Altman D, Egger M, Pocock S, Gøtzsche PC, Vandenbroucke J. **The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement: guidelines for reporting observational studies**. *Epidemiology* (2007) **18** 800-804. DOI: 10.1097/EDE.0b013e3181577654
35. 35MathWorks20202023-02-06
https://www.mathworks.com/company/jobs/resources/locations/us-natick.html. *MathWorks* (2020)
36. O'Brien E, Coats A, Owens P, Petrie J, Padfield PL, Littler WA, de Swiet M, Mee F. **Use and interpretation of ambulatory blood pressure monitoring: recommendations of the British hypertension society**. *BMJ* (2000) **320** 1128-1134. DOI: 10.1136/bmj.320.7242.1128
37. Bhalke RD, Giri MA, Anil RY, Balasaheb M, Nanasaheb PA, Pande VV. **Hypotension?: a comprehensive review**. *J Pharmacogn Phytochem* (2021) **10** 1945-1947
38. Abildso CG, Zizzi S, Fitzpatrick SJ. **Predictors of clinically significant weight loss and participant retention in an insurance-sponsored community-based weight management program**. *Health Promot Pract* (2013) **14** 580-588. DOI: 10.1177/1524839912462393
39. Mendoza-Velásquez JJ, Flores-Vázquez JF, Barrón-Velázquez E, Sosa-Ortiz AL, Illigens BMW, Siepmann T. **Autonomic dysfunction in α-synucleinopathies**. *Front Neurol* (2019) **10** 363. DOI: 10.3389/fneur.2019.00363
40. Harrison JK, Van Der Wardt V, Conroy SP, Stott DJ, Dening T, Gordon AL, Logan P, Welsh TJ, Taggar J, Harwood R, Gladman JRF. **New horizons: the management of hypertension in people with dementia**. *Age Ageing* (2016) **45** 740-746. DOI: 10.1093/ageing/afw155
41. Peters R, Beckett N, Forette F, Tuomilehto J, Clarke R, Ritchie C, Waldman A, Walton I, Poulter R, Ma S, Comsa M, Burch L, Fletcher A, Bulpitt C. **Incident dementia and blood pressure lowering in the hypertension in the very elderly trial cognitive function assessment (HYVET-COG): a double-blind, placebo controlled trial**. *Lancet Neurol* (2008) **7** 683-689. DOI: 10.1016/s1474-4422(08)70143-1
42. Dinesen B, Haesum LK, Soerensen N, Nielsen C, Grann O, Hejlesen O, Toft E, Ehlers L. **Using preventive home monitoring to reduce hospital admission rates and reduce costs: a case study of telehealth among chronic obstructive pulmonary disease patients**. *J Telemed Telecare* (2012) **18** 221-225. DOI: 10.1258/jtt.2012.110704
43. Dirikgil E, Roos R, Groeneveld GH, Heringhaus C, Silven AV, Petrus AHJ, Villalobos-Quesada M, Tsonaka R, van der Boog PJ, Rabelink TJ, Bos WJW, Chavannes NH, Atsma DE, Teng YO. **Home monitoring reduced short stay admissions in suspected COVID-19 patients: COVID-box project**. *Eur Respir J* (2021) **58** 2100636. DOI: 10.1183/13993003.00636-2021
44. Howard R, Gathercole R, Bradley R, Harper E, Davis L, Pank L, Lam N, Talbot E, Hooper E, Winson R, Scutt B, Ordonez Montano V, Nunn S, Lavelle G, Bateman A, Bentham P, Burns A, Dunk B, Forsyth K, Fox C, Poland F, Leroi I, Newman S, O'Brien J, Henderson C, Knapp M, Woolham J, Gray R. **The effectiveness and cost-effectiveness of assistive technology and telecare for independent living in dementia: a randomised controlled trial**. *Age Ageing* (2021) **50** 882-890. DOI: 10.1093/ageing/afaa284
|
---
title: Association of Treatment With Nirmatrelvir and the Risk of Post–COVID-19 Condition
authors:
- Yan Xie
- Taeyoung Choi
- Ziyad Al-Aly
journal: JAMA Internal Medicine
year: 2023
pmcid: PMC10037200
doi: 10.1001/jamainternmed.2023.0743
license: CC BY 4.0
---
# Association of Treatment With Nirmatrelvir and the Risk of Post–COVID-19 Condition
## Key Points
### Question
Is treatment with nirmatrelvir in the acute phase of SARS-CoV-2 infection associated with a lower risk of post–COVID-19 condition (PCC)?
### Findings
In this cohort study of 281 793 people with SARS-CoV-2 infection who had at least 1 risk factor for progression to severe COVID-19 illness, compared with 246 076 who had no treatment, nirmatrelvir use in the acute phase ($$n = 35$$ 717) was associated with reduced risk of PCC, including reduced risk of 10 of 13 post–acute sequelae in various organ systems, as well as reduced risk of post–acute death and post–acute hospitalization. Nirmatrelvir was associated with reduced risk of PCC in people who were unvaccinated, vaccinated, and boosted, and in people with primary SARS-CoV-2 infection and reinfection.
### Meaning
In people with SARS-CoV-2 infection and at least 1 risk factor for progression to severe COVID-19 illness, treatment with nirmatrelvir during the acute phase of COVID-19 was associated with reduced risk of PCC.
## Abstract
This cohort study examines the association of treatment with nirmatrelvir in the acute phase of COVID-19 and risk of post–COVID-19 condition.
### Importance
Post–COVID-19 condition (PCC), also known as long COVID, affects many individuals. Prevention of PCC is an urgent public health priority.
### Objective
To examine whether treatment with nirmatrelvir in the acute phase of COVID-19 is associated with reduced risk of PCC.
### Design, Setting, and Participants
This cohort study used the health care databases of the US Department of Veterans Affairs (VA) to identify patients who had a SARS-CoV-2 positive test result between January 3, 2022, and December 31, 2022, who were not hospitalized on the day of the positive test result, who had at least 1 risk factor for progression to severe COVID-19 illness, and who had survived the first 30 days after SARS-CoV-2 diagnosis. Those who were treated with oral nirmatrelvir within 5 days after the positive test ($$n = 35$$ 717) and those who received no COVID-19 antiviral or antibody treatment during the acute phase of SARS-CoV-2 infection (control group, $$n = 246$$ 076) were identified.
### Exposures
Treatment with nirmatrelvir or receipt of no COVID-19 antiviral or antibody treatment based on prescription records.
### Main Outcomes and Measures
Inverse probability weighted survival models were used to estimate the association of nirmatrelvir (vs control) with post–acute death, post–acute hospitalization, and a prespecified panel of 13 post–acute COVID-19 sequelae (components of PCC) and reported in relative scale as relative risk (RR) or hazard ratio (HR) and in absolute scale as absolute risk reduction in percentage at 180 days (ARR).
### Results
A total of 281 793 patients (mean [SD] age, 61.99 [14.96]; 242 383 [$86.01\%$] male) who had a positive SARS-CoV-2 test result and had at least 1 risk factor for progression to severe COVID-19 illness were studied. Among them, 246 076 received no COVID-19 antiviral or antibody treatment during the acute phase of SARS-CoV-2 infection, and 35 717 received oral nirmatrelvir within 5 days after the positive SARS-CoV-2 test result. Compared with the control group, nirmatrelvir was associated with reduced risk of PCC (RR, 0.74; $95\%$ CI, 0.72-0.77; ARR, $4.51\%$; $95\%$ CI, 4.01-4.99), including reduced risk of 10 of 13 post–acute sequelae (components of PCC) in the cardiovascular system (dysrhythmia and ischemic heart disease), coagulation and hematologic disorders (pulmonary embolism and deep vein thrombosis), fatigue and malaise, acute kidney disease, muscle pain, neurologic system (neurocognitive impairment and dysautonomia), and shortness of breath. Nirmatrelvir was also associated with reduced risk of post–acute death (HR, 0.53; $95\%$ CI, 0.46-0.61); ARR, $0.65\%$; $95\%$ CI, 0.54-0.77), and post–acute hospitalization (HR, 0.76; $95\%$ CI, 0.73-0.80; ARR, $1.72\%$; $95\%$ CI, 1.42-2.01). Nirmatrelvir was associated with reduced risk of PCC in people who were unvaccinated, vaccinated, and boosted, and in people with primary SARS-CoV-2 infection and reinfection.
### Conclusions and Relevance
This cohort study found that in people with SARS-CoV-2 infection who had at least 1 risk factor for progression to severe disease, treatment with nirmatrelvir within 5 days of a positive SARS-CoV-2 test result was associated with reduced risk of PCC across the risk spectrum in this cohort and regardless of vaccination status and history of prior infection; the totality of findings suggests that treatment with nirmatrelvir during the acute phase of COVID-19 may reduce the risk of post–acute adverse health outcomes.
## Introduction
Post–COVID-19 condition (PCC), also known as long COVID, is the disease encompassing the post–acute sequelae of SARS-CoV-2 infection, and it affects millions of people around the world.1,2,3 Despite PCC affecting a substantial portion of the patient population, there is no approved medication for the prevention or treatment of PCC. Several hypotheses have been proposed to explain the underlying mechanisms of PCC including persistence of the virus (or its fragments) or intensity of the inflammation during the acute phase of the disease.4 The antiviral nirmatrelvir (in combination with ritonavir, marketed under the name Paxlovid) that has been shown to reduce the risk of progression to severe acute COVID-19 has been suggested as a candidate drug that may reduce the risk of developing PCC.5,6 In December 2021, oral nirmatrelvir was approved in the US for the treatment of acute SARS-CoV-2 infection (typically within 5 days of symptom onset) in nonhospitalized people at risk of progression to severe COVID-19 illness. Millions of people in the US have since received treatment with nirmatrelvir. Urgent calls have been made to evaluate whether treatment with nirmatrelvir in the acute phase of COVID-19 reduces the risk of PCC—but data have thus far been lacking.5 *Addressing this* question will guide treatment approaches of SARS-CoV-2 infections and will inform the effort to develop and optimize prevention and treatment strategies for PCC.
In this cohort study, we used the health care databases of the US Department of Veterans Affairs (VA) to identify patients who had a SARS-CoV-2 positive test result between January 3, 2022, and December 31, 2022, who were not hospitalized on the day of the positive test, who had at least 1 risk factor for progression to severe COVID-19 illness, and who had survived the first 30 days after SARS-CoV-2 diagnosis. We identified those who were prescribed oral nirmatrelvir within 5 days after the positive test and did not receive other outpatient COVID-19 antiviral or antibody treatment within 30 days after the positive test (nirmatrelvir group, $$n = 35$$ 717) and those who received no outpatient COVID-19 antiviral or antibody treatment within 30 days after the positive test (control group, $$n = 246$$ 076). We then used the inverse probability weighting approach to balance the characteristics of the groups and evaluate whether treatment with oral nirmatrelvir vs the control was associated with reduced risk of post–acute outcomes, including PCC (from a set of 13 prespecified post–acute sequelae of SARS-CoV-2 infection), post–acute death, post–acute hospitalization, and each individual post–acute sequela.
## Setting
The VA operates the largest integrated health care system in the US; the system comprises 1283 health care facilities (including 171 VA medical centers and 1112 outpatient sites) located across the US. The VA provides comprehensive health care to discharged veterans of the US armed forces including preventative and health maintenance, outpatient care, inpatient hospital care, prescriptions, mental health care, home health care, primary care, specialty care, geriatric and extended care, medical equipment, and prosthetics.
## Data Sources
The cohort study was conducted using the VA health care databases. The VA health care data are updated daily and include individual-level demographic information and data on health care encounters, comorbidities, procedures, and surgeries. Data domains included outpatient encounters, inpatient encounters, inpatient and outpatient medications, laboratory results and non-VA care program integrity tools were used. The VA COVID-19 Shared Data Resource was used to collect information on patients with COVID-19 and vaccination status. The Area Deprivation Index (ADI)—which is a composite measure of income, education, employment, and housing—was used as a summary measure of contextual disadvantage at participants’ residential locations.7
## Cohort
A flowchart and a timeline of cohort construction are provided in eFigures 1 and 2 in Supplement 1, respectively. There were 332 256 participants who had a positive SARS-CoV-2 test result between January 3, 2022, and December 31, 2022, when their first date of positive test was set to be T0. A total of 37 466 participants prescribed nirmatrelvir within 5 days of T0 were selected into the nirmatrelvir group. We then selected 36 641 participants with at least 1 risk factor of progression to severe acute COVID-19 illness, which included being older than 60 years, a body mass index (BMI) of greater than 25 (calculated as weight in kilograms divided by height in meters squared), current smoker, cancer, cardiovascular disease, kidney disease, chronic lung disease, diabetes, immune dysfunction, and hypertension. We also excluded participants with liver disease, end-stage kidney disease, estimated glomerular filtration rate (eGFR; a measure of kidney function) of less than 30 mL/min/1.73m2 (to convert to mL/s/m2, multiply by 0.0167), and/or prescription fill of a medication that precluded them from receiving nirmatrelvir ($$n = 35$$ 815). Participants who did not use other outpatient COVID-19 antiviral or antibody treatments within 30 days after T0 were selected ($$n = 35$$ 776). To examine the post–acute events, only participants alive 30 days after T0 were included in the nirmatrelvir group (final $$n = 35$$ 717).
A control group of participants was constructed from the 332 256 participants who had a positive SARS-CoV-2 test result between January 3, 2022, and December 31, 2022, when their first date positive test was set to be T0. From those who were not prescribed nirmatrelvir within 5 days of T0 ($$n = 294$$ 790), we selected 278 965 participants with at least 1 risk factor of progression to severe acute COVID-19 illness, and excluded participants with liver disease, end stage kidney disease, eGFR of less than 30 mL/min/1.73m2, and/or prescription fill of a medication that precluded them from receiving nirmatrelvir ($$n = 260$$ 093). Participants who did not use any outpatient COVID-19 antiviral or antibody treatments within 30 days after T0 ($$n = 249$$ 443) were further selected. To examine the post–acute events, only participants alive 30 days after T0 were included in the control group (final $$n = 246$$ 076).
The final cohort consisted of 281 793 participants, of which 35 717 were in the nirmatrelvir group and 246 076 in the control group. The cohort was followed-up until February 2, 2023.
## Exposure and Outcomes
We defined exposure as a filled nirmatrelvir prescription within 5 days of SARS-CoV-2 positive test result. We examined the risk of post–acute death and hospitalization and a composite outcome of death or hospitalization. We also studied individual post–acute sequelae—which were selected based on prior evidence1,2,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23—including ischemic heart disease, dysrhythmia, deep vein thrombosis, pulmonary embolism, fatigue and malaise, liver disease, acute kidney injury, muscle pain, diabetes, neurocognitive impairment, dysautonomia, and shortness of breath and cough. Individual sequelae were defined based on inpatient and outpatient International Statistical Classification of Diseases and Related Health Problems, Tenth Revision (ICD-10) diagnosis codes and laboratory results; death was defined based on vital status data; and hospitalization was defined based on inpatient encounter data. The PCC score was built by assigning weights to each individual sequela following the methods of the Global Burden of Disease Long COVID Collaborators (weights provided at https://github.com/yxie618/Nirmatrelvir_PASC)24; we then constructed the PCC score for each participant as the sum of weights of all the incident sequelae in that participant during follow up.25 Incident outcomes were assessed within those without history of the related outcome within 3 years before T0. All outcomes were ascertained 30 days after T0.
## Covariates
We identified baseline characteristics that may be associated with the use of treatment and the occurrence or assessment of outcomes based on literature review and prior knowledge.1,19,25,26 All covariates were assessed within 3 years before study enrollment unless otherwise specified. Predefined covariates included age, race (White, Black, and other), ethnicity (Hispanic and non-Hispanic), sex, ADI, BMI, smoking status (current, former, and never), history of SARS-CoV-2 infection, use of steroids, use of long-term care, eGFR, systolic and diastolic blood pressure, cancer, chronic lung disease, dementia, diabetes, hyperlipidemia, and immune dysfunction. We adjusted for medications that would have drug-drug interaction with nirmatrelvir-ritonavir based on 3 levels (require temporary hold of concomitant medication; require adjustment of concomitant medication dosing; and require monitoring for adverse effects). We also considered health care utilization parameters including number of outpatient and inpatient encounters, number of laboratory encounters and number of outpatient medications received within 1 year before study enrollment and influenza vaccination status. Continuous variables were transformed into restricted cubic spline functions to account for potential nonlinear associations.
## Statistical Analysis
Baseline characteristics were reported as mean and standard deviation or frequency and percentage. Covariate balance between groups was evaluated by the absolute standardized differences where an absolute standardized difference of less than 0.1 was considered evidence of good balance.
To examine the risk of incident outcomes, for each outcome besides death or hospitalization, we conducted analysis on a subcohort of participants without the history of the outcome within 3 years before T0. An inverse probability weighting method was used to balance the differences in baseline characteristics between the nirmatrelvir and control groups. Logistic regression was built to estimate the probability of receiving nirmatrelvir given covariates. The probability was then used as the propensity score. We then constructed the inverse probability weights as a value of 1 for those in the nirmatrelvir group and as propensity score divided by (1−propensity score) for those in the control group to estimate the association within population with same baseline characteristics as the nirmatrelvir group. Weights larger than 10 would be truncated at 10 to reduce the influence of extreme weights (in the present study, no weights were larger than 10, and none were truncated). The inverse probability weights were then applied to a Cox survival model in order to estimate the association of nirmatrelvir with individual outcomes. Hazard ratio (HR) and survival probability for both groups at 180 days were estimated. Absolute risk reduction (ARR) at 180 days was computed as the difference of survival probability in the nirmatrelvir group compared with the control group. To estimate the association of nirmatrelvir with PCC score, the inverse probability weighted zero inflated Poisson regression was used, and the relative risk (RR) and ARR in percentage at 180 days were estimated.
The association of nirmatrelvir with the risk of PCC was further examined within prespecified subgroups by age (≤60 years, >60 years to ≤70 years, and >70 years), race (White and Black), sex, smoking status (current smoker, former smoker and never smoker), cancer, cardiovascular disease, chronic kidney disease, chronic lung disease, diabetes, immune dysfunction, and hypertension. We also examined the association within populations with different vaccination status (unvaccinated, 1-2 doses of vaccine, and boosted) and infection status (with primary SARS-CoV-2 infection and reinfection). To examine the association within populations with different baseline risks, we also defined subgroups based on the number of baseline risk factors (1-2, 3-4, or ≥5) of progression to severe acute COVID-19 illness, where risk factors included age of older than 60 years, BMI greater than 25, current smoker, cancer, cardiovascular disease, kidney disease, chronic lung disease, diabetes, immune dysfunction, and hypertension.
We challenged the robustness of findings in multiple sensitivity analyses, including [1] application of the overlap weighting method to balance baseline characteristics in the treatment and control groups (whereas, in the primary approach, we used the inverse probability weighing approach to balance the groups); [2] application of the doubly robust approach to additionally adjust for covariates in the inverse probability weighted survival models (whereas, in the primary approach, we used inverse probability weighted survival models); [3] application of the high-dimensional variable selection algorithm to additionally identify 100 covariates from data domains including diagnoses, medications, and laboratory test results that were used along with predefined variables to construct the weights (whereas, in the primary approach, we used only predefined covariates); [4] application of inverse probability of censoring weight to account for those who died during the acute phase of infection (within 30 days of infection) (whereas, in the primary approach, we removed those who died during acute phase from the analyses); [5] defined outcomes based on events that occurred 90 days after infection (whereas, in the primary approach, outcomes were defined based on events that occurred 30 days after infection); [6] defined incident outcome as occurrence of the outcome in those without history of the related outcome within 5 years before infection (whereas, in the primary approach, the washout period was 3 years before infection); [7] additionally adjusted for hospitalization, intensive care unit admission and ventilator use during the acute phase of infection (whereas, in the primary approach, we did not adjust for information after exposure); [8] defined outcomes based on PCC ICD-10 code U09.9 (whereas, in the primary definition, this was based on a set of 13 predefined post–acute sequelae of COVID-19); and [9] defined outcomes based on the first occurrence of any individual sequela (whereas, in the primary definition, this accounted for both the number of sequelae and the influence of each sequela on health).
Analyses were performed with SAS Enterprise Guide, version 8.2 (SAS Institute). Data visualizations were performed in R 4.0.4 (R Project for Statistical Computing). The robust sandwich variance estimator was used to estimate variance in weighted analyses. Risk on relative scale with a $95\%$ CI that does not cross 1 and risk on absolute scale with a $95\%$ CI that does not cross 0 was considered statistically significant. The study was approved by the VA St Louis Health Care System institutional review board, which granted a waiver of informed consent because of the retrospective nature of the study. The study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.
## Results
The cohort included 281 793 participants; 35 717 were in the nirmatrelvir group, and 246 076 were in the control group that received no COVID-19 antiviral or antibody treatment within the first 30 days after infection. The demographic and health characteristics before weighting are provided in eTable 1 in Supplement 1; characteristics after weighting are provided in Table 1. Absolute standardized mean differences between the the nirmatrelvir group and the control group after application of inverse probability weighting were all below 0.1—suggesting good balance (Table 1). The unadjusted event rates are presented in eTable 2 in Supplement 1.
**Table 1.**
| Characteristic | No. (%) | No. (%).1 | No. (%).2 | SMD between nirmatrelvir and control groupa |
| --- | --- | --- | --- | --- |
| Characteristic | Overall cohort (n = 281 793) | Control group (n = 246 076) | Nirmatrelvir group (n = 35 717) | SMD between nirmatrelvir and control groupa |
| Age, mean (SD), y | 65.70 (13.39) | 65.76 (13.39) | 65.64 (13.38) | 0.01 |
| Race | Race | Race | Race | Race |
| Black | 57 768 (20.50) | 50 465 (20.51) | 7316 (20.48) | 0.001 |
| White | 208 696 (74.06) | 182 367 (74.11) | 26 431 (74.00) | 0.002 |
| Otherb | 15 358 (5.45) | 13 246 (5.38) | 1970 (5.52) | 0.006 |
| Ethnicity | Ethnicity | Ethnicity | Ethnicity | Ethnicity |
| Hispanic | 22 121 (7.85) | 19 169 (7.79) | 2823 (7.90) | 0.004 |
| Non-Hispanic | 259 672 (92.15) | 226 907 (92.21) | 32 894 (92.10) | 0.004 |
| Sex | Sex | Sex | Sex | Sex |
| Female | 34 294 (12.17) | 29 741 (12.09) | 4378 (12.26) | 0.005 |
| Male | 247 499 (87.83) | 216 335 (87.91) | 31 339 (87.74) | 0.005 |
| Smoking status | Smoking status | Smoking status | Smoking status | Smoking status |
| Never | 121 622 (43.16) | 105 825 (43.01) | 15 470 (43.31) | 0.006 |
| Former | 116 719 (41.42) | 102 289 (41.57) | 14 744 (41.28) | 0.006 |
| Current | 43 452 (15.42) | 37 965 (15.43) | 5503 (15.41) | 0.001 |
| Area Deprivation Index, mean (SD)c | 51.07 (19.70) | 51.07 (19.74) | 51.07 (19.67) | <0.001 |
| Long-term care | 1873 (0.64) | 1584 (0.64) | 289 (0.64) | <0.001 |
| Vaccination | Vaccination | Vaccination | Vaccination | Vaccination |
| Without prior vaccination | 47 144 (16.73) | 41 129 (16.71) | 5982 (16.75) | 0.001 |
| With 1 shot of vaccination | 10 173 (3.61) | 8861 (3.60) | 1296 (3.63) | 0.002 |
| With 2 shots of vaccination | 60 191 (21.36) | 52 372 (21.28) | 7655 (21.43) | 0.004 |
| With booster | 164 285 (58.30) | 143 711 (58.40) | 20 784 (58.19) | 0.004 |
| BMI, mean (SD) | 30.86 (6.37) | 30.86 (6.44) | 30.86 (6.31) | <0.001 |
| eGFR, mean (SD), ml/min/1.73m2d | 78.56 (18.37) | 78.53 (18.41) | 78.59 (18.34) | <0.001 |
| Blood pressure, mean (SD), mm Hg | Blood pressure, mean (SD), mm Hg | Blood pressure, mean (SD), mm Hg | Blood pressure, mean (SD), mm Hg | Blood pressure, mean (SD), mm Hg |
| Systolic | 134.14 (11.34) | 134.16 (11.34) | 134.13 (11.34) | <0.001 |
| Diastolic | 78.29 (6.97) | 78.26 (6.98) | 78.33 (6.96) | 0.01 |
| History of SARS-CoV-2 infection | 47 933 (17.01) | 42 069 (17.10) | 6046 (16.93) | 0.004 |
| Use of steroid | 9271 (3.29) | 8167 (3.32) | 1162 (3.25) | 0.004 |
| Medications that would have drug-drug interaction with nirmatrelvir-ritonavir | Medications that would have drug-drug interaction with nirmatrelvir-ritonavir | Medications that would have drug-drug interaction with nirmatrelvir-ritonavir | Medications that would have drug-drug interaction with nirmatrelvir-ritonavir | Medications that would have drug-drug interaction with nirmatrelvir-ritonavir |
| On concomitant medication that requires temporary hold | 146 138 (51.86) | 128 353 (52.16) | 18 411 (51.55) | 0.01 |
| On concomitant medication that requires dosing adjustment | 109 617 (38.90) | 96 523 (39.23) | 13 780 (38.58) | 0.01 |
| On concomitant medication that requires monitoring for adverse events | 107 307 (38.08) | 946,38 (38.46) | 13 467 (37.71) | 0.02 |
| Cancer | 50 864 (18.05) | 44 569 (18.11) | 6425 (17.99) | 0.003 |
| Chronic lung disease | 63 403 (22.50) | 55 468 (22.54) | 8019 (22.45) | 0.002 |
| Dementia | 20 120 (7.14) | 17 358 (7.05) | 2582 (7.23) | 0.007 |
| Diabetes type 2 | 100 318 (35.60) | 87 820 (35.69) | 12 684 (35.51) | 0.004 |
| Cardiovascular disease | 83 157 (29.51) | 734,09 (29.83) | 10 426 (29.19) | 0.01 |
| Hyperlipidemia | 111 889 (39.66) | 97 741 (39.72) | 14 148 (39.61) | 0.002 |
| Immune dysfunction | 15 978 (5.67) | 14036 (5.70) | 2010 (5.63) | 0.003 |
| No. of hospital admissions, mean (SD)e | 0.16 (0.61) | 0.17 (0.61) | 0.16 (0.61) | 0.01 |
| No. of outpatient visits, mean (SD)e | 3.20 (1.39) | 3.22 (1.39) | 3.19 (1.39) | 0.02 |
| No. of blood panel tests, mean (SD)e | 7.69 (8.04) | 7.76 (8.10) | 7.63 (7.98) | 0.02 |
| No. of medications, mean (SD)e | 10.31 (7.27) | 10.38 (7.31) | 10.23 (7.23) | 0.02 |
| No. of hospital admissions from Medicare | 0.03 (0.21) | 0.03 (0.21) | 0.03 (0.22) | <0.001 |
| No. of outpatient visits from Medicare | 0.14 (0.58) | 0.14 (0.58) | 0.14 (0.58) | <0.001 |
| Influenza vaccine | 195 142 (69.25) | 170 796 (69.41) | 24 676 (69.09) | 0.007 |
| Calendar wk of study enrollment, mean (SD) | 31.98 (11.75) | 32.07 (11.65) | 31.89 (11.85) | 0.02 |
In this study, measurements of risk are provided on both relative and absolute scale: [1] RR or HR of nirmatrelvir in comparison to the control group and [2] ARR in percentage at 180 days; the latter represents the event rate reduction in the nirmatrelvir group compared with the control group at 180 days.
## Risk of PCC
Compared with the control group, nirmatrelvir was associated with reduced risk of PCC (RR, 0.74; $95\%$ CI, 0.72-0.77); the event rate was $12.99\%$ ($95\%$ CI, 12.52-13.49) and $17.51\%$ ($95\%$ CI, 17.08-17.94) at 180 days in the nirmatrelvir and the control groups, respectively. This corresponded to an ARR of $4.51\%$ ($95\%$ CI, 4.01-4.99) at 180 days (Figures 1A and 2A; eTable 3 in Supplement 1).
**Figure 1.:** *Relative and Absolute Risk Reduction of Nirmatrelvir Compared With the No-Treatment Control GroupA, Post–COVID-19 condition (PCC), death, hospitalization, and composite outcome of death or hospitalization. B, Individual post–acute sequelae (components of PCC). Outcomes were ascertained 30 days after the SARS-CoV-2 positive test result until the end of follow-up. The nirmatrelvir group consisted of 35 717 patients, and the control group consisted of 246 076 patients. Adjusted hazard ratios and 95% CIs are presented. Absolute risk reduction of nirmatrelvir compared with the control group per 100 persons at 180 days and associated 95% CIs were estimated based on the difference of survival probability in the nirmatrelvir group compared with the control group. Statistically significant results are presented in light blue, and results that lacked statistical significance are presented in orange.* **Figure 2.:** *Event Rates of Post–Acute Outcomes in Nirmatrelvir and No-Treatment Control GroupA, Post–COVID-19 condition (PCC). B, Death. C, Hospitalization. D, Composite outcome of death or hospitalization. Outcomes were ascertained 30 days after the SARS-CoV-2 positive test until the end of follow-up. Event rates in percentage presented for the nirmatrelvir group (blue, n = 35 717) and the control group (orange, n = 246 076). Shaded areas are 95% CIs.*
## Risk of Post–Acute Death and Post–Acute Hospitalization
Compared with the control group, nirmatrelvir was associated with reduced risk of post–acute death (HR, 0.53; $95\%$ CI, 0.46-0.61; ARR, $0.65\%$; $95\%$ CI, 0.54-0.77), post–acute hospitalization (HR, 0.76; $95\%$ CI, 0.73-0.80; ARR, $1.72\%$; $95\%$ CI, 1.42-2.01), and the composite outcome of post–acute death or hospitalization (HR, 0.74; $95\%$ CI, 0.70-0.77; ARR, $2.15\%$; $95\%$ CI, 1.85-2.46) (Figures 1A and 2B, 2C, and 2D; eTable 3 in Supplement 1).
## Risk of Individual Post–Acute Sequelae
Compared with the control group, nirmatrelvir was associated with reduced risk of 10 of the 13 prespecified post–acute sequelae evaluated in this analysis. Nirmatrelvir was associated with reduced risk of sequelae in the cardiovascular system (dysrhythmia and ischemic heart disease), coagulation and hematologic disorders (pulmonary embolism and deep vein thrombosis), fatigue and malaise, liver disease, acute kidney disease, muscle pain, neurologic system (neurocognitive impairment and dysautonomia), and shortness of breath (Figure 1B, Table 2). There was lack of a statistically significant association between nirmatrelvir and other post–acute sequelae, including new-onset diabetes, liver disease, and cough (Figure 1B, Table 2).
**Table 2.**
| Organ system | Outcome | Hazard ratio (95% CI) | Event rate, % at 180 d (95% CI) | Event rate, % at 180 d (95% CI).1 | Event rate, % at 180 d (95% CI).2 |
| --- | --- | --- | --- | --- | --- |
| Organ system | Outcome | Hazard ratio (95% CI) | Nirmatrelvir groupa | Control groupb | Absolute risk reduction |
| Cardiovascular | Dysrhythmia | 0.73 (0.68 to 0.78) | 2.86 (2.67 to 3.06) | 3.91 (3.82 to 4.00) | 1.05 (0.84 to 1.26) |
| Cardiovascular | Ischemic heart disease | 0.71 (0.63 to 0.79) | 1.23 (1.10 to 1.36) | 1.74 (1.68 to 1.80) | 0.51 (0.37 to 0.65) |
| Coagulation and hematologic | Pulmonary embolism | 0.61 (0.51 to 0.74) | 0.43 (0.35 to 0.51) | 0.70 (0.66 to 0.74) | 0.27 (0.18 to 0.36) |
| Coagulation and hematologic | Deep vein thrombosis | 0.72 (0.56 to 0.93) | 0.26 (0.20 to 0.32) | 0.36 (0.33 to 0.39) | 0.10 (0.03 to 0.17) |
| Fatigue and malaise | Fatigue and malaise | 0.79 (0.73 to 0.84) | 3.53 (3.29 to 3.77) | 4.47 (4.37 to 4.58) | 0.94 (0.69 to 1.20) |
| Gastrointestinal | Liver disease | 0.91 (0.81 to 1.02) | 1.31 (1.17 to 1.45) | 1.45 (1.39 to 1.51) | 0.14 (−0.02 to 0.29) |
| Kidney | Acute kidney injury | 0.67 (0.58 to 0.77) | 0.79 (0.68 to 0.89) | 1.17 (1.12 to 1.22) | 0.38 (0.27 to 0.5) |
| Musculoskeletal | Muscle pain | 0.65 (0.58 to 0.72) | 1.35 (1.22 to 1.49) | 2.08 (2.01 to 2.14) | 0.72 (0.57 to 0.88) |
| Metabolic | Diabetes | 0.98 (0.86 to 1.11) | 1.49 (1.31 to 1.67) | 1.52 (1.45 to 1.59) | 0.03 (−0.16 to 0.22) |
| Neurological | Neurocognitive impairment | 0.74 (0.67 to 0.83) | 1.29 (1.16 to 1.42) | 1.74 (1.68 to 1.80) | 0.45 (0.30 to 0.59) |
| Neurological | Dysautonomia | 0.86 (0.75 to 0.97) | 0.97 (0.86 to 1.09) | 1.14 (1.09 to 1.19) | 0.16 (0.04 to 0.29) |
| Pulmonary | Shortness of breath | 0.89 (0.83 to 0.95) | 4.18 (3.91 to 4.44) | 4.69 (4.58 to 4.79) | 0.51 (0.23 to 0.79) |
| Pulmonary | Cough | 0.96 (0.88 to 1.04) | 3.47 (3.21 to 3.73) | 3.62 (3.52 to 3.72) | 0.15 (−0.13 to 0.43) |
## Risk of PCC in Subgroups
Compared with the control group, people treated with nirmatrelvir exhibited reduced risk of PCC in subgroups based on age, race, sex, smoking, cancer, cardiovascular disease, chronic kidney disease, chronic lung disease, diabetes, immune dysfunction and hypertension (Figure 3A; eTable 4 in Supplement 1).
**Figure 3.:** *Relative Risk of Post–COVID-19 Condition in the Nirmatrelvir vs No-Treatment GroupsA, By demographic and disease subgroups included age (≤60 years, >60 years to ≤70 years, and >70 years), race (White and Black), sex, smoking status (current smoker, former smoker, and never smoker), cancer, cardiovascular disease, chronic kidney disease, chronic lung disease, diabetes, immune dysfunction, and hypertension. B, By number of baseline risk factors (1 to 2, 3 to 4, ≥5), vaccination status (unvaccinated, 1 to 2 doses of vaccine, and boosted), and SARS-CoV-2 infection status (with primary SARS-CoV-2 infection and reinfection). Baseline risk factors of progression to severe acute COVID-19 illness included age of older than 60 years, body mass index of more than 25 (calculated as weight in kilograms divided by height in meters squared), current smoker, cancer, cardiovascular disease, kidney disease, chronic lung disease, diabetes, immune dysfunction, and hypertension. Outcomes were ascertained 30 days after the SARS-CoV-2 positive test until the end of follow-up.*
Because nirmatrelvir is prescribed to people with at least 1 baseline risk factor for progression to severe acute COVID-19 illness, and to better understand the association between nirmatrelvir and the risk of PCC in people with different baseline risk strata, the association between nirmatrelvir and the risk of PCC was tested according to the number of baseline risk factors for progression to severe acute COVID-19 illness. Nirmatrelvir was associated with reduced risk of PCC in people with 1 to 2, 3 to 4, and 5 or more baseline risk factors (Figure 3B; eTable 4 in Supplement 1).
Examination of the association between nirmatrelvir and risk of PCC by vaccine status suggested that nirmatrelvir was associated with reduced risk of PCC in people who were unvaccinated, vaccinated, and those who received a booster vaccine (Figure 3B; eTable 4 in Supplement 1). Nirmatrelvir was associated with reduced risk of PCC in people with primary SARS-CoV-2 infection and in people with reinfection (Figure 3B; eTable 4 in Supplement 1).
## Sensitivity Analyses
To assess the robustness of the present study’s findings, multiple sensitivity analyses were conducted: [1] the overlap weighting method was applied to balance baseline characteristics in the treatment and control groups instead of the inverse probability weighing approach used in the primary analyses; [2] the doubly robust approach was applied to additionally adjust for covariates in the inverse probability weighted survival models, compared with the primary approach which used inverse probability weighted survival models; [3] a high-dimensional variable selection algorithm was used to identify an additional 100 covariates that were then used, along with a predefined set of covariates, to construct the weights, compared with the primary approach that used predefined covariates; [4] the inverse probability of censoring weight was applied to account for those who died during the acute phase of infection, compared with the primary approach, in which those who died during acute phase were removed from the analyses; [5] outcomes were defined based on events that occurred 90 days after infection, compared with the primary approach, in which outcomes were defined based on events that occurred 30 days after infection; [6] incident outcome was defined as occurrence of the outcome in those without history of the related outcome within 5 years before infection, compared with the primary approach, in which the washout period was 3 years before infection; [7] hospitalization, intensive care unit admission, and ventilator use during the acute phase of infection were additionally adjusted for, compared with the primary approach that did not adjust for information after exposure; [8] outcomes were defined based on PCC ICD-10 code (U09.9), compared with the primary definition that was based on a set of 13 predefined post–acute sequelae of COVID-19; and [9] outcomes were defined based on the first occurrence of any individual sequela, compared with the primary definition that accounted for both the number of sequelae and the influence of each sequela on health. All sensitivity analyses yielded results that are consistent (in both direction and magnitude) to those obtained using the primary approach (eTable 5 in Supplement 1).
## Discussion
In this cohort study of 281 793 people with SARS-CoV-2 infection who had at least 1 risk factor for progression to severe COVID-19 illness, compared with the control group of people who did not receive antiviral or antibody treatment during the acute phase of SARS-CoV-2 infection, treatment with nirmatrelvir within 5 days of a positive SARS-CoV-2 test was associated with reduced risk of PCC, including reduced risk of 10 of 13 post–acute sequelae examined. Nirmatrelvir was also associated with reduced risk of post–acute death and hospitalization at 180 days. Nirmatrelvir was associated with reduced risk of PCC in subgroups based on age, race, sex, smoking, cancer, cardiovascular disease, chronic kidney disease, chronic lung disease, diabetes, immune dysfunction, and hypertension. Nirmatrelvir was associated with reduced risk of PCC across strata of baseline risk, and in people who were unvaccinated, vaccinated, and boosted, and in people with primary SARS-CoV-2 infection and reinfection. Altogether, the findings suggest that treatment with nirmatrelvir during the acute phase reduces the risk of post–acute adverse health outcomes.
These results show that the salutary effect of nirmatrelvir may extend to the post–acute phase of COVID-19; nirmatrelvir was associated with reduced risk of PCC in the overall cohort and in various subgroups, including those across risk strata, vaccination status, and prior history of COVID-19. These findings are coupled with the observation that among 281 793 people with acute SARS-CoV-2 infection who had at least 1 risk factor for progression to severe disease who would be eligible for treatment with nirmatrelvir, 35 717 ($12.67\%$) patients were treated with nirmatrelvir, and 246 076 ($87.33\%$) patients received no antiviral treatment. The totality of evidence suggests that improving the uptake and use of nirmatrelvir in the acute phase as a means of not only preventing progression to severe acute disease but also reducing the risk of post–acute adverse health outcomes may be beneficial.
Nirmatrelvir was associated with $26\%$ less risk of PCC, $47\%$ less risk of post–acute death, and $24\%$ less risk of post–acute hospitalization; the magnitude of risk reduction on the absolute scale is also substantial amounting to 4.51, 0.65, and 1.72 less cases of PCC, post–acute death, and post–acute hospitalization for every 100 treated persons between 30 to 180 days of infection. These findings should be contextualized within the broader body of evidence showing effectiveness of nirmatrelvir in also reducing risk of hospitalization or death in the acute phase.6 The clinical decision to initiate treatment with nirmatrelvir should consider its overall effectiveness in reducing burden of death and disease in both the acute and post–acute phases of COVID-19.
Nirmatrelvir was approved in the US for the treatment of acute COVID-19 illness in people with 1 or more risk factors for progression to severe disease. Whether the salutary benefit of nirmatrelvir extends to people without risk factors for progression to severe disease (who would not qualify for nirmatrelvir prescription under the current US Food and Drug Administration emergency use authorization and were not included in the present study’s analyses) remains to be tested in future randomized clinical trials.
We note that the present results suggested risk reduction for some but not all the prespecified post–acute sequelae in this analysis. It is possible that various sequelae are mediated by various mechanisms including some that may be affected by the receipt of antivirals and others that may not. Participants in the current study were treated in the acute phase with a 5-day course of nirmatrelvir; it remains unclear whether longer duration of treatment, a higher treatment dose, or both may have resulted in more reduced risk of post–acute sequelae. It is also unclear whether initiation of treatment in the post–acute phase of COVID-19 reduces the risk of PCC.
While we examined nirmatrelvir in this work, other antivirals that have also been shown to have efficacy and effectiveness in the acute phase (eg, molnupiravir) should also be tested to understand whether the association reported here extends to other antivirals.27 This approach will help expand clinicians’ armamentarium and reduce reliance on a single agent—especially with a rising risk of antiviral resistance.28,29,30 This study has several strengths. The VA operates the largest integrated health care system in the US, and the vast and rich national health care databases of the VA—with a large number of treated and untreated patients followed longitudinally over time—allows the evaluation of outcomes that were not assessed in randomized clinical trials. The VA data contain comprehensive information about participants, including COVID-19 testing results, medication use, vaccination records, hospitalization records, death records, and other attributes, which allows the comprehensive capture of covariates from different domains, such as demographics, diagnoses, laboratory test results, medications, vital signs, health care utilization, and contextual factors. We tested robustness of the present findings in multiple sensitivity analyses that yielded consistent results.
## Limitations
This study has several limitations. The demographic composition of the cohort (majority older, White, male adults) and accessibility to VA health care may limit generalizability of study findings. We used the electronic health care databases of the VA to conduct this study, and although we took care to adjust the analyses for a large set of predefined variables, we cannot completely rule out misclassification bias and residual confounding. We relied on filled prescription records to assign exposure, and filling of nirmatrelvir prescription does not necessarily guarantee use. We did not capture nirmatrelvir use outside the VA system; if a large number of people in the control group used nirmatrelvir outside the VA, this may bias the results toward the null. These data do not capture hospitalization and diagnoses which may have occurred outside the VA. We focused the present analyses on a prespecified set of 13 sequelae and did not examine all possible components of PCC. We examined the association of nirmatrelvir with PCC at 180 days after infection, and randomized clinical trials with longer follow-up would help further support the findings. Finally, as the virus continues to mutate, new variants emerge, and vaccine uptake improves, it is possible that the effectiveness of nirmatrelvir may also change over time.
## Conclusions
This cohort study found that in people with SARS-CoV-2 infection who had at least 1 risk factor for progression to severe disease, treatment with nirmatrelvir within 5 days of a positive SARS-CoV-2 test was associated with a reduced risk of PCC across the risk spectrum in this cohort and regardless of vaccination status and history of prior infection. These findings suggest that the salutary benefit of nirmatrelvir may extend to the post–acute phase of COVID-19.
## References
1. Al-Aly Z, Xie Y, Bowe B. **High-dimensional characterization of post-acute sequelae of COVID-19**. *Nature* (2021) **594** 259-264. DOI: 10.1038/s41586-021-03553-9
2. Al-Aly Z, Bowe B, Xie Y. **Long COVID after breakthrough SARS-CoV-2 infection**. *Nat Med* (2022) **28** 1461-1467. DOI: 10.1038/s41591-022-01840-0
3. Davis HE, McCorkell L, Vogel JM, Topol EJ. **Long COVID: major findings, mechanisms and recommendations**. *Nat Rev Microbiol* (2023) **21** 133-146. DOI: 10.1038/s41579-022-00846-2
4. Merad M, Blish CA, Sallusto F, Iwasaki A. **The immunology and immunopathology of COVID-19**. *Science* (2022) **375** 1122-1127. DOI: 10.1126/science.abm8108
5. Ledford H. **Long-COVID treatments: why the world is still waiting**. *Nature* (2022) **608** 258-260. DOI: 10.1038/d41586-022-02140-w
6. Hammond J, Leister-Tebbe H, Gardner A. **Oral nirmatrelvir for high-risk, nonhospitalized adults with Covid-19**. *N Engl J Med* (2022) **386** 1397-1408. DOI: 10.1056/NEJMoa2118542
7. Kind AJH, Buckingham WR. **Making neighborhood-disadvantage metrics accessible—the neighborhood atlas**. *N Engl J Med* (2018) **378** 2456-2458. DOI: 10.1056/NEJMp1802313
8. Cohen K, Ren S, Heath K. **Risk of persistent and new clinical sequelae among adults aged 65 years and older during the post-acute phase of SARS-CoV-2 infection: retrospective cohort study**. *BMJ* (2022) **376**. DOI: 10.1136/bmj-2021-068414
9. Bull-Otterson L, Baca S, Saydah S. **Post–COVID conditions among adult COVID-19 survivors aged 18-64 and ≥65 years—United States, March 2020-November 2021**. *MMWR Morb Mortal Wkly Rep* (2022) **71** 713-717. DOI: 10.15585/mmwr.mm7121e1
10. Daugherty SE, Guo Y, Heath K. **Risk of clinical sequelae after the acute phase of SARS-CoV-2 infection: retrospective cohort study**. *BMJ* (2021) **373** n1098. DOI: 10.1136/bmj.n1098
11. Ayoubkhani D, Khunti K, Nafilyan V. **Post-COVID syndrome in individuals admitted to hospital with COVID-19: retrospective cohort study**. *BMJ* (2021) **372** n693. DOI: 10.1136/bmj.n693
12. Carfì A, Bernabei R, Landi F, Gemelli Against C-P-ACSG. **Persistent symptoms in patients after acute COVID-19**. *JAMA* (2020) **324** 603-605. DOI: 10.1001/jama.2020.12603
13. Xie Y, Bowe B, Al-Aly Z. **Burdens of post–acute sequelae of COVID-19 by severity of acute infection, demographics and health status**. *Nat Commun* (2021) **12** 6571. DOI: 10.1038/s41467-021-26513-3
14. Taquet M, Dercon Q, Luciano S, Geddes JR, Husain M, Harrison PJ. **Incidence, co-occurrence, and evolution of long-COVID features: a 6-month retrospective cohort study of 273,618 survivors of COVID-19**. *PLoS Med* (2021) **18**. DOI: 10.1371/journal.pmed.1003773
15. Bowe B, Xie Y, Xu E, Al-Aly Z.. **Kidney outcomes in long COVID**. *J Am Soc Nephrol* (2021) **32** 2851-2862. DOI: 10.1681/ASN.2021060734
16. Xie Y, Xu E, Al-Aly Z. **Risks of mental health outcomes in people with COVID-19: cohort study**. *BMJ* (2022) **376**. DOI: 10.1136/bmj-2021-068993
17. Davis HE, Assaf GS, McCorkell L. **Characterizing long COVID in an international cohort: 7 months of symptoms and their impact**. *EClinicalMedicine* (2021) **38**. DOI: 10.1016/j.eclinm.2021.101019
18. Taquet M, Geddes JR, Husain M, Luciano S, Harrison PJ. **6-month neurological and psychiatric outcomes in 236 379 survivors of COVID-19: a retrospective cohort study using electronic health records**. *Lancet Psychiatry* (2021) **8** 416-427. DOI: 10.1016/S2215-0366(21)00084-5
19. Xie Y, Al-Aly Z. **Risks and burdens of incident diabetes in long COVID: a cohort study**. *Lancet Diabetes Endocrinol* (2022) **10** 311-321. DOI: 10.1016/S2213-8587(22)00044-4
20. Xu E, Xie Y, Al-Aly Z. **Long-term neurologic outcomes of COVID-19**. *Nat Med* (2022) **28** 2406-2415. DOI: 10.1038/s41591-022-02001-z
21. Xu E, Xie Y, Al-Aly Z. **Risks and burdens of incident dyslipidaemia in long COVID: a cohort study**. *Lancet Diabetes Endocrinol* (2023) **11** 120-128. DOI: 10.1016/S2213-8587(22)00355-2
22. Xu E, Xie Y, Al-Aly Z. **Long-term gastrointestinal outcomes of COVID-19**. *Nat Commun* (2023) **14**. DOI: 10.1038/s41467-023-36223-7
23. Bowe B, Xie Y, Al-Aly Z. **Acute and postacute sequelae associated with SARS-CoV-2 reinfection**. *Nat Med* (2022) **28** 2398-2405. DOI: 10.1038/s41591-022-02051-3
24. Wulf Hanson S, Abbafati C, Aerts JG. **Estimated global proportions of individuals with persistent fatigue, cognitive, and respiratory symptom clusters following symptomatic COVID-19 in 2020 and 2021**. *JAMA* (2022) **328** 1604-1615. DOI: 10.1001/jama.2022.18931
25. Xie Y, Bowe B, Maddukuri G, Al-Aly Z. **Comparative evaluation of clinical manifestations and risk of death in patients admitted to hospital with COVID-19 and seasonal influenza: cohort study**. *BMJ* (2020) **371** m4677. DOI: 10.1136/bmj.m4677
26. Bowe B, Cai M, Xie Y, Gibson AK, Maddukuri G, Al-Aly Z. **Acute kidney injury in a national cohort of hospitalized US veterans with COVID-19**. *Clin J Am Soc Nephrol* (2020) **16** 14-25. DOI: 10.2215/CJN.09610620
27. Xie Y, Bowe B, Al-Aly Z. **Molnupiravir and risk of hospitalization or death: a target trial emulation using the US Veterans Affairs healthcare databases**. *BMJ* (2023) **380**. DOI: 10.13039/100000738
28. Al-Aly Z, Agarwal A, Alwan N, Luyckx VA. **Long COVID: long-term health outcomes and implications for policy and research**. *Nat Rev Nephrol* (2022) **19** 1-2. DOI: 10.1038/s41581-022-00652-2
29. Al-Aly Z. **Diabetes after SARS-CoV-2 infection**. *Lancet Diabetes Endocrinol* (2023) **11** 11-13. DOI: 10.1016/S2213-8587(22)00324-2
30. Iketani S, Mohri H, Culbertson B. **Multiple pathways for SARS-CoV-2 resistance to nirmatrelvir**. *Nature* (2023) **613** 558-564. DOI: 10.1038/s41586-022-05514-2
|
---
title: A qualitative study assessing how reach and participation can be improved in
workplace smoking cessation programs
authors:
- Nikita L. Poole
- Gera E. Nagelhout
- Tessa Magnée
- Lotte C. I. de Haan-Bouma
- Cas Barendregt
- Onno C. P. van Schayck
- Floor A. van den Brand
journal: Tobacco Prevention & Cessation
year: 2023
pmcid: PMC10037216
doi: 10.18332/tpc/161589
license: CC BY 4.0
---
# A qualitative study assessing how reach and participation can be improved in workplace smoking cessation programs
## Abstract
### INTRODUCTION
Randomized controlled trials have demonstrated the effectiveness of workplace smoking cessation programs. However, with low participation rates reported, it is important to understand the barriers and facilitators for the reach and participation of employees in workplace smoking cessation programs. The objective of the present study is to uncover the needs of employees regarding reach and participation when implementing a workplace program to address smoking cessation.
### METHODS
We carried out 19 semi-structured qualitative interviews in 2019 based on the Reach, Effectiveness, Adoption, Implementation and Maintenance (RE-AIM) Framework with current and former smoking employees of organizations with ≥100 employees in the Netherlands. Some of the interviewees had experience with a cessation program. Data were analyzed using the Framework method.
### RESULTS
The main barriers according to employees were insufficient promotion of the cessation program, completing the program in the employee’s own time and working night shifts and peak hours. Facilitators included being actively approached to participate by a colleague, positive reactions from colleagues about employee’s participation in the program, providing the program on location and integrating the program as part of the organization’s vitality policy.
### CONCLUSIONS
Effective workplace programs for smoking cessation can stimulate cessation but implementers often experience low participation rates. Our study presents recommendations to improve the recruitment and participation of employees in a workplace smoking cessation program, such as using active communication strategies, training managers to stimulate smoking employees to participate and making the program as accessible as possible by reimbursing time spent and offering the program at the workplace or nearby. Integrating the smoking cessation program into wider company vitality policy will also aid continued provision of the program.
## INTRODUCTION
The workplace is a valuable setting for reaching a large adult population for health promotion programs, such as those that encourage smoking cessation1,2. Furthermore, workplace cessation programs are just as effective as those in other settings at stimulating cessation3 and effective for those with a lower income and educational level4-6, suggesting utility to address smoking in lower socio-economic groups, under which smoking prevalence is often higher7. Not only does a smoking employee's health gain from quitting, but employers could also recoup and prevent costs incurred due to increased sick leave or disability, productivity losses owing to smoking breaks and increased healthcare costs8,9. Reduced smoking prevalence also translates into lower healthcare costs and increased quality of life years10,11.
A problem that remains, however, is the low rate of participation often achieved in such workplace programs3,12 which may become progressively more challenging as workers become increasingly mobile due to distance-working3, a trend heightened by the COVID-19 pandemic. Part-time or temporary contracts pose additional challenges for recruitment and the scheduling of activities so that they remain accessible to all13. Furthermore, there is a dearth of literature regarding the needs of employees when implementing a workplace health promotion program as most of the attention thus far has been given to the employer’s needs.
Workplaces in the Netherlands are moving towards becoming totally smoke-free, with legislation expanding on the 2004 workplace smoking ban14 to include the removal of designated smoking rooms in all workplaces in 2022 and introduction of smoke-free outdoor grounds of institutions such as hospitals, mental health facilities and government buildings in 202515,16. With these developments, it would be prudent for employers to consider offering a smoking cessation program to their employees and therefore it is important that we learn how this can be done successfully. In 2019, $26.2\%$ and $24.9\%$ of lower and moderately educated adults in the Netherlands smoked respectively, of which $89.3\%$ and $76.3\%$ smoked daily, compared to $15.4\%$ of highly educated adults who smoked, of which $48.7\%$ smoked daily17. To better understand the barriers and facilitators in the reach and participation of employees in workplace cessation programs, we conducted a qualitative needs assessment among employees in the Netherlands, focusing on workplaces with employees with a lower level of education. Our needs assessment focused on the following research questions: 1) ‘How can employees be reached to inform and stimulate them to participate in a smoking cessation program?’, 2) ‘How do colleagues react to the participation of others in the program?’, 3) ‘What are the practical barriers and facilitators to participation in the program?’, and 4) ‘What factors should be considered when maintaining a smoking cessation program in an organization?’.
## Design
Within this qualitative study, we performed individual qualitative interviews ($$n = 19$$) among employees of organizations in the Netherlands. Interviews were performed between January and June 2019.
## Sample
Purposive sampling was used to recruit current and former smoking employees of organizations with ≥100 employees of which relatively many people have a low level of education. Purposive sampling also made sure a variation of organizations from different sectors and employee occupations were included (Table 1). Our sample included respondents who had and had not experienced a workplace smoking cessation program to explore experienced and anticipated barriers and facilitators to reach and participation. Most of the interviewees were recruited as participants from a previous RCT in which smoking cessation group programs with financial incentives were offered6. The smoking cessation program in this RCT consisted of group-based weekly sessions of 1.5 hours for 7 weeks. Interviewees included were from both the treatment group (smoking cessation group program with financial incentives for quit success) and control group (smoking cessation group program without financial incentives). Other interviewees who had not participated in the RCT were recruited via convenience sampling through company representatives. All interviewees received 20€ as compensation for their participation in the interview, which was directly deposited to their bank account. The aims of the research were shared with the interviewees via an information letter and the informed consent form. The Central Committee on Research Involving Human Subjects in the Netherlands requires no ethical approval for non-medical research. The interviewing author and the interviewees did not know each other prior to study commencement.
**Table 1**
| Characteristics | n (%) |
| --- | --- |
| Gender | |
| Man | 9 (47) |
| Woman | 10 (53) |
| Age (years) | |
| 30–39 | 5 (26) |
| 40–49 | 7 (37) |
| 50–59 | 6 (32) |
| ≥60 | 1 (5) |
| Education level† | |
| Low | 2 (11) |
| Moderate | 11 (58) |
| High | 6 (32) |
| Occupation * | |
| Managers | 3 (16) |
| Technicians and associate professionals | 2 (11) |
| Clerical support | 6 (32) |
| Services and sales | 5 (26) |
| Plant and machine operators and assemblers | 3 (16) |
| Sector | |
| Education | 3 (16) |
| Emergency services | 2 (11) |
| Financial | 1 (5) |
| Government | 3 (16) |
| Industrial (chemical, horticulture, metal, sheltered work) | 5 (26) |
| Healthcare | 1 (5) |
| Retail | 4 (21) |
| Participated in cessation program | |
| Yes – with financial incentive | 11 (58) |
| Yes – without financial incentive | 3 (16) |
| No | 5 (26) |
| Current smoker | |
| No | 11 (58) |
| Yes | 8 (42) |
## Data collection
Semi-structured interviews were conducted face-to-face at the workplace by one of the authors (CB), who is trained and experienced in qualitative interviewing. A qualitative approach was used so that employees could share their opinions and experiences in a detailed way, also allowing the interviewer to probe when new or unexpected findings were reported. See the Supplementary file for interview topic lists for employees who have and have not experienced a workplace smoking cessation program.
Interviews lasted between 30 and 80 minutes. Two of the interviews were conducted with a manager or human resources (HR) representative present.
The interview guide was semi-structured and based on the RE-AIM Framework, which stands for Reach, Effectiveness, Adoption, Implementation and Maintenance18. We focused on aspects of reach, adoption, implementation and maintenance as the effectiveness of workplace group cessation programs has been investigated in previous work3,6,19. For adoption, we asked about the acceptability of the program by employees and their colleagues.
As mentioned, not all interviewees received a workplace smoking cessation program. The cessation program was described to these respondents before asking them what barriers and facilitators they perceived to exist.
## Analysis
All interviews were audio recorded, transcribed verbatim and imported into NVivo 12 (QSR International©, Melbourne, Australia) for coding and analysis. All interviews were coded using the Framework method20. This method consists of five stages: 1) familiarization, 2) identifying the thematic framework, 3) indexing, 4) charting, and 5) mapping and interpretation. The first stage, familiarization, was undertaken by CB, LdH-B, TM and NP. CB wrote memos and an overall report of the interviews and LdH-B, TM and NP read the report and the transcripts to familiarize themselves with the data. The report was also shared with the respondents so that they had the opportunity to provide comments. LdH-B and TM completed the second stage (indexing), coding the transcripts both deductively and inductively21. The first two transcripts were double coded by LdH-B and TM, before agreeing on a final thematic framework (stage three), whereafter the remaining transcripts were coded individually by LdH-B and TM. Themes were arranged based on the RE-AIM model from which further sub-themes were made, for example, under Reach, sub-themes ‘reasons to participate or quit smoking’ and ‘recruitment general’ were made. These sub-themes were further broken down into individual factors given by the respondents. FvdB and NP created a matrix based on the RE-AIM model and responses were summarized and added to the matrix (stage 4, charting). In the final stage, mapping and interpretation, the matrix was primarily examined by NP for connections and comparisons across respondents within codes, checking the original transcripts for context. Interpretations were discussed within the research team. For only one theme data saturation was not reached, as the sub-theme ‘not suitable for temporary workers’ emerged in the penultimate interview. The data was also analyzed for patterning in responses based on characteristics such as participation in the program, gender and smoking status; however no substantial differences in responses were found.
## How can employees be reached to inform and stimulate them to participate in a cessation program?
Employees mostly expressed that a more proactive and personal approach is needed to stimulate participation. Respondents felt that more could be done to promote the smoking cessation program as some only received a mass e-mail or saw a message on the intranet (Table 2). They also mentioned that the traditional channels of communication (e-mail, intranet messages) will not necessarily reach all types of employees such as those who are not office-based. A program participant mentioned:
**Table 2**
| Barriers | Facilitators |
| --- | --- |
| How can employees be reached to inform and stimulate them to participate in a cessation program? | How can employees be reached to inform and stimulate them to participate in a cessation program? |
| Insufficient promotion of the program, employees not approached personallyBeing approached to participate with a judgmental toneShame associated with failing to quit, which prevented talking about participating with colleagues | Being personally approached by team leader/HR staffHearing about the program from a colleague who would also participateThe program being promoted with success stories from past participantsPromotional materials available in native languages of employees |
| How do colleagues react to the participation of others in the program? | How do colleagues react to the participation of others in the program? |
| Anticipation that colleagues would be negative about time reimbursementSmoking colleagues are skeptical about quit success | Colleagues react positively and supportively to participationSeeing colleagues participate stimulates interest to participate among employees who still smokeExplaining to all employees the purpose of the program and the benefits the program can have for all staff |
| What are the practical barriers and facilitators to participation in the program? | What are the practical barriers and facilitators to participation in the program? |
| Having to complete the program in own time or use annual leaveWorking night shifts and peak hoursProgram not suitable for temporary workers | Time is reimbursed by the employerThe workplace setting lowers the threshold to participateOffering alternatives to a group-based program |
| What factors should be considered when maintaining a smoking cessation program in an organization? | What factors should be considered when maintaining a smoking cessation program in an organization? |
| Program was not long enough for sustained motivation | Enthusiasm for program to be repeatedIntegrating the program as part of company vitality policyProviding longer aftercare period with follow-up session(s) |
‘I came across it on the intranet, but if I had looked a week later … I could have missed it, the message.’ ( Participant 11) In addition, more proactive promotion of the program would enable more employees to hear about the program: ‘I actually had to enquire about it myself.’ ( Participant 10) Another barrier mentioned was being approached to participate with a judgmental tone:
‘Don’t go the usual way of “Yeah, you smoke”, “Our sick leave is so high because you smoke” You know, that blaming.’ ( Participant 8) Respondents reported discussing the announcement of the cessation program with a colleague, which was an important reason as to why they signed up for the program: ‘At first I was like “no” until my colleague who was super enthusiastic said “ah, we’re going to do that”. That won me over.’ ( Participant 2) A few respondents, however, did not discuss their decision to participate with colleagues, with one employee noting that shame was a barrier to talking about participating with colleagues:
‘But [not discussing participating with colleagues] has a reason and the reason was the shame that I would feel if I didn’t succeed.’ ( Participant 7) Employees also emphasized the importance of being personally approached, although opinions on whether one should be approached specifically by a direct manager or someone from the HR department varied: ‘[if you personally approach people] I think employees would feel less like a number. If you receive an e-mail, it says “dear [name]”, but that is of program computer work. If they really visit the smokers, people personally, I think you will achieve more.’ ( Participant 6) A direct manager was thought of by the majority of respondents to be most appropriate as there is more trust built to discuss these topics and because they know which employees smoke, other respondents suggested a member of HR to convey the importance of the program.
Respondents came with their own ideas to promote the program. One respondent who did not participate in the program, from a company with many non-Dutch-speaking employees, recommended that there also needs to be attention for the language in which promotional materials are shared: ‘*It is* important that you give a printed sheet in Polish. Give it to the people.’ ( Non-participant 3) Another mentioned that past program participants could be interviewed or share their success stories on the intranet: ‘How do you make sure that people are still enthusiastic? You only get that with … a few good stories and keep promoting that.’ ( Participant 1)
## How do colleagues react to the participation of others in the program?
Colleagues were mostly positive about participation, although less so when the program was not held in the employees’ own time. Respondents mostly reported (anticipated) positive responses from their colleagues about their participation in the program and that they received social support from colleagues for their participation: ‘My experience is that my colleagues wished me the best.’ ( Participant 14) ‘[my colleagues] think it’s great that I quit. They motivated me to quit.’ ( Participant 10) Seeing colleagues participate in the program also stimulated interest in some smoking colleagues.
Whilst colleagues were typically positive about the respondent’s participation in the cessation program, some wondered whether their colleagues were less positive about their time for the program being reimbursed: ‘They were positive towards me, but I don’t know what they say behind my back: “how much time does that take?” ’ ( Participant 13) A few respondents said that they had received skeptical reactions from colleagues about how successful they would be at quitting: ‘Well, there was a bit of “Yeah, (…) you’re not going to make it. You know, you’ve tried so many times, so it won’t work this time either”.’ ( Participant 4)
Some colleagues only shared their skepticism after the program had finished.
## What are the practical barriers and facilitators to participation in the program?
Employees named several practical barriers to participation, spanning the setting and timing of the program. Many respondents emphasized the importance of the program being offered during working hours and not having to use their leave allowance: ‘I don’t think you should get into that discussion that it’s about hours, because then someone will already drop out. So I think you should fully facilitate it and [the hours] should never be an issue if you ask me.’ ( Participant 1) Respondents who were not reimbursed for their time on the program found this to be a significant barrier to participation: ‘Our employer let us participate in this, but he has actually said from the start: “We facilitate it in the sense that we make a location available, but you have to invest your own time there”. This has been a thorn in my side, I’m very honest, because I think that if you set up something like this as an employer, which is very good, you should also make people free for that.’ ( Participant 2)
The workplace setting also lowered the threshold to participate: ‘And because it was so accessible, eh, I was already here, it was an hour of your working hours, so the walk was very easy, er, I did it because I thought: well, what have I got to lose? You know, I’m here anyway, I’ll walk over there and I’ll see.’ ( Participant 14) Additional barriers were faced by people who work night shifts or experience peak hours during their shift as the timeslots for the program were sometimes not convenient or were at odds with the demands of work: ‘I don’t think there are very many people who would like to stay before or after the night shift.’ ( Nonparticipant 1)
‘I think it varies from department to department. I have a peak here from 12:00 to 14:00 (...) After or before that, I can arrange something. But don’t touch my peak, you know?’ ( Non-participant 4) Lastly, it was reported that it may not be possible to offer the program to all employees, such as those with short, temporary contracts: ‘The program isn’t for everyone. Some people work for two weeks, sometimes three weeks and then they go again. But this is suitable for permanent workers.’ ( Non-participant 3)
## What factors should be considered when maintaining a smoking cessation program in an organization?
Respondents suggested ideas for improving the program and some saw it as part of a wider movement towards a healthier lifestyle. Respondents were happy overall with the program, regardless of abstinence status at the time of the interview. Those who had not successfully quit smoking after the program shared their enthusiasm for it to be repeated for themselves or colleagues who had not yet participated: ‘I e-mailed [HR] about this: “I actually want to do [the program] again, because I didn’t stop then”. Actually yes, I did stop, but then I started again.’ ( Participant 12) The suggestion came from one respondent to integrate the program into existing company vitality policy: ‘I would perhaps make this a part of a kind of vitality program. It’s not just a healthy diet and a healthier lifestyle, but smoking is part of it too.’ ( Participant 9)
Some respondents expressed that, should the program be offered again, they would like the program to be longer than seven weeks, with continued attention paid to quitters for sustained motivation: ‘… maybe a little more attention, that you also feel the motivation more. If that fades away, you won’t have that incentive anymore.’ ( Participant 10)
## DISCUSSION
Our qualitative needs-assessment identified several barriers and facilitating factors in the reach and participation of a workplace smoking cessation program. Firstly, in reaching employees to inform them about the program, many employees felt that their employers’ efforts were not sufficient, especially when employers relied on digital communication (e-mail or intranet messages). Dutch employers also reported that some employees were not reachable through digital communication channels22. *More* generally, passive methods of recruitment for smoking cessation programs, such as public announcements, are associated with lower levels of recruitment and retention23. Instead, proactive and personal communication can increase program reach and be particularly beneficial in recruiting those with a low socioeconomic position24. In the current study, employees were far more receptive to a more personal approach, with some having decided to participate due to a conversation with a colleague. Rather than relying on incidental word-of-mouth promotion, employers could actively target team leaders or other key figures among their staff to share and promote the program. The introduction of smoke-free policy presents an opportunity to engage with employees on this topic. Although not mentioned by the participants in this study, an additional social aspect that may play a role in the decision to participate is not wanting to lose time with colleagues, which is currently spent on smoking breaks, by quitting25.
Not all groups are equally easy to reach, however. In the present study, we found that language ability could be a barrier to reaching all employees, especially where (a large proportion of) employees do not understand or speak the national language(s). With this, it is important to make sure that any materials are translated accurately26 and so testing of promotional materials would be advisable. Additionally, in organizations where this is prevalent, word-of-mouth promotion may be particularly beneficial.
Accessibility for low-threshold participation was important to the employees as they felt that the program should be offered on location or nearby, that their time should be reimbursed and fit into their working hours to the extent that that is possible. Our study found that negotiation over employees’ own time investment may deter participation, especially for those who already work unsociable hours. However, some employees suspected that colleagues would regard the reimbursement of hours as unfair, although to what extent this sentiment may have been voiced is unknown. Employers’ views on the program being reimbursed vary, as some recognized the lowered threshold to participation but others felt it was only fair that the employees invested some of their own time22. To temper potential negative reactions from colleagues, employers should clearly explain to their employees that by reimbursing time spent on the program, they are fully supporting employees to embark on a healthier lifestyle. Offering a vitality program, under which smoking cessation is one component alongside other topics such as exercise and healthy eating, also gives other employees the opportunity to engage in a healthier lifestyle.
Providing the program at the workplace has been highlighted as an important factor in deciding to participate previously27 and was also mentioned as important by the employees in our study. For some occupations, however, simply offering the program on location and reimbursing the time will not be sufficient, as additional barriers are faced by those who experience peak hours (for instance in customer- or client-facing roles) or for employees who work irregular hours, such as night shifts or who are temporary workers. Time-related barriers to participation such as a high workload, inflexibility to leave their immediate work area and competing work obligations have also been reported for other workplace programs28-31. Whilst it may not be possible for employees who work from home or who have demanding or incongruous work schedules to participate in a group-based program, these groups should not be forgotten or treated with less priority, as both they and their employer can still benefit from reduced illness and disability as a result of quitting8. Temporary workers hold a particularly unstable position in the workplace and may therefore experience more stress32, furthering their need for cessation support. Other more flexible options such as telephone, online or individual counselling could be more accessible for these groups. Individual smoking cessation counselling of a similar intensity can be just as effective as group-based counselling33 and often requires a smaller time investment from the participant as the sessions are focused on the individual rather than a larger group. This could be offered as an alternative for employees whose work obligations preclude them for joining a group-based program, with the possibility to participate during or after working hours.
Respondents were largely positive about the program, regardless of whether they had remained abstinent at the time of the interview. It is clear that for some, however, further support is needed, be that in the form of a longer program for support or the opportunity to complete the program again. The program was offered to employees as a one-off, whereas Chaiton et al.34 estimate that it can take between 6 and 30 serious quit attempts before quitting successfully (for at least one year) and so smoking employees would benefit from being offered the chance to participate in a smoking cessation program more than once. Moreover, communicating this may remove some of the shame in discussing quitting with colleagues.
In order to enable employees to participate more than once, the program would need to be maintained. For the maintenance of the program, a key step would be to entrench smoking cessation support into a greater (existing) health or vitality program offered by the employer. In this way, other barriers can be tackled such as routinising the process of recruitment and delivery of the program22. Moreover, establishing a company culture of health promotion through the creation of a comprehensive health or vitality program is a component of program success and sustainability35.
*More* generally within the workplace health promotion field, the attention thus far has been given to the employer’s needs when implementing a workplace health promotion program as program adopters, whilst to our knowledge, the needs of employees have been hardly studied. As such, we uncovered new factors that may hinder or facilitate reach and participation in a workplace health program, such as feelings of shame preventing discussion of the program with colleagues or the provision of promotional materials in other languages. Whilst our study focused on a smoking cessation workplace program, many of the findings may be applicable to other types of health promotion workplace programs in which group training sessions are given. In particular, similar findings regarding time constraints have been reported in other workplace health promotion programs28-31. However, the field overall, could benefit from more employee needs assessments being conducted with regard to other health behaviors.
Our study has highlighted some important barriers and facilitators to the recruitment and participation of employees in a workplace group-based smoking cessation program. These factors should be considered and addressed in order to ensure optimal program outcomes. To address some of the main barriers, we present the following recommendations: 1) use pro-active communication strategies such as word-of-mouth to inform about and promote the program, 2) train managers to discuss with and stimulate smoking employees to participate, 3) explain to all employees why the program is being offered and the benefits it can have for all staff, 4) make the program as accessible as possible by reimbursing time spent and offering the program at the workplace or nearby, and 5) integrate the smoking cessation program into wider company vitality policy.
## Strengths and limitations
Strengths of the study include the inclusion of employees from different sectors and occupations, increasing the validity of our results for different workplaces. Secondly, by interviewing employees with and without previous experience with a workplace smoking cessation program, we were able to see whether employees who would be hypothetically recruited for the first time would share the same concerns.
This study is not without methodological limitations. Firstly, none of the employees was from workplaces where a cessation program had been held for more than once; and so we are not able to know which factors might become relevant if the program was run repeatedly. Issues related to reaching and recruiting employees, in particular, may differ as program recruitment becomes more routinised. Whilst our sample included employees from a range of occupational roles and industries, only two participants reported having a low level of education and so certain factors that may be more relevant to this group could have been missed or underreported. Lastly, we recognize that the results of the present study may not necessarily be applicable to workplaces with younger employees (aged <30 years) or smaller workplaces (<100 employees) due to the characteristics of our sample.
## CONCLUSIONS
Workplace programs for smoking cessation are effective in stimulating cessation but implementers often experienced low participation rates. Our study presents recommendations to improve the recruitment and participation of employees in a workplace smoking cessation program, such as using active communication strategies and making the program as accessible as possible by reimbursing time spent and offering the program at the workplace or nearby. Integrating the smoking cessation program into wider company vitality policy will also aid continued provision of the program.
## CONFLICTS OF INTEREST
The authors have completed and submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest and none was reported. All authors report receiving funding from the Netherlands Organisation for Health Research and Development ZonMw (grant number #531003019). N.L. Poole also reports that she is contracted as a researcher for 190 hours by SineFuma B.V. and this contract was not for work carried out in relation to the current manuscript. F. A. van den Brand also reports being a board member of Society for Research on Nicotine and Tobacco Europe (SRNT-E).
## FUNDING
This work was supported by the Netherlands Organization for Health Research and Development ZonMw (No. 531003019).
## ETHICAL APPROVAL AND INFORMED CONSENT
The Central Committee on Research Involving Human Subjects in the Netherlands requires no ethical approval for non-medical research. Participants provided informed consent.
## DATA AVAILABILITY
The data supporting this research are available from the authors on reasonable request.
## AUTHORS’ CONTRIBUTIONS
Conceptualization: GN, FvdB and OvS. Data collection: CB. Formal analysis: CB, LdH-B, TM, FvdB and NP. Writing of original draft: NP. Reviewing of manuscript: FvdB, GN, OvS, LdH-B and TM. Funding: GN and OvS. All authors approved the final manuscript.
## PROVENANCE AND PEER REVIEW
Not commissioned; externally peer reviewed.
## References
1. Fishwick D, Carroll C, McGregor M. **Smoking cessation in the workplace**. *Occup Med (Lond)* (2013.0) **63** 526-536. DOI: 10.1093/occmed/kqt107
2. Osilla KC, Van Busum K, Schnyer C, Larkin JW, Eibner C, Mattke S. **Systematic Review of the Impact of Worksite Wellness Programs**. *Am J Manag Care* (2012.0) **18** e68-e81. PMID: 22435887
3. Cahill K, Lancaster T. **Workplace interventions for smoking cessation**. *Cochrane Database Syst Rev* (2014.0) CD003440. DOI: 10.1002/14651858.CD003440.pub4
4. Asfar T, Arheart KL, McClure LA. **Implementing a Novel Workplace Smoking Cessation Intervention Targeting Hispanic/Latino Construction Workers: A Pilot Cluster Randomized Trial**. *Health Educ Behav* (2020.0) **48** 795-804. DOI: 10.1177/1090198120960395
5. Bold KW, Kimmel L, Hanrahan TH. **A pilot clinical trial of smoking cessation services implemented in the workplace for service industry employees**. *Am J Health Promot* (2019.0) **33** 516-524. DOI: 10.1177/0890117118795987
6. van den Brand FA, Nagelhout GE, Winkens B, Chavannes NH, van Schayck OCP. **Effect of a workplace-based group training programme combined with financial incentives on smoking cessation: a cluster-randomised controlled trial**. *The Lancet Public Health* (2018.0) **3** e536-e544. DOI: 10.1016/S2468-2667(18)30185-3
7. Hiscock R, Bauld L, Amos A, Fidler JA, Munafò M. **Socioeconomic status and smoking: a review**. *Ann N Y Acad Sci* (2012.0) **1248** 107-123. DOI: 10.1111/j.1749-6632.2011.06202.x
8. Berman M, Crane R, Seiber E, Munur M. **Estimating the cost of a smoking employee**. *Tob Control* (2014.0) **23** 428-433. DOI: 10.1136/tobaccocontrol-2012-050888
9. Astrella JA. **Return on Investment: Evaluating the Evidence Regarding Financial Outcomes of Workplace Wellness Programs**. *J Nurs Adm* (2017.0) **47** 379-383. DOI: 10.1097/NNA.0000000000000499
10. Feenstra TL, Hamberg-van Reenen HH, Hoogenveen RT, Rutten-van Mölken MPMH. **Cost-Effectiveness of Face-to-Face Smoking Cessation Interventions: A Dynamic Modeling Study**. *Value Health* (2005.0) **8** 178-190. DOI: 10.1111/j.1524-4733.2005.04008.x
11. Rasmussen SR, Prescott E, Sørensen TIA, Søgaard J. **The total lifetime health cost savings of smoking cessation to society**. *Eur J Public Health* (2005.0) **15** 601-606. DOI: 10.1093/eurpub/cki024
12. Robroek SJW, van Lenthe FJ, van Empelen P, Burdorf A. **Determinants of participation in worksite health promotion programmes: a systematic review**. *Int J Behav Nutr Phys Act* (2009.0) **6** 26. DOI: 10.1186/1479-5868-6-26
13. Sargent GM, Banwell C, Strazdins L, Dixon J. **Time and participation in workplace health promotion: Australian qualitative study**. *Health Promot Int* (2016.0) **33** 436-447. DOI: 10.1093/heapro/daw078
14. Nagelhout GE, Willemsen MC, de Vries H. **The population impact of smoke-free workplace and hospitality industry legislation on smoking behaviour. Findings from a national population survey**. *Addiction* (2011.0) **106** 816-823. DOI: 10.1111/j.1360-0443.2010.03247.x
15. 15Maatregelen overheid om roken te ontmoedigenRijksoverheidAccessed March 21, 2023. https://www.rijksoverheid.nl/onderwerpen/roken/roken-ontmoedigen. *Rijksoverheid*
16. 16Ministerie van Volksgezondheid, Welzijn en SportNationaal Preventieakkoord: Naar een gezonder Nederland2018Ministerie van Volksgezondheid, Welzijn en SportAccessed November 30, 2022. https://www.rijksoverheid.nl/documenten/convenanten/2018/11/23/nationaal-preventieakkoord. *Nationaal Preventieakkoord: Naar een gezonder Nederland* (2018.0)
17. Bommelé J, Willemsen M. *KERNCIJFERS ROKEN 2020: DE LAATSTE CIJFERS OVER ROKEN, STOPPEN MET ROKEN EN HET GEBRUIK VAN ELEKTRONISCHE SIGARETTEN* (2021.0)
18. Glasgow RE, Vogt TM, Boles SM. **Evaluating the Public Health Impact of Health Promotion Interventions: The RE-AIM Framework**. *Am J Public Health* (1999.0) **89** 1322-1327. DOI: 10.2105/ajph.89.9.1322
19. Caponnetto P, Maglia M, Floresta D. **A randomized controlled trial to compare group motivational interviewing to very brief advice for the effectiveness of a workplace smoking cessation counseling intervention**. *J Addict Dis* (2020.0) **38** 465-474. DOI: 10.1080/10550887.2020.1782564
20. Ritchie J, Spencer L, Bryman A, Burgess B. **Qualitative data analysis for applied policy research**. *Analyzing Qualitative Data* (1994.0) 173-194
21. Graebner ME, Martin JA, Roundy PT. **Qualitative data: Cooking without a recipe**. *Strategic Organization* (2012.0) **10** 276-284. DOI: 10.1177/1476127012452821
22. van den Brand FA, Magnée T, de Haan-Bouma L. **Implementation of Financial Incentives for Successful Smoking Cessation in Real-Life Company Settings: A Qualitative Needs Assessment among Employers**. *Int J Environ Res Public Health* (2019.0) **16** 5135. DOI: 10.3390/ijerph16245135
23. Bricca A, Swithenbank Z, Scott N. **Predictors of recruitment and retention in randomized controlled trials of behavioural smoking cessation interventions: a systematic review and meta-regression analysis**. *Addiction* (2021.0) **117** 299-311. DOI: 10.1111/add.15614
24. Marcano Belisario JS, Bruggeling MN, Gunn LH, Brusamento S, Car J. **Interventions for recruiting smokers into cessation programmes**. *Cochrane Database Syst Rev* (2012.0) **12** CD009187. DOI: 10.1002/14651858.CD009187.pub2
25. Paul CL, Ross S, Bryant J, Hill W, Bonevski B, Keevy N. **The social context of smoking: A qualitative study comparing smokers of high versus low socioeconomic position**. *BMC Public Health* (2010.0) **10** 211. DOI: 10.1186/1471-2458-10-211
26. Bonevski B, Randell M, Paul C. **Reaching the hard-to-reach: a systematic review of strategies for improving health and medical research with socially disadvantaged groups**. *BMC Med Res Methodol* (2014.0) **14** 42. DOI: 10.1186/1471-2288-14-42
27. Kim A, Kamyab K, Zhu J, Volpp K. **Why are Financial Incentives not Effective at Influencing Some Smokers to Quit? Results of a Process Evaluation of a Worksite Trial Assessing the Efficacy of Financial Incentives for Smoking Cessation**. *J Occup Environ Med* (2011.0) **53** 62-67. DOI: 10.1097/JOM.0b013e31820061d7
28. Gilson N, McKenna J, Cooke C. **Experiences of Route and Task-Based Walking in a University Community: Qualitative Perspectives in a Randomized Control Trial**. *J Phys Act Health* (2008.0) **5** S176-S182. DOI: 10.1123/jpah.5.s1.s176
29. Estabrook B, Zapka J, Lemon SC. **Evaluating the Implementation of a Hospital Work-Site Obesity Prevention Intervention: Applying the RE-AIM Framework**. *Health Promot Pract* (2012.0) **13** 190-197. DOI: 10.1177/1524839910385897
30. Strijk JE, Proper KI, van der Beek AJ, van Mechelen W. **A process evaluation of a worksite vitality intervention among ageing hospital workers**. *Int J Behav Nutr Phys Act* (2011.0) **8** 58. DOI: 10.1186/1479-5868-8-58
31. Rojatz D, Merchant A, Nitsch M. **Factors influencing workplace health promotion intervention: a qualitative systematic review**. *Health Promot Int* (2017.0) **32** 831-839. DOI: 10.1093/heapro/daw015
32. Cheng T, Mauno S, Lee C. **The buffering effect of coping strategies in the relationship between job insecurity and employee well-being**. *Economic and Industrial Democracy* (2012.0) **35** 71-94. DOI: 10.1177/0143831X12463170
33. Stead LF, Carroll AJ, Lancaster T. **Group behaviour therapy programmes for smoking cessation**. *Cochrane Database Syst Rev* (2017.0) CD001007. DOI: 10.1002/14651858.CD001007.pub3
34. Chaiton M, Diemert L, Cohen JE. **Estimating the number of quit attempts it takes to quit smoking successfully in a longitudinal cohort of smokers**. *BMJ Open* (2016.0) **6** e011045. DOI: 10.1136/bmjopen-2016-011045
35. Goetzel RZ, Henke RM, Tabrizi M. **Do Workplace Health Promotion (Wellness) Programs Work?**. *J Occup Environ Med* (2014.0) **56** 927-934. DOI: 10.1097/JOM.0000000000000276
|
---
title: Single nucleotide variants in microRNA biosynthesis genes in Mexican individuals
authors:
- Jesús Juárez-Luis
- Moisés Canseco-Ocaña
- Miguel Angel Cid-Soto
- Xochitl H. Castro-Martínez
- Angélica Martínez-Hernández
- Lorena Orozco
- Araceli Hernández-Zavala
- Emilio J. Córdova
journal: Frontiers in Genetics
year: 2023
pmcid: PMC10037310
doi: 10.3389/fgene.2023.1022912
license: CC BY 4.0
---
# Single nucleotide variants in microRNA biosynthesis genes in Mexican individuals
## Abstract
Background: MicroRNAs (miRNAs) are important regulators in a variety of biological processes, and their dysregulation is associated with multiple human diseases. Single nucleotide variants (SNVs) in genes involved in the processing of microRNAs may alter miRNA regulation and could present high allele heterogeneity in populations from different ethnic groups. Thus, the aim of this study was to genotype 15 SNVs in eight genes involved in the miRNA processing pathway in Mexican individuals and compare their frequencies across 21 populations from five continental groups.
Methods: Genomic DNA was obtained from 399 healthy Mexican individuals. SNVs in AGO2 (rs2293939 and rs4961280), DGCR8 (rs720012), DICER (rs3742330 and rs13078), DROSHA (rs10719 and rs6877842), GEMIN3 (rs197388 and rs197414), GEMIN4 (rs7813, rs2740349, and rs4968104), TNRC6B (rs9611280), and XP05 (rs11077 and rs34324334) were genotyped using TaqMan probes. The minor allele frequency of each SNV was compared to those reported in the 1,000 Genomes database using chi-squared. Sankey plot was created in the SankeyMATIC package to visualize the frequency range of each variant in the different countries analyzed.
Results: In Mexican individuals, all 15 SNVs were found in Hardy-Weinberg equilibrium, with frequencies ranging from 0.04 to 0.45. The SNVs rs4961280, rs2740349, rs34324334, and rs720012 in Mexican individuals had the highest minor allele frequencies worldwide, whereas the minor allele frequencies of rs197388, rs10719, rs197414, and rs1107 were among the lowest in Mexican individuals. The variants had high allele heterogeneity among the sub-continental populations, ranging from monomorphic, as was the case for rs9611280 and rs34324334 in African groups, to >0.50, which was the case for variants rs11077 and rs10719 in most of the populations. Importantly, the variants rs197388, rs720012, and rs197414 had FST values > 0.18, indicating a directional selective process. Finally, the SNVs rs13078 and rs10719 significantly correlated with both latitude and longitude.
Conclusion: *These data* indicate the presence of high allelic heterogeneity in the worldwide distribution of the frequency of SNVs located in components of the miRNA processing pathway, which could modify the genetic susceptibility associated with human diseases in populations with different ancestry.
## Introduction
MicroRNAs (miRNAs) are small non-coding RNAs approximately 22 nucleotides in length that have important functions in the post-transcriptional regulation of gene expression (López-Jiménez and Andrés-León, 2021). Most miRNAs bind to the 3′UTR region of their target messenger RNA, promoting either its degradation or the repression of its translation (Annese et al., 2020). Novel functions of miRNAs in promoting transcription and enhancing translation through their binding to the 5′UTR or promoter regions have also been demonstrated recently (Younger et al., 2009). MiRNA encoding genes account for ∼$3\%$ of the human genome and at least $30\%$ of the protein-coding genes have been estimated to be regulated by miRNAs (Ni and Leng, 2015; Stavast and Erkeland, 2019).
MiRNAs are important regulators of a vast number of cellular processes, including proliferation, differentiation, intracellular signaling, and metabolism. Accordingly, alterations in the regulation of miRNAs are common in a variety of human diseases. For example, aberrant expression of miRNAs has been observed in different types of cancer, including lung, gastric, breast, and hepatocellular carcinoma (Volinia et al., 2006; Roth et al., 2011; Morishita et al., 2016; Kandettu et al., 2020). Similarly, alterations in the expression of miRNAs have been found in diabetes, cardiovascular diseases, and neuro-degenerative illnesses (Margis et al., 2011; Hashimoto and Tanaka, 2017; Yan et al., 2019).
The miRNA biogenesis pathway is a strictly regulated process composed of several enzymatic steps, starting with their transcription by RNA polymerase II as a long (∼1 Kb) primary hairpin structure, primary miRNA (pri-miRNA). The pri-miRNAs are processed in the nucleus by a protein complex composes of the ribonuclease (RNAse) III DROSHA and DiGeorge Syndrome Critical Region 8 protein (DGCR8) into a precursor miRNA (pre-miRNAs) of ∼65–70 pb. Next, the pre-miRNAs are exported to the cytoplasm through the Exportin-5 (XPO5) RAN-GTP complex, where they are cleaved by DICER, a second RNAse III protein, in complex with the trans-activating response RNA binding protein (TRBP). This processing step produces an ∼22 bp mature duplex composed of a guide miRNA strand and passenger *miRNA. The guide strand is selected and loaded into the RNA-induced silencing complex (RISC) by the Argonaute proteins (AGO1-4), whereas the *miRNA is usually degraded. The proteins GEMIN3 and GEMIN4 complete the formation and activity of the RISC (Saliminejad et al., 2019).
Recent studies have highlighted an association between single nucleotide variants (SNVs) in the miRNA processing genes and different human diseases. For example, SNVs rs197414 in GEMIN3, rs3742330 in DICER1, rs7813 in GEMIN4, and rs11077 in XP05 have been associated with increased susceptibility to bladder, colorectal, gastric, and thyroid carcinoma, respectively (Yang et al., 2008; Xie et al., 2015; Zhao et al., 2015; Wen et al., 2017). SNVs in genes involved in the miRNA biogenesis pathway have also been associated with non-malignant diseases. For example, the SNVs rs13078 in DICER, rs10719 in DROSHA, and rs720012 in DGCR8 have been associated with type 2 diabetes, primary hypertension, and late onset pre-eclampsia (Zhang et al., 2017; Huang et al., 2019a; Wen et al., 2019), respectively.
Previous studies have demonstrated that the frequency of the SNVs and strength of their association with human diseases are strongly dependent on ethnicity (Mallick et al., 2016; Gurdasani et al., 2019; Gnagnarella et al., 2021). In this sense, the minor allele frequency (MAF) of SNVs rs2237897 and rs2237892 in KCNQ1 is significantly higher in East Asian populations (0.39 and 0.38, respectively) compared to European-derived groups (0.04 and 0.06, respectively), and these SNVs have been associated with type 2 diabetes mellitus in Japanese individuals but not in European subjects (Unoki et al., 2008; Yasuda et al., 2008; Rosenberg et al., 2010). Similarly, 11 variants in the HBA$\frac{1}{2}$ locus were specifically associated with red blood cell traits in individuals with African or Amerindian ancestries, but not in European-derived populations (Hodonsky et al., 2020). Thus, genetic associations previously observed in populations with European, Asian, or African ancestry may be different in populations with high admixture levels.
The current population in *Mexico is* mainly composed of a recent admixture of original Amerindian ($56\%$), European ($41\%$) and, to a lesser extent, African individuals ($3\%$) (Moreno-Estrada et al., 2014). The complex admixture present in the Mexican population may substantially affect the frequency of variants occurring in genes involved in the miRNA biosynthesis pathway. Therefore, the present study aimed to determine the frequency of 15 variants in genes from the miRNA machinery pathway. The frequencies of these variants in Mexican individuals were also compared to those reported by the 1,000 Genome project for 21 different populations across the world.
## Study population
The sample population included 399 non-related healthy volunteers, 122 men ($30.6\%$) and 277 women ($69.4\%$), with a mean age of 43 ± 9.1 years. All participants were Mexican individuals with parents and grandparents born in Mexico and were recruited from four different geographical regions in the country: north ($$n = 100$$), central east ($$n = 100$$), south ($$n = 100$$) and south east ($$n = 99$$). Geographical regions were previously described elsewhere (Moreno-Estrada et al., 2014). Each participant signed a letter of informed consent. This study was carried out according to the Declaration of Helsinki and approved by the ethics and research committees of the National Institute of Genomic Medicine at Mexico City. In addition, we included genotype data from 21 different subpopulations belonging to five continental populations: African [Esan in Nigeria (ESN); Gambian in Western Division, Gambia (GWD); Luhya in Webuye, Kenya (LWK); Mende in Sierra Leone (MSL); and Yoruba in Ibadan, Nigeria (YRI)], Admixed Latino American [African Caribbean in Barbados (ACB); Colombian in Medellin, Colombia (CLM); Mexican ancestry in Los Angeles, California (MXL); Peruvian in Lima, Peru (PEL); and Puerto Rican in Puerto Rico (PUR)], East Asian [Chinese Dai in Xishuangbanna, China (CDX); Han Chinese in Beijing, China (CHB); Japanese in Tokyo, Japan (JPT); and Kinh in Ho Chi Minh City, Vietnam (KHV)], European [Utah residents with Northern and Western European ancestry (CEU); Finnish in Finland (FIN); British in England and Scotland (GBR); Iberian populations in Spain (IBS); and Toscani in Italy (TSI)], South Asian [Bengali in Bangladesh (BEB) and Punjabi in Lahore, Pakistan (PJL)] (Supplementary Table S1). Genotype data from all 21 subpopulations were incorporated from the International Genome Sample Resource (IGSR) (Clarke et al., 2017).
## Selection of gene variants
Fifteen SNVs in eight genes involved in the miRNA biosynthesis pathway (AGO2: rs2293939, rs4961280; DGCR8: rs720012; DICER: rs3742330, rs13078; DROSHA: rs10719, rs6877842; GEMIN3: rs197388, rs197414; GEMIN4: rs7813, rs2740349, rs4968104; TNRC6B: rs9611280; XP05: rs11077, rs34324334; Table 1) were selected from a literature search conducted in the electronic database PubMed. All variants were selected based on previous association with human diseases. The MAF of these variants was >$1\%$ in the global population according to the 1,000 Genome project database (Clarke et al., 2017).
**TABLE 1**
| Gene | Chr:Position (GRCh38) | SNV ID | Annotation | Alleles | MAF |
| --- | --- | --- | --- | --- | --- |
| AGO2 | 8:140541308 | rs2293939 | Synonymous | G/A | 0.22 |
| AGO2 | 8:140637315 | rs4961280 | Promoter | C/A | 0.37 |
| DGCR8 | 22:20111059 | rs720012 | 3′UTR | G/A | 0.45 |
| DICER1 | 14:95087025 | rs3742330 | 3′UTR | A/G | 0.18 |
| DICER1 | 14:95090410 | rs13078 | 3′UTR | T/A | 0.1 |
| DROSHA | 5:31401340 | rs10719 | 3′UTR | G/A | 0.35 |
| DROSHA | 5:31532531 | rs6877842 | 5′UTR | G/C | 0.12 |
| GEMIN3 | 1:111754860 | rs197388 | Promoter | A/T | 0.07 |
| GEMIN3 | 1:111766501 | rs197414 | Missense | C/A | 0.05 |
| GEMIN4 | 17:744946 | rs7813 | Missense | A/G | 0.28 |
| GEMIN4 | 17:745258 | rs2740349 | Missense | T/C | 0.19 |
| GEMIN4 | 17:746265 | rs4968104 | Missense | T/A | 0.09 |
| TNRC6B | 22:40156115 | rs9611280 | Missense | G/A | 0.04 |
| XPO5 | 6:43523209 | rs11077 | 3′UTR | T/G | 0.37 |
| XPO5 | 6:43567281 | rs34324334 | Missense | C/T | 0.07 |
## Sample genotyping
Genomic DNA was isolated from 10 ml of whole blood samples using the QIAamp DNA Blood Maxi kit (Qiagen, Valencia CA, United States) following the manufacturer’s protocol. Select SNVs were genotyped using TaqMan exonuclease assays on a QuantStudio 7 Flex Real-Time PCR (Applied Biosystems, Foster City, CA, United States). The genotyping call rate exceeded $95\%$ for all SNVs. Genotype validation was performed by directly sequencing a random number of samples. One hundred percent concordance was found.
## Population differentiation analysis
We measured the level of population differentiation with the Wright’s fixation index (FST) for each pair and for all populations, including our sample population of Mexican Mestizos, using Arlequin Software version 3.5.2.2 (Excoffier and Lischer, 2010). FST plots were created using the Lattice package in statistical environment R.
We estimated the FST value for all populations and all variants included in this work (Supplementary Figure S1). The highest FST values (FST = 0.23) were found for rs197388, followed by rs720012 (FST = 0.21) and rs197414 (FST = 0.19), whereas variants rs11077 (FST = 0.15), rs10719 (FST = 0.14), and rs3742330 (FST = 0.11) had intermediate global FST values among the studied populations. In contrast, rs4961280 (FST = 0.09), rs2293939 (FST = 0.07), rs7813 (FST = 0.06), and rs4968104 (FST = 0.05) had low global FST values among the studied population, and very low values were presented by variants rs2740349 (FST = 0.04), rs9611280 (FST = 0.03), rs6877842 (FST = 0.03), rs34324334 (FST = 0.02), and rs13078 (FST = 0.02).
Based on the FST analysis of the 15 variants, the African groups had the highest population differentiation compared to the other continental populations, such as the East Asian groups (FST = 0.26–0.35), South Asian group BEB (FST = 0.20–0.21), European groups CEU, FIN, and IBS (FST = 0.18–0.20), and all of the admixed Latino American groups (FST = 0.18–0.25), with exception of ACB and PUR, which have high African ancestry (Figure 3). The ACB population also exhibited important genetic differentiation with the East Asian groups (FST = 0.26–0.29). No significant differentiation was observed between the other Latino American groups and any other populations, whereas European groups exhibited slight genetic differentiation with East Asian populations (FST = 0.14–0.18).
**FIGURE 3:** *Genetic differentiation based on SNVs in components of the miRNA biosynthesis pathway. Pairwise F
ST
analysis of all variants in 21 populations with different ancestry from the 1,000 Genomes database and MEZ population. The darkest blue indicates the lowest levels of differentiation, whereas red indicates the highest levels of differentiation. Abbreviations are the same from Figure 1.*
## Correlation of gene variants with geographical characteristics
To analyze the geographical distribution of the gene variants, the correlation coefficient was estimated between the MAFs of all the variants and the latitude and longitude coordinates from each included population. Correlation was evaluated by the Pearson’s test using R version 3.4.4 statistical software. p ≤ 0.05 was considered significant.
## Statistical analysis
Genotyping data were reported as frequencies. The Hardy-*Weinberg equilibrium* was evaluated using χ2 (Genepop version 4.7; 31). The MAF of each SNV was compared to those reported in the 1,000 Genomes database (Clarke et al., 2017) using χ2 (Genepop version 4.7; 31). Significant differences were established at p ≤ 0.05. Sankey plot was created using the package SankeyMATIC in R software and show the relationship between the frequency of each variant and the number of countries that present the same frequency range.
## Allele frequencies of SNVs in the miRNA biogenesis pathway in Mexican individuals
Based on previous reports in the literature, 15 SNVs in eight genes involved in the biosynthesis of miRNAs (AGO2: rs2293939 and rs4961280; DGCR8: rs720012; DICER1: rs3742330 and rs13078; DROSHA: rs10719 and rs6877842; GEMIN3: rs197388 and rs197414; GEMIN4: rs7813, rs2740349, and rs4968104; TNRC6B: rs9611280; and XPO5: rs11077 and rs34324334) were selected for genotyping in a sample of Mexican individuals. Of these 15 SNVs, 7 were located in coding regions (6 missenses, 1 synonymous), 5 in 3′UTRs, 2 in promoters, and 1 in 5′UTRs (Table 1).
After genotyping the DNA samples from 399 healthy individuals, the analyzed SNVs showed no significant deviation from Hardy-*Weinberg equilibrium* ($p \leq 0.05$). The minor allele of SNV rs9611280 in TNRC6B showed the lowest frequency (0.04), whereas variant rs720012 in DGCR8 was the most frequent (0.45). The frequency of variants rs197388 and rs197414 in GEMIN3, rs34324334 in XPO5, rs4968104 in GEMIN4, and rs13078 in DICER1 ranged from 0.05 to 0.10, whereas rs6877842 in DROSHA, rs3742330 in DICER1, and rs2740349 in GEMIN4 presented frequencies between 0.12 and 0.19. Finally, variants rs2293939 and rs4961280 in AGO2, rs7813 in GEMIN4, rs10719 in DROSHA, and rs11077 in XPO5 presented frequencies ranging from 0.22 to 0.37 (Table 1; Supplementary Table S2). We also performed analysis of haplotypes for all variants located in the same gene (rs2293939 and rs4961280 in AGO2, rs3742330 and rs13078 in DICER1, rs10719 and rs687782 in DROSHA, rs197388 and rs197414 in GEMIN3, rs7813, rs2740349, and rs4968104 in GEMIN4 as well as rs11077 and rs343243343 in XP05). However, the linkage disequilibrium observed for each analyzed pairwise variants was below the threshold proposed by Gabriel et al. [ 2002] to consider a pair of SNVs as a haplotype, indicating that none of these variants form part of the same haplotype block.
## Allele frequencies of SNVs in Mexican individuals and other ethnic groups
After comparing our findings to those reported in the 1,000 Genomes project for 21 populations with different ethnic ancestry (Clarke et al., 2017), the MAF of SNVs rs4961280, rs2740349, rs34324334, and rs720012 in Mexican individuals was among the highest worldwide; the rs4961280 showed highest MAF across all studied populations (Figure 1; Supplementary Table S3). Similarly, the variant frequency of rs3742330 was significantly higher in Mexican individuals than in all other ethnic groups except East Asian populations. In contrast, the MAF of rs197388, rs10719, rs197414, and rs1107 in Mexican individuals was among the lowest frequencies across all analyzed populations. In addition, the frequency of variant rs6877842 in our sample was only significantly higher than the frequency in South Asian populations, whereas the frequencies of rs7813, rs4968104, rs9611280, and rs2293939 variants in Mexican individuals was intermediate between African and European populations. Finally, rs13078 showed a significant difference only with European populations (Figure 1; Supplementary Table S3). Regarding other admixed Latino American populations, the MAFs in Mexican individuals showed the greatest differences with respect to PUR and CLM.
**FIGURE 1:** *Ethnic groups with significant differences in the minor allele frequencies of SNVs with respect to Mexican individuals. The minor allele frequencies of the 15 SNVs found in the Mexican mestizo population (MEZ) in this study were compared against those reported in 21 sub-continental populations from the 1,000 Genomes database. The color gradient from light blue to dark blue indicates increasingly lower frequencies compared to MEZ (–log p-value range = 0–6). The color gradient from yellow to red indicates increasingly higher frequencies compared to MEZ (–log p-value range = 0–6). Abbreviations are ACB: African Caribbean in Barbados; BEB: Bengali in Bangladesh; CDX Chinese Dai in Xishuangbanna, China; CEU: Utah residents with Northern and Western European ancestry; CHB: Han Chinese in Beijing, China; CLM: Colombian in Medellin, Colombia; ESN: Esan in Nigeria; FIN: Finnish in Finland; GBR: British in England and Scotland; GWD: Gambian in Western Division, Gambia; IBS: Iberian populations in Spain; JPT: Japanese in Tokyo, Japan; KHV: Kinh in Ho Chi Minh City, Vietnam; LWK: Luhya in Webuye, Kenya; MSL: Mende in Sierra Leone; MXL: Mexican ancestry in Los Angeles, California; PEL: Peruvian in Lima, Peru; PJL: Punjabi in Lahore, Pakistan; PUR: Puerto Rican in Puerto Rico; TSI: Toscani in Italy and YRI: Yoruba in Ibadan, Nigeria.*
## Worldwide distribution of minor alleles in SNVs in the miRNA biosynthesis machinery
After analyzing the frequency distribution of the SNVs, variants rs9611280 and rs34324334 were monomorphic in 9 and 7 of the 22 populations reported in the 1,000 Genomes project, respectively, with MAFs ranging from 0.01 to 0.13 and 0.01 to 0.11 in the rest of the populations (Figure 2). In contrast, the frequency of rs11077 and rs10719 variants was extremely high in most of the ethnic groups, >0.50 in 7 and 10 of the 22 populations, respectively. The rest of the SNVs had a wide range of allele frequency distribution. For example, the frequency of the variant allele for rs13078, rs2740349, rs6877842, and rs4968104 ranged from 0.03–0.20, monomorphic-0.22, 0.01–0.25, and 0.03–0.31, respectively, whereas rs4961280, rs3742330, and rs197414, which were monomorphic in at least 1 out of the 22 populations, had MAFs as high as 0.37, 0.40, and 0.49, respectively. Similarly, the variant allele frequencies for rs2293939, rs7813, and rs197388 were as low as 0.02, 0.08, and 0.02, and as high as 0.44, 0.48, and 0.60, respectively. Importantly, variant rs720012 had the broadest distribution in its frequency worldwide, ranging from monomorphic to 0.58 (Figure 2; Supplementary Table S3).
**FIGURE 2:** *Variation in the frequency of the minor alleles in SNVs from components of the miRNA processing pathway in different ethnic groups. Sankey diagram showing the studied SNVs with connectors in different colors oriented towards the respective range of frequencies presented in 21 sub-continental populations from the 1,000 Genome project and the MEZ population from this study. Boxes in color below each SNV correspond to the gene in which each variant is located.*
## Correlation of gene variant frequencies with longitude and latitude
Geographical factors, such as latitude and longitude, may modify the gene variant frequency. In our analysis, the SNVs rs13078 in DICER1 and rs10719 in DROSHA significantly correlated with both latitude and longitude, with significant increases in the MAF from south to north ($r = 0.612$, $$p \leq 0.002$$) and significant decreases from west to east (r = −0.458, $$p \leq 0.032$$) in the case of rs13078, whereas the rs10719 variant significantly decreased from south to north (r = −0.490, $$p \leq 0.021$$) and significantly increased from west to east ($r = 0.510$, $$p \leq 0.015$$; Figure 4). The three SNVs in GEMIN4, rs7813, rs2740349, and rs4968104, as well as rs9611280 in TNRC6B, showed significant increases in MAF from south to north ($r = 0.625$, $$p \leq 0.002$$; $r = 0.432$, $$p \leq 0.045$$; $r = 0.637$, $$p \leq 0.001$$; and $r = 0.647$, $$p \leq 0.001$$, respectively; Figure 5). In addition, the frequency of variant rs3742330 in DICER1 significantly increased from west to east ($r = 0.466$; $$p \leq 0.029$$), whereas the MAF of rs6877842 in DROSHA and the two variants in XP05, rs34324334 and rs11077, significantly decreased from west to east (r = −0.569, $$p \leq 0.006$$; r = −0.581, $$p \leq 0.005$$; r = −0.592, $$p \leq 0.004$$, respectively; Figure 6). SNVs rs2293939, rs4961280, rs197388, rs720012, and rs197414 did not significantly correlate with latitude or longitude.
**FIGURE 4:** *Significant correlation of the minor allele frequency of SNVs with latitude and longitude. The correlation of minor allele frequencies reported in sub-continental populations from the 1,000 Genome project and those found in this study for the MEZ population with longitude and latitude was evaluated by the Pearson’s correlation test. Gene variants with p < 0.05 for latitude and longitude correlation are shown in the figure. Black dots = African groups; red dots = Admixed Latino American groups; blue dots = European groups; green dots = East Asian groups; turquoise dots = South Asian groups.* **FIGURE 5:** *Significant correlation of the minor allele frequency of SNVs with latitude. The correlation of minor allele frequencies reported in sub-continental populations from the 1,000 Genome project and those found in this study for the MEZ population with longitude and latitude was evaluated by the Pearson’s correlation test. Gene variants with p < 0.05 for latitude correlation are shown in the figure. Black dots = African groups; red dots = Admixed Latino American groups; blue dots = European groups; green dots = East Asian groups; turquoise dots = South Asian groups.* **FIGURE 6:** *Significant correlation of the minor allele frequency of SNVs with longitude. The correlation of minor allele frequencies reported in sub-continental populations from the 1,000 Genome project and those found in this study for the MEZ population with longitude and latitude was evaluated by the Pearson’s correlation test. Gene variants with p < 0.05 for longitude correlation are shown in the figure. Black dots = African groups; red dots = Admixed Latino American groups; blue dots = European groups; green dots = East Asian groups; turquoise dots = South Asian groups.*
## Distribution of gene variants in miRNA processing genes in continental populations
In African populations, variants rs720012 and rs9611280 were monomorphic, whereas variants rs34324334, rs3742330, rs2740349, and rs4961280 had MAFs <0.05 (Figure 7). Variants rs9611280 and rs34324334 remained monomorphic in East Asian populations and had MAFs ranging from 0.04 to 0.09 and 0.06 to 0.08, respectively, in the rest of the continental populations (Figure 7). In contrast, variants rs720012, rs4961280, rs3742330, and rs2740349 were common in populations outside of Africa, ranging from 0.08 to 0.52 (Figure 7). The highest frequencies for the variants rs720012 and rs3742330 were observed in East Asian populations (0.52 and 0.35, respectively), whereas the frequency peaks for rs4961280 and rs2740349 were found in Latino American groups (0.25 and 0.16, respectively).
**FIGURE 7:** *Worldwide distribution of the minor allele frequency of SNVs in components of the miRNA biosynthesis machinery in different sub-continental populations. Abbreviations in the figure are the same from Figure 1. Boxes in color denote each variant in the graphics. Numbers at the top of the bars in the graphics correspond to the minor allele frequency.*
SNVs rs4968104, rs13078, rs7813, and rs2293939 were common variants in African groups, with MAFs ranging from 0.05 to 0.09, whereas variants rs6877842, rs197414, rs197388, rs11077, and rs10719 were very common in these ethnic groups, with MAFs ranging from 0.19 to 0.67 (Figure 7). The rs13078 variant had a similar frequency in most of the continental populations compared to African groups (AMR = 0.11; EUR = 0.18; SAS = 0.05; EAS = 0.04 vs. AFR = 0.07), whereas variants rs4968104, rs7813, and rs2293939 significantly increased in frequency in the other continental populations (ranging from 0.04 to 0.18, 0.28 to 0.41, and 0.24 to 0.36, respectively). Finally, from the highly common variants in African populations, rs11077 and rs10719 remained high in all other continental populations except East Asian groups for rs11077 (0.06). In contrast, the MAF of SNVs rs6877842, rs197414, and rs197388 significantly decreased in all other continental populations (0.02–0.17, 0.01–0.13, and 0.04–0.19, respectively), particularly in East Asian populations (0.02, 0.01, and 0.04, respectively; Figure 7). Notably, SNV rs10719 had a MAF >0.5 in African and East Asian populations, whereas the MAF of rs197388 and rs11077 was >0.5 only in African populations, and rs720012, which was monomorphic in all African populations, had a MAF >0.5 in East Asian groups.
## Discussion
MiRNAs have become one of the most important regulatory systems in different biological events. Accordingly, SNVs located in components of the miRNA processing pathway have been associated with several human diseases. For example, SNVs in DROSHA have been associated with gastric cancer and susceptibility to congenital heart disease (Song et al., 2017; Borghini et al., 2021), whereas gene variants in DICER1 occur at significantly higher allele frequencies in individuals with endometrial and hepatocellular carcinoma than in the healthy population (Wang et al., 2017; Oz et al., 2018). Based on the genetic diversification of human populations caused by migration, adaptation to local environment, and genetic drift, allele frequencies for many SNVs differ depending on the ethnicity of the population group (Lan et al., 2007; Myles et al., 2008; Gurdasani et al., 2019). In the same sense, the frequency of disease-associated alleles could also change among individuals with different ancestry.
In this study, we determined the frequency of 15 variants in eight genes from the miRNA processing pathway in Mexican individuals and compared our findings to those in 21 sub-continental populations from the 1000 Genomes project. All of the analyzed SNVs were common variants in our sample population, with rs9611280 in TNRC6B occurring at the lowest frequency and rs720012 in DGCR8 being the most frequent variant. The frequencies found in our population for some of the variant alleles, such as rs10719 in DROSHA, rs2293939 in AGO2, rs7813 in GEMIN4, and rs9611280 in TNRC6B, were intermediate to the frequency previously reported in African and European groups. As the modern population in *Mexico is* composed of a recent and complex admixture of ancient Native American, European (mainly from Spain), and sub-Saharan Africans, this was an expected finding (Chacón-Duque et al., 2018; Aguilar-Velázquez and Rangel-Villalobos, 2021).
However, other variants showed different patterns of distribution, including rs13078 in DICER1 and rs11077 and rs34324334 in XP05, which showed significant differences from either European or African populations but not both. Moreover, the minor allele of rs4961280 in AGO2 had the highest frequency worldwide in our population, whereas the MAFs of rs2740349 in GEMIN4 and rs720012 in DGCR8 in Mexican individuals were among the top five in all analyzed populations. In contrast, the MAFs of rs6877842 in DROSHA, rs197414 and rs197388 in GEMIN3, and rs4968104 in GEMIN4 were among the lowest worldwide. This could indicate the presence of geographic or climate factors modifying the frequency of the derived allele in these SNVs.
Examples of derived alleles enriched in human populations by adaptation to geography, climate conditions, and lifestyle include the lactase persistence allele in the Fula population from Western Eurasia (Schaschl et al., 2022), variants rs4766578 and rs847892 in ALDH2, which are associated in European individuals with resistance to consumption of high levels of alcohol (Hodgson et al., 2014), and the protective variant in the Duffy blood group gene, which provides resistance to malaria in sub-Saharan Africans (Pierron et al., 2018; Reynolds et al., 2019). In the case of admixed Latino American populations, variant alleles in IL1R1 and MUC1, important regulators of the adaptive immune response, occurred at significantly higher frequencies in indigenous individuals from the southeastern region of the United States and from the central region of Mexico, respectively, compared to individuals with European ancestry (Ávila-Arcos et al., 2020). In addition, variants in genes associated with lipid metabolism, such as APOA5, ABCG5, and ABCA1, have strong signals of positive selection in the *Mexican indigenous* population (Villarreal-Molina et al., 2008; Acuña-Alonzo et al., 2010; Lindo et al., 2018). Moreover, variant alleles in MGAM, a gene related to starch digestion, have also been found to be enriched in Sud-American individuals compared to African and European populations (Clark et al., 2003).
As expected, the 15 variants had a high range of distribution in the different sub-continental populations evaluated. For example, rs720012 was monomorphic in all of the African populations but had a MAF ranging from 0.44 to 0.58 in East Asian populations. Similarly, the variant alleles of rs4961280 and rs3742330, which were absent in at least one African group, had frequencies ranging from 0.24 to 0.37 in admixed Latino American groups and 0.31 to 0.40 in East Asian individuals. In contrast, the variant alleles of rs9611280 and rs34324334 were monomorphic or occurred at low frequency in African and East Asian populations, with a frequency up to 0.08 and 0.11, respectively, in the rest of the analyzed sub-continental populations. Notably, rs720012, rs10719, rs197388, and rs11077 had MAFs >0.5, mainly in East Asian and African populations, suggesting the presence of selection forces in the worldwide distribution of these variants.
The functional effect of a SNV depends on its location in the structure of the gene. In this sense, the SNV rs10719 located in the 3′UTR of DROSHA disrupts miR-27b binding site leading to an overexpression of DROSHA transcript (Wen et al., 2018). Likewise, the SNV rs11077, located in the 3′UTR of XP05, is associated with an alteration in the stability of the mRNA, suggesting the regulation of this gene by miRNAs (Ding et al., 2013). In other examples of functional effects, rs9611280 missense variant in TNRC6B gene has been shown to affect the splicing of the mRNA (Martin-Guerrero et al., 2015), whereas rs4961280, located in the promoter region of AGO2 was found to upregulate the expression of the gene in prostate cancer patients (Nikolić et al., 2017).
Based on pairwise FST analysis from the 15 analyzed variants, African populations had the highest levels of genetic differentiation with respect to all other sub-continental populations. Variants rs720012 in DGCR8 (FST = 0.21) and rs197388 (FST = 0.23) and rs197414 (FST = 0.19) in GEMIN3 exhibited signals of a directional selective process according to Clark et al. [ 2003], who proposed values of FST > 0.18 as being suggestive of a selective process in human populations. These data suggest the functional importance of these variants and genes in the miRNA processing pathway.
The SNV rs720012 in DGCR8 has been previously associated with non-muscle-invasive bladder cancer, tuberculosis susceptibility, and increased risk of pre-eclampsia (Ke et al., 2013; Cheng et al., 2018; Huang et al., 2019b). The rs197388 variant in GEMIN3 has been associated with primary open-angle glaucoma in the Polish population (Molasy et al., 2018), idiopathic azoospermia in a Turkish population (Ozlem et al., 2017), and increased risk of oropharyngeal squamous cell carcinoma (Chen et al., 2016). The rs197414 variant has been associated with increased risk of bladder and esophageal cancer (Ye et al., 2008; Yang et al., 2008). The SNV rs720012 is located in the 3′UTR region of DGCR8, whereas rs197388 is located in the promoter region of GEMIN3 and rs197414 is a missense variant. Although these variants have been associated with different diseases, no functional effects have been described previously.
Another important finding in our study was the significant correlation between the MAFs of rs13078 in DICER1 and rs10719 in DROSHA with the geographical coordinates of latitude and longitude. The rs13078 variant allele has been associated with a decreased risk of developing type 2 diabetes and an increased risk of gestational hypertension and laryngeal cancer (Osuch-Wojcikiewicz et al., 2015; Wen et al., 2019; Huang et al., 2019a). Germinal and somatic mutations in DICER1 have also been associated with a rare genetic cancer prone disease called pleuropulmonary blastoma familial tumor susceptibility syndrome, or DICER1 syndrome (González et al., 2022). Variant rs10719 has been associated with increased susceptibility to malignant diseases, such as colorectal cancer and gastric carcinoma (Cho et al., 2015; Li et al., 2017). This variant has also been found to be associated with metabolic diseases, including pre-eclampsia susceptibility, primary hypertension, and ischemic stroke (Kim et al., 2018; Zhang et al., 2017; Rezaei et al., 2018).
Taken together, our data suggest that the worldwide distribution of the frequency of SNVs located in components of the miRNA processing pathway has been shaped by different adaptive forces. As all of the variants analyzed in this study have been associated with genetic risk to human diseases, populations with different ancestry would present different susceptibility to specific illnesses. Our data also indicate the importance of studying admixed populations to fully understand the genetic architecture of complex human diseases.
## Data availability statement
The original contributions presented in the study are included in the article/Supplementary Materials, further inquiries can be directed to the corresponding author.
## Ethics statement
The studies involving human participants were reviewed and approved by Ethics and research committees of the National Institute of Genomic Medicine at Mexico City. The patients/participants provided their written informed consent to participate in this study.
## Author contributions
JJ-L and MC-O contributed to experimental analysis, acquisition of samples, and data. MAC-S performed the analysis of genetic data. XHC-M and AM-H contributed to sample collection and experimental design. JJ-L, LO, AH-Z, and EC participated in conceptualization, writing, and funding acquisition.
## Conflict of interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
## Publisher’s note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
## Supplementary material
The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fgene.2023.1022912/full#supplementary-material.
## References
1. Acuña-Alonzo V., Flores-Dorantes T., Kruit J. K., Villarreal-Molina T., Arellano-Campos O., Hünemeier T.. **A functional ABCA1 gene variant is associated with low HDL-cholesterol levels and shows evidence of positive selection in native Americans**. *Hum. Mol. Genet.* (2010) **19** 2877-2885. DOI: 10.1093/hmg/ddq173
2. Aguilar-Velázquez J. A., Rangel-Villalobos H.. **Distribution of paternal lineages in Mestizo populations throughout Mexico: an in silico study based on Y-STR haplotypes**. *J. Hum. Genet.* (2021) **66** 327-332. DOI: 10.1038/s10038-020-00824-7
3. Annese T., Tamma R., De Giorgis M., Ribatti D.. **microRNAs biogenesis, functions and role in tumor angiogenesis**. *Front. Oncol.* (2020) **10** 581007. DOI: 10.3389/fonc.2020.581007
4. Ávila-Arcos M. C., McManus K. F., Sandoval K., Rodríguez-Rodríguez J. E., Villa-Islas V., Martin A. R.. **Population history and gene divergence in native Mexicans inferred from 76 human exomes human exomes**. *Mol. Biol. Evol.* (2020) **37** 994-1006. DOI: 10.1093/molbev/msz282
5. Borghini A., Vecoli C., Mercuri A., Turchi S., Andreassi M. G.. **Individual and joint effects of genetic polymorphisms in microRNA-machinery genes on congenital heart disease susceptibility**. *Cardiol. Young* (2021) **31** 965-968. DOI: 10.1017/S1047951120004874
6. Chacón-Duque J. C., Adhikari K., Fuentes-Guajardo M., Mendoza-Revilla J., Acuña-Alonzo V., Barquera R.. **Latin Americans show wide-spread Converso ancestry and imprint of local Native ancestry on physical appearance**. *Nat. Commun.* (2018) **9** 5388. DOI: 10.1038/s41467-018-07748-z
7. Chen X., Sturgis E. M., Wang C., Cao X., Li Y., Wei Q.. **Significance of microRNA-related variants in susceptibility to recurrence of oropharyngeal cancer patients after definitive radiotherapy**. *Oncotarget* (2016) **7** 35015-35025. DOI: 10.18632/oncotarget.9014
8. Cheng H., Li H., Feng Y., Zhang Z.. **Correlation analysis between SNPs in microRNA-machinery genes and tuberculosis susceptibility in the Chinese Uygur population**. *Med. Baltim.* (2018) **97** e13637. DOI: 10.1097/MD.0000000000013637
9. Cho S. H., Ko J. J., Kim J. O., Jeon Y. J., Yoo J. K., Oh J.. **3'-UTR polymorphisms in the MiRNA machinery genes DROSHA, DICER1, RAN, and XPO5 are associated with colorectal cancer risk in a Korean population**. *PLoS One* (2015) **10** e0131125. DOI: 10.1371/journal.pone.0131125
10. Clark A. G., Nielsen R., Signorovitch J., Matise T. C., Glanowski S., Heil J.. **Linkage disequilibrium and inference of ancestral recombination in 538 single-nucleotide polymorphism clusters across the human genome**. *Am. J. Hum. Genet.* (2003) **73** 285-300. DOI: 10.1086/377138
11. Clarke L., Fairley S., Zheng-Bradley X., Streeter I., Perry E., Lowy E.. **The international genome sample resource (IGSR): A worldwide collection of genome variation incorporating the 1000 genomes project data**. *Nucleic Acids Res.* (2017) **45** D854-D859. DOI: 10.1093/nar/gkw829
12. Ding C., Li C., Wang H., Li B., Guo Z.. **A miR-SNP of the XPO5 gene is associated with advanced non-small-cell lung cancer**. *Onco Targets Ther.* (2013) **6** 877-881. DOI: 10.2147/OTT.S48284
13. Excoffier L., Lischer H. E.. **Arlequin suite ver 3.5: A new series of programs to perform population genetics analyses under linux and windows**. *Mol. Ecol. Resour.* (2010) **10** 564-567. DOI: 10.1111/j.1755-0998.2010.02847.x
14. Gabriel S. B., Schaffner S. F., Nguyen H., Moore J. M., Roy J., Blumenstiel B.. **The structure of haplotype blocks in the human genome**. *Science* (2002) **296** 2225-2229. DOI: 10.1126/science.1069424
15. Gnagnarella P., Raimondi S., Aristarco V., Johansson H., Bellerba F., Corso F.. **Ethnicity as modifier of risk for Vitamin D receptors polymorphisms: Comprehensive meta-analysis of all cancer sites**. *Crit. Rev. Oncol. Hematol.* (2021) **158** 103202. DOI: 10.1016/j.critrevonc.2020.103202
16. González I. A., Stewart D. R., Schultz K. A. P., Field A. P., Hill D. A., Dehner L. P.. **DICER1 tumor predisposition syndrome: An evolving story initiated with the pleuropulmonary blastoma**. *Mod. Pathol.* (2022) **35** 4-22. DOI: 10.1038/s41379-021-00905-8
17. Gurdasani D., Barroso I., Zeggini E., Sandhu M. S.. **Genomics of disease risk in globally diverse populations**. *Nat. Rev. Genet.* (2019) **20** 520-535. DOI: 10.1038/s41576-019-0144-0
18. Hashimoto N., Tanaka T.. **Role of miRNAs in the pathogenesis and susceptibility of diabetes mellitus**. *J. Hum. Genet.* (2017) **62** 141-150. DOI: 10.1038/jhg.2016.150
19. Hodgson J. A., Pickrell J. K., Pearson L. N., Quillen E. E., Prista A., Rocha J.. **Natural selection for the Duffy-null allele in the recently admixed people of Madagascar**. *Proc. Biol. Sci.* (2014) **281** 20140930. DOI: 10.1098/rspb.2014.0930
20. Hodonsky C. J., Baldassari A. R., Bien S. A., Raffield L. M., Highland H. M., Sitlani C. M.. **Ancestry-specific associations identified in genome-wide combined-phenotype study of red blood cell traits emphasize benefits of diversity in genomics**. *BMC Genomics* (2020) **21** 228. DOI: 10.1186/s12864-020-6626-9
21. Huang X., An Y., Li X., Wang D., Tan H., Lei J.. **Genetic variants in DICER1, DROSHA, RAN, and XPO5 genes and risk of pregnancy-induced hypertension**. *Pregnancy Hypertens.* (2019a) **16** 161-166. DOI: 10.1016/j.preghy.2019.04.005
22. Huang X., Li Z., Lei J., Wang D., Zhang Y.. **Genetic polymorphism in DGCR8 is associated with late onset of preeclampsia**. *BMC Med. Genet.* (2019b) **20** 151. DOI: 10.1186/s12881-019-0887-7
23. Kandettu A., Radhakrishnan R., Chakrabarty S., Sriharikrishnaa S., Kabekkodu S. P.. **The emerging role of miRNA clusters in breast cancer progression**. *Biochim. Biophys. Acta Rev. Cancer* (2020) **1874** 188413. DOI: 10.1016/j.bbcan.2020.188413
24. Ke H. L., Chen M., Ye Y., Hildebrandt M. A., Wu W. J., Wei H.. **Genetic variations in micro-RNA biogenesis genes and clinical outcomes in non-muscle-invasive bladder cancer**. *Carcinogenesis* (2013) **34** 1006-1011. DOI: 10.1093/carcin/bgt006
25. Kim J. O., Bae J., Kim J., Oh S. H., An H. J., Han I. B.. **Association of MicroRNA biogenesis genes polymorphisms with ischemic stroke susceptibility and post-stroke mortality**. *J. Stroke* (2018) **20** 110-121. DOI: 10.5853/jos.2017.02586
26. Lan Q., Shen M., Garcia-Rossi D., Chanock S., Zheng T., Berndt S. I.. **Genotype frequency and F ST analysis of polymorphisms in immunoregulatory genes in Chinese and Caucasian populations**. *Immunogenetics* (2007) **59** 839-852. DOI: 10.1007/s00251-007-0253-3
27. Li J., Liao Y., Liao Y., Ruan S., Wan Y., Peng L.. **DROSHArs10719 T>C is associated with lymph node metastasis and clinical stage of gastric cancer patients**. *Int. J. Clin. Exp. Pathol.* (2017) **10** 7333-7339. PMID: 31966573
28. Lindo J., Haas R., Hofman C., Apata M., Moraga M., Verdugo R. A.. **The genetic prehistory of the Andean highlands 7000 years BP though European contact**. *Sci. Adv.* (2018) **4** eaau4921. DOI: 10.1126/sciadv.aau4921
29. López-Jiménez E., Andrés-León E.. **The implications of ncRNAs in the development of human diseases**. *Noncoding RNA* (2021) **7** 17. DOI: 10.3390/ncrna7010017
30. Mallick S., Li H., Lipson M., Mathieson I., Gymrek M., Racimo F.. **The simons genome diversity project: 300 genomes from 142 diverse populations**. *Nature* (2016) **538** 201-206. DOI: 10.1038/nature18964
31. Margis R., Margis R., Rieder C. R.. **Identification of blood microRNAs associated to Parkinsonĭs disease**. *J. Biotechnol.* (2011) **152** 96-101. DOI: 10.1016/j.jbiotec.2011.01.023
32. Martin-Guerrero I., Gutierrez-Camino A., Lopez-Lopez E., Bilbao-Aldaiturriaga N., Pombar-Gomez M., Ardanaz M.. **Genetic variants in miRNA processing genes and pre-miRNAs are associated with the risk of chronic lymphocytic leukemia**. *PLoS One* (2015) **10** e0118905. DOI: 10.1371/journal.pone.0118905
33. Molasy M., Walczak A., Przybyłowska-Sygut K., Zaleska-Żmijewska A., Szaflik J., Szaflik J. P.. **Analysis of the polymorphic variants of RAN and GEMIN3 genes and risk of Primary Open-Angle Glaucoma in the Polish population**. *Ophthalmic Genet.* (2018) **39** 180-188. DOI: 10.1080/13816810.2017.1381978
34. Moreno-Estrada A., Gignoux C. R., Fernández-López J. C., Zakharia F., Sikora M., Contreras A. V.. **Human genetics. The genetics of Mexico recapitulates Native American substructure and affects biomedical traits**. *Science* (2014) **344** 1280-1285. DOI: 10.1126/science.1251688
35. Morishita A., Iwama H., Fujihara S., Sakamoto T., Fujita K., Tani J.. **MicroRNA profiles in various hepatocellular carcinoma cell lines**. *Oncol. Lett.* (2016) **12** 1687-1692. DOI: 10.3892/ol.2016.4853
36. Myles S., Davison D., Barrett J., Stoneking M., Timpson N.. **Worldwide population differentiation at disease-associated SNPs**. *BMC Med. Genomics* (2008) **1** 22. DOI: 10.1186/1755-8794-1-22
37. Ni W. J., Leng X. M.. **Dynamic miRNA-mRNA paradigms: New faces of miRNAs**. *Biochem. Biophys. Rep.* (2015) **4** 337-341. DOI: 10.1016/j.bbrep.2015.10.011
38. Nikolić Z., Savić Pavićević D., Vučić N., Cerović S., Vukotić V., Brajušković G.. **Genetic variants in RNA-induced silencing complex genes and prostate cancer**. *World J. Urol.* (2017) **35** 613-624. DOI: 10.1007/s00345-016-1917-0
39. Osuch-Wojcikiewicz E., Bruzgielewicz A., Niemczyk K., Sieniawska-Buccella O., Nowak A., Walczak A.. **Association of polymorphic variants of miRNA processing genes with larynx cancer risk in a polish population**. *Biomed. Res. Int.* (2015) **2015** 298378. DOI: 10.1155/2015/298378
40. Oz M., Karakus S., Yildirim M., Bagci B., Sari I., Bagci G.. **Genetic variants in the microRNA machinery gene (Dicer) have a prognostic value in the management of endometrial cancer**. *J. Cancer Res. Ther.* (2018) **14** 1279-1284. DOI: 10.4103/0973-1482.187291
41. Ozlem I. A., Balkan M., Erdal M. E., Rustemoglu A., Atar M., Hatipoglu N. K.. **Association of microRNA-related gene polymorphisms and idiopathic azoospermia in a south-east Turkey population**. *Biotechnol. Biotechnol. Equip.* (2017) **31** 356-362. DOI: 10.1080/13102818.2017.1281759
42. Pierron D., Heiske M., Razafindrazaka H., Pereda-Loth V., Sanchez J., Alva O.. **Strong selection during the last millennium for African ancestry in the admixed population of Madagascar**. *Nat. Commun.* (2018) **9** 932. DOI: 10.1038/s41467-018-03342-5
43. Reynolds A. W., Mata-Míguez J., Miró-Herrans A., Briggs-Cloud M., Sylestine A., Barajas-Olmos F.. **Comparing signals of natural selection between three Indigenous North American populations**. *Proc. Natl. Acad. Sci. U. S. A.* (2019) **116** 9312-9317. DOI: 10.1073/pnas.1819467116
44. Rezaei M., Eskandari F., Mohammadpour-Gharehbagh A., Teimoori B., Yaghmaei M., Mokhtari M.. **The Drosha rs10719 T>C polymorphism is associated with preeclampsia susceptibility**. *Clin. Exp. Hypertens.* (2018) **40** 440-445. DOI: 10.1080/10641963.2017.1392555
45. Rosenberg N. A., Huang L., Jewett E. M., Szpiech Z. A., Jankovic I., Boehnke M.. **Genome-wide association studies in diverse populations**. *Nat. Rev. Genet.* (2010) **11** 356-366. DOI: 10.1038/nrg2760
46. Roth C., Kasimir-Bauer S., Pantel K., Schwarzenbach H.. **Screening for circulating nucleic acids and caspase activity in the peripheral blood as potential diagnostic tools in lung cancer**. *Mol. Oncol.* (2011) **5** 281-291. DOI: 10.1016/j.molonc.2011.02.002
47. Saliminejad K., Khorram Khorshid H. R., Soleymani Fard S., Ghaffari S. H.. **An overview of microRNAs: Biology, functions, therapeutics, and analysis methods**. *J. Cell Physiol.* (2019) **234** 5451-5465. DOI: 10.1002/jcp.27486
48. Schaschl H., Göllner T., Morris D. L.. **Positive selection acts on regulatory genetic variants in populations of European ancestry that affect ALDH2 gene expression**. *Sci. Rep.* (2022) **12** 4563. DOI: 10.1038/s41598-022-08588-0
49. Song X., Zhong H., Wu Q., Wang M., Zhou J., Zhou Y.. **Association between SNPs in microRNA machinery genes and gastric cancer susceptibility, invasion, and metastasis in Chinese Han population**. *Oncotarget* (2017) **8** 86435-86446. DOI: 10.18632/oncotarget.21199
50. Stavast C. J., Erkeland S. J.. **The non-canonical aspects of MicroRNAs: Many roads to gene regulation**. *Cells* (2019) **8** 1465. DOI: 10.3390/cells8111465
51. Unoki H., Takahashi A., Kawaguchi T., Hara K., Horikoshi M., Andersen G.. **SNPs in KCNQ1 are associated with susceptibility to type 2 diabetes in East Asian and European populations**. *Nat. Genet.* (2008) **40** 1098-1102. DOI: 10.1038/ng.208
52. Villarreal-Molina M. T., Flores-Dorantes M. T., Arellano-Campos O., Villalobos-Comparan M., Rodríguez-Cruz M., Miliar-García A.. **Association of the ATP-binding cassette transporter A1 R230C variant with early-onset type 2 diabetes in a Mexican population**. *Diabetes* (2008) **57** 509-513. DOI: 10.2337/db07-0484
53. Volinia S., Calin G. A., Liu C. G., Ambs S., Cimmino A., Petrocca F.. **A microRNA expression signature of human solid tumors defines cancer gene targets**. *Proc. Natl. Acad. Sci. U. S. A.* (2006) **103** 2257-2261. DOI: 10.1073/pnas.0510565103
54. Wang Z., Wang J., Wang Q., Wang J., Yuan J., Si Y.. **Genetic variant in DICER gene is associated with prognosis of hepatocellular carcinoma in a Chinese cohort**. *Hepatol. Res.* (2017) **47** 845-853. DOI: 10.1111/hepr.12824
55. Wen J., Gao Q., Wang N., Zhang W., Cao K., Zhang Q.. **Association of microRNA-related gene XPO5 rs11077 polymorphism with susceptibility to thyroid cancer**. *Med. Baltim.* (2017) **96** e6351. DOI: 10.1097/MD.0000000000006351
56. Wen J., Lv Z., Ding H., Fang X., Sun M.. **Association of miRNA biosynthesis genes DROSHA and DGCR8 polymorphisms with cancer susceptibility: A systematic review and meta-analysis**. *Biosci. Rep.* (2018) **38** BSR20180072. DOI: 10.1042/BSR20180072
57. Wen Z., Zou X., Xie X., Zheng S., Chen X., Zhu K.. **Association of polymorphisms in miRNA processing genes with type 2 diabetes mellitus and its vascular complications in a southern Chinese population**. *Front. Endocrinol. (Lausanne).* (2019) **10** 461. DOI: 10.3389/fendo.2019.00461
58. Xie Y., Wang Y., Zhao Y., Guo Z.. **Single-nucleotide polymorphisms of microRNA processing machinery genes are associated with risk for gastric cancer**. *Onco Targets Ther.* (2015) **8** 567-571. DOI: 10.2147/OTT.S79150
59. Yan B., Wang H., Tan Y., Fu W.. **microRNAs in cardiovascular disease: Small molecules but big roles**. *Curr. Top. Med. Chem.* (2019) **19** 1918-1947. DOI: 10.2174/1568026619666190808160241
60. Yang H., Dinney C. P., Ye Y., Zhu Y., Grossman H. B., Wu X.. **Evaluation of genetic variants in microRNA-related genes and risk of bladder cancer**. *Cancer Res.* (2008) **68** 2530-2537. DOI: 10.1158/0008-5472.CAN-07-5991
61. Yasuda K., Miyake K., Horikawa Y., Hara K., Osawa H., Furuta H.. **Variants in KCNQ1 are associated with susceptibility to type 2 diabetes mellitus**. *Nat. Genet.* (2008) **40** 1092-1097. DOI: 10.1038/ng.207
62. Ye Y., Wang K. K., Gu J., Yang H., Lin J., Ajani J. A.. **Genetic variations in microRNA-related genes are novel susceptibility loci for esophageal cancer risk**. *Cancer Prev. Res. (Phila)* (2008) **1** 460-469. DOI: 10.1158/1940-6207.CAPR-08-0135
63. Younger S. T., Pertsemlidis A., Corey D. R.. **Predicting potential miRNA target sites within gene promoters**. *Bioorg Med. Chem. Lett.* (2009) **19** 3791-3794. DOI: 10.1016/j.bmcl.2009.04.032
64. Zhang Y., Cao A. L., Dong C.. **rs10719 polymorphism located within DROSHA 3'-untranslated region is responsible for development of primary hypertension by disrupting binding with microRNA-27b**. *Med. Sci. Monit.* (2017) **23** 911-918. DOI: 10.12659/msm.897607
65. Zhao Y., Du Y., Zhao S., Guo Z.. **Single-nucleotide polymorphisms of microRNA processing machinery genes and risk of colorectal cancer**. *Onco Targets Ther.* (2015) **8** 421-425. DOI: 10.2147/OTT.S78647
|
---
title: 'Hypertension and Ovarian Cancer: A Case-Control Study in Saudi Arabia'
journal: Cureus
year: 2023
pmcid: PMC10037349
doi: 10.7759/cureus.35294
license: CC BY 3.0
---
# Hypertension and Ovarian Cancer: A Case-Control Study in Saudi Arabia
## Abstract
Background *There is* limited evidence that evaluates the association between hypertension and ovarian cancer. The study aims to investigate the association between ovarian cancer and hypertension, the difference in lipid profile, and the association between body mass index (BMI) and ovarian cancer.
Methods We conducted a case-control study at King Abdelaziz Medical City (KAMC), oncology department. All Saudi female patients who were diagnosed with primary ovarian cancer admitted to the oncology department at KAMC from 2016 to 2019 were selected. The data were collected from medical records of patients of the KAMC by chart review using The Ministry of National Guard Health Affairs BESTCare database.
Results A total of 137 Saudi female patients diagnosed with ovarian cancer attending to gynecology and oncology center in KAMC from 2016 to 2019 were included in this study. The mean age of participants was 57 in cases and 56 in controls with a mean BMI of 29.64 in cases and 31 in controls. There were 63 obese cases, therefore, the proportion of obesity was $46\%$. Approximately one-third of cases were overweight ($28\%$) while one-fourth ($26\%$) of them were underweight or normal weight. Roughly two-thirds of cases were hypertensive with an overall proportion of 66 % ($95\%$ confidence interval (CI) 58-74) while one-third of controls were hypertensive with an overall proportion of $32\%$. Cases were having significantly higher triglycerides ($$p \leq 0.03$$) and lower high-density lipoprotein (HDL) ($$p \leq 0.001$$) than controls. The significant variables were analyzed using logistic regression. It was found that hypertensive subjects were 10.06 times more likely ($95\%$ CI: 4.88-20.71) to be associated with the cases as compared to controls ($p \leq 0.001$). Also, an increase in BMI was significantly associated with being a case with OR = 1.07 ($95\%$ CI: 1.02-1.12; $$p \leq 0.004$$).
Conclusion In conclusion, hypertension, elevated BMI, higher triglycerides, and lower HDL were significantly associated with ovarian cancer.
## Introduction
Ovarian cancer occurs when cells grow abnormally forming tumors in the ovary. Ovarian cancer has been estimated to be the seventh most common cancer diagnosed among women in the world and the fifth most common cancer-related death among women in the United States. Although ovarian cancer is less prevalent than breast cancer, it is three times more lethal [1,2]. The prevalence of ovarian cancer globally was estimated to be $3.6\%$ of cancer cases and $4.3\%$ of cancer-related deaths in female patients [3]. The number of identified cases of ovarian cancer among Saudi female patients was estimated to be 991 from 2001 to 2008, and Riyadh was the highest incidence rate of ovarian cancer, while the lowest incidence rate of ovarian cancer was in the northern region of Saudi Arabia [4].
Unfortunately, the incidence rate of ovarian cancer around the world is continuously increasing due to an increase in risk factors, in Saudi Arabia for example, the incidence rate increased by up to $5\%$ from 2001 to 2008 [4,5]. There are potential risk factors for developing ovarian cancer. Family history of ovarian cancer, age, infertility therapy, and hormonal replacement after menopause are possible factors that favor developing ovarian cancer [5,6]. Metabolic syndrome increases the risk of developing cancer types and cancer-related morbidity and mortality [7]. Metabolic syndrome is a collection of risk factors encompassing high blood pressure, elevated blood glucose, raised triglyceride (TG), low high-density lipoprotein (HDL), and abdominal obesity. Obesity is associated with an increased risk of ovarian cancer, particularly in women who are extremely obese and pre-menopausal, however, there was no significant association noticed in post-menopausal women based on a meta-analysis study [8]. On the other hand, a recent meta-analysis study declares that there was an $80\%$ increased risk of ovarian cancer in obese patients in post-menopausal women compared to normal weight, particularly those who do not use hormonal replacement therapy [9]. Multiple studies have investigated the association between lipid profile and ovarian cancer but ended up with different conclusions. For example, a systematic review and meta-analysis study included 12 studies and found that high total cholesterol is significantly associated with ovarian cancer [10]. Contrariwise, total cholesterol and HDL were significantly lower among ovarian cancer patients in multiple studies, based on a meta-analysis study published in 2020 [11]. Finally, there is a moderate increase in the risk of ovarian cancer in diabetic patients based on a meta-analysis study [12].
To our knowledge, there are limited studies that investigate the association between hypertension and ovarian cancer as searched using PubMed and using the following terms (hypertension and ovarian cancer) and/or (metabolic syndrome and ovarian cancer), which necessitates further studies, especially in different populations. The authors aimed to investigate the association between ovarian cancer and hypertension, the difference in lipid profile, and the association between body mass index (BMI) and ovarian cancer.
## Materials and methods
We conducted a case-control study at the oncology department of King Abdelaziz Medical City (KAMC) in 2019. It is a dynamic and progressive entity in comprehensive cancer care. It currently has five medical divisions: Adult Hematology, Adult Medical Oncology, Gynecology Oncology, Pediatric Hematology/Oncology, and Palliative Care. The data were collected from medical records of patients of the KAMC by chart review using The Ministry of National Guard Health Affairs BESTCare database. All Saudi female patients who were diagnosed with primary ovarian cancer admitted to the oncology department at KAMC from 2016 to 2019 were selected. We chose this period because the gynecology oncology center was established in 2016. To assess the association between ovarian cancer and hypertension as well as the difference in lipids and the association of obesity between cases and controls, we selected controls who attended KAMC for the same period of cases with matching age (+/- 5 years) without ovarian cancer and diabetes using the consecutive technique. The independent variables were age and ovarian cancer, whereas the dependent variables were hypertension, overweight, and obesity. Being overweight was defined as a BMI of 25-29.9 kg/m2, whereas obesity was defined as a BMI of 30 kg/m2 or higher. Both overweight and obesity were identified based on the World Health Organization Classification. Hypertension was defined as a systolic blood pressure of 130 or above and/or diastolic blood pressure of 80 and above based on the American Health Association Classification. Data were analyzed using SPSS (statistical package for social sciences) analysis software version 22 (IBM Corp, Armonk, New York, USA). Continuous variables were described using means and standard deviations, whereas categorical variables were presented as numbers and percentages. we used a T-test to assess if there is a significant difference between the means of the two groups. We used logistic regression analysis to assess the association of hypertension and BMI with ovarian cancer. The study was approved by the institution review board of King Abdullah International Medical Research Center (KAIMRC) (Approval number: SP$\frac{20}{362}$/R).
## Results
A total of 137 Saudi female patients diagnosed with ovarian cancer attending to gynecology and oncology center in KAMC from 2016 to 2019 were included in this study. The mean age of participants was 57 in cases and 56 in controls with a mean BMI of 29.64 in cases and 31 in controls. There were 63 obese cases, therefore, the proportion of obesity was $46\%$ (Table 1). Approximately one-third of cases were overweight ($28\%$) while one-fourth ($26\%$) of them were underweight or normal. Roughly two-thirds of cases were hypertensive with an overall proportion of $66\%$ while one-third of controls were hypertensive with an overall proportion of $32\%$ (Table 2). The means of TG in cases and controls were 1.34 mmol/L and 1.16 mmol/L, respectively, while the mean of HDL in cases and controls was 1.13 mmol/L and 1.29 mmol/L, respectively (Table 3). The mean of cholesterol and low-density lipoprotein (LDL) were similar as described in Table 3.
Comparing the mean TG levels between cases and controls showed that cases had significantly higher TG levels than controls ($$p \leq 0.03$$). Furthermore, cases were having significantly lower HDL than controls ($$p \leq 0.001$$) (Table 3).
The significant variables were analyzed using logistic regression. It was found that hypertensive subjects were 10.06 times more likely ($95\%$ CI: 4.88-20.71) to be associated with the cases as compared to controls ($p \leq 0.001$). Also, an increase in BMI was significantly associated with being a case with OR = 1.07 ($95\%$ CI: 1.02-1.12; $$p \leq 0.004$$) (Table 4).
**Table 4**
| Unnamed: 0 | OR | 95% CI for OR | 95% CI for OR.1 | p-value |
| --- | --- | --- | --- | --- |
| | OR | Lower | Upper | p-value |
| Hypertensive | 10.06 | 4.88 | 20.71 | <0.001 |
| BMI | 1.07 | 1.02 | 1.12 | 0.004 |
## Discussion
This study investigates the association between ovarian cancer and hypertension, the difference in lipid profile, and the association between BMI and ovarian cancer. The data obtained from this study showed that obesity was significantly associated with ovarian cancer (OR = 1.07 $95\%$ CI: 1.02-1.12; $$p \leq 0.004$$). This finding was in agreement with the previous studies [8,9]. Although the exact etiology of obesity’s increased risk of developing ovarian cancer is unknown, it is hypothesized that excess body mass increases ovarian cancer partly through the estrogenic effect by increasing the synthesis of estrogen levels in adipocytes. In addition, cases had significantly higher TG levels ($$p \leq 0.03$$). Although a significantly higher TG level in ovarian cancer compared to controls was observed in other studies, when age-stratified is applied TG level was insignificant [11]. furthermore, the authors observed that HDL was significantly lower than controls ($$p \leq 0.001$$) which is consistent with previous studies [11]. It could be partly explained that since cancer cells have a strong affinity for sterols and lipids, lipid metabolism is a crucial component of cancer signaling [13,14].
Finally, in the present study, hypertension was significantly associated with ovarian cancer (OR: 10 $95\%$ CI: 4.88-20.71). Similarly, a case-control study conducted at Tianjin Medical University, China enrolled 573 epithelial ovarian cancer patients and 1146 matched controls and revealed that hypertension was significantly associated with ovarian cancer (OR = 2.423; $95\%$ CI: 1.963-1.2.990) [15]. In addition, a prospective cohort study that enrolled 287320 women from Austria, Norway, and Sweden found that during the follow-up, 644 epithelial ovarian cancer and 388 death from ovarian cancer, and they concluded that there is no association between metabolic syndrome and epithelial ovarian cancer, however, increasing in blood pressure and blood cholesterol increase the risk of mucinous and endometrioid tumors, respectively [16]. On the other hand, a network of case-control study which conducted in Italy enrolled 970 ovarian cancer and 3045 controls suggesting no association between ovarian cancer and hypertension [17]. Hypertension is associated with certain types of cancer. For example, a systematic review and meta-analysis study investigated the association between breast cancer and hypertension and suggested that hypertension is significantly associated with an increased risk of breast cancer (RR: 1.15; $95\%$ CI: 1.08-1.22) [18].
Moreover, in a large prospective pooled cohort study, both treated and untreated, hypertension was associated with an increased likelihood of developing cancer compared with normotensive individuals [19]. Additionally, based on a prospective cohort study by Stocks et al. of roughly 577,800 adults followed for 12 years, there was an association between hypertension and cancer incidence in men and between hypertension and higher cancer mortality in both men and women [19]. The exact mechanism by which hypertension causes cancer is unknown. Animal models suggest dysregulation of apoptosis due to elevated blood pressures contributing to cancer [20,21]. Another possible explanation for overburdened hypertension among ovarian cancer patients is that using certain classes of chemotherapeutic agents is associated with increased blood pressure [22]. For example, bevacizumab (anti-vascular endothelial growth factor) can induce endothelial dysfunction along with decreased nitric oxide bioavailability causing an increase in the vascular tone which contributes to the elevation of blood pressure [21].
This study has methodological limitations as in other studies. Unfortunately, there was no documented information about medications such as chemotherapy and hormonal replacement therapy which can be a confounder. Because of that, it may result in reporting bias. Additionally, diabetes could not be investigated in our study since most of the cases did not have documented blood glucose which also makes it impossible to evaluate the association between metabolic syndrome and ovarian cancer.
## Conclusions
In conclusion, hypertension was significantly associated with ovarian cancer. In addition, low HDL, high TG, and elevated BMI were significantly associated with ovarian cancer.
## References
1. Momenimovahed Z, Tiznobaik A, Taheri S, Salehiniya H. **Ovarian cancer in the world: epidemiology and risk factors**. *Int J Womens Health* (2019) **11** 287-299. PMID: 31118829
2. **American Cancer Society: Ovarian Cancer Statistics | How Common is Ovarian Cancer**. (2020) **7** 20
3. Torre LA, Bray F, Siegel RL, Ferlay J, Lortet-Tieulent J, Jemal A. **Global cancer statistics, 2012**. *CA Cancer J Clin* (2015) **65** 87-108. PMID: 25651787
4. Alghamdi IG, Hussain II, Alghamdi MS, Alghamdi MM, Dohal AA, El-Sheemy MA. **Incidence rate of ovarian cancer cases in Saudi Arabia: an observational descriptive epidemiological analysis of data from Saudi Cancer Registry 2001-2008**. *Int J Womens Health* (2014) **6** 639-645. PMID: 25028565
5. Gates MA, Rosner BA, Hecht JL, Tworoger SS. **Risk factors for epithelial ovarian cancer by histologic subtype**. *Am J Epidemiol* (2010) **171** 45-53. PMID: 19910378
6. Devita VT, Hellman S, Rosenberg SA. **Cancer: Principles & Practice of Oncology, 6th edition**. *Cancer: principles & practice of oncology[M* (2005)
7. Cowey S, Hardy RW. **The metabolic syndrome: a high-risk state for cancer?**. *Am J Pathol* (2006) **169** 1505-1522. PMID: 17071576
8. Leitzmann MF, Koebnick C, Danforth KN. **Body mass index and risk of ovarian cancer**. *Cancer* (2009) **115** 812-822. PMID: 19127552
9. Liu Z, Zhang TT, Zhao JJ, Qi SF, Du P, Liu DW, Tian QB. **The association between overweight, obesity and ovarian cancer: a meta-analysis**. *Jpn J Clin Oncol* (2015) **45** 1107-1115. PMID: 26491203
10. Zhang D, Xi Y, Feng Y. **Ovarian cancer risk in relation to blood lipid levels and hyperlipidemia: a systematic review and meta-analysis of observational epidemiologic studies**. *Eur J Cancer Prev* (2021) **30** 161-170. PMID: 32483012
11. Onwuka JU, Okekunle AP, Olutola OM, Akpa OM, Feng R. **Lipid profile and risk of ovarian tumours: a meta-analysis**. *BMC Cancer* (2020) **20** 200. PMID: 32164586
12. Lee JY, Jeon I, Kim JW, Song YS, Yoon JM, Park SM. **Diabetes mellitus and ovarian cancer risk: a systematic review and meta-analysis of observational studies**. *Int J Gynecol Cancer* (2013) **23** 402-412. PMID: 23354371
13. Cruz PM, Mo H, McConathy WJ, Sabnis N, Lacko AG. **The role of cholesterol metabolism and cholesterol transport in carcinogenesis: a review of scientific findings, relevant to future cancer therapeutics**. *Front Pharmacol* (2013) **4** 1-7. PMID: 23346057
14. Gorin A, Gabitova L, Astsaturov I. **Regulation of cholesterol biosynthesis and cancer signaling**. *Curr Opin Pharmacol* (2012) **12** 710-716. PMID: 22824431
15. Chen Y, Zhang L, Liu W, Wang K. **Case-control study of metabolic syndrome and ovarian cancer in Chinese population**. *Nutr Metab* (2017) **14** 1-9
16. Bjørge T, Lukanova A, Tretli S. **Metabolic risk factors and ovarian cancer in the Metabolic Syndrome and Cancer project**. *Int J Epidemiol* (2011) **40** 1667-1677. PMID: 21984693
17. Soler M, Chatenoud L, Negri E, Parazzini F, Franceschi S, la Vecchia C. **Hypertension and hormone-related neoplasms in women**. *Hypertension* (1999) **34** 320-325. PMID: 10454461
18. Han H, Guo W, Shi W, Yu Y, Zhang Y, Ye X, He J. **Hypertension and breast cancer risk: a systematic review and meta-analysis**. *Sci Rep* (2017) **7** 1-9. PMID: 28127051
19. Stocks T, Van Hemelrijck M, Manjer J. **Blood pressure and risk of cancer incidence and mortality in the Metabolic Syndrome and Cancer Project**. *Hypertension* (2012) **59** 802-810. PMID: 22353615
20. Harding JL, Sooriyakumaran M, Anstey KJ. **Hypertension, antihypertensive treatment and cancer incidence and mortality: a pooled collaborative analysis of 12 Australian and New Zealand cohorts**. *J Hypertens* (2016) **34** 149-155. PMID: 26630217
21. Hamet P, Richard L, Dam TV. **Apoptosis in target organs of hypertension**. *Hypertension* (1995) **26** 642-648. PMID: 7558225
22. Maitland ML, Bakris GL, Black HR. **Initial assessment, surveillance, and management of blood pressure in patients receiving vascular endothelial growth factor signaling pathway inhibitors**. *J Natl Cancer Inst* (2010) **102** 596-604. PMID: 20351338
|
---
title: '“Black People Like Me”: A virtual conference series to engage underserved
patients with asthma in patient centered outcomes research'
authors:
- LeRoy Graham
- Mary Hart
- Michael Stinson
- Rhoda Moise
- Lynda Mitchell
- Tonya A. Winders
- Donna D. Gardner
journal: Research Involvement and Engagement
year: 2023
pmcid: PMC10037352
doi: 10.1186/s40900-023-00428-3
license: CC BY 4.0
---
# “Black People Like Me”: A virtual conference series to engage underserved patients with asthma in patient centered outcomes research
## Abstract
There are more Black people with asthma in the US than White people. Black people have more hospital stays or deaths from asthma than White people. This highlights the need for Black people to be involved in research. Black people are missing from research. Patient centered outcomes research (PCOR) looks at patient treatment results. A webinar series titled “Black People Like Me” (BPLM) was developed for the Black community. This was a discussion about asthma, COVID-19, and PCOR between patients and experts. BPLM was a free, one-hour, virtual meeting held once a month for 6 months. BPLM was developed by Black patients, clergy, physicians, and a program evaluator. The goal was to meet the needs of the Black community. An average of 658 people participated in each virtual meeting. BPLM found that $77\%$ of attendees were highly satisfied with the sessions. BPLM increased attendee knowledge of asthma, COVID-19, and PCOR. Attendees reported an increased willingness to be involved in PCOR. Including Black clergy helped BPLM be successful by encouraging trust in the program within the Black community. Future programs like BPLM may be developed to help people make decisions about being involved in their health care and PCOR.
### Background
In response to racial inequity in asthma, asthma-related research among diverse patients is vital. However, people from historically marginalized groups are underrepresented in clinical and patient-centered outcomes research (PCOR). The “Black People Like Me” (BPLM) virtual conference series was developed to: [1] engage Black patients with asthma and their caregivers in education and discussions about asthma, and [2] encourage involvement in PCOR. Education about COVID-19 and COVID-19 vaccination was also incorporated.
### Methods
The Project Advisory Group consisting of Black patients, clergy, physicians, and a program evaluator met monthly to develop BPLM. The program consisted of free one-hour virtual sessions held monthly for 6 months. BPLM was promoted through the Allergy & Asthma Network website, emails, social media, and personal contacts with a recruitment goal of ≥ 100 Black patients with asthma or caregivers. Program evaluations, interactive polling questions during each session, and participant pre- and post-session tests were conducted.
### Results
Sessions averaged 658 participants including Black patients, family members, caregivers, Black clergy, health care providers, and other concerned community. Overall, $77\%$ of participants strongly agreed with satisfaction with the sessions. Pre- and post-tests demonstrated that participants exhibited growth in knowledge regarding asthma risk, PCOR, and PCOR research opportunities for patients, exhibited preexisting and sustained knowledge regarding COVID-19 vaccination and side effects, and demonstrated an increased sense of empowerment during healthcare visits.
### Conclusions
BPLM demonstrated that a virtual platform can successfully engage Black communities. Incorporating clergy and religious organizations was critical in developing the trust of the Black community towards BPLM.
## Background
Asthma poses a considerable burden on patients and the health care system. There is inequity in asthma prevalence in the US between individuals who identify as Black compared with those who identify as White ($10.4\%$ vs $7.9\%$, respectively) [1]. In addition, Black individuals are approximately 3-times more likely to die from asthma-related causes or visit an emergency department for asthma, and almost 4-times more likely be admitted to the hospital for asthma compared with White individuals [2].
Asthma is a heterogenous disease, and related research among diverse patient cohorts is vital. Patient populations and outcomes investigated in randomized controlled trials may not reflect real-world situations [3–5]. Patient-centered outcomes research (PCOR) can fill in some of the gaps associated with clinical trials by comparing outcomes relevant to daily life among many different treatment options and among diverse populations [5, 6]. Importantly, PCOR explores having the patient involved in different roles of research—individuals are not just a subject but a research partner [7].
People from historically marginalized groups are underrepresented in clinical trial research and PCOR [8–10]. Barriers to participation in clinical trials for people in historically marginalized groups include mistrust, lack of comfort with the clinical trial process, lack of information about clinical trials, time and resource constraints associated with trial participation, and lack of clinical trial awareness [11]. To significantly impact inequities in asthma outcomes, effective efforts to increase participation levels of patients from historically marginalized groups in research need to include patient engagement programs that address the many barriers. “ Not One More Life” (NOML) is a non-profit program driven by health care volunteers and church leaders that engages historically marginalized communities in education, health screening, and health care related to asthma [12]. Based on learnings from NOML, the “Black People Like Me” (BPLM) virtual conference series was developed to: [1] engage Black patients living with asthma and their caregivers in education and discussions about asthma, and [2] encourage involvement in PCOR. With the start of the COVID-19 pandemic, a need was also seen to educate the Black community about the virus and COVID-19 vaccination, which was therefore incorporated into the BPLM series. A critical feature of the BPLM program is that it incorporated clergy and religious organizations to help develop trust in the Black community, based on learnings from NOML. Involvement of religious organizations had the added benefit of leveraging existing assets to facilitate meaningful engagement and discussion with the community.
## Program development
Funding for the development of the BPLM conference series was through a Eugene Washington PCORI Engagement Award received by the Allergy & Asthma Network (AAN) from the Patient-Centered Outcomes Research Institute (PCORI). AAN is a non-profit patient education and advocacy organization for people with asthma, allergies, and related conditions. NOML currently operates as a program of AAN.
The first step in the BPLM conference series development was the establishment of a Project Advisory Group (PAG). The PAG consisted of Black patients, clergy, physicians with experience in PCOR and COVID-19, and a PCORI-experienced program evaluator. There was also a Project Lead (Dr. LeRoy Graham) and a Project Manager (Mary Hart). Meetings of the PAG occurred monthly where they recommended and discussed appropriately worded titles, session objectives, recruitment strategies, proposed images, suggestions for speakers, and topics of interest to the Black community. The conference platform, dates, times, and participant honorariums were all selected by the PAG to best meet the needs and expectations of the Black community. A title for the conference series of “Black People Like Me” was developed by the PAG, with a corresponding branded Powerpoint template created by the PAG and AAN. Colors, logo, and images used were chosen by the PAG to look like “Black People Like Me” and be diverse in sex, age, and community (Fig. 1).Fig. 1“Black People Like Me” virtual conference session Powerpoint template example
## Virtual conference sessions
BPLM was a six-month series of one-hour virtual sessions held each month (Table 1). Registration for the conference was free. A virtual conference was necessary as the BPLM program was developed during the COVID-19 pandemic. The virtual format had the additional advantage of being less expensive than hosting an in-person conference. Zoom™ was chosen as the virtual platform because that was the platform the Black community was most comfortable with using to stay connected and attend religious services during COVID-19–related lockdown and social distancing procedures. In addition, Zoom™ could be used from a smartphone and did not require installation of an app. Table 1“Black People Like Me” virtual conference session titles and attendanceMonth, session noTopicMeeting duration (Minutes)Registered participantsAttendance N (Rate %)Question and answer, NDecember, Session 1Asthma, COVID-19, Questions We Need Answered: Black People Like Me Series108N = 1412Patient/family, $$n = 113$$$n = 24$HCP, $$n = 377$$Clergy, $$n = 9$$Unknown, $$n = 81$$Total, 580 ($41\%$)January, Session 2African Americans and Research: Making it Work for Folks Like Us89N = 960Patient/family, $$n = 176$$$n = 26$HCP, $$n = 225$$Clergy, $$n = 6$$Unknown, $$n = 84$$Total, 491 ($51\%$)February, Session 3COVID-19 and Black Folk: Changing the Game, Changing the Outcome103N = 1006Patient/family, $$n = 217$$$n = 37$HCP, $$n = 203$$Clergy, $$n = 11$$Unknown, $$n = 80$$Total, 511 ($51\%$)March, Session 4Asthma and COVID-19: Is Research Really that Important To Me?79N = 812Patient/family, $$n = 182$$$n = 18$HCP, $$n = 105$$Clergy, $$n = 8$$Unknown, $$n = 238$$Total, 533 ($66\%$)April, Session 5Asthma and COVID-19: My Journey with Asthma, COVID or Other Health Issues…Letting Researchers Know What Questions We Need Answered96N = 1733Patient/family, $$n = 560$$$n = 58$HCP, $$n = 60$$Clergy, $$n = 12$$Unknown, $$n = 415$$Total, 1047 ($60\%$)May, Session 6Patient Advocacy in Asthma & COVID-19: Where Do We Go from Here?105N = 1668Patient/family, $$n = 616$$$n = 41$HCP, $$n = 61$$Clergy, $$n = 18$$Unknown, $$n = 88$$Total, 783 ($47\%$)Average100N = 1265658 ($52\%$)$$n = 34$$HCP health care provider Information from the designated speaker was transferred to the BPLM template and polling questions about the topic being discussed were inserted in the presentation to gain immediate feedback from the participants. The format for each session included a welcome from the program moderator and Project Lead, and an introduction of the topic, agenda, and objectives. Speakers presented their topic(s) while interactive polling questions, chat, and Q&A took place throughout the session. The Project Manager or AAN staff communicated the questions from the chat and Q&A to the moderator to ask the speakers and panelists. AAN support staff used Twitter (#BlackPeopleLikeMe) to send out real-time messages throughout each session. All sessions were recorded and uploaded to the AAN website at: http://blackpeoplelikeme.com/.
Patient Advisor speakers for the session were from diverse backgrounds and were allowed to tell their stories from their experience, from their heart. No prepared scripts were used. Practice sessions with the moderator and all speakers were held a few days before the conference sessions to ensure everyone was familiar with the platform, camera and microphone use, program flow, polling questions, and Q&A, and go over the backup plan if anyone lost connection, etc.
To further engage the community and maintain momentum between sessions, a #BlackPeopleLikeMe Facebook group was created and managed by AAN. Session participants were encouraged to join and those who did were asked to share their experiences living with asthma, COVID-19, and related conditions by posting images, quotes, or poems to describe their feelings and journey this past year.
## Recruitment
The recruitment goal for BPLM was to engage at least 100 Black patients with asthma or caregivers of patients with asthma. *The* general public was also invited. AAN posted the event on their website, sent emails to 40,240 patients, families, and health care providers, posted Twitter tweets, and advertised the sessions in all their social media groups (e.g., LinkedIn, Instagram, and Facebook). In addition, PAG members recruited participants through personal emails, phone calls, and by contacting clergy and physician offices. Participation incentives were provided at the end of each session in the form of an electronic emailed gift card to the first 100 attendees from the US who indicated they were patients with asthma and who stayed for the entire session.
## Program evaluations
Program evaluations included both quantitative and qualitative methods to comprehensively assess impact. Quantitative methods included written evaluations that were administered after every session to rate the overall satisfaction of program materials, speakers, topics, and use of virtual program format for registration and live virtual sessions. In addition, interactive polling questions related to the session topic were administered during each session. Participants also took a pre- and post-session test to assess knowledge of COVID-19, health risk for people with asthma and related symptoms, health inequity, disease management, preventative measures, and research engagement. The links to the pre- and post-tests were emailed to participants when they registered for the program and after the program. Qualitative evaluations included assessment of the Q&A portion of each session and comments posted in the chat box.
## Session participation and responses
Session participants included Black patients, family members, or caregivers, Black clergy, health care providers, and other concerned community from across the US, with some non-US participants. The sessions averaged 658 participants with an average duration of approximately an hour and 40 min each (Table 1). Sessions ran over the allotted 1 h because participants remained engaged by submitting comments and questions for discussion.
Program evaluations found that overall, $77\%$ of participants strongly agreed with satisfaction with various aspects of the sessions (Table 2). Participants mainly shared positive feedback highlighting their delight in the interactivity of the sessions, educational and informational enhancement opportunities, inclusion of both expert physicians and researchers, and experience from community members and patients. Participants desired more engagement through handouts and shared access to other outlets such as church groups. Additionally, comments included room for improvement with technology and sensitivity around race and time. Overall, the chat box themes highlighted the safe space created with the BPLM program along with participants’ appreciation for the information and education shared. Participant questions submitted during the Q&A portion centered around the COVID-19 vaccine, disease management, health equity, and outreach. Sample questions from the sessions are shown in Table 3.Table 2“Black People Like Me” virtual conference session evaluation resultsEvaluation questionAnswer ChoicesDecember $$n = 178$$*ResponsesN (%)JanuaryN = 183*ResponsesN (%)FebruaryN = 168*ResponsesN (%)MarchN = 144*ResponsesN (%)AprilN = 319*ResponsesN (%)MayN = 233*ResponsesN (%)It was easy to register for the programStrongly Agree159 ($90\%$)154 ($84\%$)151 ($90\%$)120 ($83\%$)240 ($75\%$)173 ($74\%$)Agree17 ($10\%$)28 ($15\%$)17 ($10\%$)23 ($16\%$)76 ($24\%$)57 ($24\%$)Disagree1 ($1\%$)–––3 ($1\%$)1 ($1\%$)Strongly Disagree–1 ($1\%$)–1 ($1\%$)–2 ($1\%$)I felt encouraged and safe to participate during the sessionStrongly Agree140 ($79\%$)138 ($76\%$)134 ($78\%$)115 ($80\%$)242 ($77\%$)176 ($77\%$)Agree37 ($21\%$)42 ($23\%$)34 ($20\%$)29 ($20\%$)70 ($22\%$)52 ($23\%$)Disagree––––1 ($1\%$)–Strongly Disagree–2 ($1\%$)––2 ($1\%$)2 ($1\%$)The speakers were experts and taught me a lot about COVID and AsthmaStrongly Agree126 ($71\%$)111 ($61\%$)133 ($80\%$)105 ($73\%$)239 ($76\%$)185 ($79\%$)Agree47 ($26\%$)66 ($36\%$)32 ($19\%$)38 ($26\%$)74 ($23\%$)48 ($21\%$)Disagree5 ($3\%$)5 ($3\%$)2 ($1\%$)1 ($1\%$)2 ($1\%$)–Strongly Disagree––––1 ($1\%$)1 ($1\%$)I was satisfied with the topics presentedStrongly Agree134 ($75\%$)123 ($68\%$)129 ($78\%$)108 ($76\%$)227 ($72\%$)178 ($76\%$)Agree41 ($23\%$)57 ($31\%$)36 ($22\%$)31 ($22\%$)86 ($27\%$)54 ($23\%$)Disagree3 ($2\%$)1 ($1\%$)1 ($1\%$)2 ($1\%$)2 ($1\%$)–Strongly Disagree–1 ($1\%$)––2 ($1\%$)1 ($1\%$)The Zoom Webinar was a good way to present this informationStrongly Agree139 ($78\%$)133 ($73\%$)135 ($80\%$)106 ($75\%$)237 ($75\%$)164 ($71\%$)Agree39 ($22\%$)49 ($27\%$)33 ($19\%$)34 ($25\%$)79 ($25\%$)64 ($28\%$)Disagree––––2 ($1\%$)2 ($1\%$)Strongly Disagree–1 ($1\%$)–––1 ($1\%$)I felt like the presenters answered the questions at the endStrongly Agree87 ($50\%$)114 ($63\%$)113 ($68\%$)98 ($69\%$)193 ($62\%$)149 ($64\%$)Agree75 ($43\%$)65 ($36\%$)49 ($29\%$)42 ($30\%$)110 ($35\%$)78 ($34\%$)Disagree11 ($6\%$)2 ($1\%$)5 ($3\%$)2 ($1\%$)9 ($3\%$)3 ($1\%$)Strongly Disagree––––1 ($1\%$)2 ($1\%$)I liked having the polling questions during the presentationStrongly Agree125 ($71\%$)137 ($76\%$)134 ($80\%$)99 ($69\%$)194 ($61\%$)166 ($71\%$)Agree51 ($29\%$)42 ($23\%$)33 ($20\%$)43 ($30\%$)116 ($37\%$)64 ($27\%$)Disagree–2 ($1\%$)–1 ($1\%$)6 ($2\%$)2 ($1\%$)Strongly Disagree–––––1 ($1\%$)I was satisfied with the overall sessionStrongly Agree135 ($77\%$)135 ($74\%$)136 ($81\%$)108 ($76\%$)232 ($73\%$)167 ($72\%$)Agree40 ($23\%$)45 ($25\%$)31 ($18\%$)32 ($22\%$)82 ($26\%$)63 ($27\%$)Disagree1 ($1\%$)2 ($1\%$)1 ($1\%$)2 ($1\%$)2 ($1\%$)1 ($1\%$)Strongly Disagree–1 ($1\%$)–1 ($1\%$)–1 ($1\%$)*Monthly totals may not be uniform due to percentile rounding to whole numbers and participants skipping questionsTable 3Sample questions posed by participants during the “Black People Like Me” virtual conference sessionsThemesSub-themesSample responsesVaccineSafetyWhat if you are allergic to aspirin? Is the vaccine safe?DosageAre there any repercussions to missing the second COVID-19 vaccine?LongevityWill we need a booster vaccine in the fall?AccessWhere can I get vaccinated?EfficacyWith the vaccine, can you still get and spread the virus?Side effectsWhat are the long-term effects of the vaccine? Infertility? Can the vaccine cause loss of taste, smell, or both?HesitancyHow do you instill trust with Black Americans that the vaccine is safe?Allergy interactionsThose who are denied a vaccine due to allergies what option of defense due they have?VariantsIf we've had the two vaccines, is it safe to join society again in light of the variants?Disease managementChildrenWhat’s the best way to help children with asthma make their lungs stronger (or make asthma less of a problem)? Exercise? Certain foods?AgeWhat is the lifespan of people living with asthma on medication vs without? How long do they live? What specific age groups are meant by "elderly" versus "younger"?RecommendationsIt would be great if an app is developed to provide feedback regarding controller medication use to the provider and the parent, similar to diabetes type 1 careInterrelated factorsIs there any significant link between depression and asthma? Asthma and COPD, what is the difference and is there any link? Does stress trigger asthma? Is having asthma a risk factor in COVID-19? *Is asthma* a disability?Treatment optionsIs there a long-term asthma treatment under development or testing that may change lives of the asthmatic community? *Is asthma* curable like malaria? *Why is* it important for asthma patients to stick to an asthma control action plan?Health equityEnvironmentWhat role does environmental racism play in the disproportionate numbers in Black Americans with asthma?Race and comorbidityIn the study with African Americans and Latinos with asthma, were comorbidities a factor in emergency room visits and asthma related deaths?RaceI've personally not taken the COVID-19 vaccine. Do you think we should take it, as Blacks, or we should abstain? I feel we may be given substandard ones. Why are Black people more affected by COVID-19 than White people?SexWhy does COVID-19 tend to affect more men than females?OutreachEducationWhat are some ways that we can educate our communities about asthma?Webinar distributionWhen will this webinar be available for rebroadcast? *Will this* session be archived so that we can share it with others who were not able to log on?Patient engagementHow can someone become a patient partner in a study?
Pre- and post-tests conducted for session 1 helped the PAG with restructuring evaluations for future sessions. Highlights of the pre- and post-test results from sessions 2–6 are shown in Table 4. Session 2 pre- and post-tests demonstrated that participants exhibited growth in knowledge regarding asthma risk and research opportunities for patients, and in session 3, participants exhibited preexisting and sustained knowledge regarding COVID-19 vaccination procedures and side effects. For session 4, participants exhibited improvements regarding knowledge of PCOR, sustained awareness of PCORI training opportunities for patient involvement in research, and increased implementation of COVID-19 vaccination. Polling questions in session 4 had a greater number of responses than the pre- and post-tests and reflected an increased willingness by participants after the expert presentations to take the COVID-19 vaccine and to participate in clinical research. In session 5, participants demonstrated an increased sense of empowerment in reference to participants perceived knowledge in the engagement process and treatment decisions during health care visits. There was an increase in the perception that getting health care or treatment for a health issue is a problem for some Black communities ($88\%$ [$$n = 135$$/155] pre-test to $98\%$ [$$n = 64$$/65] post-test). In the last session, participants exhibited an increased knowledge of COVID-19 long hauler health issues, higher rates of vaccination, and awareness of PCORI research opportunities. Participants also reported increased interest in serving as a patient advisor with AAN ($25\%$ [$$n = 27$$/117] pre-test to $31\%$ [$$n = 17$$/68] post-test).Table 4“Black People Like Me” virtual conference session pre- and post-test result highlightsMonth, session noAssessment questionAnswer choicesPre-testResponsesN (%)*Post-testResponsesN (%)*January, Session 2N = 71N = 40“Asthma exacerbation” is when asthma symptoms get worse. You may need steroids or to go to the hospitalTrue67 ($94\%$)39 ($100\%$)False4 ($6\%$)0The death rate for African Americans with Asthma compared to Caucasians is:No difference2 ($3\%$)02 times less7 ($10\%$)1 ($3\%$)2–3 times greater43 ($61\%$)31 ($78\%$)10 times greater19 ($27\%$)8 ($20\%$)Patient Centered Outcomes Research (PCOR)Is how research findings move into real world practice13 ($19\%$)6 ($15\%$)Helps people and their caregivers communicate and make informed decisions, allowing their voices to be heard in assessing the value of health care41 ($59\%$)28 ($70\%$)Research facts that make results true04 ($10\%$)The study of patient experience16 ($23\%$)2 ($5\%$)Asthma patients can participate as research advisorsTrue68 ($96\%$)36 ($90\%$)False3 ($4\%$)4 ($10\%$)I don’t know00February, Session 3N = 93N = 116Once I have taken the COVID-19 vaccine I no longer need to wear a mask or distance myself from others because I can’t catch COVID-19True2 ($2\%$)1 ($1\%$)False90 ($98\%$)115 ($99\%$)*It is* important to take ________ doses of the Moderna COVID-19 vaccine for it to properly work? ( choose only one)1 dose1 ($1\%$)1 ($1\%$)2 doses91 ($98\%$)115 ($99\%$)3 doses1 ($1\%$)04 doses00The most common side effects people are reporting from getting the COVID-19 vaccine are: (select all that apply)Sore arm80 ($86\%$)107 ($92\%$)Flu like symptoms for 24–48 h61 ($66\%$)81 ($70\%$)Nose bleeds01 ($1\%$)Tiredness/fatigue60 ($65\%$)84 ($72\%$)March, Session 4N = 68N = 38PCOR is an acronym for:Project center on research2 ($3\%$)0Patient-centered outcomes research63 ($93\%$)36 ($95\%$)Patient centers on research3 ($4\%$)2 ($5\%$)None of the above00PCORI has training sessions/videos to help train patients to learn how they can be more involved in all stages of researchTrue68 ($100\%$)38 ($100\%$)False00There is still hesitancy in the Black community to get the COVID-19 vaccine. Will you take it or have you taken it?Yes, I will take it when it is available to me33 ($49\%$)12 ($32\%$)Yes, I have already taken it23 ($34\%$)19 ($50\%$)No, I will not take it12 ($18\%$)7 ($18\%$)April, Session 5N = 155N = 65Who has the greatest knowledge in the engagement process?Government officials, researchers30 ($19\%$)5 ($8\%$)Doctors, nurses64 ($41\%$)17 ($26\%$)Community – you54 ($35\%$)41 ($63\%$)I don’t know7 ($5\%$)2 ($3\%$)When you visit your doctor or health care provider, how do you generally decide what treatment or medicines you will take for your condition?The doctor knows best about what I need and prescribes it53 ($34\%$)9 ($14\%$)The doctor explains all the options available to me, we discuss the options together and I make the decision based on what is right for me82 ($53\%$)51 ($79\%$)I request to take what my neighbor is taking because it is less expensive3 ($2\%$)0The decision is based on what my insurance company will cover13 ($8\%$)3 ($5\%$)Other3 ($2\%$)2 ($3\%$)Is getting health care or treatment for a health issue a problem for some Black communities?Yes135 ($88\%$)64 ($98\%$)No12 ($8\%$)1 ($2\%$)I don’t know6 ($4\%$)0May, Session 6N = 117N = 68What are some commonly reported lingering health issues from having COVID-19? ( Choose all that apply)Anxiety60 ($51\%$)46 ($68\%$)Fatigue86 ($74\%$)56 ($82\%$)Racing heartbeat49 ($42\%$)42 ($62\%$)Brain fog – problems with memory or concentration51 ($44\%$)51 ($75\%$)Shortness of breath99 ($85\%$)57 ($84\%$)Have you received the COVID-19 vaccine? ( Choose only one)Yes79 ($68\%$)54 ($81\%$)No26 ($22\%$)7 ($11\%$)No access to the vaccine where I live6 ($5\%$)0Do not plan to take it1 ($1\%$)1 ($1\%$)Waiting for my second shot5 ($4\%$)5 ($7\%$)Any member of a community may participate in a PCORI research study or engagement award. ( Choose only one)True92 ($79\%$)57 ($84\%$)False25 ($21\%$)11 ($16\%$)What are some examples of how you would like to get “involved” after being a part of these sessions? ( Choose all that apply)Help out with Community asthma, COVID outreach/education85 ($77\%$)38 ($69\%$)Sign up to become an Allergy & Asthma Network Volunteer41 ($37\%$)19 ($35\%$)Sign up as an Allergy & Asthma Network Patient Advisor27 ($25\%$)17 ($31\%$)Join the Allergy & Asthma Network Asthma360 Research Registry49 ($45\%$)17 ($31\%$)Contact PCORI about becoming a PCORI Patient Advisor25 ($23\%$)12 ($22\%$)Sign up and participate in a *Research focus* group or study62 ($56\%$)26 ($47\%$)Contact Allergy & Asthma Network for more information about how to become more involved in helping others, telling my story, etc43 ($39\%$)19 ($35\%$)*Monthly totals may not be uniform due to percentile rounding to whole numbers and participants skipping questions
## Social media engagement
Facebook received more than 12,000 impressions and an engagement of over 225 accounts during the BPLM series, Instagram received over 6900 impressions and an engagement of over 100 accounts during the BPLM series, and Twitter received over 53,000 impressions and engagement of over 580 accounts during the BPLM series. The private BPLM Facebook group currently has 132 members, and from March 2022-March 2023, 94 posts have been created, there have been 311 comments from members, and 566 reactions from members on the posts.
## Discussion
BPLM demonstrated that a virtual platform can be successful in engaging Black communities. The virtual conference far exceeded the goal of 100 participants. Clearly the community was curious about getting involved in research, and participants took time out of their day to attend because the sessions focused on questions that were important to them. Current, real-world conversations were held that had an impact on participant perceptions. One of the sessions was provided by a representative from PCORI that explained how and where to find more information about engaging as a patient partner in PCOR and clinical research. Many people had not considered their potential role as a partner in research, and this topic could be a focus for future sessions. Increasing patient involvement in research can identify new areas of concern or for health care improvement that may not be considered by health care providers and researchers [13]. The sessions also changed perceptions about how misinformed or uninformed participants were about COVID-19 and COVID-19 vaccines.
The intention of BPLM was to inform participants how they could become involved in PCOR. After the program, participants were added to a distribution list for the AAN research newsletter, and when opportunities to participate in research were identified, this information was sent to participants using the same distribution list. The BPLM Facebook group also posts information about research participation opportunities. To date there has been no follow-up to see if any participants of BPLM actually chose to become engaged in PCOR, and this may be a subject of future research.
There were some challenges to the BPLM program. Early on the patient advisors from the PAG were not as engaged as expected. The Project Lead and Manager held one-on-one meetings with each of the patient advisors and held a few monthly meetings with just the patient advisors to get their feedback about the project and to help promote more engagement in the group. Eventually the group was more vocal and bonded with time. Technical issues were the main challenge with the virtual sessions. The PAG learned from the first session that more direct and detailed instructions to presenters and panelists was needed in how to use the virtual platform. A speakers instruction list was created for the team to use with each session and the number of presentation and technical issues was reduced for the rest of the sessions.
There was a concern that the virtual session format would not keep the attention of the participants. Different types and styles of educating and communicating with the participants (e.g., patient storytelling, chat, interactive polling questions) were used to help keep participants from getting distracted when listening or watching a session online rather than in-person. However, the virtual format cannot replace the full personal approach of an in-person conference. Judging from the response and attendance with these virtual sessions, a hybrid of both virtual and in-person conferences may be the best approach for future conferences.
Gaining the trust of the community was a goal of BPLM. Research shows that compared with White individuals, Black individuals are more likely to trust health information from community stakeholders, such as religious organizations and charitable organizations [14]. Therefore, such organizations should be considered to assist in dissemination of health care information targeted toward the Black community. Using the principles from NOML, clergy assisted and led the way in engaging their community and building a foundation of trust in BPLM. In addition, the patient advisors in the PAG proved critical to creating genuine, honest, and relatable sessions that people responded to and were excited about throughout all six sessions.
## Conclusions
BPLM successfully engaged Black communities in the topics of asthma, COVID-19, and PCOR using a virtual platform. Incorporating clergy and religious organizations was critical in developing the trust of the Black community towards BPLM. It is our hope that BPLM will have far reaching and long-lasting effects in the Black community, such as increasing patient involvement in PCOR and increasing the sense of empowerment during an individual’s health care journey. We anticipate that learnings from BPLM will lead to future engaging programs.
## Disclaimer
The statements presented in this publication are solely the responsibility of the author(s) and do not necessarily represent the views of the Patient-Centered Outcomes Research Institute® (PCORI®), its Board of Governors or Methodology Committee.”
## References
1. 1.National Center for Health Statistics. Percentage of current asthma for adults aged 18 and over, United States, 2019.. National Health Interview Survey; 2019 [cited 2022 June 14]. Available from: https://wwwn.cdc.gov/NHISDataQueryTool/SHS_2019_ADULT3/index.html.
2. 2.Asthma and African Americans. U.S. Department of Health and Human Services Office of Minority Health; 2021 [cited 2022 June 14]. Available from: https://minorityhealth.hhs.gov/omh/browse.aspx?lvl=4&lvlid=15.
3. Hsu S, Rosen KJ, Cupertino A, Temple L, Fleming F. **Generalizability of randomized controlled trials in rectal cancer**. *J Gastrointest Surg* (2022.0) **26** 453-465. DOI: 10.1007/s11605-021-05192-x
4. Scierka LE, Jelani QU, Smolderen KG, Gosch K, Spertus JA, Mena-Hurtado C. **Patient representativeness of a peripheral artery disease cohort in a randomized control trial versus a real-world cohort: the CLEVER trial versus the PORTRAIT registry**. *Contemp Clin Trials* (2022.0) **112** 106624. DOI: 10.1016/j.cct.2021.106624
5. Luce BR, Simeone JC. **How different is research done by the Patient-centered Outcomes Research Institute, and what difference does it make?**. *J Comp Eff Res* (2019.0) **8** 1239-1251. DOI: 10.2217/cer-2019-0054
6. Bingham CO, Bartlett SJ, Merkel PA, Mielenz TJ, Pilkonis PA, Edmundson L. **Using patient-reported outcomes and PROMIS in research and clinical applications: experiences from the PCORI pilot projects**. *Q Life Res* (2016.0) **25** 2109-2116. DOI: 10.1007/s11136-016-1246-1
7. 7.Patient Centered Outcomes Research Institute. 2022 [cited 2022 June 14]. Available from: https://www.pcori.org/about/about-pcori.
8. 8.2020 Drug trials snapshots summary report. US Food and Drug Administration; 2021 [cited 2022 June 14]. Available from: https://www.fda.gov/media/145718/download.
9. Hutchings A, Neuburger J, van der Meulen J, Black N. **Estimating recruitment rates for routine use of patient reported outcome measures and the impact on provider comparisons**. *BMC Health Serv Res* (2014.0) **14** 66. DOI: 10.1186/1472-6963-14-66
10. 10.National Academies of Sciences, Engineering, and Medicine Improving representation in clinical trials and research: Building research equity for women and underrepresented groups 2022. Washington, DC: The National Academies Press. 10.17226/26479.
11. Clark LT, Watkins L, Piña IL, Elmer M, Akinboboye O, Gorham M. **Increasing diversity in clinical trials: overcoming critical barriers**. *Curr Probl Cardiol* (2019.0) **44** 148-172. DOI: 10.1016/j.cpcardiol.2018.11.002
12. Harris D, Graham L, Teague WG. **Not one more life: a health and faith partnership engaging at-risk African Americans with Asthma in Atlanta**. *Ann Am Thorac Soc* (2019.0) **16** 421-425. DOI: 10.1513/AnnalsATS.201803-166IP
13. Moise RK, Srithanaviboonchai K, Alsolami A, Soares MH, Elewonibi BR, Tolmasky ME. **The patient-provider continuum of care: narratives of people living with comorbid HIV and diabetes in Northern Thailand**. *J Patient Exp* (2020.0) **7** 749-757. DOI: 10.1177/2374373519882226
14. Fareed N, Swoboda CM, Jonnalagadda P, Walker DM, Huerta TR. **Differences between races in health information seeking and trust over time: evidence from a cross-sectional, pooled analyses of HINTS data**. *Am J Health Promot* (2021.0) **35** 84-92. DOI: 10.1177/0890117120934609
|
---
title: Changes, differences, and factors of parenthood in high-risk pregnant women
and their partners in Japan
authors:
- Eriko Kawamura
- Midori Asano
journal: BMC Pregnancy and Childbirth
year: 2023
pmcid: PMC10037369
doi: 10.1186/s12884-023-05519-3
license: CC BY 4.0
---
# Changes, differences, and factors of parenthood in high-risk pregnant women and their partners in Japan
## Abstract
### Background
Various stressors exists for pregnant women worldwide, especially negative social and environmental influences that can increase the number of high-risk pregnant women. These may cause a difficult transition to parenthood for women and their partners. However, limited studies have focused on and examined parenthood. Therefore, this study aimed to identify the changes in parenthood from pregnancy to post-discharge after childbirth among high-risk pregnant women and their partners, as well as the presence or absence of gender differences and the factors associated with parenthood.
### Methods
This longitudinal quantitative study used a self-administered anonymous questionnaire distributed among 127 pregnant women and their partners who visited a high-risk pregnant outpatient clinic. The Scale of Early Childrearing Parenthood (SECP; three subareas, 33 items) was administered thrice: during pregnancy (T1), after childbirth (T2), and after discharge (T3).
### Results
The analysis included 85 T1 (37 fathers and 48 mothers), 36 T2 (13 fathers and 23 mothers), and 31 T3 (11 fathers and 20 mothers) responses. There was a significant increase in the SECP scores for both parents from T1 to T3. Mothers had a greater increase in the SECP scores from T1 to T2 than fathers. In addition, fathers’ mean SECP scores at T1 and T2 were higher compared with those of the mothers. Mothers’ and fathers’ SECP scores at each time point showed no significant differences.
At all time points, the SECP scores were commonly and significantly associated with infertility treatment, physical and mental condition, postpartum depression at T2, and parenting stress at T3.
### Conclusions
Because parenthood in the infertility treatment group was significantly higher throughout the series, we need to support such couples so that childbirth does not become their main goal. We suggest interventions for factors that impede parenthood development, understand the various backgrounds of the parents, and support the couple individually while also considering them as a unit.
### Supplementary Information
The online version contains supplementary material available at 10.1186/s12884-023-05519-3.
## Background
For referencing purposes, the global total fertility rate (TFR) was forecasted to be 1.66 in 2100, and involved 23 countries, including Japan, Thailand, and Spain, forecasted to have a population decline of < $50\%$ from 2017 to 2100, leading to severe population decline worldwide [1]. In Japan, the TFR is expected to be as low as 1.34 in 2020, and declining fertility is a major problem [2]. With the increase in the maternal age at delivery and the development of advanced reproductive medical technology, "high-risk pregnant women" with some risk of health problems, worsening complications, or death for both mother and child is increasing [3]. Specifically, the increasing trend toward late marriages has also increased reliance on reproductive healthcare, with $18.2\%$ of couples currently undergoing testing and treatment for infertility [4]. Evidently, infertility is not a women-only problem, as approximately half of the causes of infertility rely on men [5]. In addition, the environment surrounding pregnant women today is likely to lead to increased isolation and the burden of child care [6, 7]. Particularly in high-risk pregnant women, such social, physical, and psychological factors as health concerns about themselves and their foetus may increase stress and difficulties in childcare. Since the coronavirus disease 2019 pandemic, fathers cannot be involved in childcare from birth until discharge due to limited access to medical facilities and experience increased stress in childcare [8–10]. Japan faces the pressing and important issues of declining fertility and childrearing difficulties. Among the various barriers, how the transition to parenthood relates to both men and women must be examined.
Thus, we focused on ‘parenthood’ among high-risk pregnant women and their partners to examine parental awareness and support in acquiring and developing their parenting competencies. Belsky [11] identified three parenting determinants: parental psychological resources, child characteristics, and contextual sources of support. Parenthood is a parenting role that is common to men and women and leads to improved parenting skills [12, 13]. It is characterized by love and respect for self as well as compassion and tenderness toward the child, which develops with the progression of life stages and is demonstrated in the protection and care of the child during pregnancy, delivery, and child rearing [13]. Pregnancy and the transition to parenthood are major adjustment periods for many adults, with important implications for new parents, couple relationships, and infant development [14]. Previous studies with high-risk pregnant women report postpartum depression (PPD) [15–17]; specifically, the association between depressed parents and parenting stress [18–20] and attachment and feelings toward the child [21, 22]. However, studies focusing on the parenthood of high-risk pregnant women are limited. Furthermore, studies focusing on fathers as partners are limited, and even fewer have identified differences in parenthood between mothers and fathers and the factors associated with parenthood.
Our objectives were as follows: [1] To examine how fathers’ and mothers’ parenthood changed from the gestational period to one month after the child’s discharge from the hospital and [2] to examine whether there were differences in parenthood between fathers and mothers during pregnancy, within seven days of delivery, and one-month post-discharge. Furthermore, we examined the factors associated with parenthood in each period. For mothers, parent–child attachment is formed during pregnancy [23], and while gaining a sense of togetherness through pregnancy, they can gradually develop affection for their children through the experience of delivery and breastfeeding [24]. Notably, paternal education does not include fathers, leading to a lack of preparation for parenthood for them [14]. Therefore, considering hypotheses for [1] and [2], we expect that changes in the development of parentage will be significantly greater for mothers. Furthermore, we expect that multiple factors are associated with the elements of parenthood [11–14], as indicated above since they are not composed solely of factors such as parenting ability and feelings toward the child.
These findings will aid nursing care that promotes the acquisition and development of parenthood from the early prenatal period. This may positively impact the child’s direct and indirect growth and development, as well as the parents’ growth and development. In addition, in clinical settings where there is a chronic lack of human resources and time limitations, validation of an effective parenthood development will contribute to the family support skills of healthcare workers and the well-being of children and their families. Furthermore, it is important to examine differences in the parenthood of mothers and fathers, as studying gender differences can lead to gender innovation [25].
## Design
This longitudinal research study used a self-administered anonymous questionnaire survey.
## Participants and procedures
Participants were recruited from two high-risk pregnant women’s hospitals in Japan between April and December 2021. Pregnant women and their partners (Japanese) who visited the hospital were considered eligible to be candidates for research collaboration by the outpatient physician, and those who agreed to cooperate were recruited as participants. We excluded pregnant women’s partners who were not related to the foetus.
We requested the target facilities’ co-operation for this study and explained the study in writing and orally. After consent was obtained from the facilities, approval was obtained from the ethical review committee of each facility. Subsequently, we asked the attending physicians in the outpatient clinics to list the potential participants, which are all couples who fulfilled the above criterion.
Finally, the researcher explained the study in writing and orally to the patients who agreed to participate, obtained their consent, and gave them a questionnaire to complete anonymously.
Participants completed a questionnaire at three-time points: during pregnancy (Time point 1 [T1]), within seven days of delivery (Time point 2 [T2]), and within one month after the child’s discharge (Time point 3 [T3]), and mailed it each time.
## Participant background
We surveyed 24 items divided into three categories of participant backgrounds. The items were as follows: [1] Parents’ Characteristics: gender, age, job status, education level, family patterns, years of marriage, infertility treatment, number of children, pregnancy weeks at the time of the survey, mental and physical conditions, and fostered environment; [2] Child characteristics: birth weeks, birth weight, sex, multiple births, NICU hospitalisation, and length of hospitalisation; [3] Childcare and housework: child handling experience, anxiety for childcare, anxiety except for childcare, husbands work- wives do housework, male participation in childcare, childcare leave, and participation in childcare during hospitalisation.
## Measurement Scales
The roots used in this study are shown in Table 1. The Scale of Early Childrearing Parenthood (SECP) [26, 27] was administered at T1, T2, and T3. The scale was developed by Ohashi and Asano [26] based on the maternal role attainment theory of Mercer [28], and its reliability and validity have been confirmed. The SECP has been increasingly used in Japan in recent years. The SECP assessed parental qualities and divided them into awareness of the child and self-awareness. Awareness of oneself comprised the state of the parental and non-parental roles, combined with an awareness of the child, to form the scale’s three subdomains. This self-administered questionnaire consisted of 33 items: 13, 9, and 11 items on the state of the parental role, non-parental role, and awareness of the child, respectively. A high total score indicated high awareness of oneself and the child, and there was no cut-off value. Table 1 Tools in useTool NameNumber of ItemsSubdomains（Number of Items）Answer FormatCronbach’s αSECP33The state of the parental role[13]5-point Likert scale0.82 -0.94The state of the non-parental role[9]Awareness of the child[11]EPDS10–4-point scale0.82PSI-SF19Parent[10]5-point Likert scale0.66 - 0.88Child[9] The state of the parental role comprised parental satisfaction, concern for the child, and the relationship with the child. The state of the non-parental role comprised feelings of satisfaction with self-independence as a parent, self-acceptance, and relationship with society. Awareness of the child covered affection towards the child, understanding of the child and their developmental growth, and child-rearing ability. The items were rated on a five-point Likert scale with the following response options: strongly agree, agree, neither, disagree, and strongly disagree. The total scores were calculated by adding each subscale’s scores, ranging from 33 to 165. Cronbach’s alpha for the SECP in the present study ranged from 0.82 to 0.94.
The Edinburgh Postnatal Depression Scale (EPDS) [29] was administered at T2. The Japanese version of the EPDS [30], originally developed by Cox, Holden, and Sagovsky [28], assessed PPD. This self-administered questionnaire consisted of ten items, each scored on a four-point scale (0–3). The scale’s reliability and validity have been well established [30]. If the total score was ≥ 9 points, or if the question item ‘I have had thoughts of hurting myself’ was ≥ 1 point, it was screened as possibly depressed. Cronbach’s alpha for the EPDS in this study was 0.82.
The Parenting Stress Index-Short Form Scale (PSI-SF) [31] was administered at T3. The Japanese version of the PSI-SF [31], originally developed by Abidin [32] and based on the full-length PSI [33], was used to assess parenting stress. This scale has been widely used, both nationally and internationally. This self-administered questionnaire consisted of 19 items and two subdomains: ten and nine items on the parent and child, respectively. The items were rated on a five-point Likert scale with the following response options: strongly agree, agree, neither, disagree, and strongly disagree. Higher scores indicated greater parenting stress. The total scores were calculated by adding each subscale’s scores, ranging from 19 to 95. The scale’s reliability and validity have been well established [29]. Cronbach’s alpha for the PSI-SF in the present study ranged from 0.66 to 0.88.
## Statistical analysis
Descriptive statistics were calculated to provide an overview of the participants. Descriptive data were expressed as means (range) for continuous variables and the number of persons (%) for nominal variables.
First, a paired t-test was performed to determine the change in the fathers’ and mothers’ SECP total and three subscale scores during T1, T2, and T3. Next, a t-test was conducted to determine the differences between the fathers’ and mothers’ SECP total and the three subscale scores at T1, T2, and T3. Finally, to identify the factors associated with the SECP at T1, T2, and T3, t-tests were conducted to compare two categories (e.g., father vs. mother), Pearson’s product-moment correlation analysis for continuous variables (e.g., age, EPDS, and PSI-SF), while one-way analysis of variance was used with more than two categories (e.g., job status). Assumptions of normality were evaluated. For items that showed significant differences from the SECP, Cohen’s d was calculated as an effect size to clarify the size of the difference in substantive means [34]. Effect sizes of 0.20, 0.50, and 0.80 were considered small, moderate, and large, respectively [34]. The criterion for statistical significance was set at a two-sided significance probability of less than $5\%$. All statistical analyses were performed using SPSS (IBM SPSS Statistics 28 for Windows, Tokyo, Japan).
This study was approved by the Bioethics Review Committee of the Nagoya University Graduate School of Medicine (2020–0322) and Ogaki municipal hospital Institutional Review Board (20,200,528–1). All methods were carried out in accordance with relevant guidelines and regulations (such as the Declaration of Helsinki).
## Results
The recruitment process is shown in Fig. 1. In total, 127 questionnaires were distributed. Of these, 86 ($67.7\%$) responses were collected, 85 ($98.8\%$) were valid at T1, 37 ($29.1\%$) were collected, 36 ($28.3\%$) were valid at T2, and 31 ($24.4\%$) valid responses were collected at T3.Fig. 1Flow chart of participants. High-risk pregnant women who visit the research facilities are defined as those who may be at some risk of health problems, worsening complications, or death for both mother and child during pregnancy, childbirth, or after delivery
## Participants’ background
Table 2 shows the background information of each participant at T1, T2, and T3. In T1, more than half ($\frac{48}{85}$, $56.5\%$) in T1, more than $60\%$ ($\frac{23}{36}$, $63.9\%$) in T2, and more than $60\%$ ($\frac{20}{31}$, $64.5\%$) in T3 were mothers. Approximately half ($\frac{44}{85}$, $51.8\%$) were the first child in T1, less than half ($\frac{17}{36}$, $47.2\%$) in T2, and approximately half ($\frac{16}{31}$, $47.2\%$) in T3 were the first child. One quarter ($\frac{9}{36}$, $25.0\%$) were screened for risk of PPD (hereafter referred to as EPDS positive) and admitted to the NICU in T1 and T2, respectively, and less than $20\%$ ($\frac{5}{31}$, $25.0\%$) were admitted to the NICU in T3. The mean PSI-SF scores were 37.77 (22–55) total points. Table 2 Participants’ BackgroundT1a ($$n = 85$$)T2a ($$n = 36$$)T3a ($$n = 31$$)n(%)M(range)n(%)M(range)n(%)M(range)Parents'Characteristics Gender Father37(43.5)13(36.1)11(35.5) Mother48(56.5)23(63.9)20(64.5)Age（years old)32.61[21-50]33.94[22-50]33.84[22-50] Job Status b Regular 54(63.5)22(61.1)20(64.5) Non-regular9(10.6)5(13.9)3(9.7) Self-employment4(4.7)2(5.6)2(6.4) Unemployed 18(21.2)7(19.4)6(19.4) Education Level Less than high School diploma22(25.9)9(25.0)7(22.6) College/University or higher63(74.1)27(75.0)24(77.4) Family Patterns Nuclear 75(88.2)32(88.9)29(93.5) Extend10(11.8)4(11.1)2(6.5) Years of Marriage Less than 3 years35(41.2)12(33.3)11(35.5) More than 3 years50(58.8)24(66.7)20(64.5) Infertility Treatment c Yes29(34.1)11(30.6)9(29.0) Non-regular56(65.9)25(69.4)22(71.0) Number of Children First child44(51.8)17(47.2)16(51.6) Second child or more41(48.2)19(52.8)15(48.4) Pregnancy Weeks at Time of Survey Less than 28 weeks31(36.5)−− More than 28 weeks54(63.5)−− Mental and Physical Condition Both in good condition65(76.5)26(72.2)21(67.7) Either or both in poor condition20(23.5)10(27.8)10(32.3) Fostered Environment Attachment82(96.5)34(94.4)30(96.8) No attachment3(3.5)2(5.6)1(3.2) Child Characteristics Birth Weeks (weeks)37.78[26-41]37.87[26-41] Birth Weight Less than 2500g−9(25.0)6(19.4) More than 2500g−27(75.0)25(80.6) Sex Boy−21(58.3)19(61.3) Girl−15(41.7)12(38.7) Multiple Births Yes−3(8.3)0(0.0) No−33(91.7)31(100.0) NICU Hospitalization Yes−9(25.0)5(16.1) No−27(75.0)26(83.9) Length of Hospitalization Less than 1 week−−23(74.2) From 1 week to 1 month−−5(16.1) More than 1 month−−3(9.7)Childcare and Housework Child Handling Experience Yes63(74.1)30(83.3)25(80.6) No22(25.9)6(16.7)6(19.4) Anxiety for Childcare Yes40(47.1)13(36.1)13(41.9) No45(52.9)23(63.9)18(58.1) Anxiety Except for Childcare Yes21(24.7)6(16.7)6(19.4) No64(78.3)30(83.3)25(80.6) Husbands Work, Wives Do Housework Agree14(16.5)8(22.2)4(12.9) Against71(83.5)28(77.8)27(87.1) Male Participation in Childcare Very much agree66(77.6)28(77.8)24(77.4) More or less agree19(22.4)8(22.2)7(22.6) Childcare Leave Yes−19(52.8)18(58.1) No−17(47.2)13(41.9) Participation in Childcare during Hospitalization Yes−−28(90.3) No−−3(9.7) EPDS Score−4.75[0-14]− Negative−27(75.0)− Positive−9(25.0)− PSI-SF Overall Score−−37.77[22-55] Child Aspects−−18.84[9-27] Parent Aspects−−18.94[10-30]a"T" is the assessment time point, where T1 is during pregnancy, T2 is seven days after delivery, and T3 is within one month after the child’s dischargebRegular employment includes company employees and civil servants. Unemployed include housewives and househusbandscFertility treatment includes various methods such as hormonal therapy, timing methods, AIH (Artificial insemination), IVF (In vitro fertilisation), and ICSI (intra-cytoplasmic sperm injection) *The data* analysis addressed the three major aims. First, we addressed how father-mother parenthood changed from pregnancy to post-discharge of the infants. We also examined whether fathers and mothers had differences in parenthood during the three-time points of pregnancy, postpartum, and post-discharge. The factors associated with parenthood at each time point were also examined.
## Changes in parenthood for fathers and mothers
The 31 participants who participated in all three periods (fathers, 11; mothers, 20) were included in the analysis. Changes in the SECP scores for fathers and mothers from T1 to T3 are shown in Table 3 (Supplemental Fig. S1- a, b, c, and d).Table 3 Comparison of SECP scores across assessment times for father and motherOverall ScoresState of the Parental RoleState of the Non-Parental RoleAwareness of the ChildM (SD) T1T2M (SD)T1T2M (SD)T1T2M (SD)T1T2pdpdpdppdppdpdAll（$$n = 31$$） T1112.74(13.12)−−−−48.65(6.04)−−−32.06(5.90)−−−32.03(6.66)−−−− T2125.48(13.52)＜0.0011.02−−54.55(6.22)＜0.0011.02−34.13(5.13)0.0220.44−37.13(7.44)＜0.0010.90−− T3129.48(15.46)＜0.0011.080.0370.39055.74(6.39)＜0.0011.050.21034.58(5.88)0.0030.580.37939.32(6.98)＜0.0010.930.0300.41Father（$$n = 11$$） T1117.18(11.32)−−−−49.55(6.44)−−−33.36(4.70)−−−34.27(6.48)−−−− T2125.82(11.55)0.0290.77−−52.27(6.62)0.052−−34.18(5.40)0.624−−39.36(6.27)0.0170.87−− T3129.00(11.03)0.0250.790.282−54.55(6.50)0.0230.790.13135.09(5.59)0.271−0.35639.09(5.75)0.0440.690.841−Mother（$$n = 20$$） T1110.30(13.66)−−−−48.15(5.91)−−−31.35(6.47)−−−30.80(6.58)−−−− T2125.30(14.45)＜0.0011.17−55.80(5.78)＜0.0011.30−34.10(5.12)0.0110.6335.90(7.89)＜0.0010.89−− T3130.00(18.17)＜0.0011.210.068−56.40(6.39)＜0.0011.190.63234.30(6.15)0.0500.720.74339.45(7.71)＜0.0011.070.010.66The statistical method is a paired T-test From T1 to T2, the ‘overall score’ and ‘awareness of the child’ increased significantly for both fathers and mothers by more than moderate margins. In contrast, the ‘state of the parental role’ and the ‘state of the non-parental role’ increased significantly by more than moderate margins only for mothers.
From T2 to T3, ‘overall score,’ ‘state of the parental role,’ and ‘state of the non-parental role’ did not increase significantly for either parent. However, ‘awareness of the child’ increased significantly by a moderate margin only for mothers.
From T1 to T3, the ‘overall score,’ ‘state of the parental role,’ and ‘awareness of the child’ increased significantly for both parents, with more than moderate differences.
In contrast, the ‘state of the non-parental role’ increased only for mothers by moderate margins, while fathers did not change.
## Differences in parenthood between fathers and mothers
The details are presented in Table 4. The mean SECP scores at T1, T2, and T3 were not significantly different between parents for ‘overall score’ and all three subscale scores. Table 4 Differences in SECP scores between father and mother at each assessment time-pointOverall ScoresState of the Parental RoleState of the Non-Parental RoleAwareness of the ChildrenM (SD)pM (SD)pM (SD)pM (SD)pT1($$n = 85$$) Father120.68(15.83)0.48650.27(6.82)0.46234.03(5.57)0.16736.38(7.97)0.311 Mother118.15(17.04)51.35(6.61)32.17(6.48)34.63(7.70)T2($$n = 36$$) Father125.62(10.60)0.97652.31(6.07)0.08434.08(5.14)0.85039.23(5.82)0.239 Mother125.48(13.86)55.83(5.49)33.74(5.08)36.35(7.64)T3($$n = 31$$) Father129.00(11.03)0.90054.55(6.50)0.44935.09(5.59)0.72639.09(5.75)0.897 Mother130.00(18.17)56.40(6.39)34.30(6.15)39.45(7.71)The statistical method is a T-test; SD Standard deviation, M Mean
## Factors related to parenthood
Table 5 shows the overall SECP scores and associated factors at each assessment. The following two items significantly differed from the total SECP scores for T1, T2, and T3: infertility treatment (higher score for ‘yes’) and mental and physical condition (higher score for ‘both in good condition’).Table 5 Factors associated with SECP (overall score) at each assessment time pointAllT1T2T3M (SD) pdM (SD) pdM (SD) pdrrr119.25(16.48)125.53(12.62)129.48(15.46)Infertility Treatmenta Yes126.69(15.45)0.0020.72132.82(14.37)0.0190.89140.78(14.21)0.0071.15 Non-regular115.39(15.77)122.32(10.54)124.86(13.69)Pregnancy Weeks at Time of Survea Less than 28 weeks125.32(15.14)0.0050.61−− More than 28 weeks115.76(16.32)−−Mental and Physical Conditiona Both in good condition121.97(15.98)0.0050.73128.42(12.69)0.0240.88133.95(14.03)0.0170.97 Either or both in poor condition110.40(15.25)118.00(9.12)120.10(14.63)Anxiety for Childcarea Yes113.88(17.36)0.0040.66120.08(10.10)0.050−123.78(14.25)0.191− No124.02(14.20)128.61(13.05)132.05(16.09)Anxiety Except for Childcarea Yes110.81(16.34)0.0060.71121.17(9.50)0.361−120.33(20.75)0.107− No122.02(16.67)126.40(13.11)131.68(13.52)EPDSb Score−-0.430.009−PSI-SFb Overall Score−−-0.65＜0.001 Child Aspects−−-0.57＜0.001 Parent Aspects−−-0.59＜0.001aT-test; bPearson product-moment correlation analysis; SD Standard deviation, M Mean, EPDS The Edinburgh Postnatal Depression Scale, PSI-SF The Parenting Stress Index-Short Form Scale For T1, three additional items were significantly different from the SECP: Pregnancy weeks at the time of the survey (higher score for ‘less than 28 weeks’), anxiety regarding childcare (higher score for ‘no’), and anxiety, except for childcare (higher score for ‘no’).
For T2, the EPDS (negative correlation) and for T3, the PSI-SF (negative correlation for both the overall score and scores on the children’s and parents’ aspects) were significantly different from the SECP.
## Discussion
This is the first study to measure parenthood from pregnancy for high-risk pregnant women and their partners, which discussed the changes in parenthood and the associated factors for fathers and mothers.
In this study, both fathers’ and mothers’ SECP scores increased significantly from pregnancy till the child’s discharge. During pregnancy, the SECP scores for mothers were lower than for fathers. However, the SECP scores for mothers increased significantly after childbirth, especially in the ‘state of the parental role,’ which was considerably higher than that of fathers. This may be because of the mothers’ experiences of childbirth. The results of the present study were consistent with the fact that the development of parenthood towards infants was more pronounced in mothers than in fathers and that mothers who were more involved in child-rearing underwent greater changes as a result of becoming parents [35]. Contrary to the hypothesis, an interesting finding of this study was that the mean SECP scores (which excluded ‘state of the parental role’) during pregnancy and after birth were higher for fathers than for mothers. Although it was generally believed that mothers were more parental than fathers [36], the opposite was true in this study. This may be characteristic of high-risk pregnant couples. The promotion of communication between the couple during the pregnancy [37, 38] and the enhancement of support for the marital relationship [39] have contributed to the men’s adjustment to fatherhood. In contrast, wives perceived parenthood as more constraining and burdensome than their husbands, which may be because wives experienced pregnancy and childbirth, undertook physiological changes, and were subject to actual constraints [40]. In particular, it has been reported that the mother’s illness and severe fatigue in older primiparas were related to higher parental stress two months after childbirth [41]. In this study, the fathers’ parenthood was higher than that of the mothers, possibly because the couples had communicated and solved various problems before the child’s birth. Furthermore, high-risk pregnant women were especially subject to greater physical and emotional strain than the general pregnant population.
There was no significant difference in the SECP scores between high-risk pregnant women and their partners during pregnancy, after childbirth, and after discharge, which indicated no gender difference in parenthood before and after childbirth. This result was consistent with previous studies that showed no gender differences in SECP scores from immediately after childbirth to the early parenting period [42, 43]. Thus, both fathers and mothers followed similar parental transition processes. However, these results were obtained only after childbirth. This result indicated no gender differences in parenthood during pregnancy or even before birth, which was a novel finding.
Finally, we found that fertility treatment and mental and physical conditions were commonly associated with the parenthood of high-risk pregnant women and their partners throughout the continuum from pregnancy to their child’s discharge. In addition, their parenthood was also related to the EPDS score after childbirth and the PSI-SF score after the child’s discharge.
The infertility treatment rate for our participants in this study was > $30\%$, which was higher than that of the general group. Women who conceived and gave birth after infertility treatment viewed their infertility experiences as meaningful [44]. Furthermore, for patients who underwent infertility treatment, ‘having a child’ was sometimes described as a goal to be fulfilled [45]. The increased self-esteem gained from fulfilling goals may have influenced the parenthood of infertility-treated couples. In addition, the shared stress of infertility may even stabilize marital relationships [46–48], resulting in increased parenthood. This study showed significantly higher SECP scores in the infertility treatment group at all time points from conception to post-discharge. However, it cannot be denied that childbirth tended to be their goal. Most studies indicated no difference in anxiety regarding pregnancy, foetal development, and delivery between those who underwent fertility treatment and those who conceived naturally. However, when the items were examined, there were some differences, and the results were inconsistent [49]. In addition, in the infertility experience, overall, people were more likely to experience poorer well-being (e.g., higher depression and negative affect) when they faced a blocked parenthood goal [50]. An examination of the congruence between partners’ perceived infertility-related stress and its relationship to marital adjustment and depression in infertile couples showed that couple incongruence was unrelated to depression in males and incongruence over relationship concerns. Furthermore, the need for parenthood was related to female depression [46]. Additionally, studies indicated that partners who underwent in vitro fertilization (IVF) might not have enough support from their closest social environments [51, 52]. Previous research has shown that infertility is stressful. Thus, while infertility experience may strengthen the couple’s bonds, it increases the stress caused by the vulnerability of support. Hence, whether infertility experience directly affects the development of parenthood and how parenthood might change should be examined.
The results of mental and physical conditions were similar to those of previous studies [53]. The support for mental health is important, especially since the EPDS and parenting stress were relevant for postpartum parenthood. Recently, PPD in fathers and mothers has been increasing, and its prevalence was not significantly different between men and women [54]. Paternal PPD is associated with relationships and physical health and has negative effects on children [55, 56]. Furthermore, it is also associated with PPD in mothers, and there was concern that the child-rearing environment may deteriorate if the couple suffered from mental illness simultaneously [55]. In addition, parenting stress more significantly affects anxiety than anger [57]. Therefore, physical and mental conditions and depressive symptoms, including parenting stress, should be considered together.
In the present study, only a few children were admitted to the NICU, who were low birth-weight babies. Japan has drastically reduced rates of maternal, perinatal, and neonatal or infant deaths, making it the safest country to give birth and raise a child [3]. Thus, the results indicate that the high-risk pregnant outpatient clinic functioned effectively and avoided risk. Moreover, there were no significant differences between parenthood and child factors. However, the child’s health was a major concern for high-risk pregnant women and their partners. According to previous studies, the risk of parental PPD was four to eighteen and three to nine times higher for mothers and fathers of VLBW infants, respectively, compared with mothers and fathers of term infants [16]. Therefore, it is necessary to emphasize the importance of assessing the parents’ mental state and the child’s health problems and providing nursing interventions.
## Practical implications
We suggest that couples treated for infertility in high-risk pregnant outpatient clinics require support so that childbirth does not become the goal and that parenthood, which increases through childbirth, does not decrease. For mothers, there is a need to intervene with an awareness of the importance of birth review as the delivery experience and immediate contact with the child may affect subsequent parenthood. Prior research showed that fathers perceived that perinatal health professionals viewed ‘mothers as a priority’ [58]. Thus, we suggest interventions that stimulate fathers with experiences that increase their awareness of parenthood, especially during the transition to parenthood before and after birth, since, unlike mothers, they undergo fewer physical changes. Furthermore, hospitalized high-risk obstetrical patients may commonly experience depression or anxiety symptoms and not receive treatment [59] and, therefore, may not be intervening despite the predicted high-risk factors.
As described above, parenthood may be impeded without third-party intervention for high-risk pregnant women and their partners. Therefore, we believe that healthcare providers must be aware of these issues and recognize the requirement for long-term involvement in individual problems, considering the various backgrounds of pregnant women and their partners. In particular, mental health assessments during the postpartum period should be promoted for both mothers and fathers to prevent parental psychological distress [60]. We suggest viewing parents as one unit rather than separate, as couples interacting with each other regarding mental health issues.
Finally, a significant solution to the global population decline, including Japan, is sustaining and enhancing female reproductive health [1]. The main target groups of reproductive health care are women, mothers, foetuses, and children, but it also includes men as reproductive and child-rearing partners [61]. Therefore, ongoing support for women and their partners, starting at the stage of fertility treatment prior to pregnancy, will contribute to the decline in fertility.
## Strengths and limitations
The strength of this study was that although support for expectant mothers was strengthened and focused on ‘seamless support’ from pregnancy, data on fathers were valuable in emphasizing the need for support. Including both fathers and mothers provided a more dynamic view of the family system. In addition, including data from the pregnancy period was novel as data using the SECP only covered the period from birth to the first year of life. Furthermore, this study added to the knowledge of parents with a history of infertility treatment. In Japan, the government started a system to provide universal health insurance coverage for infertility treatment in 2022 [62]. Since this demand is expected to increase in modern society, this study will support couples who overcome difficulties in becoming new parents.
This study has some limitations. First, the sample size was inadequate due to the rarity of the participants. During the survey period, Japan fell into the seventh wave of the coronavirus disease 2019 (COVID-19) pandemic, and direct recruitment has been impossible since because of the ban on researchers entering and leaving the facilities. Moreover, for the reasons stated above, the inability to directly remind participants at the time of delivery may have been the reason for the significant decrease in frequency rates during the T2 phase. Hence, the possibility that some items failed to show significant differences due to the insufficient sample size cannot be ruled out. In particular, the small number of treatment groups in the NICU was unexpected, and this requires further examination.
Second, in Japan, there is still an underlying cultural value of "men work, women raise children," a gender role division of labour. Therefore, these values may affect parenthood and may not be transferable to other cultures in other countries worldwide. This study focused on Japanese nationals; of all new-borns in Japan in 2017, less than $3\%$ had a foreign mother, a $26\%$ increase from the previous two decades [63]. Furthermore, this number is expected to increase. Therefore, to realize seamless support for all pregnant women giving birth in Japan, data from foreign-born participants are required, and must be considered by future studies.
## Conclusion
In this study, there was no difference in the parenthood of high-risk pregnant women and their partners, but the father's parenthood was higher than the mother's during pregnancy and after childbirth. Throughout the series, from pregnancy to discharge, parenthood was commonly associated with infertility treatment and physical and mental conditions.
The parenthood of the fertility treatment group was significantly higher during pregnancy, after delivery, and post-discharge. Hence, for couples who received infertility treatment at high-risk pregnancy outpatient clinics, we suggest interventions for factors that impede parenthood development, understand the various backgrounds of the participants, and provide individualised long-term support so that childbirth does not become the goal of the couple’s life. In particular, this study suggests the need to support postpartum mental health by considering couples as a unit.
## Supplementary Information
Additional file 1: Figure S1. Changes in the SECP scores for fathers and mothers. Blue lines and letters represent fathers; red lines and letters represent mothers. Each represents ***$p \leq 0.001$ **$p \leq 0.01$ *$p \leq 0.05.$ The ‘d’ in the figure indicates the effect size.
## Authors’ information
EK was a doctoral graduate student at Nagoya University when this study was conducted. MA is a Professor and Vice Director at Nagoya University Graduate School of Medicine. She is also a member of the Science Council of Japan.
## References
1. Vollset SE, Goren E, Yuan CW, Cao J, Smith AE, Hsiao T. **Fertility, mortality, migration, and population scenarios for 195 countries and territories from 2017 to 2100: a forecasting analysis for the Global Burden of Disease Study**. *Lancet* (2020.0) **396** 1285-1306. DOI: 10.1016/S0140-6736(20)30677-2
2. 2.Nihon Keizai Shimbun20-year birth rate 1.34, 5th consecutive year of decline, lowest level in 13 years2021. *20-year birth rate 1.34, 5th consecutive year of decline, lowest level in 13 years* (2021.0)
3. 3.Japan Ministry of Health, Labour and WelfareNinsanpu no shinryou no genjou to kadai (Current status and issues of medical care in for pregnant and postpartum woman)2019. *Ninsanpu no shinryou no genjou to kadai (Current status and issues of medical care in for pregnant and postpartum woman)* (2019.0)
4. 4.Japan Ministry of Health, Labour and Welfare. Handbook of Support for Balancing Infertility Treatment and Work. https://www.mhlw.go.jp/bunya/koyoukintou/pamphlet/dl/30l.pdf. Accessed 2 Feb 2022.
5. Tsuge A, Irahara M, Kato K. *Seishoku iryou no hatten to Kokusai chowa, Igakukai Sinbun (Development of reproductive health care and international harmonisation)* (2021.0)
6. Preis H, Mahaffey B, Heiselman C, Lobel M. **Vulnerability and resilience to pandemic-related stress among U.S. women pregnant at the start of the COVID-19 pandemic**. *Soc Sci Med* (2020.0) **266** 113348. DOI: 10.1016/j.socscimed.2020.113348
7. Barbosa-Leiker C, Smith CL, Crespi EJ, Brooks O, Burduli E, Ranjo S. **Stressors, coping, and resources needed during the COVID-19 pandemic in a sample of perinatal women**. *BMC Pregnancy Childbirth* (2021.0) **21** 171. DOI: 10.1186/s12884-021-03665-0
8. Sase Y. **A survey on website information regarding novel coronavirus infection at regional medical care support hospitals**. *J JSHA* (2020.0) **57** 105-112
9. Vance AJ, Malin KJ, Miller J, Shuman CJ, Moore TA, Benjamin A. **Parents’ pandemic NICU experience in the United States: a qualitative study**. *BMC Pediatr* (2021.0) **21** 558. DOI: 10.1186/s12887-021-03028-w
10. Taubman–Ben-Ari O, Ben-Yaakov O, Chasson M. **Parenting stress among new parents before and during the COVID-19 pandemic**. *Child Abuse Negl* (2021.0) **117** 105080. DOI: 10.1016/j.chiabu.2021.105080
11. Belsky J. **Determinants of parenting: a process model**. *Child Dev* (1984.0) **55** 83-96. DOI: 10.1111/j.1467-8624.1984.tb00275.x
12. Ogasawara M. **Parenthood of men, who have become a parent: a literature review**. *Bull Kansai Univ Nurs Health Sci* (2010.0) **2** 11-21
13. Ohashi Y, Asano M. **Literature in Japan on “Parenthood” and similar words: aiming at clarification of the “Parenthood” concept**. *Jpn J Res Fam Nurs* (2009.0) **14** 57-65
14. Deave T, Johnson D, Ingram J. **Transition to parenthood: the needs of parents in pregnancy and early parenthood**. *BMC Pregnancy Childbirth* (2008.0) **8** 30. DOI: 10.1186/1471-2393-8-30
15. Hinkle SN, Buck Louis GM, Rawal S, Zhu Y, Albert PS, Zhang C. **A longitudinal study of depression and gestational diabetes in pregnancy and the postpartum period**. *Diabetologia* (2016.0) **59** 2594-2602. DOI: 10.1007/s00125-016-4086-1
16. Helle N, Barkmann C, Bartz-Seel J, Diehl T, Ehrhardt S, Hendel A. **Very low birth-weight as a risk factor for postpartum depression four to six weeks postbirth in mothers and fathers: Cross-sectional results from a controlled multicentre cohort study**. *J Affect Disord* (2015.0) **180** 154-161. DOI: 10.1016/j.jad.2015.04.001
17. Goecke TW, Voigt F, Faschingbauer F, Spangler G, Beckmann MW, Beetz A. **The association of prenatal attachment and perinatal factors with pre- and postpartum depression in first-time mothers**. *Arch Gynecol Obstet* (2012.0) **286** 309-316. DOI: 10.1007/s00404-012-2286-6
18. Dong S, Dong Q, Chen H, Yang S. **Mother's Parenting Stress and Marital Satisfaction During the Parenting Period: Examining the Role of Depression, Solitude, and Time Alone**. *Front Psychol.* (2022.0) **13** 847419. DOI: 10.3389/fpsyg.2022.847419
19. Tognasso G, Gorla L, Ambrosini C, Figurella F, De Carli P, Parolin L. **Parenting Stress, Maternal Self-Efficacy and Confidence in Caretaking in a Sample of Mothers with Newborns (0–1 Month)**. *Int J Environ Res Public Health* (2022.0) **19** 9651. DOI: 10.3390/ijerph19159651
20. Tedgård E, Tedgård U, Råstam M, Johansson BA. **Parenting stress and its correlates in an infant mental health unit: a cross-sectional study**. *Nord J Psychiatry* (2020.0) **74** 30-39. DOI: 10.1080/08039488.2019.1667428
21. Kucharska M. **Selected predictors of maternal-fetal attachment in pregnancies with congenital disorders, other complications, and in healthy pregnancies**. *Health Psychol Rep* (2021.0) **9** 193-206. DOI: 10.5114/hpr.2020.97295
22. Kakiguchi M, Terasaki N, Morifuji K, Yamamoto N, Nakao Y, Tanaka H. **Promoting positive maternal feelings toward low-birth-weight infants admitted to the neonatal intensive care unit**. *Health Sci Res* (2014.0) **26** 7-13
23. Narita S, Maehara S. **The development of maternal-fetal attachment during pregnancy**. *J Jpn Acad Nurs Sci* (1993.0) **13** 1-9. DOI: 10.5630/jans1981.13.2_1
24. Matsuda Y. **Creation a scale of father identity in the transition period and examining its reliability and validity**. *J Jpn Acad Nurs Sci* (2018.0) **38** 9-17. DOI: 10.5630/jans.38.9
25. Sims ST, Stefanick ML, Kronenberg F, Sachedina NA, Schiebinger L. **Gendered Innovations: A New Approach for Nursing Science**. *Bio Res Nurs* (2010.0) **12** 156-161. DOI: 10.1177/1099800410375108
26. Ohashi Y, Asano M. **Development of a scale to measure early childhood parenting: an examination of its reliability and validity**. *J Jpn Soc Nurs Res* (2010.0) **33** 45-53
27. Ohashi Y, Asano M. **Transition to early parenthood, and family functioning relationships in Japan: a longitudinal study**. *Nurs Health Sci* (2012.0) **14** 140-147. DOI: 10.1111/j.1442-2018.2011.00669.x
28. Mercer RT. *Becoming a Mother* (1995.0)
29. Okano T. **Validation and reliability of the Japanese version of the EPDS (Edinburgh Postnatal Depression Scale)**. *Arch Psychiatr Diagn Clin Eval* (1996.0) **7** 525-533
30. Cox JL, Holden JM, Sagovsky R. **Detection of postnatal depression. Development of the 10-item Edinburgh Postnatal Depression Scale**. *Br J Psychiatry* (1987.0) **150** 782-6. DOI: 10.1192/bjp.150.6.782
31. Araki A, Kanematsu Y, Yokosawa S, Arayashiki A, Aizumi I, Hujishima K. **A study for developing parenting stress index-short form scale**. *J Child Health* (2005.0) **64** 408-416
32. Abidin R. *Parenting stress index: Manual* (1995.0)
33. Narama M, Kanematsu Y, Araki A, Maru M, Nakamura N, Takeda J. **Validity and reliability of the Japanese version of the Parenting Stress Index**. *J Child Health* (1999.0) **58** 610-616
34. Cohen J. *Statistical power analysis for the behavioral sciences* (1988.0)
35. Oikawa Y. **An analytic study of factors associated with the development of parenthood: the study of the environmental factors influencing development of parenthood for babies and infants**. *Bull Saitama Prefectural Univ* (2005.0) **7** 1-7
36. Abbasi E, Tahmasbi H, Hasani S, Takami-Gholam-Reza N. **Comparison of material and Paternal fatal attachment in couples referred to health care center in Sari**. *J Breeze Health* (2012.0) **1** 13-18
37. Iwata H, Mori E. **Promoting men’s adjustment to fatherhood: review of nursing care**. *J CANS* (2004.0) **10** 49-55
38. Nakajima K, Tokiwa Y. **A review of the literature on the marital relationship and supports to a pregnant wife by her husband**. *Acad Knowledge Arch Gunma Institutes* (2008.0) **29** 111-119
39. Baldwin S, Malone M, Sandall J, Bick D. **Mental health and wellbeing during the transition to fatherhood: a systematic review of first time fathers’ experiences**. *JBI Database System Rev Implement Rep* (2018.0) **16** 2118-2191. DOI: 10.11124/JBISRIR-2017-003773
40. Sasaki K, Ueda A, Suzuki Y, Maeda T, Kakayama R. **Does shaping parental consciousness relate to concern for the unborn child and marital relationship during pregnancy?**. *J Yonago Med Ass* (2004.0) **55** 142-150
41. Maehara K, Mori E, Tsuchiya M, Sakajo A, Hiroko Iwata H, Ozawa H. **Predictors of parenting stress in older primiparous mothers 2 months after childbirth**. *J GS Nrg ChibaU* (2015.0) **37** 27-35
42. 42.Ohashi Y. Shussan kara 12 kagetugo made no oyasei no henka to oyasei to kazokukinou no Kanren (Changes in parenthood from birth to 12 months after birth and the relationship between parenthood and family functioning): PhD thesis, Nagoya University, Graduate School of Medicine; 2012. http://hdl.handle.net/2237/16442.
43. Solmeyer AR, Feinberg ME. **Mother and father adjustment during early parenthood: The roles of infant temperament and coparenting relationship quality**. *Infant Behav Dev* (2011.0) **34** 504-514. DOI: 10.1016/j.infbeh.2011.07.006
44. Araki Y, Sakamoto S, Kunikiyo K, Tokiwa Y, Nakajima K. **Understanding the infertility experience of women who became pregnant and delivered a child after infertility treatment**. *J Jpn Soc Fertil Nurs* (2011.0) **8** 23-31
45. Yanagihara Y. **Desire to have children as an issue of gestational surrogacy**. *J Death Life Stud* (2012.0) **17** 204-29
46. Peterson BD, Newton CR, Rosen KH. **Examining congruence between partners’ perceived infertility-related stress and its relationship to marital adjustment and depression in infertile couples**. *Fam Process* (2003.0) **42** 59-70. DOI: 10.1111/j.1545-5300.2003.00059.x
47. Repokari L, Punamäki R, Unkila-Kallio L, Vilska S, Poikkeus P, Sinkkonen J. **Infertility treatment and marital relationships: A 1-Year prospective study among successfully treated ART couples and their controls**. *Hum Reprod* (2007.0) **22** 1481-1491. DOI: 10.1093/humrep/dem013
48. Ying LY, Wu LH, Loke AY. **The Experience of Chinese Couples Undergoing In Vitro Fertilization Treatment: Perception of the Treatment Process and Partner Support**. *PLoS One.* (2015.0) **10** e0139691. DOI: 10.1371/journal.pone.0139691
49. Sasaki N. **Literature research concerning psychological responses of people experienced fertility treatment in Japan**. *J Yamaguchi Prefectural Univ* (2014.0) **7** 49-56
50. da Silva SM, Boivin J, Gameiro S. **Self-regulation and wellbeing when facing a blocked parenthood goal: a systematic review and meta-analysis**. *PLoS One.* (2016.0) **11** e0157649. DOI: 10.1371/journal.pone.0157649
51. Malina A, Pooley JA. **Psychological consequences of IVF fertilization – Review of research**. *Ann Agric Environ Med* (2017.0) **24** 554-558. DOI: 10.5604/12321966.1232085
52. Peterson B, Newton C, Rosen K, Skaggs G. **Gender differences in how men and women who are referred for IVF cope with infertility stress**. *Hum Reprod* (2006.0) **21** 2443-2449. DOI: 10.5604/12321966.1232085
53. Asano M. **Kosodateki no Kazoku no Well-Being wo mezashite. (Toward Family Well-Being in Child Care)**. *GRL Stud* (2020.0) **3** 48-51
54. Tokumitsu K, Sugawara N, Maruo K, Suzuki T, Yasui-Furukori N, Shimoda K. **Prevalence of perinatal depression among Japanese men: a meta-analysis**. *Ann Gen Psychiatry* (2020.0) **19** 1-65. DOI: 10.1186/s12991-020-00316-0
55. Takehara K. *Chicioya no sanzen sango no utsu no jittai to sono shien, Igakukai Sinbun (The reality of prenatal and postpartum depression in fathers and its support)* (2021.0)
56. Domoney J, Trevillion K, Challacombe FL. **Developing an intervention for paternal perinatal depression: an international Delphi study**. *J Affective Disord Rep.* (2020.0) **2** 100033. DOI: 10.1016/j.jadr.2020.100033
57. Takahashi Y. **Parenting stress and associated factors in mothers of infants**. *J Faculty Nurs Iwate Prefectural Univ* (2007.0) **9** 31-41
58. Hambidge S, Cowell A, Arden-Close E, Mayers A. **“What kind of man gets depressed after having a baby?” Fathers’ experiences of mental health during the perinatal period**. *BMC Pregnancy Childbirth* (2021.0) **21** 463. DOI: 10.1186/s12884-021-03947-7
59. Byatt N, Hicks-Courant K, Davidson A, Levesque R, Mick E, Allison J. **Depression and anxiety among high-risk obstetric inpatients**. *Gen Hosp Psychiatry* (2014.0) **36** 644-649. DOI: 10.1016/j.genhosppsych.2014.07.011
60. Takehara K, Suto M, Kato T. **Parental psychological distress in the postnatal period in Japan: a population-based analysis of a national cross-sectional survey**. *Sci Rep* (2020.0) **10** 13770. DOI: 10.1038/s41598-020-70727-2
61. 61.JOICFP. 2023. https://www.joicfp.or.jp/jpn/. Accessed 7 Feb 2023.
62. 62.Japan Ministry of Health, Labour and WelfareFunin chiryou ni kansuru torikumi (Initiatives related to fertility treatment)2022. *Funin chiryou ni kansuru torikumi (Initiatives related to fertility treatment)* (2022.0)
63. 63.Nihon Keizai ShimbunMothers are foreigners, 25,000 per year2019. *Mothers are foreigners, 25,000 per year* (2019.0)
|
---
title: 'Movement Behavior and Health Outcomes among Sedentary Adults: A Cross-Sectional
Study'
authors:
- Federico Arippa
- Athena Nguyen
- Massimiliano Pau
- Carisa Harris-Adamson
journal: International Journal of Environmental Research and Public Health
year: 2023
pmcid: PMC10037417
doi: 10.3390/ijerph20054668
license: CC BY 4.0
---
# Movement Behavior and Health Outcomes among Sedentary Adults: A Cross-Sectional Study
## Abstract
Background: Sedentary behavior, which is highly prevalent among office workers, is associated with multiple health disorders, including those of the musculoskeletal and cardiometabolic systems. Although prior studies looked at postures or physical activity during work or leisure time, few analyzed both posture and movement throughout the entire day. Objective: This cross-sectional pilot study examined the movement behavior of sedentary office workers during both work and leisure time to explore its association with musculoskeletal discomfort (MSD) and cardiometabolic health indicators. Methods: Twenty-six participants completed a survey and wore a thigh-based inertial measuring unit (IMU) to quantify the time spent in different postures, the number of transitions between postures, and the step count during work and leisure time. A heart rate monitor and ambulatory blood pressure cuff were worn to quantify cardiometabolic measures. The associations between movement behavior, MSD, and cardiometabolic health indicators were evaluated. Results: The number of transitions differed significantly between those with and without MSD. Correlations were found between MSD, time spent sitting, and posture transitions. Posture transitions had negative correlations with body mass index and heart rate. Conclusions: Although no single behavior was highly correlated with health outcomes, these correlations suggest that a combination of increasing standing time, walking time, and the number of transitions between postures during both work and leisure time was associated with positive musculoskeletal and cardiometabolic health indicators among sedentary office workers and should be considered in future research.
## 1. Introduction
Sedentary time has been steadily increasing while physical activity has been decreasing worldwide [1]. The World Health Organization (WHO) reported that $31\%$ of people who are 15 years or older take part in less than 2.5 h per week of moderate activity and roughly 3.2 million deaths per year are associated with such sedentary lifestyles [2]. The impact of sedentary behavior is vast and includes higher healthcare costs, loss in productivity, and increased disability-adjusted life-years (DALYs) [3]. Further, sedentary behavior continues to grow as leisure activities include computers or screen watching [4,5,6,7] and the number of sedentary jobs increases at the expense of physically active ones [8]. With an average of 8 to 12 h per day [9,10], office employees are among those workers that are most sedentary, accounting for almost $81.8\%$ of their work hours.
Prior literature indicates that sedentary time is highly associated with musculoskeletal discomfort (MSD) [11,12,13,14,15] and adverse cardiometabolic outcomes [16,17,18,19,20,21]. Prolonged sitting time has a negative impact on resting heart rate, adiposity, vascular function [17,20], plasma glucose, HDL-cholesterol, and triacylglycerol [19]. In addition to cardiometabolic outcomes, prolonged sitting was associated with diminished endothelial function in the leg vasculature and more frequent urinary tract symptoms [22,23,24,25]. Prolonged sitting also leads to increased and sustained intradiscal pressure since compressive forces are higher when sitting compared with standing [13,15,26], which is detrimental to the hydration and nutrition of the intervertebral disc [14], thus having a negative impact on the low back both in terms of musculoskeletal pain and biomechanical load when sustained for long durations [11,27]. However, periodic breaks in sedentary time that include brief episodes of standing or walking, as well as higher total physical activity, were found to be helpful to nourish the nucleus pulposus and intervertebral disc [26] and to reduce adiposity [19].
Studies documented the inverse relationship between increased standing time and all-cause mortality rates among people who primarily sit [28,29]; even 10 breaks per day were associated with positive cardiometabolic outcomes, including decreased waist circumference, systolic blood pressure, triglycerides, glucose, and insulin and with increased HDL-cholesterol [16]. Thus, experts recommended that office-based workers should include at least 2 h per day of standing and walking during working hours [30]. Nonetheless, it is important to distinguish between periodic and prolonged standing, the latter of which has harmful effects as well [31,32,33,34,35,36,37,38]. Since numerous studies have associated prolonged standing with poor cardiometabolic and musculoskeletal outcomes [34], the recommendation to increase standing time is unique to those who spend most of their time in a seated position, and even then, frequent short intervals of standing time are recommended throughout the day versus one long bout.
Increasing movement through posture transitions, walking, or even micromovements while seated, which is defined by small changes in one’s position without large changes in postures, was suggested as a potential way to mitigate the negative health effects on the musculoskeletal and cardiometabolic systems associated with prolonged sedentary behavior [39,40,41,42,43,44]. WHO guidelines recommend that adults should undertake 150 to 300 min of moderate-intensity physical activity, 75–150 min of vigorous-intensity physical activity, or some equivalent combination of moderate-intensity and vigorous-intensity aerobic physical activity per week [45]. Walking was identified as the most effective method to improve MSD and cardiometabolic outcomes among sedentary workers [46,47,48] since walking changes the demands on the musculoskeletal system and increases energy expenditure compared with both sitting and standing [49]. Previous research shows that more frequent walking during a sedentary time was negatively associated with body mass index (BMI), waist circumference, 2 h plasma glucose, triglycerides [50,51,52,53], and MSD [54]. However, the frequency and duration required for maximum benefit remain unclear. In addition, transitioning between postures may be beneficial to both musculoskeletal and cardiometabolic health. Sit–stand workstations were implemented as a way to support transitions between sitting and standing [11], and task-based walking, such as “walk and talk” meetings, was implemented as a way to transition between sitting or standing and walking. Micromovements can be quantitatively assessed, for instance, by tracking the trunk’s center of pressure using force platforms or pressure-sensitive mats [42,43,55,56,57]. Although micromovements were suggested as a coping strategy to reduce discomfort once it has developed [42,43,57,58], people who perform a higher number of micromovements proactively have a lower probability of developing low back pain; thus, interventions designed to increase micromovements before pain develops may prevent or prolong MSD from developing [59,60,61,62,63].
## A Comprehensive Movement Behavior Model
Considering prior studies on movement, MSD, and cardiometabolic outcomes [11,12,13,14,15], we present a framework (Figure 1) to assess daily movement behavior among sedentary workers to determine which measurements are most important to optimize their health outcomes. The first strategy was to determine the total time spent sitting, standing, and walking each day. This was based on the evidence that prolonged sitting time is associated with MSD symptoms [13,15,26], and that bouts of standing and walking of suitable duration can mitigate discomfort and improve cardiometabolic measures [26,30,48,52,53,64,65,66,67]. These variables are interrelated; as one sits less, they will stand or walk more.
The second strategy was to describe the movement that occurs while sitting, standing, or walking. While sitting and standing, micromovements can be quantified by sway patterns (sway path and area), mean pressure, and in-posture movements that capture quick shifts of the body [42]. Walking, which is a general term that can differ substantially by person or environment, can also be further defined by quantifying step count and cadence, which captures the rate of steps.
A third strategy to quantify movement behavior was to measure the number of transitions between sitting, standing, and walking, which provides feedback on the pattern of whole-body posture changes. Transitions are defined by moving from sitting to standing, standing to walking, or sitting to walking (Figure 1). As noted above, numerous studies identified benefits associated with transitions and more work is being done to understand how to optimize their frequency and timing [13,51,59,68,69]. Although transitions are often prompted by discomfort, this reactive approach may be “too little too late” to mitigate MSD. Ideally, transitions should occur proactively as part of a comprehensive strategy to prevent MSD and contribute to one’s overall daily energy expenditure.
Although prior research investigated the relationships between certain measures of movement behavior, musculoskeletal discomfort, and adverse cardiometabolic outcomes [11,12,13,14,15,26,30,48,52,53,64,65,66,67], it is still unclear which, singularly or in combination, are the most impactful. Additionally, little is known about whether different movement strategies are more important during work or leisure time.
In summary, although multiple strategies were identified as ways to mitigate the negative health effects associated with prolonged sitting among sedentary office workers, a comprehensive approach to quantifying these movements is needed to understand which combinations and patterns of movement are most important for optimizing musculoskeletal and cardiometabolic health. By measuring movement behavior more comprehensively and consistently during both work and leisure time, we can help sedentary office workers to optimize their movement behavior throughout their day. Furthermore, consistent measurement of movement behaviors may increase our understanding of the effectiveness and efficacy of interventions designed to increase movement in sedentary office workers. Therefore, in addition to developing the comprehensive Movement Behavior Model used to quantify sedentary postures and movement, the aim of this pilot study was twofold: [1] investigate the existence of differences in movement behavior metrics between those with MSD and those without MSD; and [2] examine the relationships between movement behavior metrics (during work and leisure time), MSD, and cardiometabolic outcomes.
## 2.1. Participants
In total, 31 office workers at the University of California, Berkeley applied to take part in this cross-sectional pilot study. Among them, twenty-six met the inclusion criteria and were enrolled in this study. Recruitment methods included posting flyers throughout campus and sending emails through department listservs. The inclusion criteria specified that participants must have a sit–stand desk, work at the desk for at least thirty hours per week, and be capable of standing for at least twenty minutes. Exclusion criteria included any MSD or illness that would prevent the worker from standing while working at their desk. This study was approved by the University of California, Berkeley Committee for Protection of Human Subjects (protocol code 2019-10-12607).
## 2.2. Procedure
All data were collected at the UC *Berkeley campus* in participants’ offices at the beginning of their work shifts. Upon arrival, participants signed an informed consent form and anthropometric measurements were collected. Participants were informed that they could withdraw at any time during the study without consequences. Subjects were asked to complete a baseline survey through a Qualtrics link sent via SMS text message. The survey gathered data on demographic characteristics, physical activity, and MSD using the 0 to 10 Numeric Pain Rating Scale (NRS).
Additionally, participants were instructed to wear the activPAL monitor (Glasgow, UK) on their thighs for at least 48 h to record activity and posture data [70], and wore an Actiheart heart rate monitor (Boerne, TX, USA) and Spacelabs blood pressure cuff (Snoqualmie, WA, USA) for 24 h.
## 2.3. Measures
Height and weight were measured by means of an ultrasonic digital height meter (Soehnle 5003, Soehnle, Germany) and a digital scale (RE310, Wunder, Italy), respectively. BMI was calculated by dividing the individual’s body mass (expressed in kilograms) by their stature (meters squared); values of BMI lower than 18.5 identify underweight, between 18.5 and 25 is considered in the healthy range, while 25 to 30 and higher than 30 fall in the overweight and obese ranges, respectively. Regarding the hip–waist measurement, the WHO protocol was followed [71]; a lower hip–waist ratio indicates a healthier distribution of body fat and lower risks of health problems (0.95 or less for men and 0.80 or less for women [72]).
Activity and posture were quantified using step count and the duration of time spent sitting, standing, and walking. Postural transitions were defined as changes between sitting, standing, and walking. Activity and posture data were sampled at a rate of 20 Hz.
The MSD scores were grouped into four regions: [1] head, neck, and shoulders; [2] upper and lower back; [3] hips, knees, feet, and ankles; and [4] elbows, hands, and wrists. A composite MSD score was generated by summing the NRS scores across the four regions for a maximum score of 40.
Cardiometabolic data, including heart rate (HR), mean arterial pressure (MAP), and pulse pressure (PP) was based on 1 min and 30 min (6 a.m.–10 p.m.) or 60 min (10 p.m.–6 a.m.) sampling rates, respectively. Resting HR was calculated by taking the average of the five lowest values throughout the day, while the average HR was calculated using the data from the 24 h period; a lower HR (optimal range 60–80 bpm [73]) is an indicator of cardiovascular health. The MAP, which is the average blood pressure per cardiac cycle, was calculated by doubling the diastole measurement, adding the systole measurement, and dividing the value by three [74]; similarly to HR, a low MAP identifies healthier subjects. The PP represents the difference between systolic blood pressure and diastolic blood pressure [75]; with regular physical activity, the elasticity of blood vessels improves, resulting in a lower PP, both during rest and exercise.
## 2.4. Analysis
According to the self-reported questionnaire on MSD, people with a composite pain score of 2 or greater were categorized in the group titled “MSD”, whereas those who had a composite MSD score less than 2 in all four regions were sorted into the “NO-MSD” group. Self-reported work hours from the baseline survey were also used to stratify the movement behavior by work and leisure time for each participant.
Two-tailed independent-sample t-tests were used to assess differences in demographics, cardiometabolic outcomes, and movement activity (daily, work, and leisure) between the MSD and NO-MSD groups.
Spearman correlation coefficients were used to understand whether activity levels at work matched those during leisure time, and to explore the relationship between motor behavior, MSD, and cardiometabolic data. For absolute values of r, 0.0 to 0.39 was considered weak, 0.4 to 0.69 was moderate, 0.7 to 0.99 was strong, and 1 was perfect [76].
All analyses were completed using SPSS v26. An alpha of 0.05 was used as the threshold for significance and all independent variables were between subjects.
## 3. Results
Demographic analysis (Table 1) showed that most of the participants were female, and the average age of the subjects was 33.2 ± 9.4 years old. Only two participants had a current medical condition and one subject had diabetes. One participant had a previous injury.
Scores for MSD reported during the baseline survey are reported in Table 2. Daily, work, and leisure activity data of participants are presented in Table 3. There were minimal differences between the MSD and NO-MSD activity levels; however, participants in the MSD group transitioned significantly more frequently compared with those in the NO-MSD group during working hours.
The activity data (Table 3) and composite MSD score (Table 2) were used for the analyses in Table 4. Moderate positive correlations were found between the composite MSD score and transitions, while a negative relationship was evidenced with time spent standing (Table 4).
Correlations between movement behavior metrics during work and leisure are presented in Table 5. Time spent in a seated position during leisure was negatively associated with the number of steps and the standing and walking time at work. Sleeping time was positively correlated with the number of steps, transitions, and walking time at work.
The average cardiometabolic measurements showed moderate correlations with activity measures (Table 6). In particular, transitions were negatively correlated with BMI and heart rate; walking time during leisure was positively associated with average pulse pressure.
## 4. Discussion
In this cross-sectional pilot study, we looked at the relationships between movement behavior measures, pain, and cardiometabolic outcomes to inform future studies on which measurements may be important indicators of health during work and leisure time.
Overall, the results showed that participants were primarily sedentary, sitting for an average of more than 9 h per day. People with MSD transitioned significantly more throughout their working time relative to people without MSD. This is consistent with previous publications stating that people suffering from back pain tended to move more frequently than those without pain, as frequent posture changes provided relief and rest for passive and active structures that accumulate pressure during static postures, especially of the spine [11,26]. However, despite the participants in the MSD group transitioning more frequently, they also spent one hour less standing compared with those in the NO-MSD group, though this difference was not statistically significant. It is unclear whether the reduced amount of standing contributed to increased MSD, or whether perceived MSD contributed to less standing. The average amount of standing time per hour at work was 18 min in the MSD group and 24 min in the NO-MSD group. Therefore, on average, both groups stood more than what was reported in other studies, and what was recommended in previous papers [77,78]. Since the development of pain, especially low back pain, can occur during relatively short standing bouts [79,80,81,82,83,84,85,86], it is possible that participants with MSD spent less time in a standing posture because the development of pain was quicker than those without MSD [87]. This trend was also confirmed when investigating the correlation between MSD score and movement behavior indicators since MSD was positively correlated with postural transitions at work and negatively correlated with daily standing time. It is possible that the recommendation for people with MSD should focus on more posture transitions to avoid static standing postures that can result in increased pain and discomfort levels.
Regarding the relationship between movement behavior and cardiometabolic outcomes, we found negative associations for leisure and daily (leisure and work) transitions with BMI, resting heart rate, and average heart rate. This may indicate that in addition to reducing MSD, changing posture more frequently throughout the day (particularly during leisure time) may be important for cardiometabolic health outcomes. Our findings are consistent with prior studies that examined the relationships between breaks that interrupt sedentary time and improve cardiometabolic outcomes [50,51,88] emphasizing the importance of "sitting less and moving more". Although the time spent standing and walking did not have any relationship with cardiometabolic outcomes in our population, it is possible that our analysis was underpowered and included a generally healthy population that stood and walked more than other study populations [30,47,48,52,53,64]. Overall, in addition to time spent sitting, standing, and walking [89], our findings indicate that quantifying daily leisure and work transition metrics might be beneficial to optimizing MSD and cardiometabolic health. While further investigation is needed to ascertain the relative importance of these variables, these data and the developed Movement Behavior Model may serve as a general model for quantifying and further clarifying activity levels, MSD, and cardiometabolic risk in office workers.
In this study, it was also evaluated whether participants’ movement behavior during work was consistent with their movement behavior during leisure time since movement behavior during both work and leisure time was proposed as an important strategy to reduce the risk of cardiometabolic disorders and MSD [90,91]. The results showed that participants who were more sedentary at work were also more sedentary during leisure and had less sleep duration than those who were active. Although this is consistent with prior studies that found moderate physical activity to contribute toward improving both sleep quality and duration, the relationship here was not bound to moderate physical activity and included all non-sitting time [92]. If, in fact, more non-sitting time and transitions throughout the day improve sleep, this could be an additional benefit to encourage more movement during work and leisure time, especially in a sedentary population. Further studies should be performed to understand how interventions that reduce sedentary behavior at work also impact sedentary behavior during leisure time, and whether the increase in movement impacts sleep.
It should be noted that our Movement Behavior Model did not quantify the intensity of physical activity, which is something that was identified as important for cardiometabolic health and sleep quality [90,91]. Future studies should quantify heart rate while walking and performing other physical activity for exercise during the day (Figure 2). Further, it is possible that tracking work and leisure time metrics separately may facilitate more consistent movement throughout the day, thereby reducing prolonged sitting bouts. While further investigation is needed to ascertain the relative importance of the metrics presented here, these data and the Movement Behavior Model may serve as a general model for quantifying and further clarifying the metrics that are most important for optimizing musculoskeletal and cardiometabolic health among sedentary office workers.
Some considerations can be drawn on the basis of the obtained results. First, since time is limited in nature, spending more time on one activity (such as sitting, standing, or walking) will result in less time available for other activities. More time standing means less time sitting or walking. Increased time spent working reduces leisure time and the likelihood of physical activity during leisure time. This means that these measures are somewhat dependent on one another. Larger datasets could allow for a more robust statistical approach that accounts for data dependency, as well as their pattern of occurrence throughout the day. Perhaps, rather than the total amount of time spent on each activity, the timing, frequency, and duration of each occurrence are more important when studying their impact on health outcomes. For example, the association of posture transitions with MSD and cardiometabolic outcomes may differ if the analysis focuses on the pattern of posture transitions throughout the day instead of just the overall number of occurrences. The same could be true for patterns of sitting, standing, and walking. This should be explored in future studies. Some limitations of the study should be acknowledged. Data collection began at the end of 2019; therefore, given the COVID-19 pandemic, our sample size was limited. Larger populations are needed to fully understand the relationships investigated. Further, our sample of convenience from the university was a more active cohort than many sedentary workers, having a higher average standing time than cohorts in other studies [72]. More variance in the movement behaviors of participants may be needed in order to strengthen the observed correlations and to generalize the results presented. Moreover, as previously mentioned, the intensity of physical activity was not included in our analysis; its inclusion would likely improve the presented model and should be considered in future work (Figure 2). Furthermore, while our Movement Behavior Model features in-chair fidgeting, sway patterns, and mean pressure, these metrics were largely investigated in previous work [43]. Exploring all of these metrics together in a larger cohort may be needed to better understand the relative importance of each metric. Lastly, the cross-sectional design of this study did not allow for investigating causality between movement behavior measures, MSD, and cardiometabolic outcomes. Future research may include longitudinal studies to track changes in these parameters over time to better characterize the causal relationships that exist between them. This could contribute to the development of predictive models or activity scores that help sedentary workers take a comprehensive approach to increasing movement throughout the day.
## 5. Conclusions
This cross-sectional pilot study offers preliminary insights into the relationship between work and leisure time movement behavior measures, MSD, and cardiometabolic outcomes. The findings may be helpful for further study of these relationships and their application to interventions that prevent adverse health outcomes associated with sedentary behavior. Further investigation of a larger cohort may help to quantify the relative contribution of each measure and should consider the additional benefit from engagement in physical activities of different intensities. Ideally, the most important variables could be integrated into a user-friendly index, the impact of which could be investigated as an intervention.
## References
1. Owen N., Healy G.N., Dempsey P.C., Salmon J., Timperio A., Clark B.K., Goode A.D., Koorts H., Ridgers N.D., Hadgraft N.T.. **Sedentary Behavior and Public Health: Integrating the Evidence and Identifying Potential Solutions**. *Annu. Rev. Public Health* (2020) **41** 265-287. DOI: 10.1146/annurev-publhealth-040119-094201
2. Hall G., Laddu D.R., Phillips S.A., Lavie C.J., Arena R.. **A Tale of Two Pandemics: How Will COVID-19 and Global Trends in Physical Inactivity and Sedentary Behavior Affect One Another?**. *Prog. Cardiovasc. Dis.* (2021) **64** 108-110. DOI: 10.1016/j.pcad.2020.04.005
3. Ding D., Lawson K.D., Kolbe-Alexander T.L., Finkelstein E.A., Katzmarzyk P.T., van Mechelen W., Pratt M.. **The Economic Burden of Physical Inactivity: A Global Analysis of Major Non-Communicable Diseases**. *Lancet* (2016) **388** 1311-1324. DOI: 10.1016/S0140-6736(16)30383-X
4. Jans M.P., Proper K.I., Hildebrandt V.H.. **Sedentary Behavior in Dutch Workers: Differences between Occupations and Business Sectors**. *Am. J. Prev. Med.* (2007) **33** 450-454. DOI: 10.1016/j.amepre.2007.07.033
5. Saidj M., Menai M., Charreire H., Weber C., Enaux C., Aadahl M., Kesse-Guyot E., Hercberg S., Simon C., Oppert J.-M.. **Descriptive Study of Sedentary Behaviours in 35,444 French Working Adults: Cross-Sectional Findings from the ACTI-Cités Study**. *BMC Public Health* (2015) **15**. DOI: 10.1186/s12889-015-1711-8
6. Hadgraft N.T., Lynch B.M., Clark B.K., Healy G.N., Owen N., Dunstan D.W.. **Excessive Sitting at Work and at Home: Correlates of Occupational Sitting and TV Viewing Time in Working Adults**. *BMC Public Health* (2015) **15**. DOI: 10.1186/s12889-015-2243-y
7. Bontrup C., Taylor W.R., Fliesser M., Visscher R., Green T., Wippert P.-M., Zemp R.. **Low Back Pain and Its Relationship with Sitting Behaviour among Sedentary Office Workers**. *Appl. Ergon.* (2019) **81**. DOI: 10.1016/j.apergo.2019.102894
8. Genin P.M., Dessenne P., Finaud J., Pereira B., Dutheil F., Thivel D., Duclos M.. **Effect of Work-Related Sedentary Time on Overall Health Profile in Active vs. Inactive Office Workers**. *Front. Public Health* (2018) **6** 279. DOI: 10.3389/fpubh.2018.00279
9. Mann S., Hamid Hamad A., Kumbhare D.. **The Problem of Sedentary Behaviour in the Office Workspace: A Structured Exercise Program for Primary Prevention**. *J. Nov. Physiother.* (2018) **8** 392. DOI: 10.4172/2165-7025.1000392
10. Parry S., Straker L.. **The Contribution of Office Work to Sedentary Behaviour Associated Risk**. *BMC Public Health* (2013) **13**. DOI: 10.1186/1471-2458-13-296
11. Agarwal S., Steinmaus C., Harris-Adamson C.. **Sit-Stand Workstations and Impact on Low Back Discomfort: A Systematic Review and Meta-Analysis**. *Ergonomics* (2018) **61** 538-552. DOI: 10.1080/00140139.2017.1402960
12. Foley B., Engelen L., Gale J., Bauman A., Mackey M.. **Sedentary Behavior and Musculoskeletal Discomfort Are Reduced When Office Workers Trial an Activity-Based Work Environment**. *J. Occup. Environ. Med.* (2016) **58** 924-931. DOI: 10.1097/JOM.0000000000000828
13. Karakolis T., Barrett J., Callaghan J.P.. **A Comparison of Trunk Biomechanics, Musculoskeletal Discomfort and Productivity during Simulated Sit-Stand Office Work**. *Ergonomics* (2016) **59** 1275-1287. DOI: 10.1080/00140139.2016.1146343
14. Marras W.S., Lavender S.A., Leurgans S.E., Fathallah F.A., Ferguson S.A., Allread W.G., Rajulu S.L.. **Biomechanical Risk Factors for Occupationally Related Low Back Disorders**. *Ergonomics* (1995) **38** 377-410. DOI: 10.1080/00140139508925111
15. Nachemson A.. **Intravital Dynamic Pressure Measurements in Lumbar Discs: A Study of Common Movements, Maneuvers and Exercises**. *Scand. J. Rehabil. Med.* (1970) **1** 1-40
16. Carson V., Wong S.L., Winkler E., Healy G.N., Colley R.C., Tremblay M.S.. **Patterns of Sedentary Time and Cardiometabolic Risk among Canadian Adults**. *Prev. Med.* (2014) **65** 23-27. DOI: 10.1016/j.ypmed.2014.04.005
17. Carter S., Hartman Y., Holder S., Thijssen D.H., Hopkins N.D.. **Sedentary Behavior and Cardiovascular Disease Risk: Mediating Mechanisms**. *Exerc. Sport Sci. Rev.* (2017) **45** 80-86. DOI: 10.1249/JES.0000000000000106
18. Healy G.N., Dunstan D.W., Salmon J., Shaw J.E., Zimmet P.Z., Owen N.. **Television Time and Continuous Metabolic Risk in Physically Active Adults**. *Med. Sci. Sports Exerc.* (2008) **40** 639-645. DOI: 10.1249/MSS.0b013e3181607421
19. Henson J., Yates T., Biddle S.J.H., Edwardson C.L., Khunti K., Wilmot E.G., Gray L.J., Gorely T., Nimmo M.A., Davies M.J.. **Associations of Objectively Measured Sedentary Behaviour and Physical Activity with Markers of Cardiometabolic Health**. *Diabetologia* (2013) **56** 1012-1020. DOI: 10.1007/s00125-013-2845-9
20. Huynh Q.L., Blizzard C.L., Sharman J.E., Magnussen C.G., Dwyer T., Venn A.J.. **The Cross-Sectional Association of Sitting Time with Carotid Artery Stiffness in Young Adults**. *BMJ Open* (2014) **4**. DOI: 10.1136/bmjopen-2013-004384
21. Sugiyama T., Hadgraft N., Clark B.K., Dunstan D.W., Owen N.. **Sitting at Work & Waist Circumference—A Cross-Sectional Study of Australian Workers**. *Prev. Med.* (2020) **141**. DOI: 10.1016/j.ypmed.2020.106243
22. Morishima T., Restaino R.M., Walsh L.K., Kanaley J.A., Fadel P.J., Padilla J.. **Prolonged Sitting-Induced Leg Endothelial Dysfunction Is Prevented by Fidgeting**. *Am. J. Physiol.-Heart Circ. Physiol.* (2016) **311** H177-H182. DOI: 10.1152/ajpheart.00297.2016
23. Padilla J., Fadel P.J.. **Prolonged Sitting Leg Vasculopathy: Contributing Factors and Clinical Implications**. *Am. J. Physiol.-Heart Circ. Physiol.* (2017) **313** H722-H728. DOI: 10.1152/ajpheart.00326.2017
24. Park H.J., Park C.H., Chang Y., Ryu S.. **Sitting Time, Physical Activity and the Risk of Lower Urinary Tract Symptoms: A Cohort Study**. *BJU Int.* (2018) **122** 293-299. DOI: 10.1111/bju.14147
25. Restaino R.M., Walsh L.K., Morishima T., Vranish J.R., Martinez-Lemus L.A., Fadel P.J., Padilla J.. **Endothelial Dysfunction Following Prolonged Sitting Is Mediated by a Reduction in Shear Stress**. *Am. J. Physiol.-Heart Circ. Physiol.* (2016) **310** H648-H653. DOI: 10.1152/ajpheart.00943.2015
26. Callaghan J.P., McGill S.M.. **Low Back Joint Loading and Kinematics during Standing and Unsupported Sitting**. *Ergonomics* (2001) **44** 280-294. DOI: 10.1080/00140130118276
27. Huang M., Hajizadeh K., Gibson I., Lee T.. **Analysis of Compressive Load on Intervertebral Joint in Standing and Sitting Postures**. *Technol. Health Care* (2016) **24** 215-223. DOI: 10.3233/THC-151100
28. van der Ploeg H.P., Chey T., Ding D., Chau J.Y., Stamatakis E., Bauman A.E.. **Standing Time and All-Cause Mortality in a Large Cohort of Australian Adults**. *Prev. Med.* (2014) **69** 187-191. DOI: 10.1016/j.ypmed.2014.10.004
29. Katzmarzyk P.T.. **Standing and Mortality in a Prospective Cohort of Canadian Adults**. *Med. Sci. Sports Exerc.* (2014) **46** 940-946. DOI: 10.1249/MSS.0000000000000198
30. Buckley J.P., Hedge A., Yates T., Copeland R.J., Loosemore M., Hamer M., Bradley G., Dunstan D.W.. **The Sedentary Office: An Expert Statement on the Growing Case for Change towards Better Health and Productivity**. *Br. J. Sports Med.* (2015) **49** 1357-1362. DOI: 10.1136/bjsports-2015-094618
31. Tüchsen F., Krause N., Hannerz H., Burr H., Kristensen T.S.. **Standing at Work and Varicose Veins**. *Scand. J. Work. Environ. Health* (2000) **26** 414-420. DOI: 10.5271/sjweh.562
32. Krause N., Lynch J.W., Kaplan G.A., Cohen R.D., Salonen R., Salonen J.T.. **Standing at Work and Progression of Carotid Atherosclerosis**. *Scand. J. Work. Environ. Health* (2000) **26** 227-236. DOI: 10.5271/sjweh.536
33. Smith P., Ma H., Glazier R.H., Gilbert-Ouimet M., Mustard C.. **The Relationship Between Occupational Standing and Sitting and Incident Heart Disease Over a 12-Year Period in Ontario, Canada**. *Am. J. Epidemiol.* (2018) **187** 27-33. DOI: 10.1093/aje/kwx298
34. Waters T.R., Dick R.B.. **Evidence of Health Risks Associated with Prolonged Standing at Work and Intervention Effectiveness**. *Rehabil. Nurs. J.* (2015) **40** 148-165. DOI: 10.1002/rnj.166
35. Chester M.R., Rys M.J., Konz S.A.. **Leg Swelling, Comfort and Fatigue When Sitting, Standing, and Sit/Standing**. *Int. J. Ind. Ergon.* (2002) **29** 289-296. DOI: 10.1016/S0169-8141(01)00069-5
36. Balasubramanian V., Adalarasu K., Regulapati R.. **Comparing Dynamic and Stationary Standing Postures in an Assembly Task**. *Int. J. Ind. Ergon.* (2009) **39** 649-654. DOI: 10.1016/j.ergon.2008.10.017
37. Andersen J.H., Haahr J.P., Frost P.. **Risk Factors for More Severe Regional Musculoskeletal Symptoms: A Two-Year Prospective Study of a General Working Population**. *Arthritis Rheum.* (2007) **56** 1355-1364. DOI: 10.1002/art.22513
38. Baker R., Coenen P., Howie E., Lee J., Williamson A., Straker L.. **A Detailed Description of the Short-Term Musculoskeletal and Cognitive Effects of Prolonged Standing for Office Computer Work**. *Ergonomics* (2018) **61** 877-890. DOI: 10.1080/00140139.2017.1420825
39. Genaidy A.M., Al-Shedi A.A., Karwowski W.. **Postural Stress Analysis in Industry**. *Appl. Ergon.* (1994) **25** 77-87. DOI: 10.1016/0003-6870(94)90068-X
40. Liao M.H., Drury C.G.. **Posture, Discomfort and Performance in a Vdt Task**. *Ergonomics* (2000) **43** 345-359. DOI: 10.1080/001401300184459
41. McLean L., Tingley M., Scott R.N., Rickards J.. **Computer Terminal Work and the Benefit of Microbreaks**. *Appl. Ergon.* (2001) **32** 225-237. DOI: 10.1016/S0003-6870(00)00071-5
42. Arippa F., Leban B., Pau M.. **Postural Strategies of Bus Drivers during a Regular Work Shift in Urban Area: A Pilot Study**. *Proc. Hum. Factors Ergon. Soc. Annu. Meet.* (2019) **63** 967-971. DOI: 10.1177/1071181319631429
43. Arippa F., Nguyen A., Pau M., Harris-Adamson C.. **Postural Strategies among Office Workers during a Prolonged Sitting Bout**. *Appl. Ergon.* (2022) **102**. DOI: 10.1016/j.apergo.2022.103723
44. Sammonds G.M., Mansfield N.J., Fray M.. **Improving Long Term Driving Comfort by Taking Breaks—How Break Activity Affects Effectiveness**. *Appl. Ergon.* (2017) **65** 81-89. DOI: 10.1016/j.apergo.2017.05.008
45. Bull F.C., Al-Ansari S.S., Biddle S., Borodulin K., Buman M.P., Cardon G., Carty C., Chaput J.-P., Chastin S., Chou R.. **World Health Organization 2020 Guidelines on Physical Activity and Sedentary Behaviour**. *Br. J. Sports Med.* (2020) **54** 1451. DOI: 10.1136/bjsports-2020-102955
46. Sitthipornvorakul E., Janwantanakul P., Lohsoonthorn V.. **The Effect of Daily Walking Steps on Preventing Neck and Low Back Pain in Sedentary Workers: A 1-Year Prospective Cohort Study**. *Eur. Spine J.* (2015) **24** 417-424. DOI: 10.1007/s00586-014-3577-3
47. Sitthipornvorakul E., Klinsophon T., Sihawong R., Janwantanakul P.. **The Effects of Walking Intervention in Patients with Chronic Low Back Pain: A Meta-Analysis of Randomized Controlled Trials**. *Musculoskelet. Sci. Pract.* (2018) **34** 38-46. DOI: 10.1016/j.msksp.2017.12.003
48. Gallagher K.M., Payne M., Daniels B., Caldwell A.R., Ganio M.S.. **Walking Breaks Can Reduce Prolonged Standing Induced Low Back Pain**. *Hum. Mov. Sci.* (2019) **66** 31-37. DOI: 10.1016/j.humov.2019.03.012
49. Levine J.A., Schleusner S.J., Jensen M.D.. **Energy Expenditure of Nonexercise Activity**. *Am. J. Clin. Nutr.* (2000) **72** 1451-1454. DOI: 10.1093/ajcn/72.6.1451
50. Healy G.N., Dunstan D.W., Salmon J., Cerin E., Shaw J.E., Zimmet P.Z., Owen N.. **Breaks in Sedentary Time: Beneficial Associations with Metabolic Risk**. *Diabetes Care* (2008) **31** 661-666. DOI: 10.2337/dc07-2046
51. Owen N., Healy G.N., Matthews C.E., Dunstan D.W.. **Too Much Sitting: The Population-Health Science of Sedentary Behavior**. *Exerc. Sport Sci. Rev.* (2010) **38** 105-113. DOI: 10.1097/JES.0b013e3181e373a2
52. Bailey D.P., Locke C.D.. **Breaking up Prolonged Sitting with Light-Intensity Walking Improves Postprandial Glycemia, but Breaking up Sitting with Standing Does Not**. *J. Sci. Med. Sport* (2015) **18** 294-298. DOI: 10.1016/j.jsams.2014.03.008
53. Lin Y.-P., Lin C.-C., Chen M.-M., Lee K.-C.. **Short-Term Efficacy of a “Sit Less, Walk More” Workplace Intervention on Improving Cardiometabolic Health and Work Productivity in Office Workers**. *J. Occup. Environ. Med.* (2017) **59** 327-334. DOI: 10.1097/JOM.0000000000000955
54. Kar G., Hedge A.. **Effects of a Sit-Stand-Walk Intervention on Musculoskeletal Discomfort, Productivity, and Perceived Physical and Mental Fatigue, for Computer-Based Work**. *Int. J. Ind. Ergon.* (2020) **78**. DOI: 10.1016/j.ergon.2020.102983
55. Fenety A., Putnam C., Walker J.M.. **In-Chair Movement: Validity, Reliability and Implications for Measuring Sitting Discomfort**. *Appl. Ergon.* (2000) **31** 383-393. DOI: 10.1016/S0003-6870(00)00003-X
56. Leban B., Fancello G., Fadda P., Pau M.. **Changes in Trunk Sway of Quay Crane Operators during Work Shift: A Possible Marker for Fatigue?**. *Appl. Ergon.* (2017) **65** 105-111. DOI: 10.1016/j.apergo.2017.06.007
57. Sammonds G.M., Fray M., Mansfield N.J.. **Effect of Long Term Driving on Driver Discomfort and Its Relationship with Seat Fidgets and Movements (SFMs)**. *Appl. Ergon.* (2017) **58** 119-127. DOI: 10.1016/j.apergo.2016.05.009
58. Hermann S., Bubb H.. *Development of an Objective Measure to Quantify Automotive Discomfort over Time* (2007) 2824-2830
59. Davis K.G., Kotowski S.E.. **Postural Variability: An Effective Way to Reduce Musculoskeletal Discomfort in Office Work**. *Hum. Factors* (2014) **56** 1249-1261. DOI: 10.1177/0018720814528003
60. Vergara M., Page Á.. **Relationship between Comfort and Back Posture and Mobility in Sitting-Posture**. *Appl. Ergon.* (2002) **33** 1-8. DOI: 10.1016/S0003-6870(01)00056-4
61. Aarås A., Horgen G., Ro O.. **Work with the Visual Display Unit: Health Consequences**. *Int. J. Hum.–Comput. Interact.* (2000) **12** 107-134. DOI: 10.1207/S15327590IJHC1201_5
62. Pynt J., Higgs J., Mackey M.. **Seeking the Optimal Posture of the Seated Lumbar Spine**. *Physiother. Theory Pract.* (2001) **17** 5-21. DOI: 10.1080/09593980151143228
63. Srinivasan D., Mathiassen S.E.. **Motor Variability in Occupational Health and Performance**. *Clin. Biomech.* (2012) **27** 979-993. DOI: 10.1016/j.clinbiomech.2012.08.007
64. Cocate P.G., de Oliveira A., Hermsdorff H.H.M., Alfenas R.d.C.G., Amorim P.R.S., Longo G.Z., Peluzio M.d.C.G., Faria F.R., Natali A.J.. **Benefits and Relationship of Steps Walked per Day to Cardiometabolic Risk Factor in Brazilian Middle-Aged Men**. *J. Sci. Med. Sport* (2014) **17** 283-287. DOI: 10.1016/j.jsams.2013.04.017
65. Duvivier B.M.F.M., Schaper N.C., Koster A., van Kan L., Peters H.P.F., Adam J.J., Giesbrecht T., Kornips E., Hulsbosch M., Willems P.. **Benefits of Substituting Sitting with Standing and Walking in Free-Living Conditions for Cardiometabolic Risk Markers, Cognition and Mood in Overweight Adults**. *Front. Physiol.* (2017) **8** 353. DOI: 10.3389/fphys.2017.00353
66. Pulsford R.M., Blackwell J., Hillsdon M., Kos K.. **Intermittent Walking, but Not Standing, Improves Postprandial Insulin and Glucose Relative to Sustained Sitting: A Randomised Cross-over Study in Inactive Middle-Aged Men**. *J. Sci. Med. Sport* (2017) **20** 278-283. DOI: 10.1016/j.jsams.2016.08.012
67. Chastin S.F.M., Egerton T., Leask C., Stamatakis E.. **Meta-Analysis of the Relationship between Breaks in Sedentary Behavior and Cardiometabolic Health**. *Obesity* (2015) **23** 1800-1810. DOI: 10.1002/oby.21180
68. Callaghan J.P., De Carvalho D., Gallagher K., Karakolis T., Nelson-Wong E.. **Is Standing the Solution to Sedentary Office Work?**. *Ergon. Des.* (2015) **23** 20-24. DOI: 10.1177/1064804615585412
69. Davis K.G., Kotowski S.E., Sharma B., Herrmann D., Krishnan A.P.. **Combating the Effects of Sedentary Work: Postural Variability Reduces Musculoskeletal Discomfort**. *Proc. Hum. Factors Ergon. Soc. Annu. Meet.* (2009) **53** 884-886. DOI: 10.1177/154193120905301406
70. Ferguson T., Rowlands A.V., Olds T., Maher C.. **The Validity of Consumer-Level, Activity Monitors in Healthy Adults Worn in Free-Living Conditions: A Cross-Sectional Study**. *Int. J. Behav. Nutr. Phys. Act.* (2015) **12** 42. DOI: 10.1186/s12966-015-0201-9
71. 71.
World Health Organization
Waist Circumference and Waist-Hip Ratio: Report of a WHO Expert Consultation, Geneva, Switzerland, 8–11 December 2008World Health OrganizationGeneva, Switzerland2011. *Waist Circumference and Waist-Hip Ratio: Report of a WHO Expert Consultation, Geneva, Switzerland, 8–11 December 2008* (2011)
72. Gelber R.P., Gaziano J.M., Orav E.J., Manson J.E., Buring J.E., Kurth T.. **Measures of Obesity and Cardiovascular Risk Among Men and Women**. *J. Am. Coll. Cardiol.* (2008) **52** 605-615. DOI: 10.1016/j.jacc.2008.03.066
73. Laukkanen J.A., Zaccardi F., Khan H., Kurl S., Jae S.Y., Rauramaa R.. **Long-Term Change in Cardiorespiratory Fitness and All-Cause Mortality: A Population-Based Follow-up Study**. *Mayo Clin. Proc.* (2016) **91** 1183-1188. DOI: 10.1016/j.mayocp.2016.05.014
74. DeMers D., Wachs D.. **Physiology, Mean Arterial Pressure**. *StatPearls* (2020)
75. Zheng L., Sun Z., Li J., Zhang R., Zhang X., Liu S., Li J., Xu C., Hu D., Sun Y.. **Pulse Pressure and Mean Arterial Pressure in Relation to Ischemic Stroke Among Patients With Uncontrolled Hypertension in Rural Areas of China**. *Stroke* (2008) **39** 1932-1937. DOI: 10.1161/STROKEAHA.107.510677
76. Akoglu H.. **User’s Guide to Correlation Coefficients**. *Turk. J. Emerg. Med.* (2018) **18** 91-93. DOI: 10.1016/j.tjem.2018.08.001
77. Chambers A.J., Robertson M.M., Baker N.A.. **The Effect of Sit-Stand Desks on Office Worker Behavioral and Health Outcomes: A Scoping Review**. *Appl. Ergon.* (2019) **78** 37-53. DOI: 10.1016/j.apergo.2019.01.015
78. Robertson M.M., Ciriello V.M., Garabet A.M.. **Office Ergonomics Training and a Sit-Stand Workstation: Effects on Musculoskeletal and Visual Symptoms and Performance of Office Workers**. *Appl. Ergon.* (2013) **44** 73-85. DOI: 10.1016/j.apergo.2012.05.001
79. Fewster K.M., Gallagher K.M., Howarth S.H., Callaghan J.P.. **Low Back Pain Development Differentially Influences Centre of Pressure Regularity Following Prolonged Standing**. *Gait Posture* (2020) **78** e1-e6. DOI: 10.1016/j.gaitpost.2017.06.005
80. Sorensen C.J., Norton B.J., Callaghan J.P., Hwang C.T., Van Dillen L.R.. **Is Lumbar Lordosis Related to Low Back Pain Development during Prolonged Standing?**. *Man. Ther.* (2015) **20** 553-557. DOI: 10.1016/j.math.2015.01.001
81. Gregory D.E., Callaghan J.P.. **Prolonged Standing as a Precursor for the Development of Low Back Discomfort: An Investigation of Possible Mechanisms**. *Gait Posture* (2008) **28** 86-92. DOI: 10.1016/j.gaitpost.2007.10.005
82. Nelson-Wong E., Callaghan J.P.. **Is Muscle Co-Activation a Predisposing Factor for Low Back Pain Development during Standing? A Multifactorial Approach for Early Identification of at-Risk Individuals**. *J. Electromyogr. Kinesiol.* (2010) **20** 256-263. DOI: 10.1016/j.jelekin.2009.04.009
83. Nelson-Wong E., Gregory D.E., Winter D.A., Callaghan J.P.. **Gluteus Medius Muscle Activation Patterns as a Predictor of Low Back Pain during Standing**. *Clin. Biomech.* (2008) **23** 545-553. DOI: 10.1016/j.clinbiomech.2008.01.002
84. Marshall P.W.M., Patel H., Callaghan J.P.. **Gluteus Medius Strength, Endurance, and Co-Activation in the Development of Low Back Pain during Prolonged Standing**. *Hum. Mov. Sci.* (2011) **30** 63-73. DOI: 10.1016/j.humov.2010.08.017
85. Gallagher K.M., Callaghan J.P.. **Early Static Standing Is Associated with Prolonged Standing Induced Low Back Pain**. *Hum. Mov. Sci.* (2015) **44** 111-121. DOI: 10.1016/j.humov.2015.08.019
86. Viggiani D., Callaghan J.P.. **A Hip Abduction Exercise Prior to Prolonged Standing Increased Movement While Reducing Cocontraction and Low Back Pain Perception in Those Initially Reporting Low Back Pain**. *J. Electromyogr. Kinesiol.* (2016) **31** 63-71. DOI: 10.1016/j.jelekin.2016.09.005
87. Nelson-Wong E., Gallagher K., Johnson E., Antonioli C., Ferguson A., Harris S., Johnson H., Miller J.B.. **Increasing Standing Tolerance in Office Workers with Standing-Induced Back Pain**. *Ergonomics* (2020) **63** 804-817. DOI: 10.1080/00140139.2020.1761034
88. Healy G.N., Winkler E.A.H., Owen N., Anuradha S., Dunstan D.W.. **Replacing Sitting Time with Standing or Stepping: Associations with Cardio-Metabolic Risk Biomarkers**. *Eur. Heart J.* (2015) **36** 2643-2649. DOI: 10.1093/eurheartj/ehv308
89. Hawari N.S.A., Al-Shayji I., Wilson J., Gill J.M.R.. **Frequency of Breaks in Sedentary Time and Postprandial Metabolic Responses**. *Med. Sci. Sports Exerc.* (2016) **48** 2495-2502. DOI: 10.1249/MSS.0000000000001034
90. Westgaard R.H.. **Work-Related Musculoskeletal Complaints: Some Ergonomics Challenges upon the Start of a New Century**. *Appl. Ergon.* (2000) **31** 569-580. DOI: 10.1016/S0003-6870(00)00036-3
91. Fullick S., Grindey C., Edwards B., Morris C., Reilly T., Richardson D., Waterhouse J., Atkinson G.. **Relationships between Leisure-Time Energy Expenditure and Individual Coping Strategies for Shift-Work**. *Ergonomics* (2009) **52** 448-455. DOI: 10.1080/00140130802707725
92. Wang F., Boros S.. **The Effect of Physical Activity on Sleep Quality: A Systematic Review**. *Eur. J. Physiother.* (2021) **23** 11-18. DOI: 10.1080/21679169.2019.1623314
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.